private void UpdateRow(TrainFilesCarrier.TrainFileRow row)
{
string file = VCDir.Instance.TrainDirAudio + row.Path;
if (File.Exists(file))
{
_yourChanged = true;
_yourPath = file;
_yourWav = new WavFileWrapper(_yourPath);
if (_yourWav.Load())
{
_yourWav.NormalizeWave(1.0);
LogUtil.Info("Load Wave: {0} -- OK\n", _yourPath);
option = VCContext.Instance.MFCCOptions;
if (option.ShiftSampleToZero)
{
LogUtil.Info("Shift Sample To Zero: -- -- OK\n");
_yourWav.ShifToZero();
}
waveViewer.WaveData = _yourWav.FullData;
waveViewer.FitToScreen();
_label = row.Word;
int size = _yourWav.FullData.Count;
_startSelected = (float)row.Start / size;
_endSelected = (float)row.End / size;
waveViewer.LeftSlider = _startSelected;
waveViewer.RightSlider = _endSelected;
}
}
}
public string Reg(string path, int start, int end)
{
if (_words == null || _words.Count == 0)
{
return(" Models Failed");
}
WavFileWrapper Wav = new WavFileWrapper(path);
LogUtil.Info("*******************Reg*******************\n");
LogUtil.Info("Load Wave: {0}\n", path);
if (Wav.Load())
{
Wav.NormalizeWave(1.0f);
LogUtil.Info("Load Wave: {0} -- OK\n", path);
if (VCContext.Instance.MFCCOptions.ShiftSampleToZero)
{
LogUtil.Info("Shift Sample To Zero: -- -- OK\n");
Wav.ShifToZero();
}
Wav.SelectedWave((uint)start, (uint)end);
MFCCWrapper Mfcc = new MFCCWrapper(Wav, VCContext.Instance.MFCCOptions.TimeFrame, VCContext.Instance.MFCCOptions.TimeShift, VCContext.Instance.MFCCOptions.CepFilter, VCContext.Instance.MFCCOptions.LowFreq, VCContext.Instance.MFCCOptions.HighFreq, VCContext.Instance.MFCCOptions.NumCeps, 2);
bool res = Mfcc.Process();
List <List <double> > data = null;
if (res)
{
switch (VCContext.Instance.MFCCOptions.TrainCofficientType)
{
case 0:
data = Mfcc.Mfcc;
break;
case 1:
data = Mfcc.DetalMfcc;
break;
case 2:
data = Mfcc.DoubleDetalMfcc;
break;
default:
data = Mfcc.Mfcc;
break;
}
LogUtil.Info("Load Wave: {0} Process MFCC -- OK\n", path);
int reg = 0;
double max = _models[reg].LogProbability(data);
LogUtil.Info("HModel = {0} Log Value = {1}\n", _words[reg], max);
for (int i = 1; i < _models.Count; i++)
{
double cur = _models[i].LogProbability(data);
LogUtil.Info("HModel = {0} Log Value = {1}\n", _words[i], cur);
if (max < cur)
{
max = cur;
reg = i;
}
}
LogUtil.Info("HModel = {0} Log Value = {1} Reg = {2}\n", _words[reg], max, _words[reg]);
return(_words[reg]);
}
else
{
LogUtil.Info("Load Wave: {0} -- FAILED\n", path);
}
}
return("NONE");
}
public bool Train()
{
_words = new List <string>();
_hmms = new List <string>();
_models = new List <HMMWrapper>();
foreach (TrainFilesCarrier.TrainFileRow ent in tbEntry)
{
if (!_words.Contains(ent.Word))
{
_words.Add(ent.Word);
LogUtil.Info("Word : {0}\n", ent.Word);
}
}
foreach (string word in _words)
{
List <string> files = new List <string>();
LogUtil.Info("List File for Word : {0}\n", word);
VCDir.CreateDirectory(VCDir.Instance.TrainDirMFCC + word);
VCDir.CreateDirectory(VCDir.Instance.TrainDirHMM);
foreach (TrainFilesCarrier.TrainFileRow ent in tbEntry)
{
if (word.Equals(ent.Word))
{
string audio = VCDir.Instance.TrainDirAudio + ent.Path;
string mfcc = VCDir.Instance.TrainDirMFCC + ent.Path;
WavFileWrapper Wav = new WavFileWrapper(audio);
if (Wav.Load())
{
Wav.NormalizeWave(1.0f);
if (VCContext.Instance.MFCCOptions.ShiftSampleToZero)
{
Wav.ShifToZero();
}
Wav.SelectedWave((uint)ent.Start, (uint)ent.End);
MFCCWrapper Mfcc = new MFCCWrapper(Wav, VCContext.Instance.MFCCOptions.TimeFrame, VCContext.Instance.MFCCOptions.TimeShift, VCContext.Instance.MFCCOptions.CepFilter, VCContext.Instance.MFCCOptions.LowFreq, VCContext.Instance.MFCCOptions.HighFreq, VCContext.Instance.MFCCOptions.NumCeps, 2);
Mfcc.UserStandardization = VCContext.Instance.MFCCOptions.UseStandardization;
bool res = Mfcc.Process();
res &= Mfcc.SaveMFCC(mfcc + ".Mfcc" + ".xml");
res &= Mfcc.SaveDeltaMFCC(mfcc + ".Delta" + ".xml");
res &= Mfcc.SaveDoubleMFCC(mfcc + ".Double" + ".xml");
if (res)
{
switch (VCContext.Instance.MFCCOptions.TrainCofficientType)
