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

Merge pull request #83 from alibaba-damo-academy/dev_lzr

Browse Source 
remove useless vars and fix bug in predictor tail_process_fn
This commit is contained in:
hnluo 2023-02-09 15:26:57 +08:00 committed by GitHub
commit 579b998ded
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 8 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
funasr/bin/asr_inference.py
View File

@ -464,16 +464,6 @@ def inference_modelscope(
return _forward return _forward
def set_parameters(language: str = None,
sample_rate: Union[int, Dict[Any, int]] = None):
if language is not None:
global global_asr_language
global_asr_language = language
if sample_rate is not None:
global global_sample_rate
global_sample_rate = sample_rate
def get_parser(): def get_parser():
parser = config_argparse.ArgumentParser( parser = config_argparse.ArgumentParser(
description="ASR Decoding", description="ASR Decoding",

6
funasr/models/predictor/cif.py
View File

@ -68,7 +68,8 @@ class CifPredictor(nn.Module):
mask_2 = torch.cat([ones_t, mask], dim=1) mask_2 = torch.cat([ones_t, mask], dim=1)
mask = mask_2 - mask_1 mask = mask_2 - mask_1
tail_threshold = mask * tail_threshold tail_threshold = mask * tail_threshold
alphas = torch.cat([alphas, tail_threshold], dim=1) alphas = torch.cat([alphas, zeros_t], dim=1)
alphas = torch.add(alphas, tail_threshold)
else: else:
tail_threshold = torch.tensor([tail_threshold], dtype=alphas.dtype).to(alphas.device) tail_threshold = torch.tensor([tail_threshold], dtype=alphas.dtype).to(alphas.device)
tail_threshold = torch.reshape(tail_threshold, (1, 1)) tail_threshold = torch.reshape(tail_threshold, (1, 1))
@ -597,7 +598,8 @@ class CifPredictorV3(nn.Module):
mask_2 = torch.cat([ones_t, mask], dim=1) mask_2 = torch.cat([ones_t, mask], dim=1)
mask = mask_2 - mask_1 mask = mask_2 - mask_1
tail_threshold = mask * tail_threshold tail_threshold = mask * tail_threshold
alphas = torch.cat([alphas, tail_threshold], dim=1) alphas = torch.cat([alphas, zeros_t], dim=1)
alphas = torch.add(alphas, tail_threshold)
else: else:
tail_threshold = torch.tensor([tail_threshold], dtype=alphas.dtype).to(alphas.device) tail_threshold = torch.tensor([tail_threshold], dtype=alphas.dtype).to(alphas.device)
tail_threshold = torch.reshape(tail_threshold, (1, 1)) tail_threshold = torch.reshape(tail_threshold, (1, 1))

16
funasr/utils/asr_utils.py
View File

@ -186,23 +186,12 @@ def recursion_dir_all_wav(wav_list, dir_path: str) -> List[str]:
return wav_list return wav_list
def set_parameters(language: str = None):
if language is not None:
global global_asr_language
global_asr_language = language
def compute_wer(hyp_list: List[Any], def compute_wer(hyp_list: List[Any],
ref_list: List[Any], ref_list: List[Any],
lang: str = None) -> Dict[str, Any]: lang: str = None) -> Dict[str, Any]:
assert len(hyp_list) > 0, 'hyp list is empty' assert len(hyp_list) > 0, 'hyp list is empty'
assert len(ref_list) > 0, 'ref list is empty' assert len(ref_list) > 0, 'ref list is empty'
if lang is not None:
global global_asr_language
global_asr_language = lang
rst = { rst = {
'Wrd': 0, 'Wrd': 0,
'Corr': 0, 'Corr': 0,
@ -216,12 +205,15 @@ def compute_wer(hyp_list: List[Any],
'wrong_sentences': 0 'wrong_sentences': 0
} }
if lang is None:
lang = global_asr_language
for h_item in hyp_list: for h_item in hyp_list:
for r_item in ref_list: for r_item in ref_list:
if h_item['key'] == r_item['key']: if h_item['key'] == r_item['key']:
out_item = compute_wer_by_line(h_item['value'], out_item = compute_wer_by_line(h_item['value'],
r_item['value'], r_item['value'],
global_asr_language) lang)
rst['Wrd'] += out_item['nwords'] rst['Wrd'] += out_item['nwords']
rst['Corr'] += out_item['cor'] rst['Corr'] += out_item['cor']
rst['wrong_words'] += out_item['wrong'] rst['wrong_words'] += out_item['wrong']