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
update code
维石9377eed41e5e
diff --git a/funasr/datasets/llm_datasets/datasets.py b/funasr/datasets/llm_datasets/datasets.pyindex b66055412..61cadedd1 100644--- a/funasr/datasets/llm_datasets/datasets.py+++ b/funasr/datasets/llm_datasets/datasets.py@@ -64,8 +64,6 @@ class AudioLLMNARDataset(torch.utils.data.Dataset):def __getitem__(self, index):item = self.index_ds[index]- # import pdb;- # pdb.set_trace()source = item["source"]data_src = load_audio_text_image_video(source, fs=self.fs)if self.preprocessor_speech:diff --git a/funasr/datasets/llm_datasets_qwenaudio/datasets.py b/funasr/datasets/llm_datasets_qwenaudio/datasets.pyindex b56e624ac..569665a1d 100644--- a/funasr/datasets/llm_datasets_qwenaudio/datasets.py+++ b/funasr/datasets/llm_datasets_qwenaudio/datasets.py@@ -66,8 +66,6 @@ class AudioLLMQwenAudioDataset(torch.utils.data.Dataset):def __getitem__(self, index):item = self.index_ds[index]- # import pdb;- # pdb.set_trace()source = item["source"]data_src = load_audio_text_image_video(source, fs=self.fs)if self.preprocessor_speech:diff --git a/funasr/datasets/llm_datasets_vicuna/datasets.py b/funasr/datasets/llm_datasets_vicuna/datasets.pyindex 04fa514a1..cde29a94a 100644--- a/funasr/datasets/llm_datasets_vicuna/datasets.py+++ b/funasr/datasets/llm_datasets_vicuna/datasets.py@@ -66,8 +66,6 @@ class AudioLLMVicunaDataset(torch.utils.data.Dataset):def __getitem__(self, index):item = self.index_ds[index]- # import pdb;- # pdb.set_trace()source = item["source"]data_src = load_audio_text_image_video(source, fs=self.fs)if self.preprocessor_speech:diff --git a/funasr/datasets/sense_voice_datasets/datasets.py b/funasr/datasets/sense_voice_datasets/datasets.pyindex 690a1c56c..0117182c2 100644--- a/funasr/datasets/sense_voice_datasets/datasets.py+++ b/funasr/datasets/sense_voice_datasets/datasets.py@@ -71,8 +71,6 @@ class SenseVoiceDataset(torch.utils.data.Dataset):return len(self.index_ds)def __getitem__(self, index):- # import pdb;- # pdb.set_trace()output = Nonefor idx in range(self.retry):diff --git a/funasr/frontends/default.py b/funasr/frontends/default.pyindex 462da8633..68bd6fb92 100644--- a/funasr/frontends/default.py+++ b/funasr/frontends/default.py@@ -235,7 +235,6 @@ class MultiChannelFrontend(nn.Module):self, input: torch.Tensor, input_lengths: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:# 1. Domain-conversion: e.g. Stft: time -> time-freq- # import pdb;pdb.set_trace()if self.stft is not None:input_stft, feats_lens = self._compute_stft(input, input_lengths)else:diff --git a/funasr/models/bicif_paraformer/cif_predictor.py b/funasr/models/bicif_paraformer/cif_predictor.pyindex 3739c76f4..ca98cdc2b 100644--- a/funasr/models/bicif_paraformer/cif_predictor.py+++ b/funasr/models/bicif_paraformer/cif_predictor.py@@ -198,7 +198,7 @@ class CifPredictorV3(torch.nn.Module):output2 = self.upsample_cnn(_output)output2 = output2.transpose(1, 2)output2, _ = self.self_attn(output2, mask)- # import pdb; pdb.set_trace()+alphas2 = torch.sigmoid(self.cif_output2(output2))alphas2 = torch.nn.functional.relu(alphas2 * self.smooth_factor2 - self.noise_threshold2)# repeat the mask in T demension to match the upsampled lengthdiff --git a/funasr/models/contextual_paraformer/decoder.py b/funasr/models/contextual_paraformer/decoder.pyindex 