deepgeek/CosyVoice
2
0
Fork 0
mirror of https://github.com/FunAudioLLM/CosyVoice.git synced 2026-02-05 18:09:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix white space

Browse Source
This commit is contained in:
root
2025-10-09 14:49:22 +08:00
parent 807bb6ee0b
commit 8811e9f33a
7 changed files with 27 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
runtime/triton_trtllm/model_repo/token2wav_dit/1/token2wav_dit.py
View File

@@ -133,7 +133,6 @@ class CosyVoice2_Token2Wav(torch.nn.Module):
option.intra_op_num_threads = 1 option.intra_op_num_threads = 1
self.spk_model = onnxruntime.InferenceSession(f"{model_dir}/campplus.onnx", sess_options=option, self.spk_model = onnxruntime.InferenceSession(f"{model_dir}/campplus.onnx", sess_options=option,
providers=["CPUExecutionProvider"]) providers=["CPUExecutionProvider"])
self.audio_tokenizer = s3tokenizer.load_model(f"{model_dir}/speech_tokenizer_v2_25hz.onnx").to(self.device).eval() self.audio_tokenizer = s3tokenizer.load_model(f"{model_dir}/speech_tokenizer_v2_25hz.onnx").to(self.device).eval()
gpu="l20" gpu="l20"
@@ -319,13 +318,11 @@ class CosyVoice2_Token2Wav(torch.nn.Module):
): ):
assert all(sample_rate == 16000 for sample_rate in prompt_audios_sample_rate) assert all(sample_rate == 16000 for sample_rate in prompt_audios_sample_rate)
prompt_speech_tokens_list, prompt_mels_for_flow, prompt_mels_lens_for_flow, spk_emb_for_flow = self.prepare_prompt_audio(prompt_audios_list, prompt_audios_sample_rate) prompt_speech_tokens_list, prompt_mels_for_flow, prompt_mels_lens_for_flow, spk_emb_for_flow = self.prepare_prompt_audio(prompt_audios_list, prompt_audios_sample_rate)
generated_mels, generated_mels_lens = self.forward_flow(prompt_speech_tokens_list, generated_speech_tokens_list, prompt_mels_for_flow, prompt_mels_lens_for_flow, spk_emb_for_flow) generated_mels, generated_mels_lens = self.forward_flow(prompt_speech_tokens_list, generated_speech_tokens_list, prompt_mels_for_flow, prompt_mels_lens_for_flow, spk_emb_for_flow)
generated_wavs = self.forward_hift(generated_mels, generated_mels_lens, prompt_mels_lens_for_flow) generated_wavs = self.forward_hift(generated_mels, generated_mels_lens, prompt_mels_lens_for_flow)
return generated_wavs return generated_wavs
def prepare_prompt_audio( def prepare_prompt_audio(
@@ -333,13 +330,11 @@ class CosyVoice2_Token2Wav(torch.nn.Module):
): ):
assert all(sample_rate == 16000 for sample_rate in prompt_audios_sample_rate) assert all(sample_rate == 16000 for sample_rate in prompt_audios_sample_rate)
prompt_speech_tokens_list = self.prompt_audio_tokenization(prompt_audios_list) prompt_speech_tokens_list = self.prompt_audio_tokenization(prompt_audios_list)
prompt_mels_for_flow, prompt_mels_lens_for_flow = self.get_prompt_mels(prompt_audios_list, prompt_audios_sample_rate) prompt_mels_for_flow, prompt_mels_lens_for_flow = self.get_prompt_mels(prompt_audios_list, prompt_audios_sample_rate)
spk_emb_for_flow = self.get_spk_emb(prompt_audios_list) spk_emb_for_flow = self.get_spk_emb(prompt_audios_list)
return prompt_speech_tokens_list, prompt_mels_for_flow, prompt_mels_lens_for_flow, spk_emb_for_flow return prompt_speech_tokens_list, prompt_mels_for_flow, prompt_mels_lens_for_flow, spk_emb_for_flow

1
runtime/triton_trtllm/offline_inference.py
View File

@@ -365,7 +365,6 @@ def main(args):
runner = None runner = None
else: else:
raise ValueError(f"Unsupported backend: {args.backend}") raise ValueError(f"Unsupported backend: {args.backend}")
if 'Step-Audio-2-mini' in args.token2wav_path: if 'Step-Audio-2-mini' in args.token2wav_path:
from token2wav_dit import CosyVoice2_Token2Wav from token2wav_dit import CosyVoice2_Token2Wav
else: else: