/// <inheritdoc/>
public async Task <AudioClip> ConvertTextToSpeech(string text)
{
string filename = Configuration.GetUniqueTextToSpeechFilename(text);
string filePath = GetPathToFile(filename);
AudioClip audioClip;
if (IsFileCached(filePath))
{
byte[] bytes = GetCachedFile(filePath);
float[] sound = TextToSpeechUtils.ShortsInByteArrayToFloats(bytes);
audioClip = AudioClip.Create(text, channels: 1, frequency: 48000, lengthSamples: sound.Length, stream: false);
audioClip.SetData(sound, 0);
}
else
{
audioClip = await FallbackProvider.ConvertTextToSpeech(text);
if (Configuration.SaveAudioFilesToStreamingAssets)
{
CacheAudio(audioClip, filePath);
}
}
if (audioClip == null)
{
throw new CouldNotLoadAudioFileException("AudioClip is null.");
}
return(audioClip);
}
private float[] Synthesize(string text, string outputPath, string language, string voice)
{
// Despite the fact that SpVoice.AudioOutputStream accepts values of type ISpeechBaseStream,
// the single type of a stream that is actually working is a SpFileStream.
SpFileStream stream = PrepareFileStreamToWrite(outputPath);
SpVoice synthesizer = new SpVoice {
AudioOutputStream = stream
};
string ssmlText = string.Format(ssmlTemplate, language, voice, text);
synthesizer.Speak(ssmlText, SpeechVoiceSpeakFlags.SVSFIsXML);
synthesizer.WaitUntilDone(-1);
stream.Close();
byte[] data = File.ReadAllBytes(outputPath);
float[] sampleData = TextToSpeechUtils.ShortsInByteArrayToFloats(data);
float[] cleanData = RemoveArtifacts(sampleData);
ClearCache(outputPath);
return(cleanData);
}