0b30c990d..ba2ce9ade 100644--- a/funasr/models/contextual_paraformer/decoder.py+++ b/funasr/models/contextual_paraformer/decoder.py@@ -424,7 +424,6 @@ class ContextualParaformerDecoderExport(torch.nn.Module):# contextual_mask = myutils.sequence_mask(contextual_length, device=memory.device)[:, None, :]contextual_mask = self.make_pad_mask(contextual_length)contextual_mask, _ = self.prepare_mask(contextual_mask)- # import pdb; pdb.set_trace()contextual_mask = contextual_mask.transpose(2, 1).unsqueeze(1)cx, tgt_mask, _, _, _ = self.bias_decoder(x_self_attn, tgt_mask, bias_embed, memory_mask=contextual_maskdiff --git a/funasr/models/lcbnet/model.py b/funasr/models/lcbnet/model.pyindex d3df25a91..7b2038e6f 100644--- a/funasr/models/lcbnet/model.py+++ b/funasr/models/lcbnet/model.py@@ -23,8 +23,6 @@ from funasr.utils import postprocess_utilsfrom funasr.utils.datadir_writer import DatadirWriterfrom funasr.register import tables-import pdb-@tables.register("model_classes", "LCBNet")class LCBNet(nn.Module):diff --git a/funasr/models/llm_asr/model.py b/funasr/models/llm_asr/model.pyindex 4345f696d..399dfe751 100644--- a/funasr/models/llm_asr/model.py+++ b/funasr/models/llm_asr/model.py@@ -164,8 +164,6 @@ class LLMASR(nn.Module):text: (Batch, Length)text_lengths: (Batch,)"""- # import pdb;- # pdb.set_trace()if len(text_lengths.size()) > 1:text_lengths = text_lengths[:, 0]if len(speech_lengths.size()) > 1:diff --git a/funasr/models/llm_asr_nar/model.py b/funasr/models/llm_asr_nar/model.pyindex 8c0c3ff18..192c1990b 100644--- a/funasr/models/llm_asr_nar/model.py+++ b/funasr/models/llm_asr_nar/model.py@@ -166,8 +166,6 @@ class LLMASRNAR(nn.Module):text: (Batch, Length)text_lengths: (Batch,)"""- # import pdb;- # pdb.set_trace()if len(text_lengths.size()) > 1:text_lengths = text_lengths[:, 0]if len(speech_lengths.size()) > 1:diff --git a/funasr/models/mfcca/mfcca_encoder.py b/funasr/models/mfcca/mfcca_encoder.pyindex 19a6df92a..a0bb58ee0 100644--- a/funasr/models/mfcca/mfcca_encoder.py+++ b/funasr/models/mfcca/mfcca_encoder.py@@ -34,7 +34,6 @@ from funasr.models.transformer.utils.subsampling import Conv2dSubsampling8from funasr.models.transformer.utils.subsampling import TooShortUttErrorfrom funasr.models.transformer.utils.subsampling import check_short_uttfrom funasr.models.encoder.abs_encoder import AbsEncoder-import pdbimport math@@ -363,7 +362,6 @@ class MFCCAEncoder(AbsEncoder):t_leng = xs_pad.size(1)d_dim = xs_pad.size(2)xs_pad = xs_pad.reshape(-1, channel_size, t_leng, d_dim)- # pdb.set_trace()if channel_size < 8:repeat_num = math.ceil(8 / channel_size)xs_pad = xs_pad.repeat(1, repeat_num, 1, 1)[:, 0:8, :, :]diff --git a/funasr/models/paraformer_streaming/model.py b/funasr/models/paraformer_streaming/model.pyindex f28761494..16021ceb6 100644--- a/funasr/models/paraformer_streaming/model.py+++ b/funasr/models/paraformer_streaming/model.py@@ -50,8 +50,6 @@ class ParaformerStreaming(Paraformer):super().