Refactor, add libritts

egs/libritts/TTS/vocos/discriminators.py

@@ -3,7 +3,7 @@ from typing import List, Optional, Tuple
 import torch
 from torch import nn
 from torch.nn import Conv2d
-from torch.nn.utils import weight_norm
+from torch.nn.utils.parametrizations import weight_norm
 from torchaudio.transforms import Spectrogram
 
 

egs/libritts/TTS/vocos/export-onnx.py

@@ -0,0 +1,371 @@
+#!/usr/bin/env python3
+#
+# Copyright 2023 Xiaomi Corporation (Author: Fangjun Kuang, Wei Kang)
+# Copyright 2023 Danqing Fu (danqing.fu@gmail.com)
+
+"""
+This script exports a transducer model from PyTorch to ONNX.
+
+We use the pre-trained model from
+https://huggingface.co/Zengwei/icefall-asr-librispeech-zipformer-2023-05-15
+as an example to show how to use this file.
+
+1. Download the pre-trained model
+
+cd egs/librispeech/ASR
+
+repo_url=https://huggingface.co/Zengwei/icefall-asr-librispeech-zipformer-2023-05-15
+GIT_LFS_SKIP_SMUDGE=1 git clone $repo_url
+repo=$(basename $repo_url)
+
+pushd $repo
+git lfs pull --include "exp/pretrained.pt"
+
+cd exp
+ln -s pretrained.pt epoch-99.pt
+popd
+
+2. Export the model to ONNX
+
+./zipformer/export-onnx.py \
+  --tokens $repo/data/lang_bpe_500/tokens.txt \
+  --use-averaged-model 0 \
+  --epoch 99 \
+  --avg 1 \
+  --exp-dir $repo/exp \
+  --num-encoder-layers "2,2,3,4,3,2" \
+  --downsampling-factor "1,2,4,8,4,2" \
+  --feedforward-dim "512,768,1024,1536,1024,768" \
+  --num-heads "4,4,4,8,4,4" \
+  --encoder-dim "192,256,384,512,384,256" \
+  --query-head-dim 32 \
+  --value-head-dim 12 \
+  --pos-head-dim 4 \
+  --pos-dim 48 \
+  --encoder-unmasked-dim "192,192,256,256,256,192" \
+  --cnn-module-kernel "31,31,15,15,15,31" \
+  --decoder-dim 512 \
+  --joiner-dim 512 \
+  --causal False \
+  --chunk-size "16,32,64,-1" \
+  --left-context-frames "64,128,256,-1" \
+  --fp16 True
+It will generate the following 3 files inside $repo/exp:
+
+  - encoder-epoch-99-avg-1.onnx
+  - decoder-epoch-99-avg-1.onnx
+  - joiner-epoch-99-avg-1.onnx
+
+See ./onnx_pretrained.py and ./onnx_check.py for how to
+use the exported ONNX models.
+"""
+
+import argparse
+import logging
+from pathlib import Path
+from typing import Dict, Tuple
+
+import onnx
+import torch
+import torch.nn as nn
+from onnxconverter_common import float16
+from onnxruntime.quantization import QuantType, quantize_dynamic
+from train import add_model_arguments, get_model, get_params
+
+from icefall.checkpoint import (
+    average_checkpoints,
+    average_checkpoints_with_averaged_model,
+    find_checkpoints,
+    load_checkpoint,
+)
+from icefall.utils import make_pad_mask, num_tokens, str2bool
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--sampling-rate",
+        type=int,
+        default=24000,
+        help="The sampleing rate of libritts dataset",
+    )
+
+    parser.add_argument(
+        "--frame-shift",
+        type=int,
+        default=256,
+        help="Frame shift.",
+    )
+
+    parser.add_argument(
+        "--frame-length",
+        type=int,
+        default=1024,
+        help="Frame shift.",
+    )
+
+    parser.add_argument(
+        "--epoch",
+        type=int,
+        default=28,
+        help="""It specifies the checkpoint to use for averaging.
+        Note: Epoch counts from 0.
+        You can specify --avg to use more checkpoints for model averaging.""",
+    )
+
+    parser.add_argument(
+        "--iter",
+        type=int,
+        default=0,
+        help="""If positive, --epoch is ignored and it
+        will use the checkpoint exp_dir/checkpoint-iter.pt.
+        You can specify --avg to use more checkpoints for model averaging.
+        """,
+    )
+
+    parser.add_argument(
+        "--avg",
+        type=int,
+        default=15,
+        help="Number of checkpoints to average. Automatically select "
+        "consecutive checkpoints before the checkpoint specified by "
+        "'--epoch' and '--iter'",
+    )
+
+    parser.add_argument(
+        "--use-averaged-model",
+        type=str2bool,
+        default=True,
+        help="Whether to load averaged model. Currently it only supports "
+        "using --epoch. If True, it would decode with the averaged model "
+        "over the epoch range from `epoch-avg` (excluded) to `epoch`."
+        "Actually only the models with epoch number of `epoch-avg` and "
+        "`epoch` are loaded for averaging. ",
+    )
+
+    parser.add_argument(
+        "--exp-dir",
+        type=str,
+        default="zipformer/exp",
+        help="""It specifies the directory where all training related
+        files, e.g., checkpoints, log, etc, are saved
+        """,
+    )
+
+    parser.add_argument(
+        "--fp16",
+        type=str2bool,
+        default=False,
+        help="Whether to export models in fp16",
+    )
+
+    add_model_arguments(parser)
+
+    return parser
+
+
+def add_meta_data(filename: str, meta_data: Dict[str, str]):
+    """Add meta data to an ONNX model. It is changed in-place.
+
+    Args:
+      filename:
+        Filename of the ONNX model to be changed.
+      meta_data:
+        Key-value pairs.
+    """
+    model = onnx.load(filename)
+    for key, value in meta_data.items():
+        meta = model.metadata_props.add()
+        meta.key = key
+        meta.value = value
+
+    onnx.save(model, filename)
+
+
+def export_model_onnx(
+    model: nn.Module,
+    model_filename: str,
+    opset_version: int = 13,
+) -> None:
+    """Export the joiner model to ONNX format.
+    The exported joiner model has two inputs:
+
+        - encoder_out: a tensor of shape (N, joiner_dim)
+        - decoder_out: a tensor of shape (N, joiner_dim)
+
+    and produces one output:
+
+        - logit: a tensor of shape (N, vocab_size)
+    """
+    input_tensor = torch.rand((2, 80, 100), dtype=torch.float32)
+
+    torch.onnx.export(
+        model,
+        (input_tensor,),
+        model_filename,
+        verbose=False,
+        opset_version=opset_version,
+        input_names=[
+            "features",
+        ],
+        output_names=["audio"],
+        dynamic_axes={
+            "features": {0: "N", 2: "F"},
+            "audio": {0: "N", 1: "T"},
+        },
+    )
+
+    meta_data = {
+        "model_type": "Vocos",
+        "version": "1",
+        "model_author": "k2-fsa",
+        "comment": "ConvNext Vocos",
+    }
+    logging.info(f"meta_data: {meta_data}")
+
+    add_meta_data(filename=model_filename, meta_data=meta_data)
+
+
+@torch.no_grad()
+def main():
+    args = get_parser().parse_args()
+    args.exp_dir = Path(args.exp_dir)
+
+    params = get_params()
+    params.update(vars(args))
+
+    device = torch.device("cpu")
+    params.device = device
+    logging.info(params)
+
+    logging.info("About to create model")
+    model = get_model(params)
+    model.to(device)
+
+    if not params.use_averaged_model:
+        if params.iter > 0:
+            filenames = find_checkpoints(params.exp_dir, iteration=-params.iter)[
+                : params.avg
+            ]
+            if len(filenames) == 0:
+                raise ValueError(
+                    f"No checkpoints found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            elif len(filenames) < params.avg:
+                raise ValueError(
+                    f"Not enough checkpoints ({len(filenames)}) found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            logging.info(f"averaging {filenames}")
+            model.to(device)
+            model.load_state_dict(average_checkpoints(filenames, device=device))
+        elif params.avg == 1:
+            load_checkpoint(f"{params.exp_dir}/epoch-{params.epoch}.pt", model)
+        else:
+            start = params.epoch - params.avg + 1
+            filenames = []
+            for i in range(start, params.epoch + 1):
+                if i >= 1:
+                    filenames.append(f"{params.exp_dir}/epoch-{i}.pt")
+            logging.info(f"averaging {filenames}")
+            model.to(device)
+            model.load_state_dict(average_checkpoints(filenames, device=device))
+    else:
+        if params.iter > 0:
+            filenames = find_checkpoints(params.exp_dir, iteration=-params.iter)[
+                : params.avg + 1
+            ]
+            if len(filenames) == 0:
+                raise ValueError(
+                    f"No checkpoints found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            elif len(filenames) < params.avg + 1:
+                raise ValueError(
+                    f"Not enough checkpoints ({len(filenames)}) found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            filename_start = filenames[-1]
+            filename_end = filenames[0]
+            logging.info(
+                "Calculating the averaged model over iteration checkpoints"
+                f" from {filename_start} (excluded) to {filename_end}"
+            )
+            model.to(device)
+            model.load_state_dict(
+                average_checkpoints_with_averaged_model(
+                    filename_start=filename_start,
+                    filename_end=filename_end,
+                    device=device,
+                )
+            )
+        else:
+            assert params.avg > 0, params.avg
+            start = params.epoch - params.avg
+            assert start >= 1, start
+            filename_start = f"{params.exp_dir}/epoch-{start}.pt"
+            filename_end = f"{params.exp_dir}/epoch-{params.epoch}.pt"
+            logging.info(
+                f"Calculating the averaged model over epoch range from "
+                f"{start} (excluded) to {params.epoch}"
+            )
+            model.to(device)
+            model.load_state_dict(
+                average_checkpoints_with_averaged_model(
+                    filename_start=filename_start,
+                    filename_end=filename_end,
+                    device=device,
+                )
+            )
+
+    model.eval()
+    vocos = model.generator
+
+    if params.iter > 0:
+        suffix = f"iter-{params.iter}"
+    else:
+        suffix = f"epoch-{params.epoch}"
+
+    suffix += f"-avg-{params.avg}"
+
+    opset_version = 13
+
+    logging.info("Exporting model")
+    model_filename = params.exp_dir / f"vocos-{suffix}.onnx"
+    export_model_onnx(
+        vocos,
+        model_filename,
+        opset_version=opset_version,
+    )
+    logging.info(f"Exported vocos generator to {model_filename}")
+
+    if params.fp16:
+        logging.info("Generate fp16 models")
+
+        model = onnx.load(model_filename)
+        model_fp16 = float16.convert_float_to_float16(model, keep_io_types=True)
+        model_filename_fp16 = params.exp_dir / f"vocos-{suffix}.fp16.onnx"
+        onnx.save(model_fp16, model_filename_fp16)
+
+    # Generate int8 quantization models
+    # See https://onnxruntime.ai/docs/performance/model-optimizations/quantization.html#data-type-selection
+
+    logging.info("Generate int8 quantization models")
+
+    model_filename_int8 = params.exp_dir / f"vocos-{suffix}.int8.onnx"
+    quantize_dynamic(
+        model_input=model_filename,
+        model_output=model_filename_int8,
+        op_types_to_quantize=["MatMul"],
+        weight_type=QuantType.QInt8,
+    )
+
+
+if __name__ == "__main__":
+    formatter = "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
+    logging.basicConfig(format=formatter, level=logging.INFO)
+    main()

