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 padding
雾聪d93002b5b8e0
diff --git a/runtime/onnxruntime/bin/funasr-onnx-offline-rtf.cpp b/runtime/onnxruntime/bin/funasr-onnx-offline-rtf.cppindex 4ff216a97..bb032e605 100644--- a/runtime/onnxruntime/bin/funasr-onnx-offline-rtf.cpp+++ b/runtime/onnxruntime/bin/funasr-onnx-offline-rtf.cpp@@ -144,7 +144,7 @@ int main(int argc, char *argv[])TCLAP::ValueArg<std::int32_t> thread_num("", THREAD_NUM, "multi-thread num for rtf", false, 1, "int32_t");TCLAP::ValueArg<std::string> hotword("", HOTWORD, "the hotword file, one hotword perline, Format: Hotword Weight (could be: 阿里巴巴 20)", false, "", "string");TCLAP::SwitchArg use_gpu("", INFER_GPU, "Whether to use GPU for inference, default is false", false);- TCLAP::ValueArg<std::int32_t> batch_size("", "batch-size", "batch_size for ASR model when using GPU", false, 1, "int32_t");+ TCLAP::ValueArg<std::int32_t> batch_size("", BATCHSIZE, "batch_size for ASR model when using GPU", false, 1, "int32_t");cmd.add(model_dir);cmd.add(quantize);diff --git a/runtime/onnxruntime/bin/funasr-onnx-offline.cpp b/runtime/onnxruntime/bin/funasr-onnx-offline.cppindex c8b0d4b2a..f0234c89e 100644--- a/runtime/onnxruntime/bin/funasr-onnx-offline.cpp+++ b/runtime/onnxruntime/bin/funasr-onnx-offline.cpp@@ -60,7 +60,7 @@ int main(int argc, char** argv)TCLAP::ValueArg<std::int32_t> audio_fs("", AUDIO_FS, "the sample rate of audio", false, 16000, "int32_t");TCLAP::ValueArg<std::string> hotword("", HOTWORD, "the hotword file, one hotword perline, Format: Hotword Weight (could be: 阿里巴巴 20)", false, "", "string");TCLAP::SwitchArg use_gpu("", INFER_GPU, "Whether to use GPU for inference, default is false", false);- TCLAP::ValueArg<std::int32_t> batch_size("", "batch-size", "batch_size for ASR model when using GPU", false, 5, "int32_t");+ TCLAP::ValueArg<std::int32_t> batch_size("", BATCHSIZE, "batch_size for ASR model when using GPU", false, 5, "int32_t");cmd.add(model_dir);cmd.add(quantize);diff --git a/runtime/onnxruntime/include/com-define.h b/runtime/onnxruntime/include/com-define.hindex 7160a0e47..9480b0ba3 100644--- a/runtime/onnxruntime/include/com-define.h+++ b/runtime/onnxruntime/include/com-define.h@@ -51,6 +51,7 @@ namespace funasr {#define QUANT_MODEL_NAME "model_quant.onnx"// gpu models#define INFER_GPU "gpu"+#define BATCHSIZE "batch-size"#define TORCH_MODEL_NAME "model.torchscripts"#define TORCH_QUANT_MODEL_NAME "model_quant.torchscripts"#define BLADE_MODEL_NAME "model.blade.fp16.pt"diff --git a/runtime/onnxruntime/src/paraformer-torch.cpp b/runtime/onnxruntime/src/paraformer-torch.cppindex e603e89a4..9ac0cf790 100644--- a/runtime/onnxruntime/src/paraformer-torch.cpp+++ b/runtime/onnxruntime/src/paraformer-torch.cpp@@ -281,10 +281,10 @@ std::vector<std::string> ParaformerTorch::Forward(float** din, int* len, bool inif(asr_feats.size() != 0){LfrCmvn(asr_feats);}- int32_t num_frames = asr_feats.size() / feature_dim;+ int32_t num_frames = asr_feats.size();paraformer_length.emplace_back(num_frames);- if(max_size < asr_feats.size()){- max_size = asr_feats.size();+ if(max_size < asr_feats.size()*feature_dim){+ max_size = asr_feats.size()*feature_dim;max_frames = num_frames;}diff --git a/runtime/websocket/bin/funasr-wss-server.cpp b/runtime/websocket/bin/funasr-wss-server.cppindex cc4c156a0..f395e593b 100644--- a/runtime/websocket/bin/funasr-wss-server.cpp+++ b/runtime/websocket/bin/funasr-wss-server.cpp@@ -126,7 +126,7 @@ int main(int argc, char* argv[]) {TCLAP::ValueArg<std::int32_t> fst_inc_wts("", FST_INC_WTS,"the fst hotwords incremental bias", false, 20, "int32_t");TCLAP::SwitchArg use_gpu("", INFER_GPU, "Whether to use GPU, default is false", false);- TCLAP::ValueArg<std::int32_t> batch_size("", "batch-size", "batch_size for ASR model when using GPU", false, 5, "int32_t");+ TCLAP::ValueArg<std::int32_t> batch_size("", BATCHSIZE, "batch_size for ASR model when using GPU", false, 5, "int32_t");// add filecmd.add(hotword);