ngrm_02.cc

这是一个从音频信号里提取特征参量的程序

// file: $isip/class/stat/NGramModel/ngrm_02.cc
// version: $Id: ngrm_02.cc,v 1.3 2002/07/30 05:03:25 zheng Exp $
//

// isip include files
//
#include "NGramModel.h"
#include <Console.h>

// method: diagnose
//
// arguments:
//  Integral::DEBUG level: (input) debug level for diagnostics
//
// return: a boolean value indicating status
//
boolean NGramModel::diagnose(Integral::DEBUG level_a) {

  //---------------------------------------------------------------------
  //
  // 0. preliminaries
  //
  //---------------------------------------------------------------------

  // output the class name
  //
  if (level_a > Integral::NONE) {
    String output(L"diagnosing class ");
    output.concat(CLASS_NAME);
    output.concat(L": ");
    Console::put(output);
    Console::increaseIndention();
  } 

  //--------------------------------------------------------------------
  //
  // 1. required public methods
  //
  //--------------------------------------------------------------------

  // set indentation
  //
  if (level_a > Integral::NONE) {
    Console::put(L"testing required public methods...\n");
    Console::increaseIndention();
  }

  // test destructor/constructor(s)
  //
  NGramModel ngrm0;
  NGramModel ngrm1(ngrm0);
  
  if (!ngrm1.eq(ngrm0)) {
    return Error::handle(name(), L"copy constructor", Error::TEST,
			 __FILE__, __LINE__);
  }

  // test large allocation construction and deletion
  //
  if (level_a > Integral::BRIEF) {
    
    Console::put(L"testing large chunk memory allocation and deletion:\n");
    
    // set the memory to a strange block size so we can hopefully catch any
    // frame overrun errors
    //
    NGramModel::setGrowSize((long)500);
    
    NGramModel* pft = new NGramModel();
    
    for (long j = 1; j <= 100; j++) {
      NGramModel** pfts = new NGramModel*[j * 100];
      
      // create the objects
      //
      for (long i = 0; i < j * 100; i++) {
	pfts[i] = new NGramModel();
      }
    
      // delete objects
      //
      for (long i = (j * 100) - 1; i >= 0; i--) {
	delete pfts[i];
      }
    
      delete [] pfts;
    } 
    delete pft;
  }

  // test the i/o methods 
  //
  NGramNode node2, node3;
  NGramModel rd_ngrm1, rd_ngrm2;
  HashTable<Long, NGramNode>* ht1 = ngrm0.getGramHash();
  ht1->setCapacity(3);

  ht1->insert(2, &node2);
  ht1->insert(3, &node3);
  
  // we need binary and text sof files
  //
  String tmp_filename0;
  Integral::makeTemp(tmp_filename0);
  String tmp_filename1;
  Integral::makeTemp(tmp_filename1);

  // open files in write mode
  //
  Sof tmp_file0;
  tmp_file0.open(tmp_filename0, File::WRITE_ONLY, File::TEXT);
  Sof tmp_file1;
  tmp_file1.open(tmp_filename1, File::WRITE_ONLY, File::BINARY);

  if (level_a > Integral::DETAILED) {  
    ngrm0.debug(L"ngrm0");
  }

  // write to sof file
  //
  ngrm0.write(tmp_file0, 0);
  ngrm0.write(tmp_file1, 0);

  // close the files
  //
  tmp_file0.close();
  tmp_file1.close();
  
  // open the files in read mode
  //
  tmp_file0.open(tmp_filename0);
  tmp_file1.open(tmp_filename1);

  // read the value back
  //
  rd_ngrm1.read(tmp_file0, 0);
  if (!rd_ngrm1.eq(ngrm0)) {
    return Error::handle(name(), L"read", Error::TEST, __FILE__, __LINE__);
  }

  rd_ngrm2.read(tmp_file1, 0);
  if (!rd_ngrm2.eq(ngrm0)) {
    return Error::handle(name(), L"read", Error::TEST, __FILE__, __LINE__);
  }