egs/libritts/TTS/vocos/export.py

@@ -0,0 +1,407 @@
+#!/usr/bin/env python3
+#
+# Copyright 2024 Xiaomi Corporation (Author: Wei Kang)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This script converts several saved checkpoints
+# to a single one using model averaging.
+"""
+
+Usage:
+
+Note: This is a example for libritts dataset, if you are using different
+dataset, you should change the argument values according to your dataset.
+
+(1) Export to torchscript model using torch.jit.script()
+
+
+./vocos/export.py \
+  --exp-dir ./vocos/exp \
+  --epoch 30 \
+  --avg 9 \
+  --jit 1
+
+It will generate a file `jit_script.pt` in the given `exp_dir`. You can later
+load it by `torch.jit.load("jit_script.pt")`.
+
+Check ./jit_pretrained.py for its usage.
+
+Check https://github.com/k2-fsa/sherpa
+for how to use the exported models outside of icefall.
+
+- For streaming model:
+
+./zipformer/export.py \
+  --exp-dir ./zipformer/exp \
+  --causal 1 \
+  --chunk-size 16 \
+  --left-context-frames 128 \
+  --tokens data/lang_bpe_500/tokens.txt \
+  --epoch 30 \
+  --avg 9 \
+  --jit 1
+
+It will generate a file `jit_script_chunk_16_left_128.pt` in the given `exp_dir`.
+You can later load it by `torch.jit.load("jit_script_chunk_16_left_128.pt")`.
+
+Check ./jit_pretrained_streaming.py for its usage.
+
+Check https://github.com/k2-fsa/sherpa
+for how to use the exported models outside of icefall.
+
+(2) Export `model.state_dict()`
+
+- For non-streaming model:
+
+./zipformer/export.py \
+  --exp-dir ./zipformer/exp \
+  --tokens data/lang_bpe_500/tokens.txt \
+  --epoch 30 \
+  --avg 9
+
+- For streaming model:
+
+./zipformer/export.py \
+  --exp-dir ./zipformer/exp \
+  --causal 1 \
+  --tokens data/lang_bpe_500/tokens.txt \
+  --epoch 30 \
+  --avg 9
+
+It will generate a file `pretrained.pt` in the given `exp_dir`. You can later
+load it by `icefall.checkpoint.load_checkpoint()`.
+
+- For non-streaming model:
+
+To use the generated file with `zipformer/decode.py`,
+you can do:
+
+    cd /path/to/exp_dir
+    ln -s pretrained.pt epoch-9999.pt
+
+    cd /path/to/egs/librispeech/ASR
+    ./zipformer/decode.py \
+        --exp-dir ./zipformer/exp \
+        --epoch 9999 \
+        --avg 1 \
+        --max-duration 600 \
+        --decoding-method greedy_search \
+        --bpe-model data/lang_bpe_500/bpe.model
+
+- For streaming model:
+
+To use the generated file with `zipformer/decode.py` and `zipformer/streaming_decode.py`, you can do:
+
+    cd /path/to/exp_dir
+    ln -s pretrained.pt epoch-9999.pt
+
+    cd /path/to/egs/librispeech/ASR
+
+    # simulated streaming decoding
+    ./zipformer/decode.py \
+        --exp-dir ./zipformer/exp \
+        --epoch 9999 \
+        --avg 1 \
+        --max-duration 600 \
+        --causal 1 \
+        --chunk-size 16 \
+        --left-context-frames 128 \
+        --decoding-method greedy_search \
+        --bpe-model data/lang_bpe_500/bpe.model
+
+    # chunk-wise streaming decoding
+    ./zipformer/streaming_decode.py \
+        --exp-dir ./zipformer/exp \
+        --epoch 9999 \
+        --avg 1 \
+        --max-duration 600 \
+        --causal 1 \
+        --chunk-size 16 \
+        --left-context-frames 128 \
+        --decoding-method greedy_search \
+        --bpe-model data/lang_bpe_500/bpe.model
+
+Check ./pretrained.py for its usage.
+
+Note: If you don't want to train a model from scratch, we have
+provided one for you. You can get it at
+
+- non-streaming model:
+https://huggingface.co/Zengwei/icefall-asr-librispeech-zipformer-2023-05-15
+
+- streaming model:
+https://huggingface.co/Zengwei/icefall-asr-librispeech-streaming-zipformer-2023-05-17
+
+with the following commands:
+
+    sudo apt-get install git-lfs
+    git lfs install
+    git clone https://huggingface.co/Zengwei/icefall-asr-librispeech-zipformer-2023-05-15
+    git clone https://huggingface.co/Zengwei/icefall-asr-librispeech-streaming-zipformer-2023-05-17
+    # You will find the pre-trained models in exp dir
+"""
+
+import argparse
+import logging
+from pathlib import Path
+from typing import List, Tuple
+
+import torch
+from torch import Tensor, nn
+from train import add_model_arguments, get_model, get_params
+
+from icefall.checkpoint import (
+    average_checkpoints,
+    average_checkpoints_with_averaged_model,
+    find_checkpoints,
+)
+
+from icefall.utils import str2bool
+from utils import load_checkpoint
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--sampling-rate",
+        type=int,
+        default=24000,
+        help="The sampleing rate of libritts dataset",
+    )
+
+    parser.add_argument(
+        "--frame-shift",
+        type=int,
+        default=256,
+        help="Frame shift.",
+    )
+
+    parser.add_argument(
+        "--frame-length",
+        type=int,
+        default=1024,
+        help="Frame shift.",
+    )
+
+    parser.add_argument(
+        "--epoch",
+        type=int,
+        default=30,
+        help="""It specifies the checkpoint to use for decoding.
+        Note: Epoch counts from 1.
+        You can specify --avg to use more checkpoints for model averaging.""",
+    )
+
+    parser.add_argument(
+        "--iter",
+        type=int,
+        default=0,
+        help="""If positive, --epoch is ignored and it
+        will use the checkpoint exp_dir/checkpoint-iter.pt.
+        You can specify --avg to use more checkpoints for model averaging.
+        """,
+    )
+
+    parser.add_argument(
+        "--avg",
+        type=int,
+        default=9,
+        help="Number of checkpoints to average. Automatically select "
+        "consecutive checkpoints before the checkpoint specified by "
+        "'--epoch' and '--iter'",
+    )
+
+    parser.add_argument(
+        "--use-averaged-model",
+        type=str2bool,
+        default=True,
+        help="Whether to load averaged model. Currently it only supports "
+        "using --epoch. If True, it would decode with the averaged model "
+        "over the epoch range from `epoch-avg` (excluded) to `epoch`."
+        "Actually only the models with epoch number of `epoch-avg` and "
+        "`epoch` are loaded for averaging. ",
+    )
+
+    parser.add_argument(
+        "--exp-dir",
+        type=str,
+        default="vocos/exp",
+        help="""It specifies the directory where all training related
+        files, e.g., checkpoints, log, etc, are saved
+        """,
+    )
+
+    parser.add_argument(
+        "--jit",
+        type=str2bool,
+        default=False,
+        help="""True to save a model after applying torch.jit.script.
+        It will generate a file named jit_script.pt.
+        Check ./jit_pretrained.py for how to use it.
+        """,
+    )
+
+    add_model_arguments(parser)
+
+    return parser
+
+
+class EncoderModel(nn.Module):
+    """A wrapper for encoder and encoder_embed"""
+
+    def __init__(self, encoder: nn.Module, encoder_embed: nn.Module) -> None:
+        super().__init__()
+        self.encoder = encoder
+        self.encoder_embed = encoder_embed
+
+    def forward(
+        self, features: Tensor, feature_lengths: Tensor
+    ) -> Tuple[Tensor, Tensor]:
+        """
+        Args:
+            features: (N, T, C)
+            feature_lengths: (N,)
+        """
+        x, x_lens = self.encoder_embed(features, feature_lengths)
+
+        src_key_padding_mask = make_pad_mask(x_lens)
+        x = x.permute(1, 0, 2)  # (N, T, C) -> (T, N, C)
+
+        encoder_out, encoder_out_lens = self.encoder(x, x_lens, src_key_padding_mask)
+        encoder_out = encoder_out.permute(1, 0, 2)  # (T, N, C) ->(N, T, C)
+
+        return encoder_out, encoder_out_lens
+
+
+@torch.no_grad()
+def main():
+    args = get_parser().parse_args()
+    args.exp_dir = Path(args.exp_dir)
+
+    params = get_params()
+    params.update(vars(args))
+
+    device = torch.device("cpu")
+    params.device = device
+    logging.info(f"device: {device}")
+
+    logging.info(params)
+
+    logging.info("About to create model")
+    model = get_model(params)
+
+    if not params.use_averaged_model:
+        if params.iter > 0:
+            filenames = find_checkpoints(params.exp_dir, iteration=-params.iter)[
+                : params.avg
+            ]
+            if len(filenames) == 0:
+                raise ValueError(
+                    f"No checkpoints found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            elif len(filenames) < params.avg:
+                raise ValueError(
+                    f"Not enough checkpoints ({len(filenames)}) found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            logging.info(f"averaging {filenames}")
+            model.load_state_dict(average_checkpoints(filenames, device=device))
+        elif params.avg == 1:
+            load_checkpoint(f"{params.exp_dir}/epoch-{params.epoch}.pt", model)
+        else:
+            start = params.epoch - params.avg + 1
+            filenames = []
+            for i in range(start, params.epoch + 1):
+                if i >= 1:
+                    filenames.append(f"{params.exp_dir}/epoch-{i}.pt")
+            logging.info(f"averaging {filenames}")
+            model.load_state_dict(average_checkpoints(filenames, device=device))
+    else:
+        if params.iter > 0:
+            filenames = find_checkpoints(params.exp_dir, iteration=-params.iter)[
+                : params.avg + 1
+            ]
+            if len(filenames) == 0:
+                raise ValueError(
+                    f"No checkpoints found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            elif len(filenames) < params.avg + 1:
+                raise ValueError(
+                    f"Not enough checkpoints ({len(filenames)}) found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            filename_start = filenames[-1]
+            filename_end = filenames[0]
+            logging.info(
+                "Calculating the averaged model over iteration checkpoints"
+                f" from {filename_start} (excluded) to {filename_end}"
+            )
+            model.load_state_dict(
+                average_checkpoints_with_averaged_model(
+                    filename_start=filename_start,
+                    filename_end=filename_end,
+                    device=device,
+                )
+            )
+        else:
+            assert params.avg > 0, params.avg
+            start = params.epoch - params.avg
+            assert start >= 1, start
+            filename_start = f"{params.exp_dir}/epoch-{start}.pt"
+            filename_end = f"{params.exp_dir}/epoch-{params.epoch}.pt"
+            logging.info(
+                f"Calculating the averaged model over epoch range from "
+                f"{start} (excluded) to {params.epoch}"
+            )
+            model.load_state_dict(
+                average_checkpoints_with_averaged_model(
+                    filename_start=filename_start,
+                    filename_end=filename_end,
+                    device=device,
+                )
+            )
+
+    model.eval()
+
+    model = model.generator
+
+    if params.jit is True:
+        model.encoder = EncoderModel(model.encoder, model.encoder_embed)
+        filename = "jit_script.pt"
+
+        logging.info("Using torch.jit.script")
+        model = torch.jit.script(model)
+        model.save(str(params.exp_dir / filename))
+        logging.info(f"Saved to {filename}")
+    else:
+        logging.info("Not using torchscript. Export model.state_dict()")
+        # Save it using a format so that it can be loaded
+        # by :func:`load_checkpoint`
+        filename = params.exp_dir / "generator.pt"
+        torch.save({"model": model.state_dict()}, str(filename))
+        logging.info(f"Saved to {filename}")
+
+
+if __name__ == "__main__":
+    formatter = "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
+
+    logging.basicConfig(format=formatter, level=logging.INFO)
+    main()

egs/libritts/TTS/vocos/generator.py

@@ -1,122 +1,154 @@
-import torch
-from torch import nn
-
+import logging
 from typing import Optional
 
+import numpy as np
+import torch
+from torch import nn
+from torch.autograd import Variable
+from torch.nn import functional as F
 
-class AdaLayerNorm(nn.Module):
+
+def window_sumsquare(
+    window: torch.Tensor,
+    n_samples: int,
+    hop_length: int = 256,
+    win_length: int = 1024,
+):
     """
-    Adaptive Layer Normalization module with learnable embeddings per `num_embeddings` classes
-
-    Args:
-        num_embeddings (int): Number of embeddings.
-        embedding_dim (int): Dimension of the embeddings.
+    Compute the sum-square envelope of a window function at a given hop length.
+    This is used to estimate modulation effects induced by windowing
+    observations in short-time fourier transforms.
+    Parameters
+    ----------
+    window : string, tuple, number, callable, or list-like
+        Window specification, as in `get_window`
+    n_samples : int > 0
+        The number of expected samples.
+    hop_length : int > 0
+        The number of samples to advance between frames
+    win_length :
+        The length of the window function.
+    Returns
+    -------
+    wss : torch.Tensor, The sum-squared envelope of the window function.
     """
 
-    def __init__(self, num_embeddings: int, embedding_dim: int, eps: float = 1e-6):
-        super().__init__()
-        self.eps = eps
-        self.dim = embedding_dim
-        self.scale = nn.Embedding(
-            num_embeddings=num_embeddings, embedding_dim=embedding_dim
-        )
-        self.shift = nn.Embedding(
-            num_embeddings=num_embeddings, embedding_dim=embedding_dim
-        )
-        torch.nn.init.ones_(self.scale.weight)
-        torch.nn.init.zeros_(self.shift.weight)
+    n_frames = (n_samples - win_length) // hop_length + 1
+    output_size = (n_frames - 1) * hop_length + win_length
+    device = window.device
 
-    def forward(self, x: torch.Tensor, cond_embedding_id: torch.Tensor) -> torch.Tensor:
-        scale = self.scale(cond_embedding_id)
-        shift = self.shift(cond_embedding_id)
-        x = nn.functional.layer_norm(x, (self.dim,), eps=self.eps)
-        x = x * scale + shift
-        return x
+    # Window envelope
+    window_sq = window.square().expand(1, n_frames, -1).transpose(1, 2)
+    window_envelope = torch.nn.functional.fold(
+        window_sq,
+        output_size=(1, output_size),
+        kernel_size=(1, win_length),
+        stride=(1, hop_length),
+    ).squeeze()
+    window_envelope = torch.nn.functional.pad(
+        window_envelope, (0, n_samples - output_size)
+    )
+    return window_envelope
 
 
-class ISTFT(nn.Module):
-    """
-    Custom implementation of ISTFT since torch.istft doesn't allow custom padding (other than `center=True`) with
-    windowing. This is because the NOLA (Nonzero Overlap Add) check fails at the edges.
-    See issue: https://github.com/pytorch/pytorch/issues/62323
-    Specifically, in the context of neural vocoding we are interested in "same" padding analogous to CNNs.
-
-    Args:
-        n_fft (int): Size of Fourier transform.
-        hop_length (int): The distance between neighboring sliding window frames.
-        win_length (int): The size of window frame and STFT filter.
-        padding (str, optional): Type of padding. Options are "center" or "same". Defaults to "same".
-    """
+class ISTFT(torch.nn.Module):
+    """adapted from Prem Seetharaman's https://github.com/pseeth/pytorch-stft"""
 
     def __init__(
-        self, n_fft: int, hop_length: int, win_length: int, padding: str = "same"
+        self,
+        filter_length: int = 1024,
+        hop_length: int = 256,
+        win_length: int = 1024,
+        padding: str = "none",
+        window_type: str = "povey",
+        max_samples: int = 1440000,  # 1440000 / 24000 = 60s
     ):
-        super().__init__()
-        if padding not in ["center", "same"]:
-            raise ValueError("Padding must be 'center' or 'same'.")
-        self.padding = padding
-        self.n_fft = n_fft
+        super(ISTFT, self).__init__()
+        self.filter_length = filter_length
         self.hop_length = hop_length
         self.win_length = win_length
-        window = torch.hann_window(win_length)
-        self.register_buffer("window", window)
+        self.padding = padding
+        scale = self.filter_length / self.hop_length
+        fourier_basis = np.fft.fft(np.eye(self.filter_length))
 
