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 libtorch inference

Browse Source
This commit is contained in:
shixian.shi 2023-03-20 19:14:01 +08:00
parent 442bd1edaf
commit a9683afbb2
1 changed files with 13 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
funasr/runtime/python/libtorch/torch_paraformer/paraformer_bin.py
View File

@ -62,26 +62,28 @@ class Paraformer():
am_scores, valid_token_lens = outputs[0], outputs[1]
if len(outputs) == 4:
# for BiCifParaformer Inference
us_alphas, us_cif_peak = outputs[2], outputs[3]
us_alphas, us_peaks = outputs[2], outputs[3]
else:
us_alphas, us_cif_peak = None, None
us_alphas, us_peaks = None, None
except:
#logging.warning(traceback.format_exc())
logging.warning("input wav is silence or noise")
preds = ['']
else:
am_scores, valid_token_lens = am_scores.detach().cpu().numpy(), valid_token_lens.detach().cpu().numpy()
preds = self.decode(am_scores, valid_token_lens)
if us_cif_peak is None:
if us_peaks is None:
for pred in preds:
pred = sentence_postprocess(pred)
asr_res.append({'preds': pred})
else:
for pred, us_cif_peak_ in zip(preds, us_cif_peak):
text, tokens = pred
timestamp, timestamp_total = time_stamp_lfr6_onnx(us_cif_peak_, copy.copy(tokens))
for pred, us_peaks_ in zip(preds, us_peaks):
raw_tokens = pred
timestamp, timestamp_raw = time_stamp_lfr6_onnx(us_peaks_, copy.copy(raw_tokens))
text_proc, timestamp_proc, _ = sentence_postprocess(raw_tokens, timestamp_raw)
# logging.warning(timestamp)
if len(self.plot_timestamp_to):
self.plot_wave_timestamp(waveform_list[0], timestamp_total, self.plot_timestamp_to)
asr_res.append({'preds': text, 'timestamp': timestamp})
self.plot_wave_timestamp(waveform_list[0], timestamp, self.plot_timestamp_to)
asr_res.append({'preds': text_proc, 'timestamp': timestamp_proc, "raw_tokens": raw_tokens})
return asr_res
def plot_wave_timestamp(self, wav, text_timestamp, dest):
@ -182,6 +184,6 @@ class Paraformer():
# Change integer-ids to tokens
token = self.converter.ids2tokens(token_int)
token = token[:valid_token_num-self.pred_bias]
texts = sentence_postprocess(token)
return texts
# texts = sentence_postprocess(token)
return token