|
|
import os
|
|
|
import subprocess
|
|
|
import torchaudio
|
|
|
import soundfile
|
|
|
import numpy as np
|
|
|
from glob import glob
|
|
|
from loguru import logger
|
|
|
from huggingface_hub import snapshot_download
|
|
|
|
|
|
from VietTTS.utils.vad import get_speech
|
|
|
|
|
|
import torchaudio
|
|
|
import os
|
|
|
import subprocess
|
|
|
import tempfile
|
|
|
|
|
|
|
|
|
def convert_to_wav(input_filepath: str, target_sr: int) -> str:
|
|
|
"""
|
|
|
Convert an input audio file to WAV format with the desired sample rate using FFmpeg.
|
|
|
|
|
|
Args:
|
|
|
input_filepath (str): Path to the input audio file.
|
|
|
target_sr (int): Target sample rate.
|
|
|
|
|
|
Returns:
|
|
|
str: Path to the converted WAV file.
|
|
|
"""
|
|
|
temp_wav_file = tempfile.NamedTemporaryFile(delete=False, suffix=".wav")
|
|
|
temp_wav_filepath = temp_wav_file.name
|
|
|
temp_wav_file.close()
|
|
|
|
|
|
ffmpeg_command = [
|
|
|
"ffmpeg", "-y",
|
|
|
"-loglevel", "error",
|
|
|
"-i", input_filepath,
|
|
|
"-ar", str(target_sr),
|
|
|
"-ac", "1",
|
|
|
temp_wav_filepath
|
|
|
]
|
|
|
|
|
|
result = subprocess.run(ffmpeg_command, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
|
|
|
if result.returncode != 0:
|
|
|
os.unlink(temp_wav_filepath)
|
|
|
raise RuntimeError(f"FFmpeg conversion failed: {result.stderr.decode()}")
|
|
|
|
|
|
return temp_wav_filepath
|
|
|
|
|
|
|
|
|
def load_wav(filepath: str, target_sr: int):
|
|
|
"""
|
|
|
Load an audio file in any supported format, convert it to WAV, and load as a tensor.
|
|
|
|
|
|
Args:
|
|
|
filepath (str): Path to the audio file in any format.
|
|
|
target_sr (int): Target sample rate.
|
|
|
|
|
|
Returns:
|
|
|
Tensor: Loaded audio tensor resampled to the target sample rate.
|
|
|
"""
|
|
|
|
|
|
if not filepath.lower().endswith(".wav"):
|
|
|
logger.info(f"Converting {filepath} to WAV format")
|
|
|
filepath = convert_to_wav(filepath, target_sr)
|
|
|
|
|
|
|
|
|
speech, sample_rate = torchaudio.load(filepath)
|
|
|
speech = speech.mean(dim=0, keepdim=True)
|
|
|
if sample_rate != target_sr:
|
|
|
assert sample_rate > target_sr, f'WAV sample rate {sample_rate} must be greater than {target_sr}'
|
|
|
speech = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=target_sr)(speech)
|
|
|
|
|
|
return speech
|
|
|
|
|
|
|
|
|
def save_wav(wav: np.ndarray, sr: int, filepath: str):
|
|
|
soundfile.write(filepath, wav, sr)
|
|
|
|
|
|
|
|
|
def load_prompt_speech_from_file(filepath: str, min_duration: float=3, max_duration: float=5, return_numpy: bool=False):
|
|
|
wav = load_wav(filepath, 16000)
|
|
|
|
|
|
if wav.abs().max() > 0.9:
|
|
|
wav = wav / wav.abs().max() * 0.9
|
|
|
|
|
|
wav = get_speech(
|
|
|
audio_input=wav.squeeze(0),
|
|
|
min_duration=min_duration,
|
|
|
max_duration=max_duration,
|
|
|
return_numpy=return_numpy
|
|
|
)
|
|
|
return wav
|
|
|
|
|
|
|
|
|
def load_voices(voice_dir: str):
|
|
|
files = glob(os.path.join(voice_dir, '*.wav')) + glob(os.path.join(voice_dir, '*.mp3'))
|
|
|
voice_name_map = {
|
|
|
os.path.basename(f).split('.')[0]: f
|
|
|
for f in files
|
|
|
}
|
|
|
return voice_name_map
|
|
|
|
|
|
|
|
|
def download_model(save_dir: str):
|
|
|
snapshot_download(repo_id="duyv/viet-tts", local_dir=save_dir) |