-    def forward(self, spec: torch.Tensor) -> torch.Tensor:
-        """
-        Compute the Inverse Short Time Fourier Transform (ISTFT) of a complex spectrogram.
+        cutoff = int((self.filter_length / 2 + 1))
+        fourier_basis = np.vstack(
+            [np.real(fourier_basis[:cutoff, :]), np.imag(fourier_basis[:cutoff, :])]
+        )
+        inverse_basis = torch.FloatTensor(
+            np.linalg.pinv(scale * fourier_basis).T[:, None, :]
+        )
 
-        Args:
-            spec (Tensor): Input complex spectrogram of shape (B, N, T), where B is the batch size,
-                            N is the number of frequency bins, and T is the number of time frames.
+        assert filter_length >= win_length
+        # Consistence with lhotse, search "create_frame_window" in https://github.com/lhotse-speech/lhotse
+        assert window_type in [
+            "hanning",
+            "povey",
+        ], f"Only 'hanning' and 'povey' windows are supported, given {window_type}."
+        fft_window = torch.hann_window(win_length, periodic=False)
+        if window_type == "povey":
+            fft_window = fft_window.pow(0.85)
 
-        Returns:
-            Tensor: Reconstructed time-domain signal of shape (B, L), where L is the length of the output signal.
-        """
+        if filter_length > win_length:
+            pad_size = (filter_length - win_length) // 2
+            fft_window = torch.nn.functional.pad(fft_window, (pad_size, pad_size))
+
+        window_sum = window_sumsquare(
+            window=fft_window,
+            n_samples=max_samples,
+            hop_length=hop_length,
+            win_length=filter_length,
+        )
+
+        inverse_basis *= fft_window
+
+        self.register_buffer("inverse_basis", inverse_basis.float())
+        self.register_buffer("fft_window", fft_window)
+        self.register_buffer("window_sum", window_sum)
+        self.tiny = torch.finfo(torch.float16).tiny
+
+    def forward(self, magnitude, phase):
+        magnitude_phase = torch.cat(
+            [magnitude * torch.cos(phase), magnitude * torch.sin(phase)], dim=1
+        )
+        inverse_transform = F.conv_transpose1d(
+            magnitude_phase,
+            Variable(self.inverse_basis, requires_grad=False),
+            stride=self.hop_length,
+            padding=0,
+        )
+        inverse_transform = inverse_transform.squeeze(1)
+
+        window_sum = self.window_sum
+        if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+            if self.window_sum.size(-1) < inverse_transform.size(-1):
+                logging.warning(
+                    f"The precomputed `window_sumsquare` is too small, recomputing, "
+                    f"from {self.window_sum.size(-1)} to {inverse_transform.size(-1)}"
+                )
+                window_sum = window_sumsquare(
+                    window=self.fft_window,
+                    n_samples=inverse_transform.size(-1),
+                    win_length=self.filter_length,
+                    hop_length=self.hop_length,
+                )
+        window_sum = window_sum[: inverse_transform.size(-1)]
+        approx_nonzero_indices = (window_sum > self.tiny).nonzero().squeeze()
+
+        inverse_transform[:, approx_nonzero_indices] /= window_sum[
+            approx_nonzero_indices
+        ]
+
+        # scale by hop ratio
+        inverse_transform *= float(self.filter_length) / self.hop_length
+        assert self.padding in ["none", "same", "center"]
         if self.padding == "center":
-            # Fallback to pytorch native implementation
-            return torch.istft(
-                spec,
-                self.n_fft,
-                self.hop_length,
-                self.win_length,
-                self.window,
-                center=True,
-            )
+            pad_len = self.filter_length // 2
         elif self.padding == "same":
-            pad = (self.win_length - self.hop_length) // 2
+            pad_len = (self.filter_length - self.hop_length) // 2
         else:
-            raise ValueError("Padding must be 'center' or 'same'.")
-
-        assert spec.dim() == 3, "Expected a 3D tensor as input"
-        B, N, T = spec.shape
-
-        # Inverse FFT
-        ifft = torch.fft.irfft(spec, self.n_fft, dim=1, norm="backward")
-        ifft = ifft * self.window[None, :, None]
-
-        # Overlap and Add
-        output_size = (T - 1) * self.hop_length + self.win_length
-        y = torch.nn.functional.fold(
-            ifft,
-            output_size=(1, output_size),
-            kernel_size=(1, self.win_length),
-            stride=(1, self.hop_length),
-        )[:, 0, 0, :]
-
-        # Window envelope
-        window_sq = self.window.square().expand(1, T, -1).transpose(1, 2)
-        window_envelope = torch.nn.functional.fold(
-            window_sq,
-            output_size=(1, output_size),
-            kernel_size=(1, self.win_length),
-            stride=(1, self.hop_length),
-        ).squeeze()
-
-        # Normalize
-        norm_indexes = window_envelope > 1e-11
-        y[:, norm_indexes] = y[:, norm_indexes] / window_envelope[norm_indexes]
-
-        return y
+            return inverse_transform
+        return inverse_transform[:, pad_len:-pad_len]
 
 
 class ConvNeXtBlock(nn.Module):
@@ -127,8 +159,6 @@ class ConvNeXtBlock(nn.Module):
         intermediate_dim (int): Dimensionality of the intermediate layer.
         layer_scale_init_value (float, optional): Initial value for the layer scale. None means no scaling.
             Defaults to None.
-        adanorm_num_embeddings (int, optional): Number of embeddings for AdaLayerNorm.
-            None means non-conditional LayerNorm. Defaults to None.
     """
 
     def __init__(
@@ -136,20 +166,14 @@ class ConvNeXtBlock(nn.Module):
         dim: int,
         intermediate_dim: int,
         layer_scale_init_value: Optional[float] = None,
-        adanorm_num_embeddings: Optional[int] = None,
     ):
         super().__init__()
         self.dwconv = nn.Conv1d(
             dim, dim, kernel_size=7, padding=3, groups=dim
         )  # depthwise conv
-        self.adanorm = adanorm_num_embeddings is not None
-        if adanorm_num_embeddings:
-            self.norm = AdaLayerNorm(adanorm_num_embeddings, dim, eps=1e-6)
-        else:
-            self.norm = nn.LayerNorm(dim, eps=1e-6)
-        self.pwconv1 = nn.Linear(
-            dim, intermediate_dim
-        )  # pointwise/1x1 convs, implemented with linear layers
+        self.norm = nn.LayerNorm(dim, eps=1e-6)
+        # pointwise/1x1 convs, implemented with linear layers
+        self.pwconv1 = nn.Linear(dim, intermediate_dim)
         self.act = nn.GELU()
         self.pwconv2 = nn.Linear(intermediate_dim, dim)
         self.gamma = (
@@ -159,16 +183,13 @@ class ConvNeXtBlock(nn.Module):
         )
 
     def forward(
-        self, x: torch.Tensor, cond_embedding_id: Optional[torch.Tensor] = None
+        self,
+        x: torch.Tensor,
     ) -> torch.Tensor:
         residual = x
         x = self.dwconv(x)
         x = x.transpose(1, 2)  # (B, C, T) -> (B, T, C)
-        if self.adanorm:
-            assert cond_embedding_id is not None
-            x = self.norm(x, cond_embedding_id)
-        else:
-            x = self.norm(x)
+        x = self.norm(x)
         x = self.pwconv1(x)
         x = self.act(x)
         x = self.pwconv2(x)
@@ -189,28 +210,22 @@ class Generator(torch.nn.Module):
         hop_length: int = 256,
         intermediate_dim: int = 1536,
         num_layers: int = 8,
-        padding: str = "same",
-        layer_scale_init_value: Optional[float] = None,
-        adanorm_num_embeddings: Optional[int] = None,
+        padding: str = "none",
+        max_samples: int = 1440000,  # 1440000 / 24000 = 60s
     ):
         super(Generator, self).__init__()
         self.feature_dim = feature_dim
         self.embed = nn.Conv1d(feature_dim, dim, kernel_size=7, padding=3)
 
-        self.adanorm = adanorm_num_embeddings is not None
-        if adanorm_num_embeddings:
-            self.norm = AdaLayerNorm(adanorm_num_embeddings, dim, eps=1e-6)
-        else:
-            self.norm = nn.LayerNorm(dim, eps=1e-6)
+        self.norm = nn.LayerNorm(dim, eps=1e-6)
 
-        layer_scale_init_value = layer_scale_init_value or 1 / num_layers
+        layer_scale_init_value = 1 / num_layers
         self.convnext = nn.ModuleList(
             [
                 ConvNeXtBlock(
                     dim=dim,
                     intermediate_dim=intermediate_dim,
                     layer_scale_init_value=layer_scale_init_value,
-                    adanorm_num_embeddings=adanorm_num_embeddings,
                 )
                 for _ in range(num_layers)
             ]
@@ -221,7 +236,11 @@ class Generator(torch.nn.Module):
 
         self.out_proj = torch.nn.Linear(dim, n_fft + 2)
         self.istft = ISTFT(
-            n_fft=n_fft, hop_length=hop_length, win_length=n_fft, padding=padding
+            filter_length=n_fft,
+            hop_length=hop_length,
+            win_length=n_fft,
+            padding=padding,
+            max_samples=max_samples,
         )
 
     def _init_weights(self, m):
@@ -229,29 +248,17 @@ class Generator(torch.nn.Module):
             nn.init.trunc_normal_(m.weight, std=0.02)
             nn.init.constant_(m.bias, 0)
 
-    def forward(self, x: torch.Tensor, **kwargs) -> torch.Tensor:
-        bandwidth_id = kwargs.get("bandwidth_id", None)
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
         x = self.embed(x)
-        if self.adanorm:
-            assert bandwidth_id is not None
-            x = self.norm(x.transpose(1, 2), cond_embedding_id=bandwidth_id)
-        else:
-            x = self.norm(x.transpose(1, 2))
-
+        x = self.norm(x.transpose(1, 2))
         x = x.transpose(1, 2)
         for conv_block in self.convnext:
-            x = conv_block(x, cond_embedding_id=bandwidth_id)
-
+            x = conv_block(x)
         x = self.final_layer_norm(x.transpose(1, 2))
-
         x = self.out_proj(x).transpose(1, 2)
-        mag, p = x.chunk(2, dim=1)
+        mag, phase = x.chunk(2, dim=1)
         mag = torch.exp(mag)
-        mag = torch.clip(
-            mag, max=1e2
-        )  # safeguard to prevent excessively large magnitudes
-        x = torch.cos(p)
-        y = torch.sin(p)
-        S = mag * (x + 1j * y)
-        audio = self.istft(S)
+        # safeguard to prevent excessively large magnitudes
+        mag = torch.clip(mag, max=1e2)
+        audio = self.istft(mag, phase)
         return audio

egs/libritts/TTS/vocos/infer.py

@@ -20,6 +20,7 @@ import argparse
 import json
 import logging
 import math
+import time
 import os
 from functools import partial
 from pathlib import Path
@@ -29,7 +30,7 @@ import torch.nn as nn
 from lhotse.utils import fix_random_seed
 from scipy.io.wavfile import write
 from train import add_model_arguments, get_model, get_params
-from tts_datamodule import LJSpeechTtsDataModule
+from tts_datamodule import LibriTTSDataModule
 
 from icefall.checkpoint import (
     average_checkpoints,
@@ -89,7 +90,7 @@ def get_parser():
     parser.add_argument(
         "--exp-dir",
         type=str,
-        default="flow_match/exp",
+        default="vocos/exp",
         help="The experiment dir",
     )
 
@@ -128,22 +129,31 @@ def decode_one_batch(
 
     cut_ids = [cut.id for cut in batch["cut"]]
 
+    infer_time = 0
+    audio_time = 0
+
     features = batch["features"]  # (B, T, F)
     utt_durations = batch["features_lens"]
 
     x = features.permute(0, 2, 1)  # (B, F, T)
 
+    audio_time += torch.sum(utt_durations)
+
+    start = time.time()
+
     audios = model(x.to(device))  # (B, T)
 
+    infer_time += time.time() - start
+
     wav_dir = f"{params.res_dir}/{params.suffix}"
     os.makedirs(wav_dir, exist_ok=True)
 
     for i in range(audios.shape[0]):
-        audio = audios[i][
-            : int(utt_durations[i] * params.frame_shift_ms / 1000 * 22050)
-        ]
+        audio = audios[i][: int(utt_durations[i] * 256)]
         audio = audio.cpu().squeeze().numpy()
-        write(f"{wav_dir}/{cut_ids[i]}.wav", 22050, audio)
+        write(f"{wav_dir}/{cut_ids[i]}.wav", 24000, audio)
+
+    print(f"RTF : {infer_time / (audio_time * (256/24000))}")
 
 
 def decode_dataset(
@@ -173,7 +183,7 @@ def decode_dataset(
 
     with open(f"{params.res_dir}/{test_set}.scp", "w", encoding="utf8") as f:
         for batch_idx, batch in enumerate(dl):
-            texts = batch["text"]
+            # texts = batch["text"]
             cut_ids = [cut.id for cut in batch["cut"]]
 
             decode_one_batch(
@@ -182,12 +192,12 @@ def decode_dataset(
                 batch=batch,
             )
 
