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 loading torch models
雾聪2ea00f89b8e9
diff --git a/runtime/onnxruntime/include/com-define.h b/runtime/onnxruntime/include/com-define.hindex 9cb1f2c96..a16fc5062 100644--- a/runtime/onnxruntime/include/com-define.h+++ b/runtime/onnxruntime/include/com-define.h@@ -49,6 +49,9 @@ namespace funasr {// hotword embedding compile model#define MODEL_EB_NAME "model_eb.onnx"#define QUANT_MODEL_NAME "model_quant.onnx"+// gpu models+#define TORCH_MODEL_NAME "model.gpu.torchscripts"+#define TORCH_QUANT_MODEL_NAME "model.blade.fp16.pt"#define VAD_CMVN_NAME "vad.mvn"#define VAD_CONFIG_NAME "vad.yaml"#define AM_CMVN_NAME "am.mvn"diff --git a/runtime/onnxruntime/src/offline-stream.cpp b/runtime/onnxruntime/src/offline-stream.cppindex 9cdcdd2ab..8941a0583 100644--- a/runtime/onnxruntime/src/offline-stream.cpp+++ b/runtime/onnxruntime/src/offline-stream.cpp@@ -59,6 +59,12 @@ OfflineStream::OfflineStream(std::map<std::string, std::string>& model_path, intif(model_path.find(QUANTIZE) != model_path.end() && model_path.at(QUANTIZE) == "true"){am_model_path = PathAppend(model_path.at(MODEL_DIR), QUANT_MODEL_NAME);}+ if(use_gpu){+ am_model_path = PathAppend(model_path.at(MODEL_DIR), TORCH_MODEL_NAME);+ if(model_path.find(QUANTIZE) != model_path.end() && model_path.at(QUANTIZE) == "true"){+ am_model_path = PathAppend(model_path.at(MODEL_DIR), TORCH_QUANT_MODEL_NAME);+ }+ }}am_cmvn_path = PathAppend(model_path.at(MODEL_DIR), AM_CMVN_NAME);am_config_path = PathAppend(model_path.at(MODEL_DIR), AM_CONFIG_NAME);