雾聪
2024-01-15 37f078ef5f9f6b2d01a8fa0b15c460190b906c87
runtime/websocket/bin/websocket-server.cpp
@@ -68,24 +68,28 @@
                                 int audio_fs,
                                 std::string wav_format,
                                 FUNASR_DEC_HANDLE& decoder_handle) {
  scoped_lock guard(thread_lock);
  try {
    int num_samples = buffer.size();  // the size of the buf
    if (!buffer.empty() && hotwords_embedding.size() > 0) {
      std::string asr_result;
      std::string stamp_res;
      std::string asr_result="";
      std::string stamp_res="";
      std::string stamp_sents="";
      try{
        FUNASR_RESULT Result = FunOfflineInferBuffer(
            asr_handle, buffer.data(), buffer.size(), RASR_NONE, NULL,
            asr_handle, buffer.data(), buffer.size(), RASR_NONE, nullptr,
            hotwords_embedding, audio_fs, wav_format, itn, decoder_handle);
        asr_result = ((FUNASR_RECOG_RESULT*)Result)->msg;  // get decode result
        stamp_res = ((FUNASR_RECOG_RESULT*)Result)->stamp;
        FunASRFreeResult(Result);
        if (Result != nullptr){
          asr_result = FunASRGetResult(Result, 0);  // get decode result
          stamp_res = FunASRGetStamp(Result);
          stamp_sents = FunASRGetStampSents(Result);
          FunASRFreeResult(Result);
        } else{
          std::this_thread::sleep_for(std::chrono::milliseconds(20));
          LOG(ERROR) << "FUNASR_RESULT is nullptr.";
        }
      }catch (std::exception const& e) {
        LOG(ERROR) << e.what();
        return;
      }
      websocketpp::lib::error_code ec;
@@ -95,6 +99,16 @@
       jsonresult["is_final"] = false;
      if(stamp_res != ""){
        jsonresult["timestamp"] = stamp_res;
      }
      if(stamp_sents != ""){
        try{
          nlohmann::json json_stamp = nlohmann::json::parse(stamp_sents);
          jsonresult["stamp_sents"] = json_stamp;
        }catch (std::exception const &e)
        {
          LOG(ERROR)<<e.what();
          jsonresult["stamp_sents"] = "";
        }
      }
      jsonresult["wav_name"] = wav_name;
@@ -130,6 +144,7 @@
  } catch (std::exception const& e) {
    std::cerr << "Error: " << e.what() << std::endl;
  }
  scoped_lock guard(thread_lock);
  msg["access_num"]=(int)msg["access_num"]-1;
}
@@ -144,8 +159,9 @@
  data_msg->msg["wav_format"] = "pcm";
  data_msg->msg["wav_name"] = "wav-default-id";
  data_msg->msg["itn"] = true;
  data_msg->msg["audio_fs"] = 16000;
  data_msg->msg["audio_fs"] = 16000; // default is 16k
  data_msg->msg["access_num"] = 0; // the number of access for this object, when it is 0, we can free it saftly
  data_msg->msg["is_eof"]=false;
  FUNASR_DEC_HANDLE decoder_handle =
    FunASRWfstDecoderInit(asr_handle, ASR_OFFLINE, global_beam_, lattice_beam_, am_scale_);
  data_msg->decoder_handle = decoder_handle;
@@ -226,7 +242,7 @@
        data_msg->msg["is_eof"]=true;
        guard_decoder.unlock();
        to_remove.push_back(hdl);
        LOG(INFO)<<"connection is closed: "<<e.what();
        LOG(INFO)<<"connection is closed.";
        
      }
      iter++;
@@ -250,6 +266,10 @@
  auto it_data = data_map.find(hdl);
  if (it_data != data_map.end()) {
    msg_data = it_data->second;
    if(msg_data->msg["is_eof"]){
      lock.unlock();
      return;
    }
  } else{
    lock.unlock();
    return;
@@ -274,7 +294,9 @@
      }catch (std::exception const &e)
      {
        LOG(ERROR)<<e.what();
        break;
        msg_data->msg["is_eof"]=true;
        guard_decoder.unlock();
        return;
      }
      if (jsonresult["wav_name"] != nullptr) {
@@ -285,7 +307,7 @@
      }
      // hotwords: fst/nn
      if(msg_data->hotwords_embedding == NULL){
      if(msg_data->hotwords_embedding == nullptr){
        std::unordered_map<std::string, int> merged_hws_map;
        std::string nn_hotwords = "";
@@ -295,25 +317,22 @@
            nlohmann::json json_fst_hws;
            try{
              json_fst_hws = nlohmann::json::parse(json_string);
              if(json_fst_hws.type() == nlohmann::json::value_t::object){
                // fst
                try{
                  std::unordered_map<std::string, int> client_hws_map = json_fst_hws;
                  merged_hws_map.insert(client_hws_map.begin(), client_hws_map.end());
                } catch (const std::exception& e) {
                  LOG(INFO) << e.what();
                }
              }
            } catch (std::exception const &e)
            {
              LOG(ERROR)<<e.what();
              break;
            }
            if(json_fst_hws.type() == nlohmann::json::value_t::object){
              // fst
              try{
                std::unordered_map<std::string, int> client_hws_map = json_fst_hws;
                merged_hws_map.insert(client_hws_map.begin(), client_hws_map.end());
              } catch (const std::exception& e) {
                LOG(INFO) << e.what();
              }
            }else{
              // nn
              std::string client_nn_hws = jsonresult["hotwords"];
              nn_hotwords += " " + client_nn_hws;
              LOG(INFO) << "nn hotwords: " << client_nn_hws;
              // LOG(INFO) << "nn hotwords: " << client_nn_hws;
            }
          }
        }
@@ -338,8 +357,10 @@
      if (jsonresult.contains("itn")) {
        msg_data->msg["itn"] = jsonresult["itn"];
      }
      if (jsonresult["is_speaking"] == false ||
          jsonresult["is_finished"] == true) {
      if ((jsonresult["is_speaking"] == false ||
          jsonresult["is_finished"] == true) &&
          msg_data->msg["is_eof"] != true &&
          msg_data->hotwords_embedding != nullptr) {
        LOG(INFO) << "client done";
        // for offline, send all receive data to decoder engine
        std::vector<std::vector<float>> hotwords_embedding_(*(msg_data->hotwords_embedding));
@@ -376,6 +397,7 @@
    default:
      break;
  }
  guard_decoder.unlock();
}
// init asr model