Spaces:
Running
Running
File size: 2,659 Bytes
9791162 072f330 9791162 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 |
import sys,os
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
import numpy as np
import argparse
import torch
from whisper.model import Whisper, ModelDimensions
from whisper.audio import load_audio, pad_or_trim, log_mel_spectrogram
def load_model(path, device) -> Whisper:
checkpoint = torch.load(path, map_location="cpu")
dims = ModelDimensions(**checkpoint["dims"])
# print(dims)
model = Whisper(dims)
del model.decoder
cut = len(model.encoder.blocks) // 4
cut = -1 * cut
del model.encoder.blocks[cut:]
model.load_state_dict(checkpoint["model_state_dict"], strict=False)
model.eval()
if not (device == "cpu"):
model.half()
model.to(device)
# torch.save({
# 'dims': checkpoint["dims"],
# 'model_state_dict': model.state_dict(),
# }, "large-v2.pt")
return model
def pred_ppg(whisper: Whisper, wavPath, ppgPath, device):
audio = load_audio(wavPath)
audln = audio.shape[0]
ppg_a = []
idx_s = 0
while (idx_s + 15 * 16000 < audln):
short = audio[idx_s:idx_s + 15 * 16000]
idx_s = idx_s + 15 * 16000
ppgln = 15 * 16000 // 320
# short = pad_or_trim(short)
mel = log_mel_spectrogram(short).to(device)
if not (device == "cpu"):
mel = mel.half()
with torch.no_grad():
mel = mel + torch.randn_like(mel) * 0.1
ppg = whisper.encoder(mel.unsqueeze(0)).squeeze().data.cpu().float().numpy()
ppg = ppg[:ppgln,] # [length, dim=1024]
ppg_a.extend(ppg)
if (idx_s < audln):
short = audio[idx_s:audln]
ppgln = (audln - idx_s) // 320
# short = pad_or_trim(short)
mel = log_mel_spectrogram(short).to(device)
if not (device == "cpu"):
mel = mel.half()
with torch.no_grad():
mel = mel + torch.randn_like(mel) * 0.1
ppg = whisper.encoder(mel.unsqueeze(0)).squeeze().data.cpu().float().numpy()
ppg = ppg[:ppgln,] # [length, dim=1024]
ppg_a.extend(ppg)
np.save(ppgPath, ppg_a, allow_pickle=False)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("-w", "--wav", help="wav", dest="wav", required=True)
parser.add_argument("-p", "--ppg", help="ppg", dest="ppg", required=True)
args = parser.parse_args()
print(args.wav)
print(args.ppg)
wavPath = args.wav
ppgPath = args.ppg
device = "cuda" if torch.cuda.is_available() else "cpu"
whisper = load_model(os.path.join("/tmp/large-v2.pt"), device)
pred_ppg(whisper, wavPath, ppgPath, device)
|