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

update raw_text related funcs

Browse Source
This commit is contained in:
shixian.shi 2024-02-21 15:07:19 +08:00
parent 8ea5f1302d
commit 509bc88903
4 changed files with 40 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
examples/industrial_data_pretraining/seaco_paraformer/demo.py
View File

@ -11,15 +11,16 @@ model = AutoModel(model="iic/speech_seaco_paraformer_large_asr_nat-zh-cn-16k-com
vad_model_revision="v2.0.4", vad_model_revision="v2.0.4",
punc_model="damo/punc_ct-transformer_zh-cn-common-vocab272727-pytorch", punc_model="damo/punc_ct-transformer_zh-cn-common-vocab272727-pytorch",
punc_model_revision="v2.0.4", punc_model_revision="v2.0.4",
spk_model="damo/speech_campplus_sv_zh-cn_16k-common", # spk_model="damo/speech_campplus_sv_zh-cn_16k-common",
spk_model_revision="v2.0.2", # spk_model_revision="v2.0.2",
) )
# example1 # example1
res = model.generate(input="https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav", res = model.generate(input="https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav",
hotword='达摩院 魔搭', hotword='达摩院 魔搭',
# preset_spk_num=2, # return_raw_text=True, # return raw text recognition results splited by space of equal length with timestamp
# preset_spk_num=2, # preset speaker num for speaker cluster model
# sentence_timestamp=True, # return sentence level information when spk_model is not given # sentence_timestamp=True, # return sentence level information when spk_model is not given
) )
print(res) print(res)

32
funasr/auto/auto_model.py
View File

@ -380,11 +380,13 @@ class AutoModel:
else: else:
result[k] += restored_data[j][k] result[k] += restored_data[j][k]
return_raw_text = kwargs.get('return_raw_text', False)
# step.3 compute punc model # step.3 compute punc model
if self.punc_model is not None: if self.punc_model is not None:
self.punc_kwargs.update(cfg) self.punc_kwargs.update(cfg)
punc_res = self.inference(result["text"], model=self.punc_model, kwargs=self.punc_kwargs, disable_pbar=True, **cfg) punc_res = self.inference(result["text"], model=self.punc_model, kwargs=self.punc_kwargs, disable_pbar=True, **cfg)
raw_text = copy.copy(result["text"]) raw_text = copy.copy(result["text"])
if return_raw_text: result['raw_text'] = raw_text
result["text"] = punc_res[0]["text"] result["text"] = punc_res[0]["text"]
else: else:
raw_text = None raw_text = None
@ -403,26 +405,28 @@ class AutoModel:
for res, vadsegment in zip(restored_data, vadsegments): for res, vadsegment in zip(restored_data, vadsegments):
if 'timestamp' not in res: if 'timestamp' not in res:
logging.error("Only 'iic/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch' \ logging.error("Only 'iic/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch' \
and 'iic/speech_seaco_paraformer_large_asr_nat-zh-cn-16k-common-vocab8404-pytorch'\ and 'iic/speech_seaco_paraformer_large_asr_nat-zh-cn-16k-common-vocab8404-pytorch'\
can predict timestamp, and speaker diarization relies on timestamps.") can predict timestamp, and speaker diarization relies on timestamps.")
sentence_list.append({"start": vadsegment[0],\ sentence_list.append({"start": vadsegment[0],
"end": vadsegment[1], "end": vadsegment[1],
"sentence": res['text'], "sentence": res['text'],
"timestamp": res['timestamp']}) "timestamp": res['timestamp']})
elif self.spk_mode == 'punc_segment': elif self.spk_mode == 'punc_segment':
if 'timestamp' not in result: if 'timestamp' not in result:
logging.error("Only 'iic/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch' \ logging.error("Only 'iic/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch' \
and 'iic/speech_seaco_paraformer_large_asr_nat-zh-cn-16k-common-vocab8404-pytorch'\ and 'iic/speech_seaco_paraformer_large_asr_nat-zh-cn-16k-common-vocab8404-pytorch'\
can predict timestamp, and speaker diarization relies on timestamps.") can predict timestamp, and speaker diarization relies on timestamps.")
sentence_list = timestamp_sentence(punc_res[0]['punc_array'], \ sentence_list = timestamp_sentence(punc_res[0]['punc_array'],
result['timestamp'], \ result['timestamp'],
raw_text) raw_text,
return_raw_text=return_raw_text)
distribute_spk(sentence_list, sv_output) distribute_spk(sentence_list, sv_output)
result['sentence_info'] = sentence_list result['sentence_info'] = sentence_list
elif kwargs.get("sentence_timestamp", False): elif kwargs.get("sentence_timestamp", False):
sentence_list = timestamp_sentence(punc_res[0]['punc_array'], \ sentence_list = timestamp_sentence(punc_res[0]['punc_array'],
result['timestamp'], \ result['timestamp'],
raw_text) raw_text,
return_raw_text=return_raw_text)
result['sentence_info'] = sentence_list result['sentence_info'] = sentence_list
if "spk_embedding" in result: del result['spk_embedding'] if "spk_embedding" in result: del result['spk_embedding']

1
funasr/models/paraformer/model.py
View File

@ -537,7 +537,6 @@ class Paraformer(torch.nn.Module):
result_i = {"key": key[i], "text": text_postprocessed} result_i = {"key": key[i], "text": text_postprocessed}
if ibest_writer is not None: if ibest_writer is not None:
ibest_writer["token"][key[i]] = " ".join(token) ibest_writer["token"][key[i]] = " ".join(token)
# ibest_writer["text"][key[i]] = text # ibest_writer["text"][key[i]] = text

25
funasr/utils/timestamp_tools.py
View File

@ -98,7 +98,7 @@ def ts_prediction_lfr6_standard(us_alphas,
return res_txt, res return res_txt, res
def timestamp_sentence(punc_id_list, timestamp_postprocessed, text_postprocessed): def timestamp_sentence(punc_id_list, timestamp_postprocessed, text_postprocessed, return_raw_text=False):
punc_list = ['，', '。', '？', '、'] punc_list = ['，', '。', '？', '、']
res = [] res = []
if text_postprocessed is None: if text_postprocessed is None:
@ -142,15 +142,24 @@ def timestamp_sentence(punc_id_list, timestamp_postprocessed, text_postprocessed
punc_id = int(punc_id) if punc_id is not None else 1 punc_id = int(punc_id) if punc_id is not None else 1
sentence_end = timestamp[1] if timestamp is not None else sentence_end sentence_end = timestamp[1] if timestamp is not None else sentence_end
sentence_text_seg = sentence_text_seg[:-1] if sentence_text_seg[-1] == ' ' else sentence_text_seg
if punc_id > 1: if punc_id > 1:
sentence_text += punc_list[punc_id - 2] sentence_text += punc_list[punc_id - 2]
res.append({ if return_raw_text:
'text': sentence_text, res.append({
"start": sentence_start, 'text': sentence_text,
"end": sentence_end, "start": sentence_start,
"timestamp": ts_list "end": sentence_end,
}) "timestamp": ts_list,
'raw_text': sentence_text_seg,
})
else:
res.append({
'text': sentence_text,
"start": sentence_start,
"end": sentence_end,
"timestamp": ts_list,
})
sentence_text = '' sentence_text = ''
sentence_text_seg = '' sentence_text_seg = ''
ts_list = [] ts_list = []