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
fix func FunASRWfstDecoderInit
雾聪462355c00213
diff --git a/runtime/onnxruntime/include/model.h b/runtime/onnxruntime/include/model.hindex a575390ff..570de384b 100644--- a/runtime/onnxruntime/include/model.h+++ b/runtime/onnxruntime/include/model.h@@ -31,8 +31,6 @@ class Model {virtual Vocab* GetVocab() {return nullptr;};virtual Vocab* GetLmVocab() {return nullptr;};virtual PhoneSet* GetPhoneSet() {return nullptr;};-- std::shared_ptr<fst::Fst<fst::StdArc>> lm_ = nullptr;};Model *CreateModel(std::map<std::string, std::string>& model_path, int thread_num=1, ASR_TYPE type=ASR_OFFLINE);diff --git a/runtime/onnxruntime/src/funasrruntime.cpp b/runtime/onnxruntime/src/funasrruntime.cppindex b283772ef..5e2aa935e 100644--- a/runtime/onnxruntime/src/funasrruntime.cpp+++ b/runtime/onnxruntime/src/funasrruntime.cpp@@ -767,16 +767,45 @@funasr::WfstDecoder* mm = nullptr;if (asr_type == ASR_OFFLINE) {funasr::OfflineStream* offline_stream = (funasr::OfflineStream*)handle;- funasr::Model* paraformer = offline_stream->asr_handle.get();- if (paraformer->lm_)- mm = new funasr::WfstDecoder(paraformer->lm_.get(),- paraformer->GetPhoneSet(), paraformer->GetLmVocab(), glob_beam, lat_beam, am_scale);+ auto paraformer = dynamic_cast<funasr::Paraformer*>(offline_stream->asr_handle.get());+ if(paraformer !=nullptr){+ if (paraformer->lm_){+ mm = new funasr::WfstDecoder(paraformer->lm_.get(),+ paraformer->GetPhoneSet(), paraformer->GetLmVocab(), glob_beam, lat_beam, am_scale);+ }+ return mm;+ }+ #ifdef USE_GPU+ auto paraformer_torch = dynamic_cast<funasr::ParaformerTorch*>(offline_stream->asr_handle.get());+ if(paraformer_torch !=nullptr){+ if (paraformer_torch->lm_){+ mm = new funasr::WfstDecoder(paraformer_torch->lm_.get(),+ paraformer_torch->GetPhoneSet(), paraformer_torch->GetLmVocab(), glob_beam, lat_beam, am_scale);+ }+ return mm;+ }+ #endif+} else if (asr_type == ASR_TWO_PASS){funasr::TpassStream* tpass_stream = (funasr::TpassStream*)handle;- funasr::Model* paraformer = tpass_stream->asr_handle.get();- if (paraformer->lm_)- mm = new funasr::WfstDecoder(paraformer->lm_.get(),- paraformer->GetPhoneSet(), paraformer->GetLmVocab(), glob_beam, lat_beam, am_scale);+ auto paraformer = dynamic_cast<funasr::Paraformer*>(tpass_stream->asr_handle.get());+ if(paraformer !=nullptr){+ if (paraformer->lm_){+ mm = new funasr::WfstDecoder(paraformer->lm_.get(),+ paraformer->GetPhoneSet(), paraformer->GetLmVocab(), glob_beam, lat_beam, am_scale);+ }+ return mm;+ }+ #ifdef USE_GPU+ auto paraformer_torch = dynamic_cast<funasr::ParaformerTorch*>(offline_stream->asr_handle.get());+ if(paraformer_torch !=nullptr){+ if (paraformer_torch->lm_){+ mm = new funasr::WfstDecoder(paraformer_torch->lm_.get(),+ paraformer_torch->GetPhoneSet(), paraformer_torch->GetLmVocab(), glob_beam, lat_beam, am_scale);+ }+ return mm;+ }+ #endif}return mm;}