/// <summary>
/// Read all tts unit from Unit Linguistic FeatureVector file.
/// </summary>
/// <param name="filePath">Unit Linguistic FeatureVector file.</param>
/// <param name="language">Language of the unit file.</param>
/// <returns>Unit dictionary, indexing by (sentence id + index in sentence).</returns>
public static Dictionary<string, TtsUnit> ReadAllData(string filePath,
Language language)
{
Dictionary<string, TtsUnit> units = new Dictionary<string, TtsUnit>();
using (StreamReader sr = new StreamReader(filePath))
{
string line = null;
while (string.IsNullOrEmpty(line = sr.ReadLine()) != true)
{
TtsUnit unit = new TtsUnit(language);
string[] items = line.Split(new char[] { ' ' }, StringSplitOptions.RemoveEmptyEntries);
unit.Feature = new TtsUnitFeature();
unit.Feature.Parse(items, 2);
unit.MetaUnit = new TtsMetaUnit(language);
unit.MetaUnit.Name = items[items.Length - 1];
string key = items[0] + " " + items[1];
units.Add(key, unit);
}
}
return units;
}
/// <summary>
/// Calculate PosInSyllable feature for a given unit
/// Change it to public for code re-use in script sentence.
/// </summary>
/// <param name="preUnit">Previous unit of target unit to calculate.</param>
/// <param name="unit">Target unit to calculate.</param>
/// <returns>PosInSyllable feature.</returns>
public static PosInSyllable CalculatePosInSyllable(TtsUnit preUnit, TtsUnit unit)
{
PosInSyllable pis = PosInSyllable.Coda;
if (unit.Feature.PosInSyllable == PosInSyllable.Onset)
{
pis = PosInSyllable.Onset;
if (preUnit != null && (int)preUnit.TtsBreak <= (int)TtsBreak.Phone)
{
pis = PosInSyllable.OnsetNext;
}
}
else if (unit.Feature.PosInSyllable == PosInSyllable.OnsetNext)
{
pis = PosInSyllable.OnsetNext;
}
else if (unit.Feature.PosInSyllable == PosInSyllable.Coda)
{
pis = PosInSyllable.Coda;
if ((int)unit.TtsBreak <= (int)TtsBreak.Phone)
{
pis = PosInSyllable.CodaNext;
}
}
else if (unit.Feature.PosInSyllable == PosInSyllable.CodaNext)
{
pis = PosInSyllable.CodaNext;
}
else if (unit.Feature.PosInSyllable == PosInSyllable.NucleusInV ||
unit.Feature.PosInSyllable == PosInSyllable.NucleusInVC ||
unit.Feature.PosInSyllable == PosInSyllable.NucleusInCV ||
unit.Feature.PosInSyllable == PosInSyllable.NucleusInCVC)
{
if (preUnit != null && (int)preUnit.TtsBreak <= (int)TtsBreak.Phone)
{
if ((int)unit.TtsBreak <= (int)TtsBreak.Phone)
{
pis = PosInSyllable.NucleusInCVC;
}
else
{
pis = PosInSyllable.NucleusInCV;
}
}
else
{
if ((int)unit.TtsBreak <= (int)TtsBreak.Phone)
{
pis = PosInSyllable.NucleusInVC;
}
else
{
pis = PosInSyllable.NucleusInV;
}
}
}
if (unit.MetaUnit.Special)
{
pis = PosInSyllable.Onset;
}
return pis;
}
/// <summary>
/// Build unit collection for a given syllable.
/// </summary>
/// <param name="syllable">Syllable to process.</param>
/// <returns>Unit collection.</returns>
private Collection<TtsUnit> BuildUnits(ScriptSyllable syllable)
{
string syllableText = Core.Pronunciation.CleanDecorate(syllable.Text.Trim());
string[] slices = PronunciationSeparator.SplitSlices(syllableText);
PosInSyllable[] pis = EstimatePosInSyllable(slices);
Collection<TtsUnit> units = new Collection<TtsUnit>();
int vowelPhoneCount = 0;
for (int sliceIndex = 0; sliceIndex < slices.Length; sliceIndex++)
{
string slice = slices[sliceIndex].Trim();
if (string.IsNullOrEmpty(slice))
{
continue;
}
TtsUnit unit = new TtsUnit(Language);
// break level
unit.TtsBreak = (sliceIndex == slices.Length - 1) ? syllable.TtsBreak : TtsBreak.Phone;
// pos in syllable
unit.Feature.PosInSyllable = pis[sliceIndex];
// NONE: punctuation type
// emphasis
unit.Feature.TtsEmphasis = syllable.TtsEmphasis;
// stress mark
unit.Feature.TtsStress = syllable.Stress;
// fill unit name
// remove stress mark and replace white space with '+' for unit name
unit.MetaUnit.Name = Regex.Replace(slice, " +", @"+");
unit.MetaUnit.Language = unit.Language;
Phoneme phoneme = Localor.GetPhoneme(unit.Language);
foreach (TtsMetaPhone phone in unit.MetaUnit.Phones)
{
if (phoneme.TtsVowelPhones.IndexOf(phone.Name) >= 0)
{
vowelPhoneCount++;
}
}
units.Add(unit);
}
if (vowelPhoneCount > MaxVowelCountInSyllable)
{
string message = string.Format(CultureInfo.InvariantCulture,
"There are more than {0} vowel phone in this syllable [{1}], which is supposed to contain no more than one vowel phone",
MaxVowelCountInSyllable, syllable.Text);
throw new InvalidDataException(message);
}
return units;
}
/// <summary>
/// Save information of one unit into log file.
