mirror of
https://github.com/k2-fsa/icefall.git
synced 2025-08-09 10:02:22 +00:00
add README
This commit is contained in:
parent
455366418c
commit
27c1cfa851
@ -68,5 +68,54 @@ python3 valle/infer.py --output-dir demos_epoch_${epoch}_avg_${avg}_top_p_${top_
|
||||
--text-extractor pypinyin_initials_finals --top-p ${top_p}
|
||||
```
|
||||
|
||||
# [F5-TTS](https://arxiv.org/abs/2410.06885)
|
||||
|
||||
./f5-tts contains the code for training F5-TTS model.
|
||||
|
||||
Generated samples and training logs of wenetspeech basic 7k hours data can be found [here](https://huggingface.co/yuekai/f5-tts-small-wenetspeech4tts-basic/tensorboard).
|
||||
|
||||
Preparation:
|
||||
|
||||
```
|
||||
bash prepare.sh --stage 5 --stop_stage 6
|
||||
```
|
||||
|
||||
The training command is given below:
|
||||
|
||||
```
|
||||
# docker: ghcr.io/swivid/f5-tts:main
|
||||
# pip install k2==1.24.4.dev20241030+cuda12.4.torch2.4.0 -f https://k2-fsa.github.io/k2/cuda.html
|
||||
# pip install kaldialign lhotse tensorboard bigvganinference sentencepiece
|
||||
|
||||
world_size=8
|
||||
exp_dir=exp/f5-tts-small
|
||||
python3 f5-tts/train.py --max-duration 700 --filter-min-duration 0.5 --filter-max-duration 20 \
|
||||
--num-buckets 6 --dtype "bfloat16" --save-every-n 5000 --valid-interval 10000 \
|
||||
--base-lr 7.5e-5 --warmup-steps 20000 --num-epochs 60 \
|
||||
--num-decoder-layers 18 --nhead 12 --decoder-dim 768 \
|
||||
--exp-dir ${exp_dir} --world-size ${world_size}
|
||||
```
|
||||
|
||||
To inference, use:
|
||||
```
|
||||
huggingface-cli login
|
||||
huggingface-cli download --local-dir seed_tts_eval yuekai/seed_tts_eval --repo-type dataset
|
||||
huggingface-cli download --local-dir ${exp_dir} yuekai/f5-tts-small-wenetspeech4tts-basic
|
||||
huggingface-cli download nvidia/bigvgan_v2_24khz_100band_256x --local-dir bigvgan_v2_24khz_100band_256x
|
||||
|
||||
manifest=./seed_tts_eval/seedtts_testset/zh/meta.lst
|
||||
model_path=f5-tts-small-wenetspeech4tts-basic/epoch-56-avg-14.pt
|
||||
# skip
|
||||
python3 f5-tts/generate_averaged_model.py \
|
||||
--epoch 56 \
|
||||
--avg 14 --decoder-dim 768 --nhead 12 --num-decoder-layers 18 \
|
||||
--exp-dir exp/f5_small
|
||||
|
||||
|
||||
accelerate launch f5-tts/infer.py --nfe 16 --model-path $model_path --manifest-file $manifest --output-dir $output_dir --decoder-dim 768 --nhead 12 --num-decoder-layers 18
|
||||
bash local/compute_wer.sh $output_dir $manifest
|
||||
```
|
||||
|
||||
# Credits
|
||||
- [vall-e](https://github.com/lifeiteng/vall-e)
|
||||
- [VALL-E](https://github.com/lifeiteng/vall-e)
|
||||
- [F5-TTS](https://github.com/SWivid/F5-TTS)
|
||||
|
3
egs/wenetspeech4tts/TTS/f5-tts/README.md
Normal file
3
egs/wenetspeech4tts/TTS/f5-tts/README.md
Normal file
@ -0,0 +1,3 @@
|
||||
# Introduction
|
||||
Files in this folder are copied from
|
||||
https://github.com/SWivid/F5-TTS/tree/main/src/f5_tts/model
|
173
egs/wenetspeech4tts/TTS/f5-tts/generate_averaged_model.py
Normal file
173
egs/wenetspeech4tts/TTS/f5-tts/generate_averaged_model.py
Normal file
@ -0,0 +1,173 @@
|
||||
#!/usr/bin/env python3
|
||||
#
|
||||
# Copyright 2021-2022 Xiaomi Corporation (Author: Yifan Yang)
|
||||
# Copyright 2024 Yuekai Zhang
|
||||
#
|
||||
# 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.
|
||||
"""
|
||||
Usage:
|
||||
(1) use the checkpoint exp_dir/epoch-xxx.pt
|
||||
python3 bin/generate_averaged_model.py \
|
||||
--epoch 40 \
|
||||
--avg 5 \
|
||||
--exp-dir ${exp_dir}
|
||||
|
||||
It will generate a file `epoch-28-avg-15.pt` in the given `exp_dir`.