-            assert len(texts) == len(cut_ids), (len(texts), len(cut_ids))
+            # assert len(texts) == len(cut_ids), (len(texts), len(cut_ids))
 
-            for i in range(len(texts)):
-                f.write(f"{cut_ids[i]}\t{texts[i]}\n")
+            # for i in range(len(texts)):
+            # f.write(f"{cut_ids[i]}\t{texts[i]}\n")
 
-            num_cuts += len(texts)
+            # num_cuts += len(texts)
 
             if batch_idx % 50 == 0:
                 batch_str = f"{batch_idx}/{num_batches}"
@@ -200,7 +210,7 @@ def decode_dataset(
 @torch.no_grad()
 def main():
     parser = get_parser()
-    LJSpeechTtsDataModule.add_arguments(parser)
+    LibriTTSDataModule.add_arguments(parser)
     args = parser.parse_args()
     args.exp_dir = Path(args.exp_dir)
 
@@ -318,11 +328,11 @@ def main():
 
     # we need cut ids to display recognition results.
     args.return_cuts = True
-    ljspeech = LJSpeechTtsDataModule(args)
+    libritts = LibriTTSDataModule(args)
 
-    test_cuts = ljspeech.test_cuts()
+    test_cuts = libritts.test_clean_cuts()
 
-    test_dl = ljspeech.test_dataloaders(test_cuts)
+    test_dl = libritts.test_dataloaders(test_cuts)
 
     test_sets = ["test"]
     test_dls = [test_dl]

egs/libritts/TTS/vocos/model.py

@@ -19,8 +19,9 @@ class Vocos(torch.nn.Module):
         hop_length: int = 256,
         intermediate_dim: int = 1536,
         num_layers: int = 8,
-        padding: str = "same",
+        padding: str = "none",
         sample_rate: int = 24000,
+        max_seconds: int = 60,
     ):
         super(Vocos, self).__init__()
         self.generator = Generator(
@@ -31,6 +32,7 @@ class Vocos(torch.nn.Module):
             num_layers=num_layers,
             intermediate_dim=intermediate_dim,
             padding=padding,
+            max_samples=int(sample_rate * max_seconds),
         )
 
         self.mpd = MultiPeriodDiscriminator()

egs/libritts/TTS/vocos/onnx_pretrained.py

@@ -0,0 +1,268 @@
+#!/usr/bin/env python3
+# Copyright      2022  Xiaomi Corp.        (authors: Fangjun Kuang)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This script loads ONNX models and uses them to decode waves.
+You can use the following command to get the exported models:
+
+We use the pre-trained model from
+https://huggingface.co/Zengwei/icefall-asr-librispeech-zipformer-2023-05-15
+as an example to show how to use this file.
+
+1. Download the pre-trained model
+
+cd egs/librispeech/ASR
+
+repo_url=https://huggingface.co/Zengwei/icefall-asr-librispeech-zipformer-2023-05-15
+GIT_LFS_SKIP_SMUDGE=1 git clone $repo_url
+repo=$(basename $repo_url)
+
+pushd $repo
+git lfs pull --include "exp/pretrained.pt"
+
+cd exp
+ln -s pretrained.pt epoch-99.pt
+popd
+
+2. Export the model to ONNX
+
+./zipformer/export-onnx.py \
+  --tokens $repo/data/lang_bpe_500/tokens.txt \
+  --use-averaged-model 0 \
+  --epoch 99 \
+  --avg 1 \
+  --exp-dir $repo/exp \
+  --causal False
+
+It will generate the following 3 files inside $repo/exp:
+
+  - encoder-epoch-99-avg-1.onnx
+  - decoder-epoch-99-avg-1.onnx
+  - joiner-epoch-99-avg-1.onnx
+
+3. Run this file
+
+./zipformer/onnx_pretrained.py \
+  --encoder-model-filename $repo/exp/encoder-epoch-99-avg-1.onnx \
+  --decoder-model-filename $repo/exp/decoder-epoch-99-avg-1.onnx \
+  --joiner-model-filename $repo/exp/joiner-epoch-99-avg-1.onnx \
+  --tokens $repo/data/lang_bpe_500/tokens.txt \
+  $repo/test_wavs/1089-134686-0001.wav \
+  $repo/test_wavs/1221-135766-0001.wav \
+  $repo/test_wavs/1221-135766-0002.wav
+"""
+
+import argparse
+import logging
+import math
+from pathlib import Path
+from typing import List, Tuple
+
+import onnxruntime as ort
+import torch
+import torchaudio
+from torch.nn.utils.rnn import pad_sequence
+from lhotse import Fbank, FbankConfig
+from icefall.utils import str2bool
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--model-filename",
+        type=str,
+        required=True,
+        help="Path to the encoder onnx model. ",
+    )
+
+    parser.add_argument(
+        "--sampling-rate",
+        type=int,
+        default=24000,
+        help="The sampleing rate of libritts dataset",
+    )
+
+    parser.add_argument(
+        "--frame-shift",
+        type=int,
+        default=256,
+        help="Frame shift.",
+    )
+
+    parser.add_argument(
+        "--frame-length",
+        type=int,
+        default=1024,
+        help="Frame shift.",
+    )
+
+    parser.add_argument(
+        "--use-fft-mag",
+        type=str2bool,
+        default=True,
+        help="Whether to use magnitude of fbank, false to use power energy.",
+    )
+
+    parser.add_argument(
+        "--output-dir",
+        type=str,
+        default="generated_audios",
+        help="The generated will be written to.",
+    )
+
+    parser.add_argument(
+        "sound_files",
+        type=str,
+        nargs="+",
+        help="The input sound file(s) to transcribe. "
+        "Supported formats are those supported by torchaudio.load(). "
+        "For example, wav and flac are supported. "
+        "The sample rate has to be 16kHz.",
+    )
+
+    return parser
+
+
+class OnnxModel:
+    def __init__(
+        self,
+        model_filename: str,
+    ):
+        session_opts = ort.SessionOptions()
+        session_opts.inter_op_num_threads = 1
+        session_opts.intra_op_num_threads = 4
+
+        self.session_opts = session_opts
+
+        self.init_model(model_filename)
+
+    def init_model(self, model_filename: str):
+        self.model = ort.InferenceSession(
+            model_filename,
+            sess_options=self.session_opts,
+            providers=["CPUExecutionProvider"],
+        )
+
+    def run_model(
+        self,
+        x: torch.Tensor,
+    ) -> torch.Tensor:
+        """
+        Args:
+          x:
+            A 3-D tensor of shape (N, T, C)
+          x_lens:
+            A 2-D tensor of shape (N,). Its dtype is torch.int64
+        Returns:
+          Return a tuple containing:
+            - encoder_out, its shape is (N, T', joiner_dim)
+            - encoder_out_lens, its shape is (N,)
+        """
+        out = self.model.run(
+            [
+                self.model.get_outputs()[0].name,
+            ],
+            {
+                self.model.get_inputs()[0].name: x.numpy(),
+            },
+        )
+        return torch.from_numpy(out[0])
+
+
+def read_sound_files(
+    filenames: List[str], expected_sample_rate: float
+) -> List[torch.Tensor]:
+    """Read a list of sound files into a list 1-D float32 torch tensors.
+    Args:
+      filenames:
+        A list of sound filenames.
+      expected_sample_rate:
+        The expected sample rate of the sound files.
+    Returns:
+      Return a list of 1-D float32 torch tensors.
+    """
+    ans = []
+    for f in filenames:
+        wave, sample_rate = torchaudio.load(f)
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
+        # We use only the first channel
+        ans.append(wave[0])
+    return ans
+
+
+@torch.no_grad()
+def main():
+    parser = get_parser()
+    args = parser.parse_args()
+
+    output_dir = Path(args.model_filename).parent / args.output_dir
+    output_dir.mkdir(exist_ok=True)
+    args.output_dir = output_dir
+    logging.info(vars(args))
+
+    model = OnnxModel(model_filename=args.model_filename)
+
+    config = FbankConfig(
+        sampling_rate=args.sampling_rate,
+        frame_length=args.frame_length / args.sampling_rate,  # (in second),
+        frame_shift=args.frame_shift / args.sampling_rate,  # (in second)
+        use_fft_mag=args.use_fft_mag,
+    )
+    fbank = Fbank(config)
+
+    logging.info(f"Reading sound files: {args.sound_files}")
+
+    waves = read_sound_files(
+        filenames=args.sound_files, expected_sample_rate=args.sampling_rate
+    )
+    wave_lengths = [w.size(0) for w in waves]
+    waves = pad_sequence(waves, batch_first=True, padding_value=0)
+
+    logging.info(f"waves : {waves.shape}")
+
+    features = fbank.extract_batch(waves, sampling_rate=args.sampling_rate)
+
+    if features.dim() == 2:
+        features = features.unsqueeze(0)
+
+    features = features.permute(0, 2, 1)
+
+    logging.info(f"features : {features.shape}")
+
+    logging.info("Generating started")
+
+    # model forward
+    audios = model.run_model(features)
+
+    for i, filename in enumerate(args.sound_files):
+        audio = audios[i : i + 1, 0 : wave_lengths[i]]
+        ofilename = args.output_dir / filename.split("/")[-1]
+        logging.info(f"Writting audio : {ofilename}")
+        torchaudio.save(str(ofilename), audio.cpu(), args.sampling_rate)
+
+    logging.info("Generating Done")
+
+
+if __name__ == "__main__":
+    formatter = "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
+
+    logging.basicConfig(format=formatter, level=logging.INFO)
+    main()

egs/libritts/TTS/vocos/pretrained.py

@@ -0,0 +1,196 @@
+#!/usr/bin/env python3
+# Copyright      2024  Xiaomi Corp.        (authors: Wei Kang)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This script loads a checkpoint and uses it to decode waves.
+You can generate the checkpoint with the following command:
+"""
+
+
+import argparse
+import logging
+import math
+from pathlib import Path
+from typing import List
+
+import torch
+import torchaudio
+from torch.nn.utils.rnn import pad_sequence
+from train import add_model_arguments, get_model, get_params
+from lhotse import Fbank, FbankConfig
+
+from icefall.utils import str2bool
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--checkpoint",
+        type=str,
+        required=True,
+        help="Path to the checkpoint. "
+        "The checkpoint is assumed to be saved by "
+        "icefall.checkpoint.save_checkpoint().",
+    )
+
+    parser.add_argument(
+        "--sampling-rate",
+        type=int,
+        default=24000,
+        help="The sampleing rate of libritts dataset",
+    )
+
+    parser.add_argument(
+        "--frame-shift",
+        type=int,
+        default=256,
+        help="Frame shift.",
+    )
+
+    parser.add_argument(
+        "--frame-length",
+        type=int,
+        default=1024,
+        help="Frame shift.",
+    )
+
+    parser.add_argument(
+        "--use-fft-mag",
+        type=str2bool,
+        default=True,
+        help="Whether to use magnitude of fbank, false to use power energy.",
+    )
+
+    parser.add_argument(
+        "--output-dir",
+        type=str,
+        default="generated_audios",
+        help="The generated will be written to.",
+    )
+
+    parser.add_argument(
+        "sound_files",
+        type=str,
+        nargs="+",
+        help="The input sound file(s) to transcribe. "
+        "Supported formats are those supported by torchaudio.load(). "
+        "For example, wav and flac are supported. "
+        "The sample rate has to be 16kHz.",
+    )
+
+    add_model_arguments(parser)
+
+    return parser
+
+
+def read_sound_files(
+    filenames: List[str], expected_sample_rate: float
+) -> List[torch.Tensor]:
+    """Read a list of sound files into a list 1-D float32 torch tensors.
+    Args:
+      filenames:
+        A list of sound filenames.
+      expected_sample_rate:
+        The expected sample rate of the sound files.
+    Returns:
+      Return a list of 1-D float32 torch tensors.
+    """
+    ans = []
+    for f in filenames:
+        wave, sample_rate = torchaudio.load(f)
+        assert (
+            sample_rate == expected_sample_rate
+        ), f"expected sample rate: {expected_sample_rate}. Given: {sample_rate}"
+        # We use only the first channel
+        ans.append(wave[0].contiguous())
+    return ans
+
+
+@torch.no_grad()
+def main():
+    parser = get_parser()
+    args = parser.parse_args()
+
+    params = get_params()
+
+    params.update(vars(args))
+
+    device = torch.device("cpu")
+    if torch.cuda.is_available():
+        device = torch.device("cuda", 0)
+    params.device = device
+
+    output_dir = Path(params.checkpoint).parent / params.output_dir
+    output_dir.mkdir(exist_ok=True)
+    params.output_dir = output_dir
+
+    logging.info(f"{params}")
+
+    logging.info("Creating model")
+    model = get_model(params)
+
+    model = model.generator
+
+    checkpoint = torch.load(params.checkpoint, map_location="cpu")
+    model.load_state_dict(checkpoint["model"], strict=False)
+    model.to(device)
+    model.eval()
+
+    logging.info("Constructing Fbank computer")
+
+    config = FbankConfig(
+        sampling_rate=params.sampling_rate,
+        frame_length=params.frame_length / params.sampling_rate,  # (in second),
+        frame_shift=params.frame_shift / params.sampling_rate,  # (in second)
+        use_fft_mag=params.use_fft_mag,
+    )
+    fbank = Fbank(config)
+
+    logging.info(f"Reading sound files: {params.sound_files}")
+
+    waves = read_sound_files(
+        filenames=params.sound_files, expected_sample_rate=params.sampling_rate
+    )
+    wave_lengths = [w.size(0) for w in waves]
+    waves = pad_sequence(waves, batch_first=True, padding_value=0)
+
+    features = (
+        fbank.extract_batch(waves, sampling_rate=params.sampling_rate)
+        .permute(0, 2, 1)
+        .to(device)
+    )
+
+    logging.info("Generating started")
+
+    # model forward
+    audios = model(features)
+
+    for i, filename in enumerate(params.sound_files):
+        audio = audios[i : i + 1, 0 : wave_lengths[i]]
+        ofilename = params.output_dir / filename.split("/")[-1]
+        logging.info(f"Writting audio : {ofilename}")
+        torchaudio.save(str(ofilename), audio.cpu(), params.sampling_rate)
+
+    logging.info("Generating Done")
+
+
+if __name__ == "__main__":
+    formatter = "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
+    logging.basicConfig(format=formatter, level=logging.INFO)
+    main()

egs/libritts/TTS/vocos/train.py

@@ -52,9 +52,11 @@ from utils import (
     save_checkpoint,
     plot_spectrogram,
     get_cosine_schedule_with_warmup,
+    save_checkpoint_with_global_batch_idx,
 )
 
 from icefall import diagnostics
+from icefall.checkpoint import remove_checkpoints, update_averaged_model
 from icefall.dist import cleanup_dist, setup_dist
 from icefall.env import get_env_info
 from icefall.hooks import register_inf_check_hooks
@@ -91,6 +93,20 @@ def add_model_arguments(parser: argparse.ArgumentParser):
         help="Intermediate dim of ConvNeXt module.",
     )
 
+    parser.add_argument(
+        "--max-seconds",
+        type=int,
+        default=60,
+        help="""
+        The length of the precomputed normalization window sum square
+        (required by istft). This argument is only for onnx export, it determines
+        the max length of the audio that be properly normalized.
+        Note, you can generate audios longer than this value with the exported onnx model,
+        the part longer than this value will not be normalized yet.
+        The larger this value is the bigger the exported onnx model will be.
+        """,
+    )
+
 
 def get_parser():
     parser = argparse.ArgumentParser(
@@ -203,6 +219,16 @@ def get_parser():
         """,
     )
 
+    parser.add_argument(
+        "--keep-last-epoch-k",
+        type=int,
+        default=50,
+        help="""Only keep this number of checkpoints on disk.
+        For instance, if it is 3, there are only 3 checkpoints
+        in the exp-dir with filenames `epoch-xxx.pt`.
+        """,
+    )
+
     parser.add_argument(
         "--average-period",
         type=int,
@@ -290,8 +316,8 @@ def get_params() -> AttributeDict:
             "valid_interval": 500,
             "feature_dim": 80,
             "segment_size": 16384,
-            "adam_b1": 0.8,
-            "adam_b2": 0.9,
+            "adam_b1": 0.9,
+            "adam_b2": 0.99,
             "warmup_steps": 0,
             "max_steps": 2000000,
             "env_info": get_env_info(),
@@ -311,6 +337,7 @@ def get_model(params: AttributeDict) -> nn.Module:
         intermediate_dim=params.intermediate_dim,
         num_layers=params.num_layers,
         sample_rate=params.sampling_rate,
+        max_seconds=params.max_seconds,
     ).to(device)
 
     num_param_gen = sum([p.numel() for p in model.generator.parameters()])
@@ -479,11 +506,6 @@ def compute_discriminator_loss(
     info["loss_disc_mrd"] = loss_mrd.detach().cpu().item()
     info["loss_disc_mpd"] = loss_mpd.detach().cpu().item()
 
-    for i in range(len(loss_mpd_real)):
-        info[f"loss_disc_mpd_period_{i+1}"] = loss_mpd_real[i] + loss_mpd_gen[i]
-    for i in range(len(loss_mrd_real)):
-        info[f"loss_disc_mrd_resolution_{i+1}"] = loss_mrd_real[i] + loss_mrd_gen[i]
-
     return loss_disc_all, info
 
 
@@ -497,6 +519,7 @@ def train_one_epoch(
     train_dl: torch.utils.data.DataLoader,
     valid_dl: torch.utils.data.DataLoader,
     scaler: GradScaler,
+    model_avg: Optional[nn.Module] = None,
     tb_writer: Optional[SummaryWriter] = None,
     world_size: int = 1,
     rank: int = 0,
@@ -542,6 +565,7 @@ def train_one_epoch(
         save_checkpoint(
             filename=params.exp_dir / f"bad-model{suffix}-{rank}.pt",
             model=model,
+            model_avg=model_avg,
             params=params,
             optimizer_g=optimizer_g,
             optimizer_d=optimizer_d,
@@ -588,6 +612,7 @@ def train_one_epoch(
 
             loss_disc.backward()
             optimizer_d.step()
+            scheduler_d.step()
 
             optimizer_g.zero_grad()
             loss_gen, loss_gen_info = compute_generator_loss(
@@ -599,6 +624,7 @@ def train_one_epoch(
 
             loss_gen.backward()
             optimizer_g.step()
+            scheduler_g.step()
 
             loss_info = loss_gen_info + loss_disc_info
             tot_loss = (tot_loss * (1 - 1 / params.reset_interval)) + loss_gen_info
@@ -611,6 +637,39 @@ def train_one_epoch(
         if params.print_diagnostics and batch_idx == 5:
             return
 
+        if (
+            rank == 0
+            and params.batch_idx_train > 0
+            and params.batch_idx_train % params.average_period == 0
+        ):
+            update_averaged_model(
+                params=params,
+                model_cur=model,
+                model_avg=model_avg,
+            )
+
+        if (
+            params.batch_idx_train > 0
+            and params.batch_idx_train % params.save_every_n == 0
+        ):
+            save_checkpoint_with_global_batch_idx(
+                out_dir=params.exp_dir,
+                global_batch_idx=params.batch_idx_train,
+                model=model,
+                model_avg=model_avg,
+                params=params,
+                optimizer=optimizer,
+                scheduler=scheduler,
+                sampler=train_dl.sampler,
+                scaler=scaler,
+                rank=rank,
+            )
+            remove_checkpoints(
+                out_dir=params.exp_dir,
+                topk=params.keep_last_k,
+                rank=rank,
+            )
+
         if params.batch_idx_train % 100 == 0 and params.use_fp16:
             # If the grad scale was less than 1, try increasing it.    The _growth_interval
             # of the grad scaler is configurable, but we can't configure it to have different
@@ -641,8 +700,8 @@ def train_one_epoch(
                 f"Epoch {params.cur_epoch}, batch {batch_idx}, "
                 f"global_batch_idx: {params.batch_idx_train}, batch size: {batch_size}, "
                 f"loss[{loss_info}], tot_loss[{tot_loss}], "
-                f"cur_lr_g: {cur_lr_g:.2e}, "
-                f"cur_lr_d: {cur_lr_d:.2e}, "
+                f"cur_lr_g: {cur_lr_g:.4e}, "
+                f"cur_lr_d: {cur_lr_d:.4e}, "
                 + (f"grad_scale: {scaler._scale.item()}" if params.use_fp16 else "")
             )
 
@@ -685,8 +744,6 @@ def train_one_epoch(
                     tb_writer, "train/valid_", params.batch_idx_train
                 )
 
-    scheduler_g.step()
-    scheduler_d.step()
     loss_value = tot_loss["loss_gen"]
     params.train_loss = loss_value
     if params.train_loss < params.best_train_loss:
@@ -766,7 +823,7 @@ def compute_validation_loss(
                     params.sampling_rate,
                 )
 
-        logging.info(f"RTF : {infer_time / (audio_time * 10 / 1000)}")
+        logging.info(f"Validation RTF : {infer_time / (audio_time * 10 / 1000)}")
 
         if world_size > 1:
             tot_loss.reduce(device)
@@ -811,15 +868,22 @@ def run(rank, world_size, args):
     device = torch.device("cpu")
     if torch.cuda.is_available():
         device = torch.device("cuda", rank)
-    logging.info(f"Device: {device}")
     params.device = device
     logging.info(params)
-    logging.info("About to create model")
 
+    logging.info("About to create model")
     model = get_model(params)
 
+    assert params.save_every_n >= params.average_period
+    model_avg: Optional[nn.Module] = None
+    if rank == 0:
+        # model_avg is only used with rank 0
+        model_avg = copy.deepcopy(model).to(torch.float64)
+
     assert params.start_epoch > 0, params.start_epoch
-    checkpoints = load_checkpoint_if_available(params=params, model=model)
+    checkpoints = load_checkpoint_if_available(
+        params=params, model=model, model_avg=model_avg
+    )
 
     model = model.to(device)
     generator = model.generator
@@ -915,6 +979,7 @@ def run(rank, world_size, args):
         train_one_epoch(
             params=params,
             model=model,
+            model_avg=model_avg,
             optimizer_g=optimizer_g,
             optimizer_d=optimizer_d,
             scheduler_g=scheduler_g,
@@ -936,6 +1001,7 @@ def run(rank, world_size, args):
             filename=filename,
             params=params,
             model=model,
+            model_avg=model_avg,
             optimizer_g=optimizer_g,
             optimizer_d=optimizer_d,
             scheduler_g=scheduler_g,
@@ -945,28 +1011,20 @@ def run(rank, world_size, args):
             rank=rank,
         )
 
-        if params.batch_idx_train % params.save_every_n == 0:
-            filename = params.exp_dir / f"checkpoint-{params.batch_idx_train}.pt"
-            save_checkpoint(
-                filename=filename,
-                params=params,
-                model=model,
-                optimizer_g=optimizer_g,
-                optimizer_d=optimizer_d,
-                scheduler_g=scheduler_g,
-                scheduler_d=scheduler_d,
-                sampler=train_dl.sampler,
-                scaler=scaler,
-                rank=rank,
-            )
-            if rank == 0:
-                if params.best_train_epoch == params.cur_epoch:
-                    best_train_filename = params.exp_dir / "best-train-loss.pt"
-                    copyfile(src=filename, dst=best_train_filename)
+        if params.best_train_epoch == params.cur_epoch:
+            best_train_filename = params.exp_dir / "best-train-loss.pt"
+            copyfile(src=filename, dst=best_train_filename)
 
-                if params.best_valid_epoch == params.cur_epoch:
-                    best_valid_filename = params.exp_dir / "best-valid-loss.pt"
-                    copyfile(src=filename, dst=best_valid_filename)
+        if params.best_valid_epoch == params.cur_epoch:
+            best_valid_filename = params.exp_dir / "best-valid-loss.pt"
+            copyfile(src=filename, dst=best_valid_filename)
+
+        remove_checkpoints(
+            out_dir=params.exp_dir,
+            topk=params.keep_last_epoch_k,
+            prefix="epoch",
+            rank=rank,
+        )
 
     logging.info("Done!")
 

egs/libritts/TTS/vocos/utils.py

@@ -34,6 +34,69 @@ def plot_spectrogram(spectrogram):
     return fig
 
 
+def save_checkpoint_with_global_batch_idx(
+    out_dir: Path,
+    global_batch_idx: int,
+    model: Union[nn.Module, DDP],
+    model_avg: Optional[nn.Module] = None,
+    params: Optional[Dict[str, Any]] = None,
+    optimizer_g: Optional[Optimizer] = None,
+    optimizer_d: Optional[Optimizer] = None,
+    scheduler_g: Optional[LRScheduler] = None,
+    scheduler_d: Optional[LRScheduler] = None,
+    scaler: Optional[GradScaler] = None,
+    sampler: Optional[CutSampler] = None,
+    rank: int = 0,
+):
+    """Save training info after processing given number of batches.
+
+    Args:
+      out_dir:
+        The directory to save the checkpoint.
+      global_batch_idx:
+        The number of batches processed so far from the very start of the
+        training. The saved checkpoint will have the following filename:
+
+            f'out_dir / checkpoint-{global_batch_idx}.pt'
+      model:
+        The neural network model whose `state_dict` will be saved in the
+        checkpoint.
+      model_avg:
+        The stored model averaged from the start of training.
+      params:
+        A dict of training configurations to be saved.
+      optimizer:
+        The optimizer used in the training. Its `state_dict` will be saved.
+      scheduler:
+        The learning rate scheduler used in the training. Its `state_dict` will
+        be saved.
+      scaler:
+        The scaler used for mix precision training. Its `state_dict` will
+        be saved.
+      sampler:
+        The sampler used in the training dataset.
+      rank:
+        The rank ID used in DDP training of the current node. Set it to 0
+        if DDP is not used.
+    """
+    out_dir = Path(out_dir)
+    out_dir.mkdir(parents=True, exist_ok=True)
+    filename = out_dir / f"checkpoint-{global_batch_idx}.pt"
+    save_checkpoint(
+        filename=filename,
+        model=model,
+        model_avg=model_avg,
+        params=params,
+        optimizer_g=optimizer_g,
+        scheduler_g=scheduler_g,
+        optimizer_d=optimizer_d,
+        scheduler_d=scheduler_d,
+        scaler=scaler,
+        sampler=sampler,
+        rank=rank,
+    )
+
+
 def load_checkpoint(
     filename: Path,
     model: nn.Module,

