Skip to content

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

auto frontend

游雁834a8fd9e2e9
diff --git a/funasr/models/llm_asr/model.py b/funasr/models/llm_asr/model.py
index a6a713481..82ad134ff 100644
--- a/funasr/models/llm_asr/model.py
+++ b/funasr/models/llm_asr/model.py
@@ -488,6 +488,8 @@ class LLMASR2(nn.Module):
fbank_fake_len = fbank_fake_lens[batch_idx].item()
fbank_beg_idx = fbank_beg[batch_idx, 0].item()
min_len = min(fbank_fake_len, inputs_embeds.shape[1] - fbank_beg_idx)
+ fbank_fake_len = encoder_out_lens[batch_idx].item()
+ min_len = min(fbank_fake_len, inputs_embeds.shape[1] - fbank_beg_idx)
try:
inputs_embeds[batch_idx, fbank_beg_idx : fbank_beg_idx + min_len, :] = encoder_out[
batch_idx, :min_len, :