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

修复无法预测nospeech标签的问题 (#1604)

Browse Source 
Co-authored-by: 常材 <gaochangfeng.gcf@alibaba-inc.com>
This commit is contained in:
zhifu gao 2024-04-10 09:55:21 +08:00 committed by GitHub
parent c6574bf4f4
commit 112c8e6eb7
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 13 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
funasr/models/sense_voice/whisper_lib/decoding.py
View File

@ -58,18 +58,20 @@ def detect_language(
# x = torch.tensor([[tokenizer.sot]] * n_audio).to(mel.device) # [n_audio, 1]
if x is None:
x = torch.tensor([tokenizer.encode(initial_prompt, allowed_special="all")] * n_audio).to(mel.device) # [n_audio, 1]
logits = model.logits(x, mel)[:, 0]
logits = model.logits(x[:,:-1], mel)[:, -1]
# collect detected languages; suppress all non-language tokens
mask = torch.ones(logits.shape[-1], dtype=torch.bool)
mask[list(tokenizer.all_language_tokens)] = False
mask[tokenizer.no_speech] = False
logits[:, mask] = -np.inf
language_tokens = logits.argmax(dim=-1)
language_token_probs = logits.softmax(dim=-1).cpu()
language_probs = [
{
c: language_token_probs[i, j].item()
for j, c in zip(tokenizer.all_language_tokens, tokenizer.all_language_codes)
for j, c in zip(list(tokenizer.all_language_tokens) + [tokenizer.no_speech], list(tokenizer.all_language_codes) + ["nospeech"])
}
for i in range(n_audio)
]

9
funasr/models/sense_voice/whisper_lib/tokenizer.py
View File

@ -179,7 +179,12 @@ class Tokenizer:
langs = tuple(LANGUAGES.keys())[: self.num_languages]
sot_sequence = [sot]
if self.language is not None:
sot_sequence.append(sot + 1 + langs.index(self.language))
if self.language == 'nospeech':
sot_sequence.append(self.no_speech)
else:
sot_sequence.append(sot + 1 + langs.index(self.language))
# if self.language is not None:
# sot_sequence.append(sot + 1 + langs.index(self.language))
if self.task is not None:
task_token: int = transcribe if self.task == "transcribe" else translate
sot_sequence.append(task_token)
@ -432,6 +437,8 @@ def get_tokenizer(
if language not in LANGUAGES:
if language in TO_LANGUAGE_CODE:
language = TO_LANGUAGE_CODE[language]
elif language == 'nospeech':
pass
else:
raise ValueError(f"Unsupported language: {language}")