egs/ljspeech/TTS/local/evaluate_fsd.py

@@ -0,0 +1,287 @@
+"""
+Calculate Frechet Speech Distance betweeen two speech directories.
+Adapted from: https://github.com/gudgud96/frechet-audio-distance/blob/main/frechet_audio_distance/fad.py
+"""
+import argparse
+import logging
+import os
+from multiprocessing.dummy import Pool as ThreadPool
+
+import librosa
+import numpy as np
+import soundfile as sf
+import torch
+from scipy import linalg
+from tqdm import tqdm
+from transformers import Wav2Vec2FeatureExtractor, Wav2Vec2Model
+
+logging.basicConfig(level=logging.INFO)
+
+
+def get_parser():
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--real-path", type=str, help="path of the real speech directory"
+    )
+    parser.add_argument(
+        "--eval-path", type=str, help="path of the evaluated speech directory"
+    )
+    parser.add_argument(
+        "--model-path",
+        type=str,
+        default="model/huggingface/wav2vec2_base",
+        help="path of the wav2vec 2.0 model directory",
+    )
+    parser.add_argument(
+        "--real-embds-path",
+        type=str,
+        default=None,
+        help="path of the real embedding directory",
+    )
+    parser.add_argument(
+        "--eval-embds-path",
+        type=str,
+        default=None,
+        help="path of the evaluated embedding directory",
+    )
+    return parser
+
+
+class FrechetSpeechDistance:
+    def __init__(
+        self,
+        model_path="resources/wav2vec2_base",
+        pca_dim=128,
+        speech_load_worker=8,
+    ):
+        """
+        Initialize FSD
+        """
+        self.sample_rate = 16000
+        self.channels = 1
+        self.device = (
+            torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
+        )
+        logging.info("[Frechet Speech Distance] Using device: {}".format(self.device))
+        self.speech_load_worker = speech_load_worker
+        self.feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(model_path)
+        self.model = Wav2Vec2Model.from_pretrained(model_path)
+        self.model.to(self.device)
+        self.model.eval()
+        self.pca_dim = pca_dim
+
+    def load_speech_files(self, dir, dtype="float32"):
+        def _load_speech_task(fname, sample_rate, channels, dtype="float32"):
+            if dtype not in ["float64", "float32", "int32", "int16"]:
+                raise ValueError(f"dtype not supported: {dtype}")
+
+            wav_data, sr = sf.read(fname, dtype=dtype)
+            # For integer type PCM input, convert to [-1.0, +1.0]
+            if dtype == "int16":
+                wav_data = wav_data / 32768.0
+            elif dtype == "int32":
+                wav_data = wav_data / float(2**31)
+
+            # Convert to mono
+            assert channels in [1, 2], "channels must be 1 or 2"
+            if len(wav_data.shape) > channels:
+                wav_data = np.mean(wav_data, axis=1)
+
+            if sr != sample_rate:
+                wav_data = (
+                    librosa.resample(wav_data, orig_sr=sr, target_sr=sample_rate),
+                )
+
+            return wav_data
+
+        task_results = []
+
+        pool = ThreadPool(self.speech_load_worker)
+
+        logging.info("[Frechet Speech Distance] Loading speech from {}...".format(dir))
+        for fname in os.listdir(dir):
+            res = pool.apply_async(
+                _load_speech_task,
+                args=(os.path.join(dir, fname), self.sample_rate, self.channels, dtype),
+            )
+            task_results.append(res)
+        pool.close()
+        pool.join()
+
+        return [k.get() for k in task_results]
+
+    def get_embeddings(self, x):
+        """
+        Get embeddings
+        Params:
+        -- x    : a list of np.ndarray speech samples
+        -- sr   : sampling rate.
+        """
+        embd_lst = []
+        try:
+            for speech in tqdm(x):
+                input_features = self.feature_extractor(
+                    speech, sampling_rate=self.sample_rate, return_tensors="pt"
+                ).input_values.to(self.device)
+                with torch.no_grad():
+                    embd = self.model(input_features).last_hidden_state.mean(1)
+
+                if embd.device != torch.device("cpu"):
+                    embd = embd.cpu()
+
+                if torch.is_tensor(embd):
+                    embd = embd.detach().numpy()
+
+                embd_lst.append(embd)
+        except Exception as e:
+            print(
+                "[Frechet Speech Distance] get_embeddings throw an exception: {}".format(
+                    str(e)
+                )
+            )
+
+        return np.concatenate(embd_lst, axis=0)
+
+    def calculate_embd_statistics(self, embd_lst):
+        if isinstance(embd_lst, list):
+            embd_lst = np.array(embd_lst)
+        mu = np.mean(embd_lst, axis=0)
+        sigma = np.cov(embd_lst, rowvar=False)
+        return mu, sigma
+
+    def calculate_frechet_distance(self, mu1, sigma1, mu2, sigma2, eps=1e-6):
+        """
+        Adapted from: https://github.com/mseitzer/pytorch-fid/blob/master/src/pytorch_fid/fid_score.py
+
+        Numpy implementation of the Frechet Distance.
+        The Frechet distance between two multivariate Gaussians X_1 ~ N(mu_1, C_1)
+        and X_2 ~ N(mu_2, C_2) is
+                d^2 = ||mu_1 - mu_2||^2 + Tr(C_1 + C_2 - 2*sqrt(C_1*C_2)).
+        Stable version by Dougal J. Sutherland.
+        Params:
+        -- mu1   : Numpy array containing the activations of a layer of the
+                inception net (like returned by the function 'get_predictions')
+                for generated samples.
+        -- mu2   : The sample mean over activations, precalculated on an
+                representative data set.
+        -- sigma1: The covariance matrix over activations for generated samples.
+        -- sigma2: The covariance matrix over activations, precalculated on an
+                representative data set.
+        Returns:
+        --   : The Frechet Distance.
+        """
+
+        mu1 = np.atleast_1d(mu1)
+        mu2 = np.atleast_1d(mu2)
+
+        sigma1 = np.atleast_2d(sigma1)
+        sigma2 = np.atleast_2d(sigma2)
+
+        assert (
+            mu1.shape == mu2.shape
+        ), "Training and test mean vectors have different lengths"
+        assert (
+            sigma1.shape == sigma2.shape
+        ), "Training and test covariances have different dimensions"
+
+        diff = mu1 - mu2
+
+        # Product might be almost singular
+        covmean, _ = linalg.sqrtm(sigma1.dot(sigma2).astype(complex), disp=False)
+        if not np.isfinite(covmean).all():
+            msg = (
+                "fid calculation produces singular product; "
+                "adding %s to diagonal of cov estimates"
+            ) % eps
+            logging.info(msg)
+            offset = np.eye(sigma1.shape[0]) * eps
+            covmean = linalg.sqrtm(
+                (sigma1 + offset).dot(sigma2 + offset).astype(complex)
+            )
+
+        # Numerical error might give slight imaginary component
+        if np.iscomplexobj(covmean):
+            if not np.allclose(np.diagonal(covmean).imag, 0, atol=1e-3):
+                m = np.max(np.abs(covmean.imag))
+                raise ValueError("Imaginary component {}".format(m))
+            covmean = covmean.real
+
+        tr_covmean = np.trace(covmean)
+
+        return diff.dot(diff) + np.trace(sigma1) + np.trace(sigma2) - 2 * tr_covmean
+
+    def score(
+        self,
+        real_path,
+        eval_path,
+        real_embds_path=None,
+        eval_embds_path=None,
+        dtype="float32",
+    ):
+        """
+        Computes the Frechet Speech Distance (FSD) between two directories of speech files.
+
+        Parameters:
+        - real_path (str): Path to the directory containing real speech files.
+        - eval_path (str): Path to the directory containing evaluation speech files.
+        - real_embds_path (str, optional): Path to save/load real speech embeddings (e.g., /folder/bkg_embs.npy). If None, embeddings won't be saved.
+        - eval_embds_path (str, optional): Path to save/load evaluation speech embeddings (e.g., /folder/test_embs.npy). If None, embeddings won't be saved.
+        - dtype (str, optional): Data type for loading speech. Default is "float32".
+
+        Returns:
+        - float: The Frechet Speech Distance (FSD) score between the two directories of speech files.
+        """
+        # Load or compute real embeddings
+        if real_embds_path is not None and os.path.exists(real_embds_path):
+            logging.info(
+                f"[Frechet Speech Distance] Loading embeddings from {real_embds_path}..."
+            )
+            embds_real = np.load(real_embds_path)
+        else:
+            speech_real = self.load_speech_files(real_path, dtype=dtype)
+            embds_real = self.get_embeddings(speech_real)
+            if real_embds_path:
+                os.makedirs(os.path.dirname(real_embds_path), exist_ok=True)
+                np.save(real_embds_path, embds_real)
+
+        # Load or compute eval embeddings
+        if eval_embds_path is not None and os.path.exists(eval_embds_path):
+            logging.info(
+                f"[Frechet Speech Distance] Loading embeddings from {eval_embds_path}..."
+            )
+            embds_eval = np.load(eval_embds_path)
+        else:
+            speech_eval = self.load_speech_files(eval_path, dtype=dtype)
+            embds_eval = self.get_embeddings(speech_eval)
+            if eval_embds_path:
+                os.makedirs(os.path.dirname(eval_embds_path), exist_ok=True)
+                np.save(eval_embds_path, embds_eval)
+
+        # Check if embeddings are empty
+        if len(embds_real) == 0:
+            logging.info("[Frechet Speech Distance] real set dir is empty, exiting...")
+            return -10.46
+        if len(embds_eval) == 0:
+            logging.info("[Frechet Speech Distance] eval set dir is empty, exiting...")
+            return -1
+
+        # Compute statistics and FSD score
+        mu_real, sigma_real = self.calculate_embd_statistics(embds_real)
+        mu_eval, sigma_eval = self.calculate_embd_statistics(embds_eval)
+
+        fsd_score = self.calculate_frechet_distance(
+            mu_real, sigma_real, mu_eval, sigma_eval
+        )
+
+        return fsd_score
+
+
+if __name__ == "__main__":
+    parser = get_parser()
+    args = parser.parse_args()
+    FSD = FrechetSpeechDistance(model_path=args.model_path)
+    score = FSD.score(
+        args.real_path, args.eval_path, args.real_embds_path, args.eval_embds_path
+    )
+    logging.info(f"FSD score: {score:.2f}")

egs/ljspeech/TTS/local/evaluate_wer_whisper.py

@@ -0,0 +1,139 @@
+"""
+Calculate WER with Whisper model
+"""
+import argparse
+import logging
+import os
+import re
+from pathlib import Path
+from typing import List, Tuple
+
+import librosa
+import soundfile as sf
+import torch
+from num2words import num2words
+from tqdm import tqdm
+from transformers import pipeline
+
+from icefall.utils import store_transcripts, write_error_stats
+
+logging.basicConfig(level=logging.INFO)
+
+
+def get_parser():
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument("--wav-path", type=str, help="path of the speech directory")
+    parser.add_argument("--decode-path", type=str, help="path of the speech directory")
+    parser.add_argument(
+        "--model-path",
+        type=str,
+        default="model/huggingface/whisper_medium",
+        help="path of the huggingface whisper model",
+    )
+    parser.add_argument(
+        "--transcript-path",
+        type=str,
+        default="data/transcript/test.tsv",
+        help="path of the transcript tsv file",
+    )
+    parser.add_argument(
+        "--batch-size", type=int, default=64, help="decoding batch size"
+    )
+    parser.add_argument(
+        "--device", type=str, default="cuda:0", help="decoding device, cuda:0 or cpu"
+    )
+    return parser
+
+
+def post_process(text: str):
+    def convert_numbers(match):
+        return num2words(match.group())
+
+    text = re.sub(r"\b\d{1,2}\b", convert_numbers, text)
+    text = re.sub(r"[^a-zA-Z0-9']", " ", text.lower())
+    text = re.sub(r"\s+", " ", text)
+    return text
+
+
+def save_results(
+    res_dir: str,
+    results: List[Tuple[str, List[str], List[str]]],
+):
+    if not os.path.exists(res_dir):
+        os.makedirs(res_dir)
+    recog_path = os.path.join(res_dir, "recogs.txt")
+    results = sorted(results)
+    store_transcripts(filename=recog_path, texts=results)
+    logging.info(f"The transcripts are stored in {recog_path}")
+
+    errs_filename = os.path.join(res_dir, "errs.txt")
+    with open(errs_filename, "w") as f:
+        _ = write_error_stats(f, "test", results, enable_log=True)
+    logging.info("Wrote detailed error stats to {}".format(errs_filename))
+
+
+class SpeechEvalDataset(torch.utils.data.Dataset):
+    def __init__(self, wav_path: str, transcript_path: str):
+        super().__init__()
+        self.audio_name = []
+        self.audio_paths = []
+        self.transcripts = []
+        with Path(transcript_path).open("r", encoding="utf8") as f:
+            meta = [item.split("\t") for item in f.read().rstrip().split("\n")]
+        for item in meta:
+            self.audio_name.append(item[0])
+            self.audio_paths.append(Path(wav_path, item[0] + ".wav"))
+            self.transcripts.append(item[1])
+
+    def __len__(self):
+        return len(self.audio_paths)
+
+    def __getitem__(self, index: int):
+        audio, sampling_rate = sf.read(self.audio_paths[index])
+        item = {
+            "array": librosa.resample(audio, orig_sr=sampling_rate, target_sr=16000),
+            "sampling_rate": 16000,
+            "reference": self.transcripts[index],
+            "audio_name": self.audio_name[index],
+        }
+        return item
+
+
+def main(args):
+
+    batch_size = args.batch_size
+
+    pipe = pipeline(
+        "automatic-speech-recognition",
+        model=args.model_path,
+        device=args.device,
+        tokenizer=args.model_path,
+    )
+
+    dataset = SpeechEvalDataset(args.wav_path, args.transcript_path)
+
+    results = []
+    bar = tqdm(
+        pipe(
+            dataset,
+            generate_kwargs={"language": "english", "task": "transcribe"},
+            batch_size=batch_size,
+        ),
+        total=len(dataset),
+    )
+    for out in bar:
+        results.append(
+            (
+                out["audio_name"][0],
+                post_process(out["reference"][0].strip()).split(),
+                post_process(out["text"].strip()).split(),
+            )
+        )
+    save_results(args.decode_path, results)
+
+
+if __name__ == "__main__":
+    parser = get_parser()
+    args = parser.parse_args()
+    main(args)

