mirror of
https://github.com/RVC-Boss/GPT-SoVITS.git
synced 2025-05-10 06:39:00 +08:00
53cac93589
* ruff check --fix * ruff format --line-length 120 --target-version py39 * Change the link for G2PW Model * update pytorch version and colab
24 lines
688 B
Python
24 lines
688 B
Python
import torch
|
||
|
||
|
||
def get_model():
|
||
import whisper
|
||
|
||
model = whisper.load_model("small", device="cpu")
|
||
|
||
return model.encoder
|
||
|
||
|
||
def get_content(model=None, wav_16k_tensor=None):
|
||
from whisper import log_mel_spectrogram, pad_or_trim
|
||
|
||
dev = next(model.parameters()).device
|
||
mel = log_mel_spectrogram(wav_16k_tensor).to(dev)[:, :3000]
|
||
# if torch.cuda.is_available():
|
||
# mel = mel.to(torch.float16)
|
||
feature_len = mel.shape[-1] // 2
|
||
assert mel.shape[-1] < 3000, "输入音频过长，只允许输入30以内音频"
|
||
with torch.no_grad():
|
||
feature = model(pad_or_trim(mel, 3000).unsqueeze(0))[:1, :feature_len, :].transpose(1, 2)
|
||
return feature
|