mm_02.cc

这是一个从音频信号里提取特征参量的程序

  // reset indentation
  //
  if (level_a > Integral::NONE) {
    Console::decreaseIndention();
  }
  
  //---------------------------------------------------------------------------
  //
  // 3. class-specific public methods:
  //     computation methods
  //
  //---------------------------------------------------------------------------
  
  // set indentation
  //
  if (level_a > Integral::NONE) {
    Console::put(L"testing class-specific public methods: computation methods...\n");
    Console::increaseIndention();
  }

  // test the set and get methods
  //
  VectorFloat new_weights(L"79, 5");
  mix7.setWeights(new_weights);

  VectorFloat test_weights;
  if (!mix7.getWeights(test_weights) ||
      !test_weights.almostEqual(new_weights)) {
    new_weights.debug(L"expected weights");
    test_weights.debug(L"actual weights");
    return Error::handle(name(), L"setWeights/getWeights", Error::TEST,
			 __FILE__, __LINE__);
  }

  // now change mix7 to NONE mode and make sure the results hold
  //
  mix7.setMode(NONE);
  if (!mix7.getWeights(test_weights) ||
      !test_weights.almostEqual(new_weights)) {
    new_weights.debug(L"expected weights");
    test_weights.debug(L"actual weights");
    return Error::handle(name(), L"setWeights/getWeights", Error::TEST,
			 __FILE__, __LINE__);
  }

  // test the isNormalized method
  //
  if (mix7.isNormalized()) {
    return Error::handle(name(), L"isNormalized", Error::TEST,
			 __FILE__, __LINE__);
  }

  // now normalize the data
  //
  mix7.normalizeWeights();
  if (!mix7.isNormalized()) {
    mix7.debug(L"mix7 not normalized");
    return Error::handle(name(), L"normalizeWeights", Error::TEST,
			 __FILE__, __LINE__);
  }

  // now normalize to a different normalization factor which should
  // give us the original weights back
  //
  mix7.normalizeWeights(84);
  if (!mix7.isNormalized(84) ||
      !mix7.getWeights(test_weights) ||
      !test_weights.almostEqual(new_weights)) {
    mix7.debug(L"mix7 not normalized to 84");
    return Error::handle(name(), L"normalizeWeights", Error::TEST,
			 __FILE__, __LINE__);
  }

  // test all of the normalization in PRE_COMPUTE mode
  //
  mix7.setMode(PRECOMPUTE);
  if (mix7.isNormalized()) {
    return Error::handle(name(), L"isNormalized", Error::TEST,
			 __FILE__, __LINE__);
  }

  // now normalize the data
  //
  mix7.normalizeWeights();
  if (!mix7.isNormalized()) {
    mix7.debug(L"mix7 not normalized");
    return Error::handle(name(), L"normalizeWeights", Error::TEST,
			 __FILE__, __LINE__);
  }

  // now normalize to a different normalization factor which should
  // give us the original weights back
  //
  mix7.normalizeWeights(84);
  if (!mix7.isNormalized(84) ||
      !mix7.getWeights(test_weights) ||
      !test_weights.almostEqual(new_weights)) {
    mix7.debug(L"mix7 not normalized to 84");
    return Error::handle(name(), L"normalizeWeights", Error::TEST,
			 __FILE__, __LINE__);
  }

  // test the getModels and setModels methods
  //
  MixtureModel mix8;
  mix8.setModels(mix7.getModels());
  mix7.getWeights(test_weights);
  mix8.setWeights(test_weights);

  if (!mix7.eq(mix8)) {
    mix7.debug(L"expected mixture");
    mix8.debug(L"actual mixture");    
    return Error::handle(name(), L"setModels/getModels", Error::TEST,
			 __FILE__, __LINE__);    
  }
  
  // test the likelihood methods
  //
  VectorFloat mix_mean0(L"-3.639462, 1.653617, 1.787559, 4.262538, 1.369591, 2.418731, 0.2277095, 1.344989, 0.7442688, 0.8819447, 0.5029327, 0.6623194, -7.385917, -0.01989461, 0.03300723, 3.038317e-04, 7.731465e-02, 2.036626e-02, 1.392890e-02, 1.471747e-02, 2.386167e-02, -1.337681e-02, 1.569910e-02, -7.796203e-05, 9.159852e-03, -3.887357e-02, 1.891592e-02, 3.389170e-03, 6.476890e-03, -9.673509e-03, -3.964308e-03, -1.162331e-02, -5.141579e-03, -8.838848e-03, -3.854857e-03, -2.433208e-05, -3.092395e-03, -5.406926e-04, 1.159509e-02 ");
  MatrixFloat mix_cov0(39, 39, L"1.433564e+01, 1.975537e+01, 1.671448e+01, 2.977089e+01, 2.330977e+01, 2.712510e+01, 2.764725e+01, 3.032939e+01, 3.006948e+01, 2.652304e+01, 2.420893e+01, 2.028848e+01, 5.845720e+00, 2.440532e-01, 4.785362e-01, 7.022072e-01, 1.055732e+00, 1.272656e+00, 1.545274e+00, 1.733616e+00, 1.836289e+00, 1.909358e+00, 1.861139e+00, 1.752789e+00, 1.551828e+00, 2.726359e-02, 4.366880e-02, 8.581409e-02, 1.317306e-01, 1.942055e-01, 2.418336e-01, 2.950188e-01, 3.333847e-01, 3.536546e-01, 3.684548e-01, 3.604752e-01, 3.396829e-01, 3.016612e-01, 4.171541e-03", Integral::DIAGONAL);
  VectorFloat in_vec0(L"-1.05023813258004e+00,-7.77674896710878e+00,1.43041098327926e+01,-6.21318461516028e+00,-8.64903872637757e+00,1.02604097385083e+01,-5.57177352692909e+00,5.09477660171566e+00,-4.09609275292737e+00,5.96100219381628e+00,3.76151348252777e-01,-2.97277346115335e+00,-2.30355781300591e+00,7.66914363607964e-01,1.27304216160438e-01,-3.20960700374856e-01,-1.93367994699623e+00,-4.25686341458929e-01,4.12861672859121e-01,-3.16990107434180e-01,1.81895735084098e+00,1.97179456310064e+00,-5.23509954876174e-01,-2.63924777075424e-01,3.76853360820597e-03,6.07853048112972e-04,9.75499698349679e-02,5.80349863491104e-01,-4.96041065023962e-01,-5.45823432246360e-02,-1.61513044481162e-01,2.12215102626400e-01,2.76088189007932e-01,4.25378696913581e-01,-3.04615852820205e-02,-2.56536949288540e-02,2.44496596888947e-01,6.41653297530483e-01,-6.79771984752767e-02");
  Float res_score0 = -67.155225;