  // close and  delete the temporary files
  //
  tmp_file0.close();
  tmp_file1.close();
  
  File::remove(tmp_filename0);
  File::remove(tmp_filename1);

  // reset indentation
  //
  if (level_a > Integral::NONE) {
    Console::decreaseIndention();
  }
  
  //--------------------------------------------------------------------------
  //
  // 2. class-specific public methods:
  //     algorithm methods
  //
  //--------------------------------------------------------------------------

  // set indentation
  //
  if (level_a > Integral::NONE) {
    Console::put(L"testing class-specific public methods: computational methods...\n");
    Console::increaseIndention();
  }

  // build a hash table which convert String to Long
  //
  String all_symbol(L"!SENT_START !SENT_END ZERO ONE TWO THREE FOUR FIVE SIX SEVEN EIGHT NINE OH");
  String symbol;
  long pos = 0;
  long symbol_index = 0;
  Long Symbol_index;
  Vector<String> symbol_table;
  HashTable<String, Long> symbol_hash;
  Sof sof;
  String file_name;
  
  // create a symbol table and a hash table
  //
  while(all_symbol.tokenize(symbol, pos)) {
    symbol.trim();
    if (!symbol.eq(String::DEF_VALUE)) {
      symbol_table.setLength(symbol_index + 1);
      symbol_table(symbol_index).assign(symbol);
      Symbol_index.assign(symbol_index);
      symbol_hash.insert(symbol, &Symbol_index);
      symbol_index++;
    }
  }

  // open the language model file
  //
  file_name.assign(L"../../../doc/examples/data/models/tidigits_trigram.arpa");
  if (!sof.open(file_name, File::READ_ONLY)) {
    return Error::handle(file_name, L"open", Error::FILE_NOTFND,
			 __FILE__, __LINE__);
  }

  // configure ngram object and load a trigram file
  //
  NGramModel ngrm;
  ngrm.setOrder(3);
  ngrm.load(sof, 1, symbol_table);
  sof.close();
  
  // test eq method
  //
  NGramModel ngrm_tmp;
  ngrm_tmp.assign(ngrm);

  if (!ngrm_tmp.eq(ngrm)) {
    return Error::handle(name(), L"eq", Error::TEST, __FILE__, __LINE__);    
  }    

  // test getScore
  //
  String symb1(L"ONE"), symb2(L"TWO"), symb3(L"FOUR");
  VectorLong index(2);
  Float exp_res;

  // verify a bigram (ONE TWO) probability
  //
  index(0) = *symbol_hash.get(symb1);
  index(1) = *symbol_hash.get(symb2);

  Float score = ngrm.getScore(index);
  exp_res = -1.17367 * Integral::LN10;
  if (!score.almostEqual(exp_res)) {
    exp_res.debug(L"expected result for \"ONE TWO\"");
    return Error::handle(name(), L"getScore", Error::TEST, __FILE__, __LINE__);
  }

  // verify a trigram (ONE FOUR TWO) probability
  //
  index.setLength(3);
  index(1) = *symbol_hash.get(symb3);
  index(2) = *symbol_hash.get(symb2);
  
  score = ngrm.getScore(index);
  exp_res = -1.065393 * Integral::LN10;
  if (!score.almostEqual(exp_res)) {
    exp_res.debug(L"expected result for \"ONE FOUR TWO\"");
    return Error::handle(name(), L"getScore", Error::TEST, __FILE__, __LINE__);
  }

  // verify a trigram (ONE TWO OH) backoff probability
  //
  index(0) = *symbol_hash.get(symb1);
  index(1) = *symbol_hash.get(symb2);
  index(2) = *symbol_hash.get(L"OH");
  score = ngrm.getScore(index);
  exp_res = 0.04061788 * Integral::LN10 + -1.20412 * Integral::LN10;
  if (!score.almostEqual(exp_res)) {
    exp_res.debug(L"expected result for \"ONE TWO OH\"");
    return Error::handle(name(), L"getScore", Error::TEST, __FILE__, __LINE__);
  }

