tnod_05.cc

这是一个从音频信号里提取特征参量的程序

// file: $isip/class/search/TrainNode/tnod_05.cc
// version: $Id: tnod_05.cc,v 1.8 2003/01/23 20:00:32 alphonso Exp $
//

// isip include files
//
#include "TrainNode.h"

// method: assign
//
// arguments:
//  const TrainNode& copy_node: (input) node to copy
//
// return: logical error status
//
// assign TrainNode from the copy
//
boolean TrainNode::assign(const TrainNode& copy_node_a) {

  // copy the pointers
  //
  reference_d = copy_node_a.reference_d;
  stat_model_d = copy_node_a.stat_model_d;
  
  // copy the values
  //
  score_d = copy_node_a.score_d;  
  frame_d = copy_node_a.frame_d;
  alpha_d = copy_node_a.alpha_d;
  beta_d = copy_node_a.beta_d;

  is_valid_d = copy_node_a.is_valid_d;  
  is_alpha_valid_d = copy_node_a.is_alpha_valid_d;
  is_beta_valid_d = copy_node_a.is_beta_valid_d;    
  is_accum_valid_d = copy_node_a.is_accum_valid_d;
  
  // exit gracefully
  //
  return true;
}

// method: clear
//
// arguments:
//  Integral::CMODE cmode: (input) clear mode
//
// return: logical error status
//
// clear the contents of the TrainNode
//
boolean TrainNode::clear(Integral::CMODE cmode_a) {

  // clear the pointers
  //
  reference_d = (Context*)NULL;
  stat_model_d = (StatisticalModel*)NULL;
  
  // reset the values
  //
  frame_d = DEF_TIMESTAMP;
  alpha_d = Integral::DB_LOG_MIN_VALUE;
  beta_d = Integral::DB_LOG_MIN_VALUE;
  score_d = Integral::DB_LOG_MIN_VALUE;    

  is_valid_d = false;
  is_alpha_valid_d = false;
  is_beta_valid_d = false;
  is_accum_valid_d = false;
  
  // exit gracefully
  //
  return true;
}

// method: eq
//
// arguments:
//  const TrainNode& compare_node: (input) TrainNode to compare
//
// return: true if the TrainNodes are equivalent, else false
//
// compare two TrainNodes. they are equivalent if they have equivalent history
// objects
//
boolean TrainNode::eq(const TrainNode& compare_node_a) const {

  // compare the address
  //
  return (this == &compare_node_a);
}

// method: update
//
// arguments:
//   VectorFloat& varfloor: (input) variance floor
//   long min_model: (input) minimum model count
//
// return: logical status
//
// updates the models using the accumulators generated during training
//
boolean TrainNode::update(VectorFloat& varfloor_a, long min_model_a) {

  // declare local variables
  //
  boolean status = false;

  // check to make sure that the model is valid
  //
  if (stat_model_d != (StatisticalModel*)NULL) {
    
    // update: call the statistical model
    //      
    status = stat_model_d->update(varfloor_a, min_model_a);
  }

  // exit gracefully
  //
  return status;
}

// method: accumulate
//
// arguments:
//  float utter_prob: (input) utterance probability
//  Vector<VectorFloat>& data: (input) feature vectors
//  float min_mpd: (input) minimum model probability deviance
//  float min_occupancy: (input) floor on the occupancy probability
//
// return: logical status
//
// accumulates the statistics in training which are used to update the model
//
boolean TrainNode::accumulate(double utter_prob_a, Vector<VectorFloat>& data_a,
			      float min_mpd_a, float min_occupancy_a) {

  // declare local variables
  //
  VectorDouble param;
  
  boolean precomp = false;  
  boolean status = false;

  // set up the parameter list
  //
  param.setLength(6);

  // assumptions:
  //
  // param(0) - forward probability (alpha)
  // param(1) - backward probability (beta)
  // param(2) - utterance normalization, i.e., (1) above
  // param(3) - minimum model probability deviance
  // param(4) - floor on the occupancy probability    
  // param(5) - likelihood score
  
  param(0) = alpha_d;
  param(1) = beta_d;
  param(2) = utter_prob_a;
  param(3) = (double)min_mpd_a;
  param(4) = (double)min_occupancy_a;
  param(5) = score_d;

  // check to make sure the model is valid
  //
  if (stat_model_d != (StatisticalModel*)NULL) {
    
    // accumulate: call the statistical model
    //
    status = stat_model_d->accumulate(param, data_a(frame_d), precomp);
  }
  
  // exit gracefully
  //
  return status;
}