Skip to content

SOTA Open Source TTS

Format code

Lengyue4255b1d2178d
diff --git a/fine-tune.py b/fine-tune.py
index 3d57a28..a926c52 100644
--- a/fine-tune.py
+++ b/fine-tune.py
@@ -38,8 +38,14 @@ class TrainingArguments(_TrainingArguments):
use_lora: bool = field(default=False)
-def dataset_transform(batch, tokenizer: AutoTokenizer=None):
- outputs = tokenizer(batch["prompt"], padding="longest", truncation=True, max_length=512, return_tensors="pt")
+def dataset_transform(batch, tokenizer: AutoTokenizer = None):
+ outputs = tokenizer(
+ batch["prompt"],
+ padding="longest",
+ truncation=True,
+ max_length=512,
+ return_tensors="pt",
+ )
labels = outputs.input_ids.clone()
# Set the labels to -100 so that the logits are not affected by loss
@@ -51,6 +57,7 @@ def dataset_transform(batch, tokenizer: AutoTokenizer=None):
"labels": labels,
}
+
def train():
parser = HfArgumentParser((ModelArguments, DataArguments, TrainingArguments))
model_args, data_args, training_args = parser.parse_args_into_dataclasses()
@@ -87,11 +94,11 @@ def train():
try:
dataset = load_from_disk(data_args.data_path)
- if 'train' in dataset:
- dataset = dataset['train']
+ if "train" in dataset:
+ dataset = dataset["train"]
except:
dataset = load_dataset(data_args.data_path, split="train")
-
+
dataset.set_transform(partial(dataset_transform, tokenizer=tokenizer))
dataset = dataset.train_test_split(test_size=1000, seed=42)
diff --git a/preparing_data/to_flac.py b/preparing_data/to_flac.py
index 26efa43..b1f4e92 100644
--- a/preparing_data/to_flac.py
+++ b/preparing_data/to_flac.py
@@ -1,8 +1,10 @@
-from pathlib import Path
+import random
import subprocess
from multiprocessing import Pool, cpu_count
+from pathlib import Path
+
from tqdm import tqdm
-import random
+
def convert_to_flac(src_file_path):
dst_file_path = src_file_path.with_suffix(".flac")
@@ -10,7 +12,17 @@ def convert_to_flac(src_file_path):
try:
subprocess.check_call(
- ["ffmpeg", "-y", "-i", str(src_file_path), "-acodec", "flac", "-threads", "0", str(dst_file_path)],
+ [
+ "ffmpeg",
+ "-y",
+ "-i",
+ str(src_file_path),
+ "-acodec",
+ "flac",
+ "-threads",
+ "0",
+ str(dst_file_path),
+ ],
stdout=subprocess.DEVNULL,
stderr=subprocess.DEVNULL,
)
@@ -33,13 +45,15 @@ if __name__ == "__main__":
fail_counter = 0
with Pool(processes=cpu_count(), maxtasksperchild=100) as pool:
- with tqdm(pool.imap_unordered(convert_to_flac, wav_files), total=len(wav_files)) as pbar:
+ with tqdm(
+ pool.imap_unordered(convert_to_flac, wav_files), total=len(wav_files)
+ ) as pbar:
for success in pbar:
if success:
success_counter += 1
else:
fail_counter += 1
-
+
pbar.set_description(f"Success: {success_counter}, Fail: {fail_counter}")
print(f"Successfully converted: {success_counter}")
diff --git a/preparing_data/wenet_clean/clean_wenet_speech.py b/preparing_data/wenet_clean/clean_wenet_speech.py
index 13b3ec6..3d73b42 100644
--- a/preparing_data/wenet_clean/clean_wenet_speech.py
+++ b/preparing_data/wenet_clean/clean_wenet_speech.py
@@ -1,20 +1,20 @@
import json
-from pathlib import Path
+import os
import subprocess
+import tempfile
+import time
+from pathlib import Path
import librosa
import soundfile as sf
import torch
import torchaudio
from fish_audio_preprocess.utils.separate_audio import (
- separate_audio,
- merge_tracks,
init_model,
+ merge_tracks,
+ separate_audio,
)
from tqdm import tqdm
-import time
-import os
-import tempfile
rank = int(os.environ.get("SLURM_PROCID", 0))
world_size = int(os.environ.get("SLURM_NTASKS", 1))
@@ -75,7 +75,9 @@ def main():
)
# Make it 2 channels
audio = torch.cat([audio, audio], dim=0)
- tracks = separate_audio(demucs, audio, shifts=1, num_workers=0, progress=False)
+ tracks = separate_audio(
+ demucs, audio, shifts=1, num_workers=0, progress=False
+ )
audio = merge_tracks(tracks, filter=["vocals"])[0]
vocals, sr = (
torchaudio.functional.resample(