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
add batch for offline-stream
雾聪445b7ec47cbc
diff --git a/runtime/onnxruntime/include/audio.h b/runtime/onnxruntime/include/audio.hindex 9edd9c946..89a13b649 100644--- a/runtime/onnxruntime/include/audio.h+++ b/runtime/onnxruntime/include/audio.h@@ -83,6 +83,7 @@ class DLLAPI Audio {int FetchTpass(AudioFrame *&frame);int Fetch(float *&dout, int &len, int &flag);int Fetch(float *&dout, int &len, int &flag, float &start_time);+ int Fetch(float **&dout, int *&len, int *&flag, float*& start_time, int batch_size, int &batch_in);void Padding();void Split(OfflineStream* offline_streamj);void CutSplit(OfflineStream* offline_streamj);diff --git a/runtime/onnxruntime/include/funasrruntime.h b/runtime/onnxruntime/include/funasrruntime.hindex ba3cbf47a..cc9ba3d22 100644--- a/runtime/onnxruntime/include/funasrruntime.h+++ b/runtime/onnxruntime/include/funasrruntime.h@@ -96,7 +96,7 @@ _FUNASRAPI void CTTransformerFreeResult(FUNASR_RESULT result);_FUNASRAPI void CTTransformerUninit(FUNASR_HANDLE handle);//OfflineStream-_FUNASRAPI FUNASR_HANDLE FunOfflineInit(std::map<std::string, std::string>& model_path, int thread_num, bool use_gpu=false);+_FUNASRAPI FUNASR_HANDLE FunOfflineInit(std::map<std::string, std::string>& model_path, int thread_num, bool use_gpu=false, int batch_size=1);_FUNASRAPI void FunOfflineReset(FUNASR_HANDLE handle, FUNASR_DEC_HANDLE dec_handle=nullptr);// buffer_FUNASRAPI FUNASR_RESULT FunOfflineInferBuffer(FUNASR_HANDLE handle, const char* sz_buf, int n_len,diff --git a/runtime/onnxruntime/include/model.h b/runtime/onnxruntime/include/model.hindex 570de384b..9fec8fa7d 100644--- a/runtime/onnxruntime/include/model.h+++ b/runtime/onnxruntime/include/model.h@@ -22,12 +22,15 @@ class Model {virtual void InitLm(const std::string &lm_file, const std::string &lm_config, const std::string &lex_file){};virtual void InitFstDecoder(){};virtual std::string Forward(float *din, int len, bool input_finished, const std::vector<std::vector<float>> &hw_emb={{0.0}}, void* wfst_decoder=nullptr){return "";};+ virtual std::vector<std::string> Forward(float** din, int* len, bool input_finished, const std::vector<std::vector<float>> &hw_emb={{0.0}}, void* wfst_decoder=nullptr, int batch_in=1)+ {return std::vector<string>();};virtual std::string Rescoring() = 0;virtual void InitHwCompiler(const std::string &hw_model, int thread_num){};virtual void InitSegDict(const std::string &seg_dict_model){};virtual std::vector<std::vector<float>> CompileHotwordEmbedding(std::string &hotwords){return std::vector<std::vector<float>>();};virtual std::string GetLang(){return "";};virtual int GetAsrSampleRate() = 0;+ virtual int GetBatchSize() {return 0;};virtual Vocab* GetVocab() {return nullptr;};virtual Vocab* GetLmVocab() {return nullptr;};virtual PhoneSet* GetPhoneSet() {return nullptr;};diff --git a/runtime/onnxruntime/include/offline-stream.h b/runtime/onnxruntime/include/offline-stream.hindex 0bec79771..cc0f1c43b 100644--- a/runtime/onnxruntime/include/offline-stream.h+++ b/runtime/onnxruntime/include/offline-stream.h@@ -14,7 +14,7 @@namespace funasr {class OfflineStream {public:- OfflineStream(std::map<std::string, std::string>& model_path, int thread_num, bool use_gpu=false);+ OfflineStream(std::map<std::string, std::string>& model_path, int thread_num, bool use_gpu=false, int batch_size=1);~OfflineStream(){};std::unique_ptr<VadModel> vad_handle= nullptr;@@ -33,6 +33,6 @@ class OfflineStream {bool use_itn=false;};-OfflineStream *CreateOfflineStream(std::map<std::string, std::string>& model_path, int thread_num=1, bool use_gpu=false);+OfflineStream *CreateOfflineStream(std::map<std::string, std::string>& model_path, int thread_num=1, bool use_gpu=false, int batch_size=1);} // namespace funasr#endif