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
test
语帆eb92e79fb94e
diff --git a/funasr/models/conformer/encoder.py b/funasr/models/conformer/encoder.pyindex 1d252c206..443d30912 100644--- a/funasr/models/conformer/encoder.py+++ b/funasr/models/conformer/encoder.py@@ -47,7 +47,7 @@ from funasr.models.transformer.utils.subsampling import check_short_uttfrom funasr.models.transformer.utils.subsampling import Conv2dSubsamplingPadfrom funasr.models.transformer.utils.subsampling import StreamingConvInputfrom funasr.register import tables-+import pdbclass ConvolutionModule(nn.Module):"""ConvolutionModule in Conformer model.@@ -573,7 +573,7 @@ class ConformerEncoder(nn.Module):xs_pad, masks = self.embed(xs_pad, masks)else:xs_pad = self.embed(xs_pad)-+ pdb.set_trace()intermediate_outs = []if len(self.interctc_layer_idx) == 0:xs_pad, masks = self.encoders(xs_pad, masks)@@ -601,12 +601,12 @@ class ConformerEncoder(nn.Module):xs_pad = (x, pos_emb)else:xs_pad = xs_pad + self.conditioning_layer(ctc_out)-+ pdb.set_trace()if isinstance(xs_pad, tuple):xs_pad = xs_pad[0]if self.normalize_before:xs_pad = self.after_norm(xs_pad)-+ pdb.set_trace()olens = masks.squeeze(1).sum(1)if len(intermediate_outs) > 0:return (xs_pad, intermediate_outs), olens, None