提交

kongdeqiang

2026-03-13 28ccfbfc51068a663a80764e14074df5edf2b5ba

 runtime/onnxruntime/src/fsmn-vad.cpp

@@ -60,54 +60,9 @@
        LOG(ERROR) << "Error when load vad onnx model: " << e.what();
        exit(-1);
    }
    GetInputOutputInfo(vad_session_, &vad_in_names_, &vad_out_names_, &vad_allocator);
    GetInputNames(vad_session_.get(), m_strInputNames, vad_in_names_);
    GetOutputNames(vad_session_.get(), m_strOutputNames, vad_out_names_);
}

void FsmnVad::GetInputOutputInfo(
        const std::shared_ptr<Ort::Session> &session,
        std::vector<const char *> *in_names, std::vector<const char *> *out_names,
        Ort::AllocatorWithDefaultOptions *allocator) {
    Ort::AllocatorWithDefaultOptions allocator;
    // Input info
    int num_nodes = session->GetInputCount();
    in_names->resize(num_nodes);
    for (int i = 0; i < num_nodes; ++i) {
        std::unique_ptr<char, Ort::detail::AllocatedFree> name = session->GetInputNameAllocated(i, *allocator);
        Ort::TypeInfo type_info = session->GetInputTypeInfo(i);
        auto tensor_info = type_info.GetTensorTypeAndShapeInfo();
        ONNXTensorElementDataType type = tensor_info.GetElementType();
        std::vector<int64_t> node_dims = tensor_info.GetShape();
        std::stringstream shape;
        for (auto j: node_dims) {
            shape << j;
            shape << " ";
        }
        // LOG(INFO) << "\tInput " << i << " : name=" << name.get() << " type=" << type
        //           << " dims=" << shape.str();
        (*in_names)[i] = name.get();
        name.release();
    }
    // Output info
    num_nodes = session->GetOutputCount();
    out_names->resize(num_nodes);
    for (int i = 0; i < num_nodes; ++i) {
        std::unique_ptr<char, Ort::detail::AllocatedFree> name = session->GetOutputNameAllocated(i, *allocator);
        Ort::TypeInfo type_info = session->GetOutputTypeInfo(i);
        auto tensor_info = type_info.GetTensorTypeAndShapeInfo();
        ONNXTensorElementDataType type = tensor_info.GetElementType();
        std::vector<int64_t> node_dims = tensor_info.GetShape();
        std::stringstream shape;
        for (auto j: node_dims) {
            shape << j;
            shape << " ";
        }
        // LOG(INFO) << "\tOutput " << i << " : name=" << name.get() << " type=" << type
        //           << " dims=" << shape.str();
        (*out_names)[i] = name.get();
        name.release();
    }
}


void FsmnVad::Forward(
        const std::vector<std::vector<float>> &chunk_feats,
@@ -311,8 +266,6 @@
}

FsmnVad::~FsmnVad() {
    for (auto vad_in_name_item : vad_in_names_) vad_allocator.Free((void*)vad_in_name_item);
    for (auto vad_out_name_item : vad_out_names_) vad_allocator.Free((void*)vad_out_name_item);
}

FsmnVad::FsmnVad():env_(ORT_LOGGING_LEVEL_ERROR, ""),session_options_{} {