{
case 0:
files.Add(mfcc + ".Mfcc" + ".xml");
break;
case 1:
files.Add(mfcc + ".Delta" + ".xml");
break;
case 2:
files.Add(mfcc + ".Double" + ".xml");
break;
}
}
LogUtil.Info("File : W - {0} Path - {1} MFCC process - {2}\n", ent.Word, VCDir.Instance.TrainDirMFCC + ent.Path + ".xml", (res) ? "Completed" : "Failed");
}
}
}
HMMWrapper hmm = new HMMWrapper(VCContext.Instance.MFCCOptions.TrainHMMState, VCContext.Instance.MFCCOptions.TrainGMMComponent, VCContext.Instance.MFCCOptions.TrainGMMCovVar);
bool ok = hmm.Trainning(files);
LogUtil.Info("Train word: {0} - {1}\n", word, (ok)? "Completed": "Failed");
if (ok)
{
hmm.Save(VCDir.Instance.TrainDirHMM + word + ".xml");
_models.Add(hmm);
_hmms.Add(word + ".xml");
LogUtil.Info("Save {0} Model to {1}\n", word, VCDir.Instance.TrainDirHMM + word + ".xml");
}
}
Save(VCDir.Instance.TrainXmlFile);
return(true);
}
private void bgw_DoWork(object sender, DoWorkEventArgs e)
{
bool reg = (bool)e.Argument;
if (_yourChanged)
{
_yourWav = new WavFileWrapper(_yourPath);
if (_yourWav.Load())
{
_yourWav.NormalizeWave(1.0);
option = VCContext.Instance.MFCCOptions;
LogUtil.Info("Load Wave: {0} -- OK\n", _yourPath);
if (option.ShiftSampleToZero)
{
LogUtil.Info("Shift Sample To Zero: -- -- OK\n");
_yourWav.ShifToZero();
}
int size = _yourWav.FullData.Count;
uint startPnt = (uint)(_startSelected * size);
uint endPnt = (uint)(_endSelected * size);
Debug.WriteLine("Select Data voice: Start {0} End {1}", startPnt, endPnt);
_yourWav.SelectedWave(startPnt, endPnt);
_yourMfcc = new MFCCWrapper(_yourWav, option.TimeFrame, option.TimeShift, option.CepFilter, option.LowFreq, option.HighFreq, option.NumCeps, 4);
_yourMfcc.UserStandardization = option.UseStandardization;
_yourMfcc.Process();
_yourPitch = new PitchWrapper(_yourWav, option.PitchTimeFrame, option.PitchTimeShift, option.PitchLowFreq, option.PitchHighFreq, option.PitchType, option.DropUnPitch);
if (option.UseMedian)
{
_yourPitch.SetMedianWindowSize(option.MedianWindow);
}
_yourPitch.Process();
}
}
if (_yourMfcc != null && _yourMfcc.ProcessDone)
{
// TO DO: Process Bar
SetDataChart(FormTag.YOUR_WAVE);
SetDataChart(FormTag.YOUR_MFCC);
SetDataChart(FormTag.YOUR_FREQ);
SetDataChart(FormTag.YOUR_DOUBLE);
SetDataChart(FormTag.YOUR_DETAL);
SetDataChart(FormTag.YOUR_PITCH);
}
e.Result = reg;
if (reg)
{
if (_yourMfcc != null)
{
List <List <double> > data = null;
switch (VCContext.Instance.MFCCOptions.TrainCofficientType)
{
case 0:
data = _yourMfcc.Mfcc;
break;
case 1:
data = _yourMfcc.DetalMfcc;
break;
case 2:
data = _yourMfcc.DoubleDetalMfcc;
break;
}
Action act = new Action(() =>
{
reg_lb.Text = _trainTask.Reg(data);
});
Invoke(act);
}
}
}
private void showWaveSound()
{
_yourWav = new WavFileWrapper(_yourPath);
if (_yourWav.Load())
{
waveViewer.WaveData = _yourWav.FullData;
waveViewer.FitToScreen();
vadVoice();
if (_regMode)
{
uint size = (uint)_yourWav.FullData.Count;
uint begin = (uint)(_startSelected * size);
uint end = (uint)(_endSelected * size);
_yourWav.NormalizeWave(1.0f);
option = VCContext.Instance.MFCCOptions;
LogUtil.Info("Load Wave: {0} -- OK\n", _yourPath);
if (option.ShiftSampleToZero)
{
LogUtil.Info("Shift Sample To Zero: -- -- OK\n");
_yourWav.ShifToZero();
}
Debug.WriteLine("Select Data voice: Start {0} End {1}", begin, end);
_yourWav.SelectedWave(begin, end);
_yourMfcc = new MFCCWrapper(_yourWav, option.TimeFrame, option.TimeShift, option.CepFilter, option.LowFreq, option.HighFreq, option.NumCeps, 4);
_yourMfcc.UserStandardization = option.UseStandardization;
if (_yourMfcc != null && _yourMfcc.Process())
{
List <List <double> > data = null;
switch (VCContext.Instance.MFCCOptions.TrainCofficientType)
{
case 0:
data = _yourMfcc.Mfcc;
break;
case 1:
data = _yourMfcc.DetalMfcc;
break;
case 2:
data = _yourMfcc.DoubleDetalMfcc;
break;
}
Action act = new Action(() =>
{
reg_lb.Text = _trainTask.Reg(data);
});
Invoke(act);
}
else
{
MessageBox.Show(" Cant Extraction file {0}\n", _yourPath);
}
}
}
}
