deepgeek/CosyVoice
2
0
Fork 0
mirror of https://github.com/FunAudioLLM/CosyVoice.git synced 2026-02-04 09:29:25 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix flake

Browse Source
This commit is contained in:
lyuxiang.lx
2024-09-18 17:42:54 +08:00
parent ff8e63567a
commit f6b5c42823
2 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
tools/extract_embedding.py
View File

@@ -26,9 +26,9 @@ def single_job(utt):
if sample_rate != 16000:
audio = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)(audio)
feat = kaldi.fbank(audio,
num_mel_bins=80,
dither=0,
sample_frequency=16000)
num_mel_bins=80,
dither=0,
sample_frequency=16000)
feat = feat - feat.mean(dim=0, keepdim=True)
embedding = ort_session.run(None, {ort_session.get_inputs()[0].name: feat.unsqueeze(dim=0).cpu().numpy()})[0].flatten().tolist()
return utt, embedding

2
tools/extract_speech_token.py
View File

@@ -33,7 +33,7 @@ def single_job(utt):
else:
feat = whisper.log_mel_spectrogram(audio, n_mels=128)
speech_token = ort_session.run(None, {ort_session.get_inputs()[0].name: feat.detach().cpu().numpy(),
ort_session.get_inputs()[1].name: np.array([feat.shape[2]], dtype=np.int32)})[0].flatten().tolist()
ort_session.get_inputs()[1].name: np.array([feat.shape[2]], dtype=np.int32)})[0].flatten().tolist()
return utt, speech_token