  // verify a trigram (OH ZERO THREE) backoff probability
  //
  index(0) = *symbol_hash.get(L"OH");
  index(1) = *symbol_hash.get(L"ZERO");
  index(2) = *symbol_hash.get(L"THREE");
  score = ngrm.getScore(index);
  exp_res = -0.906405 * Integral::LN10 + -1.187113 * Integral::LN10;
  if (!score.almostEqual(exp_res)) {
    exp_res.debug(L"expected result for \"OH ZERO THREE\"");
    return Error::handle(name(), L"getScore", Error::TEST, __FILE__, __LINE__);

  }

  // write the instance of the object into the Sof file
  //
  file_name.assign(L"trigram.sof");
  if (!sof.open(file_name, File::WRITE_ONLY)) {
        return Error::handle(file_name, L"open", Error::FILE_NOTFND,
			 __FILE__, __LINE__);
  }
  // store it to source ngram file, and then read it back.
  // both should match.
  //
  ngrm.store(sof, 1);
  sof.close();
  
  // load it back
  //
  NGramModel new_ngrm;
  new_ngrm.setOrder(3);
  sof.open(file_name, File::READ_ONLY);
  new_ngrm.load(sof, 1, symbol_table);
  sof.close();
  
  // compare both
  //
  if (!ngrm.eq(new_ngrm)) {
    ngrm.debug(L"ngrm");
    new_ngrm.debug(L"new_ngrm");
    return Error::handle(name(), L"getScore", Error::TEST, __FILE__, __LINE__);

  }

  // remove files
  //
  if (level_a < Integral::ALL) {
    File::remove(file_name);
  }

  // test the i/o methods 
  //
  file_name.assign(L"ngram.sof");
  Sof tmp_file2;
  
  // open files in write mode
  //
  tmp_file0.open(tmp_filename0, File::WRITE_ONLY, File::TEXT);
  tmp_file1.open(tmp_filename1, File::WRITE_ONLY, File::BINARY);
  tmp_file2.open(file_name, File::WRITE_ONLY, File::TEXT);
  
  // write to sof file
  //
  ngrm.write(tmp_file0, 0);
  ngrm.write(tmp_file1, 0);
  ngrm.write(tmp_file2, 0);
  
  // close the files
  //
  tmp_file0.close();
  tmp_file1.close();
  tmp_file2.close();
  
  // open the files in read mode
  //
  tmp_file0.open(tmp_filename0);
  tmp_file1.open(tmp_filename1);

  // read the value back
  //
  rd_ngrm1.clear(Integral::RESET);
  rd_ngrm1.read(tmp_file0, 0);
  if (!rd_ngrm1.eq(ngrm)) {
    return Error::handle(name(), L"read", Error::TEST, __FILE__, __LINE__);
  }

  rd_ngrm1.clear(Integral::RESET);
  rd_ngrm2.read(tmp_file1, 0);
  if (!rd_ngrm2.eq(ngrm)) {
    return Error::handle(name(), L"read", Error::TEST, __FILE__, __LINE__);
  }

  // close and  delete the temporary files
  //
  tmp_file0.close();
  tmp_file1.close();

  // remove temp files
  //
  File::remove(tmp_filename0);
  File::remove(tmp_filename1);
  if (level_a < Integral::ALL) {
    File::remove(file_name);
  }
  
  // reset indentation
  //
  if (level_a > Integral::NONE) {
    Console::decreaseIndention();
  }

  // --------------------------------------------------------------------
  //
  // 3. print completion message
  //
  // --------------------------------------------------------------------

  // reset indentation
  //
  if (level_a > Integral::NONE) {
    Console::decreaseIndention();
  }
  
  if (level_a > Integral::NONE) {
    String output(L"diagnostics passed for class ");
    output.concat(name());
    output.concat(L"\n");
    Console::put(output);
  }

  // exit gracefully
  //
  return true;
}