From 1d7bbbffb6a024a33859b48a7a656d0455dc0be1 Mon Sep 17 00:00:00 2001
From: zhifu gao <zhifu.gzf@alibaba-inc.com>
Date: 星期一, 16 十月 2023 11:47:59 +0800
Subject: [PATCH] Update README.md

---
 funasr/runtime/onnxruntime/src/paraformer.cpp |  106 ++++++++++++++++++++++++++++++++++++-----------------
 1 files changed, 72 insertions(+), 34 deletions(-)

diff --git a/funasr/runtime/onnxruntime/src/paraformer.cpp b/funasr/runtime/onnxruntime/src/paraformer.cpp
index cbaab2d..5bbaeef 100644
--- a/funasr/runtime/onnxruntime/src/paraformer.cpp
+++ b/funasr/runtime/onnxruntime/src/paraformer.cpp
@@ -37,11 +37,11 @@
     session_options_.DisableCpuMemArena();
 
     try {
-        m_session_ = std::make_unique<Ort::Session>(env_, am_model.c_str(), session_options_);
+        m_session_ = std::make_unique<Ort::Session>(env_, ORTSTRING(am_model).c_str(), session_options_);
         LOG(INFO) << "Successfully load model from " << am_model;
     } catch (std::exception const &e) {
         LOG(ERROR) << "Error when load am onnx model: " << e.what();
-        exit(0);
+        exit(-1);
     }
 
     string strName;
@@ -65,6 +65,7 @@
     for (auto& item : m_strOutputNames)
         m_szOutputNames.push_back(item.c_str());
     vocab = new Vocab(am_config.c_str());
+    LoadConfigFromYaml(am_config.c_str());
     LoadCmvn(am_cmvn.c_str());
 }
 
@@ -89,19 +90,19 @@
     session_options_.DisableCpuMemArena();
 
     try {
-        encoder_session_ = std::make_unique<Ort::Session>(env_, en_model.c_str(), session_options_);
+        encoder_session_ = std::make_unique<Ort::Session>(env_, ORTSTRING(en_model).c_str(), session_options_);
         LOG(INFO) << "Successfully load model from " << en_model;
     } catch (std::exception const &e) {
         LOG(ERROR) << "Error when load am encoder model: " << e.what();
-        exit(0);
+        exit(-1);
     }
 
     try {
-        decoder_session_ = std::make_unique<Ort::Session>(env_, de_model.c_str(), session_options_);
+        decoder_session_ = std::make_unique<Ort::Session>(env_, ORTSTRING(de_model).c_str(), session_options_);
         LOG(INFO) << "Successfully load model from " << de_model;
     } catch (std::exception const &e) {
         LOG(ERROR) << "Error when load am decoder model: " << e.what();
-        exit(0);
+        exit(-1);
     }
 
     // encoder
@@ -152,11 +153,11 @@
 
     // offline
     try {
-        m_session_ = std::make_unique<Ort::Session>(env_, am_model.c_str(), session_options_);
+        m_session_ = std::make_unique<Ort::Session>(env_, ORTSTRING(am_model).c_str(), session_options_);
         LOG(INFO) << "Successfully load model from " << am_model;
     } catch (std::exception const &e) {
         LOG(ERROR) << "Error when load am onnx model: " << e.what();
-        exit(0);
+        exit(-1);
     }
 
     string strName;
@@ -164,16 +165,44 @@
     m_strInputNames.push_back(strName.c_str());
     GetInputName(m_session_.get(), strName,1);
     m_strInputNames.push_back(strName);
+
+    if (use_hotword) {
+        GetInputName(m_session_.get(), strName, 2);
+        m_strInputNames.push_back(strName);
+    }
     
-    GetOutputName(m_session_.get(), strName);
-    m_strOutputNames.push_back(strName);
-    GetOutputName(m_session_.get(), strName,1);
-    m_strOutputNames.push_back(strName);
+    // support time stamp
+    size_t numOutputNodes = m_session_->GetOutputCount();
+    for(int index=0; index<numOutputNodes; index++){
+        GetOutputName(m_session_.get(), strName, index);
+        m_strOutputNames.push_back(strName);
+    }
 
     for (auto& item : m_strInputNames)
         m_szInputNames.push_back(item.c_str());
     for (auto& item : m_strOutputNames)
         m_szOutputNames.push_back(item.c_str());
+}
+
+void Paraformer::LoadConfigFromYaml(const char* filename){
+
+    YAML::Node config;
+    try{
+        config = YAML::LoadFile(filename);
+    }catch(exception const &e){
+        LOG(ERROR) << "Error loading file, yaml file error or not exist.";
+        exit(-1);
+    }
+
+    try{
+        YAML::Node lang_conf = config["lang"];
+        if (lang_conf.IsDefined()){
+            language = lang_conf.as<string>();
+        }
+    }catch(exception const &e){
+        LOG(ERROR) << "Error when load argument from vad config YAML.";
+        exit(-1);
+    }
 }
 
 void Paraformer::LoadOnlineConfigFromYaml(const char* filename){
@@ -221,11 +250,11 @@
     hw_session_options.DisableCpuMemArena();
 
     try {
-        hw_m_session = std::make_unique<Ort::Session>(hw_env_, hw_model.c_str(), hw_session_options);
+        hw_m_session = std::make_unique<Ort::Session>(hw_env_, ORTSTRING(hw_model).c_str(), hw_session_options);
         LOG(INFO) << "Successfully load model from " << hw_model;
     } catch (std::exception const &e) {
         LOG(ERROR) << "Error when load hw compiler onnx model: " << e.what();
-        exit(0);
+        exit(-1);
     }
 
     string strName;
@@ -292,7 +321,7 @@
     ifstream cmvn_stream(filename);
     if (!cmvn_stream.is_open()) {
         LOG(ERROR) << "Failed to open file: " << filename;
-        exit(0);
+        exit(-1);
     }
     string line;
 
@@ -335,7 +364,7 @@
         hyps.push_back(max_idx);
     }
     if(!is_stamp){
-        return vocab->Vector2StringV2(hyps);
+        return vocab->Vector2StringV2(hyps, language);
     }else{
         std::vector<string> char_list;
         std::vector<std::vector<float>> timestamp_list;
@@ -438,11 +467,11 @@
         }
     }
     string stamp_str="";
-    for (i=0; i<timestamp_list.size(); i++) {
-        stamp_str += std::to_string(timestamp_list[i][0]);
+    for (i=0; i<timestamp_merge.size(); i++) {
+        stamp_str += std::to_string(timestamp_merge[i][0]);
         stamp_str += ", ";
-        stamp_str += std::to_string(timestamp_list[i][1]);
-        if(i!=timestamp_list.size()-1){
+        stamp_str += std::to_string(timestamp_merge[i][1]);
+        if(i!=timestamp_merge.size()-1){
             stamp_str += ",";
         }
     }
@@ -475,7 +504,9 @@
     if (char_list.back() == "</s>") {
         char_list.pop_back();
     }
-
+    if (char_list.empty()) {
+        return ;
+    }
     vector<vector<float>> timestamp_list;
     vector<string> new_char_list;
     vector<float> fire_place;
@@ -490,6 +521,9 @@
     if(num_peak != (int)char_list.size() + 1){
         float sum = std::accumulate(us_alphas.begin(), us_alphas.end(), 0.0f);
         float scale = sum/((int)char_list.size() + 1);
+        if(scale == 0){
+            return;
+        }
         cif_peak.clear();
         sum = 0.0;
         for(auto &alpha:us_alphas){
@@ -507,6 +541,11 @@
                 fire_place.push_back(i + total_offset);
             }
         }
+    }
+    
+    num_peak = fire_place.size();
+    if(fire_place.size() == 0){
+        return;
     }
 
     // begin silence
@@ -530,6 +569,10 @@
     }
 
     // tail token and end silence
+    if(timestamp_list.size()==0){
+        LOG(ERROR)<<"timestamp_list's size is 0!";
+        return;
+    }
     if (num_frames - fire_place.back() > START_END_THRESHOLD) {
         float _end = (num_frames + fire_place.back()) / 2.0;
         timestamp_list.back()[1] = _end * TIME_RATE;
@@ -641,7 +684,7 @@
                 return "";
             }
             //PrintMat(hw_emb, "input_clas_emb");
-            const int64_t hotword_shape[3] = {1, hw_emb.size(), hw_emb[0].size()};
+            const int64_t hotword_shape[3] = {1, static_cast<int64_t>(hw_emb.size()), static_cast<int64_t>(hw_emb[0].size())};
             embedding.reserve(hw_emb.size() * hw_emb[0].size());
             for (auto item : hw_emb) {
                 embedding.insert(embedding.end(), item.begin(), item.end());
@@ -658,7 +701,7 @@
         return "";
     }
 
-    string result;
+    string result="";
     try {
         auto outputTensor = m_session_->Run(Ort::RunOptions{nullptr}, m_szInputNames.data(), input_onnx.data(), input_onnx.size(), m_szOutputNames.data(), m_szOutputNames.size());
         std::vector<int64_t> outputShape = outputTensor[0].GetTensorTypeAndShapeInfo().GetShape();
@@ -686,17 +729,6 @@
         }else{
             result = GreedySearch(floatData, *encoder_out_lens, outputShape[2]);
         }
-//         int pos = 0;
-//         std::vector<std::vector<float>> logits;
-//         for (int j = 0; j < outputShape[1]; j++)
-//         {
-//             std::vector<float> vec_token;
-//             vec_token.insert(vec_token.begin(), floatData + pos, floatData + pos + outputShape[2]);
-//             logits.push_back(vec_token);
-//             pos += outputShape[2];
-//         }
-//         //PrintMat(logits, "logits_out");
-//         result = GreedySearch(floatData, *encoder_out_lens, outputShape[2]);
     }
     catch (std::exception const &e)
     {
@@ -719,6 +751,7 @@
     std::vector<int32_t> hotword_matrix;
     std::vector<int32_t> lengths;
     int hotword_size = 1;
+    int real_hw_size = 0;
     if (!hotwords.empty()) {
       std::vector<std::string> hotword_array = split(hotwords, ' ');
       hotword_size = hotword_array.size() + 1;
@@ -735,6 +768,9 @@
             chars.insert(chars.end(), tokens.begin(), tokens.end());
           }
         }
+        if(chars.size()==0){
+            continue;
+        }
         std::vector<int32_t> hw_vector(max_hotword_len, 0);
         int vector_len = std::min(max_hotword_len, (int)chars.size());
         for (int i=0; i<chars.size(); i++) {
@@ -743,8 +779,10 @@
         }
         std::cout << std::endl;
         lengths.push_back(vector_len);
+        real_hw_size += 1;
         hotword_matrix.insert(hotword_matrix.end(), hw_vector.begin(), hw_vector.end());
       }
+      hotword_size = real_hw_size + 1;
     }
     std::vector<int32_t> blank_vec(max_hotword_len, 0);
     blank_vec[0] = 1;

--
Gitblit v1.9.1