egs/ljspeech/TTS/vocos/export-onnx.py

@@ -0,0 +1 @@
+../../../libritts/TTS/vocos/export-onnx.py

egs/ljspeech/TTS/vocos/export.py

@@ -0,0 +1 @@
+../../../libritts/TTS/vocos/export.py

egs/ljspeech/TTS/vocos/infer.py

@@ -0,0 +1,340 @@
+#!/usr/bin/env python3
+# Copyright         2024  Xiaomi Corp.        (authors: Wei Kang
+#                                                       Han Zhu)
+#
+# See ../../../../LICENSE for clarification regarding multiple authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import argparse
+import json
+import logging
+import math
+import os
+from functools import partial
+from pathlib import Path
+
+import torch
+import torch.nn as nn
+from lhotse.utils import fix_random_seed
+from scipy.io.wavfile import write
+from train import add_model_arguments, get_model, get_params
+from tts_datamodule import LJSpeechTtsDataModule
+
+from icefall.checkpoint import (
+    average_checkpoints,
+    average_checkpoints_with_averaged_model,
+    find_checkpoints,
+    load_checkpoint,
+)
+from icefall.utils import AttributeDict, setup_logger, str2bool
+
+LOG_EPS = math.log(1e-10)
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter
+    )
+
+    parser.add_argument(
+        "--epoch",
+        type=int,
+        default=100,
+        help="""It specifies the checkpoint to use for decoding.
+        Note: Epoch counts from 1.
+        You can specify --avg to use more checkpoints for model averaging.""",
+    )
+
+    parser.add_argument(
+        "--iter",
+        type=int,
+        default=0,
+        help="""If positive, --epoch is ignored and it
+        will use the checkpoint exp_dir/checkpoint-iter.pt.
+        You can specify --avg to use more checkpoints for model averaging.
+        """,
+    )
+
+    parser.add_argument(
+        "--avg",
+        type=int,
+        default=10,
+        help="Number of checkpoints to average. Automatically select "
+        "consecutive checkpoints before the checkpoint specified by "
+        "'--epoch' and '--iter'",
+    )
+
+    parser.add_argument(
+        "--use-averaged-model",
+        type=str2bool,
+        default=False,
+        help="Whether to load averaged model. Currently it only supports "
+        "using --epoch. If True, it would decode with the averaged model "
+        "over the epoch range from `epoch-avg` (excluded) to `epoch`."
+        "Actually only the models with epoch number of `epoch-avg` and "
+        "`epoch` are loaded for averaging. ",
+    )
+
+    parser.add_argument(
+        "--exp-dir",
+        type=str,
+        default="flow_match/exp",
+        help="The experiment dir",
+    )
+
+    parser.add_argument(
+        "--generate-dir",
+        type=str,
+        default="generated_wavs",
+        help="Path name of the generated wavs",
+    )
+
+    add_model_arguments(parser)
+
+    return parser
+
+
+def decode_one_batch(
+    params: AttributeDict,
+    model: nn.Module,
+    batch: dict,
+):
+    """
+    Args:
+      params:
+        It's the return value of :func:`get_params`.
+      model:
+        The text-to-feature neural model.
+      batch:
+        It is the return value from iterating
+        `lhotse.dataset.K2SpeechRecognitionDataset`. See its documentation
+        for the format of the `batch`.
+    Returns:
+      Return the decoding result. See above description for the format of
+      the returned dict.
+    """
+    device = next(model.parameters()).device
+
+    cut_ids = [cut.id for cut in batch["cut"]]
+
+    features = batch["features"]  # (B, T, F)
+    utt_durations = batch["features_lens"]
+
+    x = features.permute(0, 2, 1)  # (B, F, T)
+
+    audios = model(x.to(device))  # (B, T)
+
+    wav_dir = f"{params.res_dir}/{params.suffix}"
+    os.makedirs(wav_dir, exist_ok=True)
+
+    for i in range(audios.shape[0]):
+        audio = audios[i][: (utt_durations[i] - 1) * 256 + 1024]
+        audio = audio.cpu().squeeze().numpy()
+        write(f"{wav_dir}/{cut_ids[i]}.wav", 22050, audio)
+
+
+def decode_dataset(
+    dl: torch.utils.data.DataLoader,
+    params: AttributeDict,
+    model: nn.Module,
+    test_set: str,
+):
+    """Decode dataset.
+
+    Args:
+      dl:
+        PyTorch's dataloader containing the dataset to decode.
+      params:
+        It is returned by :func:`get_params`.
+      model:
+        The text-to-feature neural model.
+      test_set:
+        The name of the test_set
+    """
+    num_cuts = 0
+
+    try:
+        num_batches = len(dl)
+    except TypeError:
+        num_batches = "?"
+
+    with open(f"{params.res_dir}/{test_set}.scp", "w", encoding="utf8") as f:
+        for batch_idx, batch in enumerate(dl):
+            texts = batch["text"]
+            cut_ids = [cut.id for cut in batch["cut"]]
+
+            decode_one_batch(
+                params=params,
+                model=model,
+                batch=batch,
+            )
+
+            assert len(texts) == len(cut_ids), (len(texts), len(cut_ids))
+
+            for i in range(len(texts)):
+                f.write(f"{cut_ids[i]}\t{texts[i]}\n")
+
+            num_cuts += len(texts)
+
+            if batch_idx % 50 == 0:
+                batch_str = f"{batch_idx}/{num_batches}"
+
+                logging.info(
+                    f"batch {batch_str}, cuts processed until now is {num_cuts}"
+                )
+
+
+@torch.no_grad()
+def main():
+    parser = get_parser()
+    LJSpeechTtsDataModule.add_arguments(parser)
+    args = parser.parse_args()
+    args.exp_dir = Path(args.exp_dir)
+
+    params = get_params()
+    params.update(vars(args))
+
+    params.res_dir = params.exp_dir / params.generate_dir
+
+    if params.iter > 0:
+        params.suffix = f"iter-{params.iter}-avg-{params.avg}"
+    else:
+        params.suffix = f"epoch-{params.epoch}-avg-{params.avg}"
+
+    if params.use_averaged_model:
+        params.suffix += "-use-averaged-model"
+
+    setup_logger(f"{params.res_dir}/log-decode-{params.suffix}")
+    logging.info("Decoding started")
+
+    device = torch.device("cpu")
+    if torch.cuda.is_available():
+        device = torch.device("cuda", 0)
+    params.device = device
+
+    logging.info(f"Device: {device}")
+
+    logging.info(params)
+    fix_random_seed(666)
+
+    logging.info("About to create model")
+    model = get_model(params)
+
+    if not params.use_averaged_model:
+        if params.iter > 0:
+            filenames = find_checkpoints(params.exp_dir, iteration=-params.iter)[
+                : params.avg
+            ]
+            if len(filenames) == 0:
+                raise ValueError(
+                    f"No checkpoints found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            elif len(filenames) < params.avg:
+                raise ValueError(
+                    f"Not enough checkpoints ({len(filenames)}) found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            logging.info(f"averaging {filenames}")
+            model.to(device)
+            model.load_state_dict(average_checkpoints(filenames, device=device))
+        elif params.avg == 1:
+            load_checkpoint(f"{params.exp_dir}/epoch-{params.epoch}.pt", model)
+        else:
+            start = params.epoch - params.avg + 1
+            filenames = []
+            for i in range(start, params.epoch + 1):
+                if i >= 1:
+                    filenames.append(f"{params.exp_dir}/epoch-{i}.pt")
+            logging.info(f"averaging {filenames}")
+            model.to(device)
+            model.load_state_dict(average_checkpoints(filenames, device=device))
+    else:
+        if params.iter > 0:
+            filenames = find_checkpoints(params.exp_dir, iteration=-params.iter)[
+                : params.avg + 1
+            ]
+            if len(filenames) == 0:
+                raise ValueError(
+                    f"No checkpoints found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            elif len(filenames) < params.avg + 1:
+                raise ValueError(
+                    f"Not enough checkpoints ({len(filenames)}) found for"
+                    f" --iter {params.iter}, --avg {params.avg}"
+                )
+            filename_start = filenames[-1]
+            filename_end = filenames[0]
+            logging.info(
+                "Calculating the averaged model over iteration checkpoints"
+                f" from {filename_start} (excluded) to {filename_end}"
+            )
+            model.to(device)
+            model.load_state_dict(
+                average_checkpoints_with_averaged_model(
+                    filename_start=filename_start,
+                    filename_end=filename_end,
+                    device=device,
+                )
+            )
+        else:
+            assert params.avg > 0, params.avg
+            start = params.epoch - params.avg
+            assert start >= 1, start
+            filename_start = f"{params.exp_dir}/epoch-{start}.pt"
+            filename_end = f"{params.exp_dir}/epoch-{params.epoch}.pt"
+            logging.info(
+                f"Calculating the averaged model over epoch range from "
+                f"{start} (excluded) to {params.epoch}"
+            )
+            model.to(device)
+            model.load_state_dict(
+                average_checkpoints_with_averaged_model(
+                    filename_start=filename_start,
+                    filename_end=filename_end,
+                    device=device,
+                )
+            )
+
+    model = model.to(device)
+    model.eval()
+
+    num_param = sum([p.numel() for p in model.parameters()])
+    logging.info(f"Number of model parameters: {num_param}")
+
+    # we need cut ids to display recognition results.
+    args.return_cuts = True
+    ljspeech = LJSpeechTtsDataModule(args)
+
+    test_cuts = ljspeech.test_cuts()
+
+    test_dl = ljspeech.test_dataloaders(test_cuts)
+
+    test_sets = ["test"]
+    test_dls = [test_dl]
+
+    for test_set, test_dl in zip(test_sets, test_dls):
+        decode_dataset(
+            dl=test_dl,
+            params=params,
+            model=model,
+            test_set=test_set,
+        )
+
+    logging.info("Done!")
+
+
+if __name__ == "__main__":
+    main()

egs/ljspeech/TTS/vocos/onnx_pretrained.py

@@ -0,0 +1 @@
+../../../libritts/TTS/vocos/onnx_pretrained.py

egs/ljspeech/TTS/vocos/pretrained.py

@@ -0,0 +1 @@
+../../../libritts/TTS/vocos/pretrained.py

egs/ljspeech/TTS/vocos/train.py

@@ -52,9 +52,11 @@ from utils import (
     save_checkpoint,
     plot_spectrogram,
     get_cosine_schedule_with_warmup,
+    save_checkpoint_with_global_batch_idx,
 )
 
 from icefall import diagnostics
+from icefall.checkpoint import remove_checkpoints, update_averaged_model
 from icefall.dist import cleanup_dist, setup_dist
 from icefall.env import get_env_info
 from icefall.hooks import register_inf_check_hooks
@@ -65,7 +67,7 @@ from icefall.utils import (
     str2bool,
     get_parameter_groups_with_lrs,
 )
-from models import Vocos
+from model import Vocos
 from lhotse import Fbank, FbankConfig
 
 
@@ -91,6 +93,20 @@ def add_model_arguments(parser: argparse.ArgumentParser):
         help="Intermediate dim of ConvNeXt module.",
     )
 
+    parser.add_argument(
+        "--max-seconds",
+        type=int,
+        default=60,
+        help="""
+        The length of the precomputed normalization window sum square
+        (required by istft). This argument is only for onnx export, it determines
+        the max length of the audio that be properly normalized.
+        Note, you can generate audios longer than this value with the exported onnx model,
+        the part longer than this value will not be normalized yet.
+        The larger this value is the bigger the exported onnx model will be.
+        """,
+    )
+
 
 def get_parser():
     parser = argparse.ArgumentParser(
@@ -203,6 +219,16 @@ def get_parser():
         """,
     )
 
+    parser.add_argument(
+        "--keep-last-epoch-k",
+        type=int,
+        default=50,
+        help="""Only keep this number of checkpoints on disk.
+        For instance, if it is 3, there are only 3 checkpoints
+        in the exp-dir with filenames `epoch-xxx.pt`.
+        """,
+    )
+
     parser.add_argument(
         "--average-period",
         type=int,
@@ -290,8 +316,8 @@ def get_params() -> AttributeDict:
             "valid_interval": 500,
             "feature_dim": 80,
             "segment_size": 16384,
-            "adam_b1": 0.8,
-            "adam_b2": 0.9,
+            "adam_b1": 0.9,
+            "adam_b2": 0.99,
             "warmup_steps": 0,
             "max_steps": 2000000,
             "env_info": get_env_info(),
@@ -311,18 +337,17 @@ def get_model(params: AttributeDict) -> nn.Module:
         intermediate_dim=params.intermediate_dim,
         num_layers=params.num_layers,
         sample_rate=params.sampling_rate,
+        max_seconds=params.max_seconds,
     ).to(device)
 