/// </summary>
/// <param name="writer">Stream writer to save the information.</param>
/// <param name="index">Index of the unit to save.</param>
/// <param name="expectedUnit">Expected unit from front-end.</param>
/// <param name="selectedNode">Selected node through unit selection.</param>
/// <returns>Next unit index.</returns>
private static int SaveUnit(StreamWriter writer, int index,
TtsUnit expectedUnit, CostNode selectedNode)
{
writer.Write(index.ToString(CultureInfo.InvariantCulture) + " ");
++index;
writer.Write(index.ToString(CultureInfo.InvariantCulture) + " ");
writer.Write(expectedUnit.MetaUnit.Id.ToString(CultureInfo.InvariantCulture) + " ");
writer.Write(selectedNode.WaveUnit.SampleOffset.ToString(CultureInfo.InvariantCulture) + " ");
writer.Write(selectedNode.WaveUnit.SampleLength.ToString(CultureInfo.InvariantCulture) + " ");
TtsUnitFeature selFea = selectedNode.WaveUnit.Features;
if (selFea != null)
{
StringBuilder builder = new StringBuilder();
builder.AppendFormat(CultureInfo.InvariantCulture,
"{0}/ {1} ",
(int)selFea.PosInSentence, (int)expectedUnit.Feature.PosInSentence);
builder.AppendFormat(CultureInfo.InvariantCulture,
"{0}/{1} ",
(int)selFea.PosInWord, (int)expectedUnit.Feature.PosInWord);
builder.AppendFormat(CultureInfo.InvariantCulture,
"{0}/{1} ",
(int)selFea.PosInSyllable, (int)expectedUnit.Feature.PosInSyllable);
builder.AppendFormat(CultureInfo.InvariantCulture,
"{0}/{1} ",
(int)selFea.LeftContextPhone, (int)expectedUnit.Feature.LeftContextPhone);
builder.AppendFormat(CultureInfo.InvariantCulture,
"{0}/{1} ",
(int)selFea.RightContextPhone, (int)expectedUnit.Feature.RightContextPhone);
builder.AppendFormat(CultureInfo.InvariantCulture,
"{0}/{1} ",
(int)selFea.LeftContextTone, (int)expectedUnit.Feature.LeftContextTone);
builder.AppendFormat(CultureInfo.InvariantCulture,
"{0}/{1} ",
(int)selFea.RightContextTone, (int)expectedUnit.Feature.RightContextTone);
builder.AppendFormat(CultureInfo.InvariantCulture,
"{0}/ {1} ",
(int)selFea.TtsStress, (int)expectedUnit.Feature.TtsStress);
builder.AppendFormat(CultureInfo.InvariantCulture,
"{0}/{1} ",
(int)selFea.TtsEmphasis, (int)expectedUnit.Feature.TtsEmphasis);
builder.AppendFormat(CultureInfo.InvariantCulture,
"{0}/{1} ",
(int)selFea.TtsWordTone, (int)expectedUnit.Feature.TtsWordTone);
writer.Write(builder.ToString());
}
writer.Write("\r\n");
return index;
}
private static TtsUnit ParseTtsUnit(string line, Language language)
{
TtsUnit unit = new TtsUnit(language);
string[] items = line.Split(new char[] { ' ' },
StringSplitOptions.RemoveEmptyEntries);
unit.Feature = new TtsUnitFeature();
unit.Feature.Parse(items, 0);
unit.MetaUnit = ParseMetaUnitForViterbi(line);
return unit;
}
/// <summary>
/// Build units from syllable.
/// </summary>
/// <param name="syllable">Syllable.</param>
/// <param name="sliceData">Slice data.</param>
/// <param name="pronunciationSeparator">Pronunciation separator.</param>
/// <returns>Units.</returns>
private static Collection<TtsUnit> BuildUnitsForSyllable(ScriptSyllable syllable,
SliceData sliceData, PronunciationSeparator pronunciationSeparator)
{
Debug.Assert(syllable != null);
Debug.Assert(sliceData != null);
string syllableText = Core.Pronunciation.RemoveStress(syllable.Text.Trim());
string[] slices = pronunciationSeparator.SplitSlices(syllableText);
PosInSyllable[] pis = EstimatePosInSyllable(slices, sliceData);
Collection<TtsUnit> units = new Collection<TtsUnit>();
for (int sliceIndex = 0; sliceIndex < slices.Length; sliceIndex++)
{
string slice = slices[sliceIndex].Trim();
if (string.IsNullOrEmpty(slice))
{
continue;
}
TtsUnit unit = new TtsUnit(sliceData.Language);
// break level
unit.TtsBreak = (sliceIndex == slices.Length - 1) ? syllable.TtsBreak : TtsBreak.Phone;
// pos in syllable
unit.Feature.PosInSyllable = pis[sliceIndex];
// NONE: punctuation type
// emphasis
unit.Feature.TtsEmphasis = syllable.TtsEmphasis;
// stress mark
unit.Feature.TtsStress = syllable.Stress;
// fill unit name
// remove stress mark and replace white space with '+' for unit name
unit.MetaUnit.Name = Regex.Replace(slice, " +", @"+");
unit.MetaUnit.Language = unit.Language;
units.Add(unit);
}
return units;
}
/// <summary>
/// Read and parse unit data from the XML text reader to utterance.
