Skip to content

Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.

fix: resolve multiple issues (#2689, #2720, #2755, #2780)

游雁a2f3d5483d5a
diff --git a/README.md b/README.md
index f80a0fe6a..4330e05b7 100644
--- a/README.md
+++ b/README.md
@@ -267,7 +267,7 @@ speech, sample_rate = soundfile.read(wav_file)
chunk_stride = chunk_size[1] * 960 # 600ms
cache = {}
-total_chunk_num = int(len((speech)-1)/chunk_stride+1)
+total_chunk_num = int((len(speech)-1)/chunk_stride+1)
for i in range(total_chunk_num):
speech_chunk = speech[i*chunk_stride:(i+1)*chunk_stride]
is_final = i == total_chunk_num - 1
@@ -307,7 +307,7 @@ speech, sample_rate = soundfile.read(wav_file)
chunk_stride = int(chunk_size * sample_rate / 1000)
cache = {}
-total_chunk_num = int(len((speech)-1)/chunk_stride+1)
+total_chunk_num = int((len(speech)-1)/chunk_stride+1)
for i in range(total_chunk_num):
speech_chunk = speech[i*chunk_stride:(i+1)*chunk_stride]
is_final = i == total_chunk_num - 1
diff --git a/README_zh.md b/README_zh.md
index fdc6a07bf..cc4d662ad 100644
--- a/README_zh.md
+++ b/README_zh.md
@@ -264,7 +264,7 @@ speech, sample_rate = soundfile.read(wav_file)
chunk_stride = chunk_size[1] * 960 # 600ms
cache = {}
-total_chunk_num = int(len((speech)-1)/chunk_stride+1)
+total_chunk_num = int((len(speech)-1)/chunk_stride+1)
for i in range(total_chunk_num):
speech_chunk = speech[i*chunk_stride:(i+1)*chunk_stride]
is_final = i == total_chunk_num - 1
@@ -306,7 +306,7 @@ speech, sample_rate = soundfile.read(wav_file)
chunk_stride = int(chunk_size * sample_rate / 1000)
cache = {}
-total_chunk_num = int(len((speech)-1)/chunk_stride+1)
+total_chunk_num = int((len(speech)-1)/chunk_stride+1)
for i in range(total_chunk_num):
speech_chunk = speech[i*chunk_stride:(i+1)*chunk_stride]
is_final = i == total_chunk_num - 1
diff --git a/funasr/datasets/audio_datasets/index_ds.py b/funasr/datasets/audio_datasets/index_ds.py
index 39ef409af..dd7c52cd4 100644
--- a/funasr/datasets/audio_datasets/index_ds.py
+++ b/funasr/datasets/audio_datasets/index_ds.py
@@ -92,7 +92,8 @@ class IndexDSJsonlRankFull(torch.utils.data.Dataset):
target = data["target"]
source_len = data.get("source_len", 1)
target_len = data.get("target_len", 0)
- if "aishell" in source:
+ text_language = data.get("text_language", "")
+ if "aishell" in source and text_language != "en":
target = target.replace(" ", "")
if (
source_len < self.min_source_length
diff --git a/funasr/models/fsmn_vad_streaming/model.py b/funasr/models/fsmn_vad_streaming/model.py
index 51581ba25..d8dd5ae1e 100644
--- a/funasr/models/fsmn_vad_streaming/model.py
+++ b/funasr/models/fsmn_vad_streaming/model.py
@@ -492,6 +492,8 @@ class FsmnVADStreaming(nn.Module):
def GetFrameState(self, t: int, cache: dict = {}):
frame_state = FrameState.kFrameStateInvalid
+ if t >= len(cache["stats"].decibel):
+ return FrameState.kFrameStateSil
cur_decibel = cache["stats"].decibel[t]
cur_snr = cur_decibel - cache["stats"].noise_average_decibel
# for each frame, calc log posterior probability of each state
diff --git a/funasr/models/sense_voice/model.py b/funasr/models/sense_voice/model.py
index 6a291813e..19427295e 100644
--- a/funasr/models/sense_voice/model.py
+++ b/funasr/models/sense_voice/model.py
@@ -930,7 +930,7 @@ class SenseVoiceSmall(nn.Module):
results.append(result_i)
continue
- logits_speech = self.ctc.softmax(encoder_out)[i, 4 : encoder_out_lens[i].item(), :]
+ logits_speech = self.ctc.log_softmax(encoder_out)[i, 4 : encoder_out_lens[i].item(), :]
pred = logits_speech.argmax(-1).cpu()
logits_speech[pred == self.blank_id, self.blank_id] = 0
align = ctc_forced_align(