-    num_param_head = sum([p.numel() for p in model.head.parameters()])
-    logging.info(f"Number of Head parameters : {num_param_head}")
-    num_param_bone = sum([p.numel() for p in model.backbone.parameters()])
-    logging.info(f"Number of Generator parameters : {num_param_bone}")
+    num_param_gen = sum([p.numel() for p in model.generator.parameters()])
+    logging.info(f"Number of Generator parameters : {num_param_gen}")
     num_param_mpd = sum([p.numel() for p in model.mpd.parameters()])
     logging.info(f"Number of MultiPeriodDiscriminator parameters : {num_param_mpd}")
     num_param_mrd = sum([p.numel() for p in model.mrd.parameters()])
     logging.info(f"Number of MultiResolutionDiscriminator parameters : {num_param_mrd}")
     logging.info(
-        f"Number of model parameters : {num_param_head + num_param_bone + num_param_mpd + num_param_mrd}"
+        f"Number of model parameters : {num_param_gen + num_param_mpd + num_param_mrd}"
     )
     return model
 
@@ -481,11 +506,6 @@ def compute_discriminator_loss(
     info["loss_disc_mrd"] = loss_mrd.detach().cpu().item()
     info["loss_disc_mpd"] = loss_mpd.detach().cpu().item()
 
-    for i in range(len(loss_mpd_real)):
-        info[f"loss_disc_mpd_period_{i+1}"] = loss_mpd_real[i] + loss_mpd_gen[i]
-    for i in range(len(loss_mrd_real)):
-        info[f"loss_disc_mrd_resolution_{i+1}"] = loss_mrd_real[i] + loss_mrd_gen[i]
-
     return loss_disc_all, info
 
 
@@ -499,6 +519,7 @@ def train_one_epoch(
     train_dl: torch.utils.data.DataLoader,
     valid_dl: torch.utils.data.DataLoader,
     scaler: GradScaler,
+    model_avg: Optional[nn.Module] = None,
     tb_writer: Optional[SummaryWriter] = None,
     world_size: int = 1,
     rank: int = 0,
@@ -544,6 +565,7 @@ def train_one_epoch(
         save_checkpoint(
             filename=params.exp_dir / f"bad-model{suffix}-{rank}.pt",
             model=model,
+            model_avg=model_avg,
             params=params,
             optimizer_g=optimizer_g,
             optimizer_d=optimizer_d,
@@ -566,10 +588,6 @@ def train_one_epoch(
             params.segment_size - params.frame_length
         ) // params.frame_shift + 1
 
-        # segment_frames = (
-        # params.segment_size + params.frame_shift // 2
-        # ) // params.frame_shift
-
         start_p = random.randint(0, features_lens.min() - (segment_frames + 1))
 
         features = features[:, start_p : start_p + segment_frames, :].permute(
@@ -594,6 +612,7 @@ def train_one_epoch(
 
             loss_disc.backward()
             optimizer_d.step()
+            scheduler_d.step()
 
             optimizer_g.zero_grad()
             loss_gen, loss_gen_info = compute_generator_loss(
@@ -605,6 +624,7 @@ def train_one_epoch(
 
             loss_gen.backward()
             optimizer_g.step()
+            scheduler_g.step()
 
             loss_info = loss_gen_info + loss_disc_info
             tot_loss = (tot_loss * (1 - 1 / params.reset_interval)) + loss_gen_info
@@ -617,6 +637,39 @@ def train_one_epoch(
         if params.print_diagnostics and batch_idx == 5:
             return
 
+        if (
+            rank == 0
+            and params.batch_idx_train > 0
+            and params.batch_idx_train % params.average_period == 0
+        ):
+            update_averaged_model(
+                params=params,
+                model_cur=model,
+                model_avg=model_avg,
+            )
+
+        if (
+            params.batch_idx_train > 0
+            and params.batch_idx_train % params.save_every_n == 0
+        ):
+            save_checkpoint_with_global_batch_idx(
+                out_dir=params.exp_dir,
+                global_batch_idx=params.batch_idx_train,
+                model=model,
+                model_avg=model_avg,
+                params=params,
+                optimizer=optimizer,
+                scheduler=scheduler,
+                sampler=train_dl.sampler,
+                scaler=scaler,
+                rank=rank,
+            )
+            remove_checkpoints(
+                out_dir=params.exp_dir,
+                topk=params.keep_last_k,
+                rank=rank,
+            )
+
         if params.batch_idx_train % 100 == 0 and params.use_fp16:
             # If the grad scale was less than 1, try increasing it.    The _growth_interval
             # of the grad scaler is configurable, but we can't configure it to have different
@@ -647,8 +700,8 @@ def train_one_epoch(
                 f"Epoch {params.cur_epoch}, batch {batch_idx}, "
                 f"global_batch_idx: {params.batch_idx_train}, batch size: {batch_size}, "
                 f"loss[{loss_info}], tot_loss[{tot_loss}], "
-                f"cur_lr_g: {cur_lr_g:.2e}, "
-                f"cur_lr_d: {cur_lr_d:.2e}, "
+                f"cur_lr_g: {cur_lr_g:.4e}, "
+                f"cur_lr_d: {cur_lr_d:.4e}, "
                 + (f"grad_scale: {scaler._scale.item()}" if params.use_fp16 else "")
             )
 
@@ -668,11 +721,10 @@ def train_one_epoch(
                         "train/grad_scale", cur_grad_scale, params.batch_idx_train
                     )
 
-        # if (
-        # params.batch_idx_train % params.valid_interval == 0
-        # and not params.print_diagnostics
-        # ):
-        if True:
+        if (
+            params.batch_idx_train % params.valid_interval == 0
+            and not params.print_diagnostics
+        ):
             logging.info("Computing validation loss")
             valid_info = compute_validation_loss(
                 params=params,
@@ -692,8 +744,6 @@ def train_one_epoch(
                     tb_writer, "train/valid_", params.batch_idx_train
                 )
 
-    scheduler_g.step()
-    scheduler_d.step()
     loss_value = tot_loss["loss_gen"]
     params.train_loss = loss_value
     if params.train_loss < params.best_train_loss:
@@ -773,7 +823,7 @@ def compute_validation_loss(
                     params.sampling_rate,
                 )
 
-        logging.info(f"RTF : {infer_time / (audio_time * 10 / 1000)}")
+        logging.info(f"Validation RTF : {infer_time / (audio_time * 10 / 1000)}")
 
         if world_size > 1:
             tot_loss.reduce(device)
@@ -818,19 +868,25 @@ def run(rank, world_size, args):
     device = torch.device("cpu")
     if torch.cuda.is_available():
         device = torch.device("cuda", rank)
-    logging.info(f"Device: {device}")
     params.device = device
     logging.info(params)
-    logging.info("About to create model")
 
+    logging.info("About to create model")
     model = get_model(params)
 
+    assert params.save_every_n >= params.average_period
+    model_avg: Optional[nn.Module] = None
+    if rank == 0:
+        # model_avg is only used with rank 0
+        model_avg = copy.deepcopy(model).to(torch.float64)
+
     assert params.start_epoch > 0, params.start_epoch
-    checkpoints = load_checkpoint_if_available(params=params, model=model)
+    checkpoints = load_checkpoint_if_available(
+        params=params, model=model, model_avg=model_avg
+    )
 
     model = model.to(device)
-    head = model.head
-    backbone = model.backbone
+    generator = model.generator
     mrd = model.mrd
     mpd = model.mpd
     if world_size > 1:
@@ -838,7 +894,7 @@ def run(rank, world_size, args):
         model = DDP(model, device_ids=[rank], find_unused_parameters=True)
 
     optimizer_g = torch.optim.AdamW(
-        itertools.chain(head.parameters(), backbone.parameters()),
+        generator.parameters(),
         params.learning_rate,
         betas=[params.adam_b1, params.adam_b2],
     )
@@ -923,6 +979,7 @@ def run(rank, world_size, args):
         train_one_epoch(
             params=params,
             model=model,
+            model_avg=model_avg,
             optimizer_g=optimizer_g,
             optimizer_d=optimizer_d,
             scheduler_g=scheduler_g,
@@ -944,6 +1001,7 @@ def run(rank, world_size, args):
             filename=filename,
             params=params,
             model=model,
+            model_avg=model_avg,
             optimizer_g=optimizer_g,
             optimizer_d=optimizer_d,
             scheduler_g=scheduler_g,
@@ -953,28 +1011,20 @@ def run(rank, world_size, args):
             rank=rank,
         )
 
-        if params.batch_idx_train % params.save_every_n == 0:
-            filename = params.exp_dir / f"checkpoint-{params.batch_idx_train}.pt"
-            save_checkpoint(
-                filename=filename,
-                params=params,
-                model=model,
-                optimizer_g=optimizer_g,
-                optimizer_d=optimizer_d,
-                scheduler_g=scheduler_g,
-                scheduler_d=scheduler_d,
-                sampler=train_dl.sampler,
-                scaler=scaler,
-                rank=rank,
-            )
-            if rank == 0:
-                if params.best_train_epoch == params.cur_epoch:
-                    best_train_filename = params.exp_dir / "best-train-loss.pt"
-                    copyfile(src=filename, dst=best_train_filename)
+        if params.best_train_epoch == params.cur_epoch:
+            best_train_filename = params.exp_dir / "best-train-loss.pt"
+            copyfile(src=filename, dst=best_train_filename)
 
-                if params.best_valid_epoch == params.cur_epoch:
-                    best_valid_filename = params.exp_dir / "best-valid-loss.pt"
-                    copyfile(src=filename, dst=best_valid_filename)
+        if params.best_valid_epoch == params.cur_epoch:
+            best_valid_filename = params.exp_dir / "best-valid-loss.pt"
+            copyfile(src=filename, dst=best_valid_filename)
+
+        remove_checkpoints(
+            out_dir=params.exp_dir,
+            topk=params.keep_last_epoch_k,
+            prefix="epoch",
+            rank=rank,
+        )
 
     logging.info("Done!")
 
@@ -997,7 +1047,8 @@ def main():
         run(rank=0, world_size=1, args=args)
 
 
+torch.set_num_threads(1)
+torch.set_num_interop_threads(1)
+
 if __name__ == "__main__":
-    torch.set_num_threads(1)
-    torch.set_num_interop_threads(1)
     main()

icefall/checkpoint.py

@@ -250,18 +250,22 @@ def save_checkpoint_with_global_batch_idx(
     )
 
 
-def find_checkpoints(out_dir: Path, iteration: int = 0) -> List[str]:
+def find_checkpoints(
+    out_dir: Path,
+    iteration: int = 0,
+    prefix: str = "checkpoint",
+) -> List[str]:
     """Find all available checkpoints in a directory.
 
-    The checkpoint filenames have the form: `checkpoint-xxx.pt`
+    The checkpoint filenames have the form: `{prefix}-xxx.pt`
     where xxx is a numerical value.
 
     Assume you have the following checkpoints in the folder `foo`:
 
-        - checkpoint-1.pt
-        - checkpoint-20.pt
-        - checkpoint-300.pt
-        - checkpoint-4000.pt
+        - {prefix}-1.pt
+        - {prefix}-20.pt
+        - {prefix}-300.pt
+        - {prefix}-4000.pt
 
     Case 1 (Return all checkpoints)::
 
@@ -290,8 +294,8 @@ def find_checkpoints(out_dir: Path, iteration: int = 0) -> List[str]:
       Return a list of checkpoint filenames, sorted in descending
       order by the numerical value in the filename.
     """
-    checkpoints = list(glob.glob(f"{out_dir}/checkpoint-[0-9]*.pt"))
-    pattern = re.compile(r"checkpoint-([0-9]+).pt")
+    checkpoints = list(glob.glob(f"{out_dir}/{prefix}-[0-9]*.pt"))
+    pattern = re.compile(rf"{prefix}-([0-9]+).pt")
     iter_checkpoints = []
     for c in checkpoints:
         result = pattern.search(c)
@@ -316,12 +320,13 @@ def find_checkpoints(out_dir: Path, iteration: int = 0) -> List[str]:
 def remove_checkpoints(
     out_dir: Path,
     topk: int,
+    prefix: str = "checkpoint",
     rank: int = 0,
 ):
     """Remove checkpoints from the given directory.
 
-    We assume that checkpoint filename has the form `checkpoint-xxx.pt`
-    where xxx is a number, representing the number of processed batches
+    We assume that checkpoint filename has the form `{prefix}-xxx.pt`
+    where xxx is a number, representing the number of processed batches/epochs
     when saving that checkpoint. We sort checkpoints by filename and keep
     only the `topk` checkpoints with the highest `xxx`.
 
@@ -330,6 +335,8 @@ def remove_checkpoints(
         The directory containing checkpoints to be removed.
       topk:
         Number of checkpoints to keep.
+      prefix:
+        The prefix of the checkpoint filename, normally `epoch`, `checkpoint`.
       rank:
         If using DDP for training, it is the rank of the current node.
         Use 0 if no DDP is used for training.
@@ -337,7 +344,7 @@ def remove_checkpoints(
     assert topk >= 1, topk
     if rank != 0:
         return
-    checkpoints = find_checkpoints(out_dir)
+    checkpoints = find_checkpoints(out_dir, prefix=prefix)
 
     if len(checkpoints) == 0:
         logging.warn(f"No checkpoints found in {out_dir}")