public void LogInfo()
{
this.LogInfo("Loading tied-state acoustic model from: " + Location);
MeansPool.LogInfo();
VariancePool.LogInfo();
MatrixPool.LogInfo();
SenonePool.LogInfo();
if (MeansTransformationMatrixPool != null)
{
MeansTransformationMatrixPool.LogInfo();
}
if (MeansTransformationVectorPool != null)
{
MeansTransformationVectorPool.LogInfo();
}
if (VarianceTransformationMatrixPool != null)
{
VarianceTransformationMatrixPool.LogInfo();
}
if (VarianceTransformationVectorPool != null)
{
VarianceTransformationVectorPool.LogInfo();
}
MixtureWeightsPool.LogInfo();
SenonePool.LogInfo();
this.LogInfo("Context Independent Unit Entries: "
+ ContextIndependentUnits.Count);
HmmManager.LogInfo();
}
private Pool <ISenone> CreateTiedSenonePool(float distFloor, float varianceFloor)
{
var pool = new Pool <ISenone>("senones");
var numMeans = MeansPool.Size;
var numVariances = VariancePool.Size;
var numGaussiansPerState = MixtureWeightsPool.GauPerState;
var numSenones = MixtureWeightsPool.StatesNum;
var numStreams = MixtureWeightsPool.StreamsNum;
this.LogInfo("Senones " + numSenones);
this.LogInfo("Gaussians Per State " + numGaussiansPerState);
this.LogInfo("Means " + numMeans);
this.LogInfo("Variances " + numVariances);
Debug.Assert(numGaussiansPerState > 0);
Debug.Assert(numVariances == _numBase * numGaussiansPerState * numStreams);
Debug.Assert(numMeans == _numBase * numGaussiansPerState * numStreams);
var meansTransformationMatrix = MeansTransformationMatrixPool == null ? null
: MeansTransformationMatrixPool.Get(0);
var meansTransformationVector = MeansTransformationVectorPool == null ? null
: MeansTransformationVectorPool.Get(0);
var varianceTransformationMatrix = VarianceTransformationMatrixPool == null ? null
: VarianceTransformationMatrixPool.Get(0);
var varianceTransformationVector = VarianceTransformationVectorPool == null ? null
: VarianceTransformationVectorPool.Get(0);
_phoneticTiedMixtures = new MixtureComponentSet[_numBase];
for (var i = 0; i < _numBase; i++)
{
var mixtureComponents = new List <PrunableMixtureComponent[]>();
for (var j = 0; j < numStreams; j++)
{
var featMixtureComponents = new PrunableMixtureComponent[numGaussiansPerState];
for (var k = 0; k < numGaussiansPerState; k++)
{
var whichGaussian = i * numGaussiansPerState * numStreams + j * numGaussiansPerState + k;
featMixtureComponents[k] = new PrunableMixtureComponent(
MeansPool.Get(whichGaussian),
meansTransformationMatrix, meansTransformationVector,
VariancePool.Get(whichGaussian),
varianceTransformationMatrix,
varianceTransformationVector, distFloor, varianceFloor, k);
}
mixtureComponents.Add(featMixtureComponents);
}
_phoneticTiedMixtures[i] = new MixtureComponentSet(mixtureComponents, _topGauNum);
}
for (var i = 0; i < numSenones; i++)
{
ISenone senone = new SetBasedGaussianMixture(MixtureWeightsPool, _phoneticTiedMixtures[Senone2Ci[i]], i);
pool.Put(i, senone);
}
return(pool);
}
/**
* /// Creates the senone pool from the rest of the pools.
* ///
* /// @param distFloor
* /// the lowest allowed score
* /// @param varianceFloor
* /// the lowest allowed variance
* /// @return the senone pool
*/
public Pool <ISenone> CreateSenonePool(float distFloor, float varianceFloor)
{
var pool = new Pool <ISenone>("senones");
var numMeans = MeansPool.Size;
var numVariances = VariancePool.Size;
var numGaussiansPerSenone = MixtureWeightsPool.GauPerState;
var numSenones = MixtureWeightsPool.StatesNum;
var numStreams = MixtureWeightsPool.StreamsNum;
var whichGaussian = 0;
this.LogInfo("Senones " + numSenones);
this.LogInfo("Gaussians Per Senone " + numGaussiansPerSenone);
this.LogInfo("Means " + numMeans);
this.LogInfo("Variances " + numVariances);
Debug.Assert(numGaussiansPerSenone > 0);
Debug.Assert(numVariances == numSenones * numGaussiansPerSenone);
Debug.Assert(numMeans == numSenones * numGaussiansPerSenone);
var meansTransformationMatrix = MeansTransformationMatrixPool == null ? null
: MeansTransformationMatrixPool.Get(0);
var meansTransformationVector = MeansTransformationVectorPool == null ? null
: MeansTransformationVectorPool.Get(0);
var varianceTransformationMatrix = VarianceTransformationMatrixPool == null ? null
: VarianceTransformationMatrixPool.Get(0);
var varianceTransformationVector = VarianceTransformationVectorPool == null ? null
: VarianceTransformationVectorPool.Get(0);
for (var i = 0; i < numSenones; i++)
{
var mixtureComponents = new MixtureComponent[numGaussiansPerSenone
* numStreams];
for (var j = 0; j < numGaussiansPerSenone; j++)
{
mixtureComponents[j] = new MixtureComponent(
MeansPool.Get(whichGaussian),
meansTransformationMatrix, meansTransformationVector,
VariancePool.Get(whichGaussian),
varianceTransformationMatrix,
varianceTransformationVector, distFloor, varianceFloor);
whichGaussian++;
}
ISenone senone = new GaussianMixture(MixtureWeightsPool, mixtureComponents, i);
pool.Put(i, senone);
}
return(pool);
}
public void Update(Transform transform, ClusteredDensityFileData clusters)
{
for (var index = 0; index < MeansPool.Size; index++)
{
var transformClass = clusters.GetClassIndex(index);
var tmean = new float[VectorLength[0]];
var mean = MeansPool.Get(index);
for (var l = 0; l < VectorLength[0]; l++)
{
tmean[l] = 0;
for (var m = 0; m < VectorLength[0]; m++)
{
tmean[l] += transform.As[transformClass][0][l][m]
* mean[m];
}
tmean[l] += transform.Bs[transformClass][0][l];
}
Array.Copy(tmean, 0, mean, 0, tmean.Length);
}
}