speech_recognition/FunASR
1
0
Fork 0
mirror of https://github.com/modelscope/FunASR synced 2025-09-15 14:48:36 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

refine decoding process, merge flow and vocoder

Browse Source
This commit is contained in:
志浩 2024-07-10 11:49:42 +08:00
parent ffb553eb8e
commit 8335712c66
1 changed files with 5 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
funasr/models/llm_asr/model.py
View File

@ -2374,15 +2374,15 @@ class LLMASR5(nn.Module):
if min_length is not None and i < min_length:
pred[:, self.codebook_size + self.ad_sos_eos] = float(np.finfo(np.float32).min)
top_ids = self.ras_sampling(pred[0], out_tokens[0, :out_token_len, 0])
if torch.any(top_ids == (self.codebook_size + self.ad_sos_eos)):
hit_eos = True
break
out_tokens[0, out_token_len, 0] = top_ids[0]
seq_input[0, prompt_len + out_token_len, :] = self.codec_embedder(top_ids)[0]
out_token_len += 1
if torch.any(out_tokens[:, out_token_len - 1] == (self.codebook_size + self.ad_sos_eos)):
hit_eos = True
out_tokens = out_tokens[:, :out_token_len, :]
break
if decoding_length is None:
return out_tokens
else: