hmm_05.cc

这是一个从音频信号里提取特征参量的程序

    String identifier;
    Filename input_file_name;    
    
    Filename input_ID;
    String output;
    String train_path;
    String train_file;    
    Filename output_file_name;
    Sdb output_sdb;    
    Vector<VectorFloat> data;

    // load the hmm models from the model file
    //
    if (!load()) {
      return Error::handle(name(), L"run: load hmm models",
			   Error::ARG, __FILE__, __LINE__);
    }

    // set the search engine mode
    //
    search_engine_d.setInitialLevel(initial_level_d);    
    search_engine_d.setSearchMode(HierarchicalSearch::TRAIN);    

    // loop over the number of iterations
    //
    for (long iter = 0; iter < num_iterations_d; iter++) {

      current_file_num = 0;

      // print the current iteration
      //
      if (verbosity_d >= Integral::BRIEF) {
	Console::increaseIndention();
	output.assign(L"\nstarting iteration: ");
	output.concat(iter);
	Console::put(output);
	Console::decreaseIndention();
      }
      
      // loop through each input utterance
      //
      for (sdb_a.gotoFirst(); sdb_a.getName(input_ID);
	   sdb_a.gotoNext()) {

	current_file_num++;

	// get audio file path
	//
	identifier.assign(input_ID);
	if (!audio_db_d.getRecord(identifier, input_file_name)) {
	  return Error::handle(name(), L"initialize - unable to find audio file for identifier", Error::ARG, __FILE__, __LINE__);	    
	}      	
	
	// print utterance processing information
	//      
	if (verbosity_d >= Integral::BRIEF) {
	  Console::increaseIndention();
	  output.assign(L"\nprocessing file ");
	  output.concat(current_file_num);
	  output.concat(L" (");
	  output.concat(identifier);
	  output.concat(L")");
	  output.concat(L": ");
	  output.concat(input_file_name);
	  Console::put(output);
	  Console::decreaseIndention();	    
	}

	// clear all data structures needed to decode
	//
	if (!search_engine_d.initializeLinearDecoder()) {
	  return Error::handle(name(), L"linearDecoder", Error::ARG, __FILE__, __LINE__);
	}
	
	// initialize the top search level with the corresponding transcription
	// we don't need a transcription database for segmented utterances
	//
	if (transcription_db_file_d.length() > 0) {
	  initTranscription(identifier, current_file_num - 1);
	}
	
	// process the utterance file by the front end
	//
	fe_d.open(input_file_name);

	// retrieve the all frames of data in advance
	//
	extractFeatures(data);

	// decode the utterance
	// 	
	if (!verify_d) {
	  search_engine_d.linearDecoder(fe_d);
	}

	// pick up the best hypothesis and determine the utterance probability
	//
	String hypotheses;
	double score = 0;
	long num_frames = 0;
	DoubleLinkedList<Trace> trace_path;
	
	if (!search_engine_d.getHypotheses(hypotheses, alignment_level_d,
					   score, num_frames, trace_path)) {

	  // if no hypothesis found
	  //
	  hypotheses.clear();
	  trace_path.clear();
	  
	  if ((!verify_d) && (verbosity_d >= Integral::BRIEF)) {
	    
	    // print the warning message
	    //
	    Console::increaseIndention();
	    Console::put(L"\nno hypothesis found");
	    Console::decreaseIndention();
	  }
	}
	else {

	  num_valid_files++;
	  total_num_frames += num_frames;
	}       

	// compute the forward probabilities (alphas)
	//
	trellis_d = (BiGraph<TrainNode>*)NULL;
	trellis_d = search_engine_d.computeForwardBackward(data, beta_threshold_d);

	// determine the utterance probability here again
	//
	computeUtterProb(score);

	// print utterance probability
	//      
	if (verbosity_d >= Integral::BRIEF) {
	  Console::increaseIndention();
	  output.assign(L"\naverage utterance probability: ");
	  output.concat(score/num_frames);
	  output.concat(L", number of frames: ");
	  output.concat(num_frames);
	  Console::put(output);
	  Console::decreaseIndention();	    
	}      
	
	// accumulate the statistics
	//
	accumulate(score, data);

	// close the front end processing
	//
	fe_d.close();

      } // end of looping through the input utterances

      // update the models using the accumulated statistics
      //
      update();

      // reset the statistical model and state transition accumulators
      //
      if (iter < (long)num_iterations_d - 1) {
	resetAccumulators();
      }

    } // end of looping over the number of iterations
    
    // clear all data structures needed to decode
    //
    if (!search_engine_d.initializeLinearPartial()) {
      return Error::handle(name(), L"linearDecoder", Error::ARG, __FILE__, __LINE__);
    }

    // clear the top level
    //
    search_engine_d.getSearchLevels()((long)initial_level_d).clear();

    // restore the top-level subgraph of the hierarchy
    //    
    LanguageModel lm_model;
    if (!lm_model.load(lm_model_file_d, ac_model_file_d, (long)initial_level_d,
		       search_engine_d.getSearchLevels())) {
      return Error::handle(name(), L"load: loading model file",
			   Error::ARG, __FILE__, __LINE__);
    }
    
    // write the models to file
    //
    if (!store()) {
      return Error::handle(name(), L"linearDecoder", Error::ARG, __FILE__, __LINE__);
    }
    
    // clean up all memory
    //
    search_engine_d.clear();

    // print the number of successfully processed files
    //
    if (verbosity_d >= Integral::BRIEF) {
      output.assign(L"\nprocessed ");
      output.concat(num_valid_files);
      output.concat(L" file(s) successfully, attempted ");
      output.concat(current_file_num);
      output.concat(L" file(s), ");
      output.concat(total_num_frames);
      output.concat(L" frame(s)\n");    
      Console::put(output);
    }
  
  } // end of TRAIN using BAUM_WELCH and NONE

  // branch on ADAPT_DECODE using SUPERVISED and INCREMENTAL
  // does not support streaming input.
  //
  else if (algorithm_d == ADAPT_DECODE && !stream_d &&
	   adapt_supervision_mode_d == SUPERVISED &&
	   adapt_sequence_mode_d == INCREMENTAL) {

    // declare local variables
    //
    long num_valid_files = 0;
    long current_file_num = 0;
    long total_num_frames = 0;

    String identifier;
    Filename input_file_name;    
    
    Filename input_ID;
    String output;
    String train_path;
    String train_file;    
    Filename output_file_name;
    Sdb output_sdb;    
    Vector<VectorFloat> data;

    // load the hmm models from the model file
    //
    if (!load()) {
      return Error::handle(name(), L"run: load hmm models",
			   Error::ARG, __FILE__, __LINE__);
    }

    // set the search engine mode
    //    
    search_engine_d.setSearchMode(HierarchicalSearch::TRAIN);    

    // loop over the number of iterations
    //
    for (long iter = 0; iter < num_iterations_d; iter++) {

      current_file_num = 0;

      // print the current iteration
      //
      if (verbosity_d >= Integral::BRIEF) {
	Console::increaseIndention();
	output.assign(L"\nstarting iteration: ");
	output.concat(iter);
	Console::put(output);
	Console::decreaseIndention();
      }
      
      // loop through each input utterance
      //
      for (sdb_a.gotoFirst(); sdb_a.getName(input_ID);
	   sdb_a.gotoNext()) {

	current_file_num++;

	// get audio file path
	//
	identifier.assign(input_ID);
	if (!audio_db_d.getRecord(identifier, input_file_name)) {
	  input_ID.debug(L"input_ID");
	  return Error::handle(name(), L"initialize - unable to find audio file for identifier", Error::ARG, __FILE__, __LINE__);	    
	}      	
	
	// print utterance processing information
	//      
	if (verbosity_d >= Integral::BRIEF) {
	  Console::increaseIndention();
	  output.assign(L"\nprocessing file ");
	  output.concat(current_file_num);
	  output.concat(L" (");
	  output.concat(identifier);
	  output.concat(L")");
	  output.concat(L": ");
	  output.concat(input_file_name);
	  Console::put(output);
	  Console::decreaseIndention();	    
	}

	// clear all data structures needed to decode
	//
	if (!search_engine_d.initializeLinearDecoder()) {
	  return Error::handle(name(), L"linearDecoder", Error::ARG, __FILE__, __LINE__);
	}
	
	// initialize the top search level with the corresponding transcription
	// we don't need a transcription database for segmented utterances
	//
	if (transcription_db_file_d.length() > 0) {
	  initTranscription(identifier, current_file_num - 1);
	}
	
	// process the utterance file by the front end
	//
	fe_d.open(input_file_name);

	// retrieve the all frames of data in advance
	//
	extractFeatures(data);

	// decode the utterance
	// 	
	if (!verify_d) {
	  search_engine_d.linearDecoder(fe_d);
	}

	// pick up the best hypothesis and determine the utterance probability
	//
	String hypotheses;
	double score = 0;
	long num_frames = 0;
	DoubleLinkedList<Trace> trace_path;
	
	if (!search_engine_d.getHypotheses(hypotheses, alignment_level_d,
					   score, num_frames, trace_path)) {

	  // if no hypothesis found
	  //
	  hypotheses.clear();
	  trace_path.clear();
	  
	  if ((!verify_d) && (verbosity_d >= Integral::BRIEF)) {
	    
	    // print the warning message
	    //
	    Console::increaseIndention();
	    Console::put(L"\nno hypothesis found");
	    Console::decreaseIndention();
	  }
	}
	else {

	  num_valid_files++;
	  total_num_frames += num_frames;
	}       

	// compute the forward probabilities (alphas)
	//
	trellis_d = (BiGraph<TrainNode>*)NULL;
	trellis_d = search_engine_d.computeForwardBackward(data, beta_threshold_d);

	// determine the utterance probability here again
	//
	computeUtterProb(score);

	// print utterance probability
	//      
	if (verbosity_d >= Integral::BRIEF) {
	  Console::increaseIndention();
	  output.assign(L"\naverage utterance probability: ");
	  output.concat(score/num_frames);
	  output.concat(L", number of frames: ");
	  output.concat(num_frames);
	  Console::put(output);
	  Console::decreaseIndention();	    
	}      
	
	// accumulate the statistics
	//
	accumulate(score, data);

	// close the front end processing
	//
	fe_d.close();

      } // end of looping through the input utterances

      // get vector of statistical models
      //
      Vector<StatisticalModel>& stat_models =
	search_engine_d.getSearchLevel((long)num_levels_d - 1).
	getStatisticalModels();
  
      // adapt the models using the accumulated statistics
      //
      adapt(stat_models);

      // reset the statistical model and state transition accumulators
      //
      if (iter < (long)num_iterations_d - 1) {
	resetAccumulators();
      }

    } // end of looping over the number of iterations
    
    // clear all data structures needed to decode
    //
    if (!search_engine_d.initializeLinearPartial()) {
      return Error::handle(name(), L"linearDecoder", Error::ARG, __FILE__, __LINE__);
    }

    // clear the top level
    //
    search_engine_d.getSearchLevels()(0).clear();
    
    // restore the top-level subgraph of the hierarchy
    //    
    LanguageModel lm_model;
    if (!lm_model.load(lm_model_file_d, ac_model_file_d, 0,
		       search_engine_d.getSearchLevels())) {
      return Error::handle(name(), L"load: loading model file",
			   Error::ARG, __FILE__, __LINE__);
    }

    // write the models to file
    //
    if (!store()) {
      return Error::handle(name(), L"linearDecoder", Error::ARG, __FILE__, __LINE__);
    }
  
    // clean up all memory
    //
    search_engine_d.clear();

    // print the number of successfully processed files
    //
    if (verbosity_d >= Integral::BRIEF) {
      output.assign(L"\nprocessed ");
      output.concat(num_valid_files);
      output.concat(L" file(s) successfully, attempted ");
      output.concat(current_file_num);
      output.concat(L" file(s), ");
      output.concat(total_num_frames);
      output.concat(L" frame(s)\n");    
      Console::put(output);
    }
  
  } // ADAPT_DECODE using SUPERVISED and INCREMENTAL
  
  // branch on the UNKNOWN options
  //
  else {    
    return Error::handle(name(), L"invalid algorithm and/or implementation", Error::ARG, __FILE__, __LINE__);
  }
  
  // gracefully exit
  //
  return true;  
}

// method: grammarDecoder
//
// arguments:
//  Sdb& sdb: (input) signal data base to run on
//
// return: a boolean value indicating status
//
// this is the run method for a grammar decoder
//
boolean HiddenMarkovModel::grammarDecoder(Sdb& sdb_a) {

  // branch on the CONTEXT_GENERATION using SYMBOL_GENERATION
  // does not support streaming input.
  //
  if (algorithm_d == CONTEXT_GENERATION && !stream_d &&
      implementation_d == SYMBOL_GENERATION && function_mode_d == NONE) {

    // declare local variables
    //
    long num_valid_files = 0;
    long current_file_num = 0;
    long total_num_frames = 0;

    String identifier;
    Filename input_file_name;    

    String output;    
    Filename input_ID;

    // load the hmm models from the model file
    //
    if (!load()) {
      return Error::handle(name(), L"run: load hmm models",
			   Error::ARG, __FILE__, __LINE__);
    }

    // set the se