/// </summary>
/// <param name="reader">XML text reader to read data from.</param>
/// <param name="utterance">Target utterance to save result units.</param>
private static void ParseUnit(XmlTextReader reader, TtsUtterance utterance)
{
TtsUnit unit = new TtsUnit(utterance.Script.Language);
unit.MetaUnit.Name = reader.GetAttribute("val");
if (reader.GetAttribute("iSyll") != null)
{
unit.Feature.PosInSyllable =
(PosInSyllable)Enum.Parse(typeof(PosInSyllable), reader.GetAttribute("iSyll"));
}
if (reader.GetAttribute("iWord") != null)
{
unit.Feature.PosInWord =
(PosInWord)Enum.Parse(typeof(PosInWord), reader.GetAttribute("iWord"));
}
if (reader.GetAttribute("iSent") != null)
{
unit.Feature.PosInSentence =
(PosInSentence)Enum.Parse(typeof(PosInSentence), reader.GetAttribute("iSent"));
}
Phoneme phoneme = Localor.GetPhoneme(utterance.Script.Language, utterance.Script.Engine);
unit.Feature.LeftContextPhone = phoneme.TtsPhone2Id(reader.GetAttribute("lPh"));
unit.Feature.RightContextPhone = phoneme.TtsPhone2Id(reader.GetAttribute("rPh"));
if (reader.GetAttribute("em") != null)
{
unit.Feature.TtsEmphasis =
(TtsEmphasis)Enum.Parse(typeof(TtsEmphasis),
reader.GetAttribute("em"));
}
if (reader.GetAttribute("st") != null)
{
unit.Feature.TtsStress =
(TtsStress)Enum.Parse(typeof(TtsStress), reader.GetAttribute("st"));
}
utterance.Script.Units.Add(unit);
}
/// <summary>
/// Convert phones of TTS unit to SR phones.
/// </summary>
/// <param name="unit">TtsUnit to be processed.</param>
/// <returns>SR phone array.</returns>
private string[] ConvertToSrPhone(TtsUnit unit)
{
List<string> retPhones = new List<string>();
// Go through each phone in this unit
foreach (TtsMetaPhone phone in unit.MetaUnit.Phones)
{
// Map phone to Speech Recognition phone(s)
string[] srPhones = Phoneme.Tts2SrPhones(phone.Name);
if (srPhones == null)
{
string message = string.Format(CultureInfo.InvariantCulture,
"Invalid TTS phone[{0}], which can not be converted to Speech Recognition Phone.",
phone.Name);
throw new InvalidDataException(message);
}
retPhones.AddRange(srPhones);
}
return retPhones.ToArray();
}
/// <summary>
/// Build one unit for mono MLF file.
/// </summary>
/// <param name="writer">Text writer to save MLF file.</param>
/// <param name="unit">Unit.</param>
private void BuildMonoMlf(TextWriter writer, TtsUnit unit)
{
string[] srPhones = ConvertToSrPhone(unit);
foreach (string srPhone in srPhones)
{
if (writer != null)
{
writer.WriteLine(srPhone);
}
}
}
/// <summary>
/// Build mlf from unit.
/// </summary>
/// <param name="unit">Unit.</param>
/// <param name="item">Script item.</param>
/// <param name="sw">Text writer.</param>
/// <param name="writeToFile">Whethe writing to file.</param>
/// <param name="phoneme">Phoneme.</param>
/// <returns>Errors.</returns>
private static ErrorSet BuildMonoMlf(TtsUnit unit, ScriptItem item, StreamWriter sw,
bool writeToFile, Phoneme phoneme)
{
Debug.Assert(unit != null);
Debug.Assert(item != null);
ErrorSet errors = new ErrorSet();
List<string> allPhones = new List<string>();
foreach (TtsMetaPhone phone in unit.MetaUnit.Phones)
{
string[] srPhones = phoneme.Tts2SrPhones(phone.Name);
if (srPhones == null)
{
string message = string.Format(CultureInfo.InvariantCulture,
"Invalid TTS phone[{0}], which can not be converted to Speech Recognition Phone.",
phone.Name);
errors.Add(ScriptError.OtherErrors, item.Id, message);
continue;
}
allPhones.AddRange(srPhones);
}
if (writeToFile)
{
foreach (string phone in allPhones)
{
sw.WriteLine(phone);
}
}
return errors;
}