  GaussianModel gauss4;
  gauss4.setMean(mix_mean0);
  gauss4.setCovariance(mix_cov0);

  // add the gaussian to the mixture model
  //
  MixtureModel mix_model0;
  mix_model0.add(gauss4);
  mix_model0.add(gauss4);

  // set the mixture model weights
  //
  //  new_weights.assign(L"2, 2");
  new_weights.assign(L"7.389056099, 7.389056099");
  mix_model0.setWeights(new_weights);

  // generate the log-likelihood
  //
  Float score0 = mix_model0.getLogLikelihood(in_vec0);

  if (!score0.almostEqual(res_score0)) {
    mix_model0.debug(L"mix_model0");
    res_score0.debug(L"expected score");
    score0.debug(L"actual score");
    return Error::handle(name(), L"getLogLikelihood", Error::TEST, __FILE__, __LINE__);
  }

  // change to NONE mode and normalize the weights so that they sum to 2.0.
  // since the weights are equal, this should put them both at 1.0
  //
  mix_model0.setMode(NONE);
  mix_model0.normalizeWeights(2.0);

  Float score1 = mix_model0.getLogLikelihood(in_vec0);
  Float res_score1 = -69.155225;

  if (!score1.almostEqual(res_score1)) {
    mix_model0.debug(L"mix_model0");
    res_score1.debug(L"expected score");
    score1.debug(L"actual score");
    return Error::handle(name(), L"getLogLikelihood", Error::TEST,
			 __FILE__, __LINE__);
  }

  Float score2 = mix_model0.getLikelihood(in_vec0);

  if (!score2.almostEqual(Integral::exp(res_score1))) {
    mix_model0.debug(L"mix_model0");
    res_score1.exp();
    res_score1.debug(L"expected score");
    score1.debug(L"actual score");
    return Error::handle(name(), L"getLogLikelihood", Error::TEST,
			 __FILE__, __LINE__);
  }

  // test the getMean
  //
  VectorFloat input_mean_mix1(L"2, 4");
  VectorFloat input_mean_mix2(L"-2, -4");
  VectorFloat exp_mean(L"0, 0");
  VectorFloat mix_weight(L"0.5, 0.5");
  VectorFloat mean;

  MatrixFloat input_cov_mix1(2, 2, L"1, 0.5", Integral::DIAGONAL);
  MatrixFloat input_cov_mix2(2, 2, L"1, 0.5", Integral::DIAGONAL);

  GaussianModel gauss5;
  gauss5.setMean(input_mean_mix1);
  gauss5.setCovariance(input_cov_mix1);

  GaussianModel gauss6;
  gauss6.setMean(input_mean_mix2);
  gauss6.setCovariance(input_cov_mix2);

  // add the gaussian to the mixture model
  //
  MixtureModel mix_model9;
  mix_model9.add(gauss5);
  mix_model9.add(gauss6);
  mix_model9.setWeights(mix_weight);
  
  mix_model9.getMean(mean);
  if (!mean.almostEqual(exp_mean)) {
    exp_mean.debug(L"expected mean");
    mean.debug(L"actual mean");
    return Error::handle(name(), L"getMean", Error::TEST, __FILE__,
			 __LINE__);
  }

  // test the getCovariance for single mixture Gaussian distribution
  //
  VectorFloat input_mean_mix3(L"2, 3");
  MatrixFloat input_cov_mix3(2, 2, L"1.33333, 2.444", Integral::DIAGONAL);
  MatrixFloat exp_cov(2, 2, L"1.33333, 2.444", Integral::DIAGONAL);
  MatrixFloat cov;
  
  GaussianModel gauss7;
  gauss7.setMean(input_mean_mix3);
  gauss7.setCovariance(input_cov_mix3);

  // add the gaussian to the mixture model
  //
  MixtureModel mix_model10;
  mix_model10.add(gauss7);
  
  mix_model10.getCovariance(cov);
  if (!cov.almostEqual(exp_cov)) {
    exp_cov.debug(L"expected covariance");
    cov.debug(L"actual covariance");
    return Error::handle(name(), L"getCovariance", Error::TEST, __FILE__,
			 __LINE__);
  }
  
  // reset indentation
  //
  if (level_a > Integral::NONE) {
    Console::decreaseIndention();
  }
  
  //---------------------------------------------------------------------------
  //
  // 4. print completion message
  //
  //---------------------------------------------------------------------------

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

  // exit gracefully
  //
  return true;
}