Skip to content

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

Fix prefix concatenation in timestamp post-processing for sensevoice (#2670)

slin00011188b464edb945
diff --git a/funasr/models/sense_voice/model.py b/funasr/models/sense_voice/model.py
index 38943a231..6a291813e 100644
--- a/funasr/models/sense_voice/model.py
+++ b/funasr/models/sense_voice/model.py
@@ -979,9 +979,9 @@ class SenseVoiceSmall(nn.Module):
timestamp_new.append([start, end])
words_new.append(word)
elif prev_word is not None and prev_word.isalpha() and prev_word.isascii() and word.isalpha() and word.isascii():
- prev_word += word
+ word = prev_word + word
timestamp_new[-1][1] = end
- words_new[-1] = prev_word
+ words_new[-1] = word
else:
# timestamp_new[-1][0] += word
timestamp_new.append([start, end])