Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.
Python asraudiochineseemotion-recognitionfunasrmcp-servermultilingual-asropenai-compatible-apiparaformerpunctuationpytorchreal-time-asrspeaker-diarizationspeech-recognitionspeech-to-textstreaming-asrtranscriptionvllmvoice-activity-detectionwhisper-alternative
auto frontend
游雁8d7f76af46cf
diff --git a/funasr/models/llm_asr/model.py b/funasr/models/llm_asr/model.pyindex 66a69eb2c..b13912346 100644--- a/funasr/models/llm_asr/model.py+++ b/funasr/models/llm_asr/model.py@@ -480,10 +480,15 @@ class LLMASR2(nn.Module):inputs_embeds = self.llm.model.get_input_embeddings()(input_ids)batch_size, token_num, dims = inputs_embeds.shape- _, l, _ = encoder_out.shape+ fbank_mask[fbank_mask < 0] = 0+ fbank_fake_lens = fbank_mask.sum(-1)+ # _, l, _ = encoder_out.shapefor batch_idx in range(batch_size):++ l = fbank_fake_lens[batch_idx].item()fbank_beg_idx = fbank_beg[batch_idx, 0].item()min_len = min(l, inputs_embeds.shape[1] - fbank_beg_idx)+inputs_embeds[batch_idx, fbank_beg_idx : fbank_beg_idx + min_len, :] = encoder_out[batch_idx, :min_len, :]