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
bugfix
游雁e42f539eb8ac
diff --git a/examples/industrial_data_pretraining/paraformer/export.py b/examples/industrial_data_pretraining/paraformer/export.pyindex 6334e3b8c..fc635cf63 100644--- a/examples/industrial_data_pretraining/paraformer/export.py+++ b/examples/industrial_data_pretraining/paraformer/export.py@@ -20,10 +20,10 @@ print(res)# # method2, inference from local path# from funasr import AutoModel-+## model = AutoModel(# model="/Users/zhifu/.cache/modelscope/hub/iic/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch"# )-+## res = model.export(type="onnx", quantize=False)# print(res)diff --git a/examples/industrial_data_pretraining/paraformer/export.sh b/examples/industrial_data_pretraining/paraformer/export.shindex 0880829be..cfec097e4 100644--- a/examples/industrial_data_pretraining/paraformer/export.sh+++ b/examples/industrial_data_pretraining/paraformer/export.sh@@ -15,10 +15,10 @@ python -m funasr.bin.export \++quantize=false-# method2, inference from local path-model="/Users/zhifu/.cache/modelscope/hub/iic/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch"--python -m funasr.bin.export \-++model=${model} \-++type="onnx" \-++quantize=false+## method2, inference from local path+#model="/Users/zhifu/.cache/modelscope/hub/iic/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch"+#+#python -m funasr.bin.export \+#++model=${model} \+#++type="onnx" \+#++quantize=falsediff --git a/examples/industrial_data_pretraining/paraformer_streaming/export.py b/examples/industrial_data_pretraining/paraformer_streaming/export.pyindex 588fd0838..aeec7a74b 100644--- a/examples/industrial_data_pretraining/paraformer_streaming/export.py+++ b/examples/industrial_data_pretraining/paraformer_streaming/export.py@@ -17,13 +17,13 @@ res = model.export(type="onnx", quantize=False)print(res)-# method2, inference from local path-from funasr import AutoModel---model = AutoModel(- model="/Users/zhifu/.cache/modelscope/hub/iic/speech_paraformer_asr_nat-zh-cn-16k-common-vocab8404-online"-)--res = model.export(type="onnx", quantize=False)-print(res)+# # method2, inference from local path+# from funasr import AutoModel+#+#+# model = AutoModel(+# model="/Users/zhifu/.cache/modelscope/hub/iic/speech_paraformer_asr_nat-zh-cn-16k-common-vocab8404-online"+# )+#+# res = model.export(type="onnx", quantize=False)+# print(res)diff --git a/examples/industrial_data_pretraining/paraformer_streaming/export.sh b/examples/industrial_data_pretraining/paraformer_streaming/export.shindex cf3e7f4c2..5bc20b136 100644--- a/examples/industrial_data_pretraining/paraformer_streaming/export.sh+++ b/examples/industrial_data_pretraining/paraformer_streaming/export.sh@@ -4,11 +4,8 @@# method1, inference from model hubexport HYDRA_FULL_ERROR=1-model="iic/speech_paraformer_asr_nat-zh-cn-16k-common-vocab8404-online"--python -m funasr.bin.export \++model=${model} \++type="onnx" \@@ -16,12 +13,12 @@ python -m funasr.bin.export \++device="cpu"-# method2, inference from local path-model="/Users/zhifu/.cache/modelscope/hub/iic/speech_paraformer_asr_nat-zh-cn-16k-common-vocab8404-online"--python -m funasr.bin.export \-++model=${model} \-++type="onnx" \-++quantize=false \-++device="cpu" \-++debug=false+## method2, inference from local path+#model="/Users/zhifu/.cache/modelscope/hub/iic/speech_paraformer_asr_nat-zh-cn-16k-common-vocab8404-online"+#+#python -m funasr.bin.export \+#++model=${model} \+#++type="onnx" \+#++quantize=false \+#++device="cpu" \+#++debug=falsediff --git a/examples/industrial_data_pretraining/sense_voice/demo.py b/examples/industrial_data_pretraining/sense_voice/demo.pyindex 3a88643c8..07046d981 100644--- a/examples/industrial_data_pretraining/sense_voice/demo.py+++ b/examples/industrial_data_pretraining/sense_voice/demo.py@@ -7,7 +7,7 @@from funasr import AutoModelfrom funasr.utils.postprocess_utils import rich_transcription_postprocess-model_dir = "iic/SenseVoiceSmall"+model_dir = "/Users/zhifu/Downloads/modelscope_models/SenseVoiceSmall" # "iic/SenseVoiceSmall"model = AutoModel(@@ -19,17 +19,30 @@ model = AutoModel(# enres = model.generate(- input=f"{model.model_path}/example/en.mp3",+ input="/Users/zhifu/Downloads/8_output.wav",cache={},language="auto", # "zn", "en", "yue", "ja", "ko", "nospeech"use_itn=True,batch_size_s=60,merge_vad=True, #- merge_length_s=15,+ merge_length_s=0.1,)text = rich_transcription_postprocess(res[0]["text"])print(text)+# en+res = model.generate(+ input="/Users/zhifu/Downloads/8_output.wav",+ cache={},+ language="auto", # "zn", "en", "yue", "ja", "ko", "nospeech"+ use_itn=True,+ batch_size_s=60,+ merge_vad=False, #+ merge_length_s=15,+)+text = rich_transcription_postprocess(res[0]["text"])+print(text)+raise "exit"# zhres = model.generate(input=f"{model.model_path}/example/zh.mp3",diff --git a/funasr/models/paraformer/export_meta.py b/funasr/models/paraformer/export_meta.pyindex db9385519..d3ace8fd6 100644--- a/funasr/models/paraformer/export_meta.py+++ b/funasr/models/paraformer/export_meta.py@@ -31,7 +31,7 @@ def export_rebuild_model(model, **kwargs):model.export_dynamic_axes = types.MethodType(export_dynamic_axes, model)model.export_name = types.MethodType(export_name, model)- model.export_name = 'model'+ model.export_name = "model"return model@@ -83,4 +83,4 @@ def export_dynamic_axes(self):def export_name(self,):- return "model.onnx"+ return "model"diff --git a/funasr/models/paraformer_streaming/export_meta.py b/funasr/models/paraformer_streaming/export_meta.pyindex 9740ecccf..681568849 100644--- a/funasr/models/paraformer_streaming/export_meta.py+++ b/funasr/models/paraformer_streaming/export_meta.py@@ -8,32 +8,6 @@ import torchfrom funasr.register import tables-def export_rebuild_model(model, **kwargs):- model.device = kwargs.get("device")- is_onnx = kwargs.get("type", "onnx") == "onnx"- encoder_class = tables.encoder_classes.get(kwargs["encoder"] + "Export")- model.encoder = encoder_class(model.encoder, onnx=is_onnx)-- predictor_class = tables.predictor_classes.get(kwargs["predictor"] + "Export")- model.predictor = predictor_class(model.predictor, onnx=is_onnx)-- decoder_class = tables.decoder_classes.get(kwargs["decoder"] + "Export")- model.decoder = decoder_class(model.decoder, onnx=is_onnx)-- from funasr.utils.torch_function import sequence_mask-- model.make_pad_mask = sequence_mask(kwargs["max_seq_len"], flip=False)-- model.forward = types.MethodType(export_forward, model)- model.export_dummy_inputs = types.MethodType(export_dummy_inputs, model)- model.export_input_names = types.MethodType(export_input_names, model)- model.export_output_names = types.MethodType(export_output_names, model)- model.export_dynamic_axes = types.MethodType(export_dynamic_axes, model)- model.export_name = types.MethodType(export_name, model)-- return model--def export_rebuild_model(model, **kwargs):# self.device = kwargs.get("device")is_onnx = kwargs.get("type", "onnx") == "onnx"@@ -64,7 +38,7 @@ def export_rebuild_model(model, **kwargs):encoder_model.export_input_names = types.MethodType(export_encoder_input_names, encoder_model)encoder_model.export_output_names = types.MethodType(export_encoder_output_names, encoder_model)encoder_model.export_dynamic_axes = types.MethodType(export_encoder_dynamic_axes, encoder_model)- encoder_model.export_name = types.MethodType(export_encoder_name, encoder_model)+ encoder_model.export_name = "model" # types.MethodType(export_encoder_name, encoder_model)# decoderdecoder_model.forward = types.MethodType(export_decoder_forward, decoder_model)@@ -72,7 +46,7 @@ def export_rebuild_model(model, **kwargs):decoder_model.export_input_names = types.MethodType(export_decoder_input_names, decoder_model)decoder_model.export_output_names = types.MethodType(export_decoder_output_names, decoder_model)decoder_model.export_dynamic_axes = types.MethodType(export_decoder_dynamic_axes, decoder_model)- decoder_model.export_name = types.MethodType(export_decoder_name, decoder_model)+ decoder_model.export_name = "decoder" # types.MethodType(export_decoder_name, decoder_model)return encoder_model, decoder_model