/// <summary>
/// Returns the next Data object, which is the mel cepstrum of the input frame. However, it can also be other Data objects like DataStartSignal.
/// </summary>
/// <returns>
/// The next available Data object, returns null if no Data object is available.
/// </returns>
/// <exception cref="System.Exception">
/// IOException closing cepstrum stream
/// or
/// IOException reading from cepstrum stream
/// </exception>
public override IData GetData()
{
IData data;
if (_curPoint == -1)
{
data = new DataStartSignal(_sampleRate);
_curPoint++;
}
else if (_curPoint == _numPoints)
{
if (_numPoints > 0)
{
_firstSampleNumber =
(_firstSampleNumber - _frameShift + _frameSize - 1);
}
// send a DataEndSignal
var numberFrames = _curPoint / _cepstrumLength;
var totalSamples = (numberFrames - 1) * _frameShift + _frameSize;
var duration = (long)
((totalSamples / (double)_sampleRate) * 1000.0);
data = new DataEndSignal(duration);
try {
if (_binary)
{
_binaryStream.Close();
}
else
{
_est.Close();
}
_curPoint++;
} catch (IOException ioe) {
throw new Exception("IOException closing cepstrum stream", ioe);
}
}
else if (_curPoint > _numPoints)
{
data = null;
}
else
{
var vectorData = new double[_cepstrumLength];
for (var i = 0; i < _cepstrumLength; i++)
{
try {
if (_binary)
{
if (_bigEndian)
{
vectorData[i] = _binaryStream.ReadFloat();
}
else
{
vectorData[i] = Utilities.ReadLittleEndianFloat(_binaryStream);
}
}
else
{
vectorData[i] = _est.GetFloat("cepstrum data");
}
_curPoint++;
} catch (IOException ioe) {
throw new Exception("IOException reading from cepstrum stream", ioe);
}
}
// System.out.println("Read: " + curPoint);
data = new DoubleData
(vectorData, _sampleRate, _firstSampleNumber);
_firstSampleNumber += _frameShift;
// System.out.println(data);
}
return(data);
}
/// <summary>
/// Creates the grammar.
/// </summary>
/// <returns>The initial node for the grammar.</returns>
protected override GrammarNode CreateGrammar()
{
GrammarNode initialNode = null;
GrammarNode finalNode = null;
// first pass create the FST nodes
int maxNodeId = CreateNodes(_path);
// create the final node:
finalNode = CreateGrammarNode(++maxNodeId, IDictionary.SilenceSpelling);
finalNode.SetFinalNode(true);
// replace each word node with a pair of nodes, which
// consists of the word node and a new dummy end node, which is
// for adding null or backoff transitions
maxNodeId = ExpandWordNodes(maxNodeId);
ExtendedStreamTokenizer tok = new ExtendedStreamTokenizer(_path, true);
// Second pass, add all of the arcs
while (!tok.IsEOF())
{
String token;
tok.Skipwhite();
token = tok.GetString();
// System.out.println(token);
if (token == null)
{
break;
}
else if (token.Equals("I"))
{
Debug.Assert(initialNode == null);
int initialID = tok.GetInt("initial ID");
String nodeName = "G" + initialID;
// TODO: FlatLinguist requires the initial grammar node
// to contain a single silence. We'll do that for now,
// but once the FlatLinguist is fixed, this should be
// returned to its former method of creating an empty
// initial grammar node
// initialNode = createGrammarNode(initialID, false);
initialNode = CreateGrammarNode(initialID, IDictionary.SilenceSpelling);
_nodes.Put(nodeName, initialNode);
// optionally add a silence node
if (_addInitialSilenceNode)
{
GrammarNode silenceNode = CreateGrammarNode(++maxNodeId, IDictionary.SilenceSpelling);
initialNode.Add(silenceNode, LogMath.LogOne);
silenceNode.Add(initialNode, LogMath.LogOne);
}
}
else if (token.Equals("T"))
{
int thisID = tok.GetInt("this id");
int nextID = tok.GetInt("next id");
GrammarNode thisNode = Get(thisID);
GrammarNode nextNode = Get(nextID);
// if the source node is an FSTGrammarNode, we want
// to join the endNode to the destination node
if (HasEndNode(thisNode))
{
thisNode = GetEndNode(thisNode);
}
float lnProb = 0f; // negative natural log
String output = tok.GetString();
if (output == null || output.Equals(","))
{
// these are epsilon (meaning backoff) transitions
if (output != null && output.Equals(","))
{
tok.GetString(); // skip the word
lnProb = tok.GetFloat("probability");
}
// if the destination node has been expanded
// we actually want to add the backoff transition
// to the endNode
if (HasEndNode(nextNode))
{
nextNode = GetEndNode(nextNode);
}
}
else
{
String word = tok.GetString(); // skip words
lnProb = tok.GetFloat("probability");
if (_ignoreUnknownTransitions && word.Equals("<unknown>"))
{
continue;
}
/*
* System.out.println(nextNode + ": " + output);
*/
Debug.Assert(HasWord(nextNode));
}
thisNode.Add(nextNode, ConvertProbability(lnProb));
}
else if (token.Equals("F"))
{
int thisID = tok.GetInt("this id");
float lnProb = tok.GetFloat("probability");
GrammarNode thisNode = Get(thisID);
GrammarNode nextNode = finalNode;
if (HasEndNode(thisNode))
{
thisNode = GetEndNode(thisNode);
}
thisNode.Add(nextNode, ConvertProbability(lnProb));
}
}
tok.Close();
Debug.Assert(initialNode != null);
return(initialNode);
}
