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

rnnt bug fix

Browse Source
This commit is contained in:
aky15 2023-04-23 19:27:17 +08:00
parent 931cac99c1
commit 490531ed51
3 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
funasr/bin/asr_inference_rnnt.py
View File

@ -174,7 +174,7 @@ class Speech2Text:
self.streaming = streaming self.streaming = streaming
self.simu_streaming = simu_streaming self.simu_streaming = simu_streaming
self.chunk_size = max(chunk_size, 0) self.chunk_size = max(chunk_size, 0)
self.left_context = max(left_context, 0) self.left_context = left_context
self.right_context = max(right_context, 0) self.right_context = max(right_context, 0)
if not streaming or chunk_size == 0: if not streaming or chunk_size == 0:

4
funasr/models/e2e_asr_transducer.py
View File

@ -531,8 +531,8 @@ class UnifiedTransducerModel(AbsESPnetModel):
sym_blank: str = "<blank>", sym_blank: str = "<blank>",
report_cer: bool = True, report_cer: bool = True,
report_wer: bool = True, report_wer: bool = True,
sym_sos: str = "<sos/eos>", sym_sos: str = "<s>",
sym_eos: str = "<sos/eos>", sym_eos: str = "</s>",
extract_feats_in_collect_stats: bool = True, extract_feats_in_collect_stats: bool = True,
lsm_weight: float = 0.0, lsm_weight: float = 0.0,
length_normalized_loss: bool = False, length_normalized_loss: bool = False,

2
funasr/modules/nets_utils.py
View File

@ -595,7 +595,7 @@ def make_chunk_mask(
mask = torch.zeros(size, size, device=device, dtype=torch.bool) mask = torch.zeros(size, size, device=device, dtype=torch.bool)
for i in range(size): for i in range(size):
if left_chunk_size <= 0: if left_chunk_size < 0:
start = 0 start = 0
else: else:
start = max((i // chunk_size - left_chunk_size) * chunk_size, 0) start = max((i // chunk_size - left_chunk_size) * chunk_size, 0)