|
||||
You can later load it by `torch.load("epoch-28-avg-15.pt")`.
|
||||
"""
|
||||
|
||||
|
||||
import argparse
|
||||
from pathlib import Path
|
||||
|
||||
import k2
|
||||
import torch
|
||||
from train import add_model_arguments, get_model
|
||||
|
||||
from icefall.checkpoint import (
|
||||
average_checkpoints,
|
||||
average_checkpoints_with_averaged_model,
|
||||
find_checkpoints,
|
||||
)
|
||||
from icefall.utils import AttributeDict
|
||||
|
||||
|
||||
def get_parser():
|
||||
parser = argparse.ArgumentParser(
|
||||
formatter_class=argparse.ArgumentDefaultsHelpFormatter
|
||||
)
|
||||
|
||||
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(
|
||||
"--exp-dir",
|
||||
type=str,
|
||||
default="zipformer/exp",
|
||||
help="The experiment dir",
|
||||
)
|
||||
add_model_arguments(parser)
|
||||
return parser
|
||||
|
||||
|
||||
@torch.no_grad()
|
||||
def main():
|
||||
parser = get_parser()
|
||||
|
||||
args = parser.parse_args()
|
||||
args.exp_dir = Path(args.exp_dir)
|
||||
|
||||
params = AttributeDict()
|
||||
params.update(vars(args))
|
||||
|
||||
if params.iter > 0:
|
||||
params.suffix = f"checkpoint-{params.iter}-avg-{params.avg}"
|
||||
else:
|
||||
params.suffix = f"epoch-{params.epoch}-avg-{params.avg}"
|
||||
|
||||
print("Script started")
|
||||
|
||||
device = torch.device("cpu")
|
||||
print(f"Device: {device}")
|
||||
|
||||
print("About to create model")
|
||||
filename = f"{params.exp_dir}/epoch-{params.epoch}.pt"
|
||||
checkpoint = torch.load(filename, map_location=device)
|
||||
args = AttributeDict(checkpoint)
|
||||
model = get_model(args)
|
||||
|
||||
if params.iter > 0:
|
||||
# TODO FIX ME
|
||||
filenames = find_checkpoints(params.exp_dir, iteration=-params.iter)[
|
||||
: params.avg + 1
|
||||
]
|
||||
if len(filenames) == 0:
|
||||
raise ValueError(
|
||||
f"No checkpoints found for --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]
|
||||
print(
|
||||
"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,
|
||||
)
|
||||
)
|
||||
filename = params.exp_dir / f"checkpoint-{params.iter}-avg-{params.avg}.pt"
|
||||
torch.save({"model": model.state_dict()}, filename)
|
||||
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"
|
||||
print(
|
||||
f"Calculating the averaged model over epoch range from "
|
||||
f"{start} (excluded) to {params.epoch}"
|
||||
)
|
||||
filenames = [
|
||||
f"{params.exp_dir}/epoch-{i}.pt" for i in range(start, params.epoch + 1)
|
||||
]
|
||||
model.to(device)
|
||||
model.load_state_dict(average_checkpoints(filenames, device=device))
|
||||
|
||||
filename = params.exp_dir / f"epoch-{params.epoch}-avg-{params.avg}.pt"
|
||||
checkpoint["model"] = model.state_dict()
|
||||
torch.save(checkpoint, filename)
|
||||
|
||||
num_param = sum([p.numel() for p in model.parameters()])
|
||||
print(f"Number of model parameters: {num_param}")
|
||||
|
||||
print("Done!")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
@ -23,7 +23,7 @@ Usage:
|
||||
# docker: ghcr.io/swivid/f5-tts:main
|
||||
# pip install k2==1.24.4.dev20241030+cuda12.4.torch2.4.0 -f https://k2-fsa.github.io/k2/cuda.html
|
||||
# pip install kaldialign lhotse tensorboard bigvganinference sentencepiece
|
||||
# huggingface-cli download nvidia/bigvgan_v2_24khz_100band_256x --local-dir bigvgan_v2_24khz_100band_256x
|
||||
|
||||
world_size=8
|
||||
exp_dir=exp/f5-tts-small
|
||||
python3 f5-tts/train.py --max-duration 700 --filter-min-duration 0.5 --filter-max-duration 20 \
|
||||
|
2545
egs/wenetspeech4tts/TTS/f5-tts/vocab.txt
Normal file
2545
egs/wenetspeech4tts/TTS/f5-tts/vocab.txt
Normal file
File diff suppressed because it is too large
Load Diff
Loading…
x
Reference in New Issue
Block a user