mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-08-19 06:53:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

minor fixes

Browse Source
This commit is contained in:
zr_jin 2024-12-06 11:51:40 +08:00
parent ce73643af6
commit 2504036f5b
2 changed files with 8 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
egs/wenetspeech4tts/TTS/valle/train.py
View File

@ -4,6 +4,7 @@
# Mingshuang Luo) # Mingshuang Luo)
# Copyright 2023 (authors: Feiteng Li) # Copyright 2023 (authors: Feiteng Li)
# Copyright 2024 (authors: Yuekai Zhang) # Copyright 2024 (authors: Yuekai Zhang)
# Copyright 2024 Tsinghua University (authors: Zengrui Jin,)
# #
# See ../../../../LICENSE for clarification regarding multiple authors # See ../../../../LICENSE for clarification regarding multiple authors
# #
@ -48,10 +49,8 @@ python3 valle/train.py --max-duration 160 --filter-min-duration 0.5 --filter-max
import argparse import argparse
import copy import copy
import logging import logging
import os
import random import random
import warnings import warnings
from contextlib import nullcontext
from pathlib import Path from pathlib import Path
from shutil import copyfile from shutil import copyfile
from typing import Any, Dict, Optional, Tuple, Union from typing import Any, Dict, Optional, Tuple, Union
@ -686,9 +685,9 @@ def compute_validation_loss(
output_dir = Path(f"{params.exp_dir}/eval/step-{params.batch_idx_train:06d}") output_dir = Path(f"{params.exp_dir}/eval/step-{params.batch_idx_train:06d}")
output_dir.mkdir(parents=True, exist_ok=True) output_dir.mkdir(parents=True, exist_ok=True)
if isinstance(model, DDP): if isinstance(model, DDP):
model.module.visualize(predicts, batch, output_dir=output_dir) model.module.visualize(predicts, batch, tokenizer, output_dir=output_dir)
else: else:
model.visualize(predicts, batch, output_dir=output_dir) model.visualize(predicts, batch, tokenizer, output_dir=output_dir)
return tot_loss return tot_loss

7
egs/wenetspeech4tts/TTS/valle/valle.py
View File

@ -23,6 +23,7 @@ import matplotlib.pyplot as plt
import numpy as np import numpy as np
import torch import torch
import torch.nn as nn import torch.nn as nn
from tokenizer import TextTokenCollater
from torch import Tensor from torch import Tensor
from torch.nn import Linear, Module from torch.nn import Linear, Module
from torch.nn import functional as F from torch.nn import functional as F
@ -1664,13 +1665,15 @@ class VALLE(nn.Module):
self, self,
predicts: Tuple[torch.Tensor], predicts: Tuple[torch.Tensor],
batch: Dict[str, Union[List, torch.Tensor]], batch: Dict[str, Union[List, torch.Tensor]],
tokenizer: TextTokenCollater,
output_dir: str, output_dir: str,
limit: int = 4, limit: int = 4,
) -> None: ) -> None:
text_tokens = batch["text_tokens"].to("cpu").detach().numpy()
text_tokens_lens = batch["text_tokens_lens"].to("cpu").detach().numpy()
audio_features = batch["audio_features"].to("cpu").detach().numpy() audio_features = batch["audio_features"].to("cpu").detach().numpy()
audio_features_lens = batch["audio_features_lens"].to("cpu").detach().numpy() audio_features_lens = batch["audio_features_lens"].to("cpu").detach().numpy()
tokens = batch["tokens"]
text_tokens, text_tokens_lens = tokenizer(tokens)
assert text_tokens.ndim == 2 assert text_tokens.ndim == 2
utt_ids, texts = batch["utt_id"], batch["text"] utt_ids, texts = batch["utt_id"], batch["text"]