__init__(*args, **kwargs)- # import pdb;- # pdb.set_trace()self.sampling_ratio = kwargs.get("sampling_ratio", 0.2)self.scama_mask = None@@ -83,8 +81,6 @@ class ParaformerStreaming(Paraformer):text: (Batch, Length)text_lengths: (Batch,)"""- # import pdb;- # pdb.set_trace()decoding_ind = kwargs.get("decoding_ind")if len(text_lengths.size()) > 1:text_lengths = text_lengths[:, 0]diff --git a/funasr/models/sense_voice/model.py b/funasr/models/sense_voice/model.pyindex 127d5a0a5..f2ea05fdf 100644--- a/funasr/models/sense_voice/model.py+++ b/funasr/models/sense_voice/model.py@@ -73,8 +73,6 @@ class SenseVoice(nn.Module):):target_mask = kwargs.get("target_mask", None)- # import pdb;- # pdb.set_trace()if len(text_lengths.size()) > 1:text_lengths = text_lengths[:, 0]if len(speech_lengths.size()) > 1:@@ -303,8 +301,6 @@ class SenseVoiceRWKV(nn.Module):):target_mask = kwargs.get("target_mask", None)- # import pdb;- # pdb.set_trace()if len(text_lengths.size()) > 1:text_lengths = text_lengths[:, 0]if len(speech_lengths.size()) > 1:@@ -648,8 +644,6 @@ class SenseVoiceFSMN(nn.Module):):target_mask = kwargs.get("target_mask", None)- # import pdb;- # pdb.set_trace()if len(text_lengths.size()) > 1:text_lengths = text_lengths[:, 0]if len(speech_lengths.size()) > 1:@@ -1052,8 +1046,6 @@ class SenseVoiceSANM(nn.Module):):target_mask = kwargs.get("target_mask", None)- # import pdb;- # pdb.set_trace()if len(text_lengths.size()) > 1:text_lengths = text_lengths[:, 0]if len(speech_lengths.size()) > 1:diff --git a/funasr/models/transformer/model.py b/funasr/models/transformer/model.pyindex 0d5ed2376..adfd52520 100644--- a/funasr/models/transformer/model.py+++ b/funasr/models/transformer/model.py@@ -145,8 +145,6 @@ class Transformer(nn.Module):text: (Batch, Length)text_lengths: (Batch,)"""- # import pdb;- # pdb.set_trace()if len(text_lengths.size()) > 1:text_lengths = text_lengths[:, 0]if len(speech_lengths.size()) > 1:diff --git a/funasr/models/whisper/model.py b/funasr/models/whisper/model.pyindex 4710b9cb1..a3321000e 100644--- a/funasr/models/whisper/model.py+++ b/funasr/models/whisper/model.py@@ -114,8 +114,6 @@ class WhisperWarp(nn.Module):result = whisper.decode(self.model, speech, language='english')# result = whisper.transcribe(self.model, speech)-- import pdb; pdb.set_trace()results = []result_i = {"key": key[0], "text": result.text}diff --git a/funasr/models/whisper_lid/model.py b/funasr/models/whisper_lid/model.pyindex 0701f6177..02cd3730a 100644--- a/funasr/models/whisper_lid/model.py+++ b/funasr/models/whisper_lid/model.py@@ -140,8 +140,6 @@ class OpenAIWhisperModel(nn.Module):text: (Batch, Length)text_lengths: (Batch,)"""- # import pdb;- # pdb.set_trace()if len(text_lengths.size()) > 1:text_lengths = text_lengths[:, 0]if len(speech_lengths.size()) > 1:diff --git a/runtime/python/onnxruntime/funasr_onnx/paraformer_bin.py b/runtime/python/onnxruntime/funasr_onnx/paraformer_bin.pyindex 871674eff..1c58d9382 100644--- a/runtime/python/onnxruntime/funasr_onnx/paraformer_bin.py+++ b/runtime/python/onnxruntime/funasr_onnx/paraformer_bin.py@@ -328,7 +328,6 @@ class ContextualParaformer(Paraformer):) -> List:# make hotword listhotwords, hotwords_length = self.proc_hotword(hotwords)- # import pdb; pdb.set_trace()[bias_embed] = self.eb_infer(hotwords, hotwords_length)# index from bias_embedbias_embed = bias_embed.transpose(1, 0, 2)@@ -376,10 +375,10 @@ class ContextualParaformer(Paraformer):return np.array(hotwords)hotword_int = [word_map(i) for i in hotwords]- # import pdb; pdb.set_trace()+hotword_int.append(np.array([1]))hotwords = pad_list(hotword_int, pad_value=0, max_len=10)- # import pdb; pdb.set_trace()+return hotwords, hotwords_lengthdef bb_infer(