/// <summary>
/// Creates a temporary audio file from the next audio chunk in the queue and writes the
/// file name to the streaming speech-to-text process so that it can send its data to the server.
/// </summary>
IEnumerator SaveAndSendNextChunk()
{
// Save recorded audio to a WAV file and convert it to RAW format.
string wavAudioFilePath = SavWav.Save(m_TempAudioComponent.TempAudioRelativePath(), m_AudioChunksQueue.Dequeue());
string rawAudioFilePath = IOUtilities.MakeFilePathUnique(Path.ChangeExtension(wavAudioFilePath, "raw"));
var audioConversionJob = new SoXAudioConversionJob(wavAudioFilePath, rawAudioFilePath, 16000, 16, 1,
AudioEncoding.EncodingType.SignedInteger, Endianness.EndiannessType.Little);
audioConversionJob.Start();
yield return(StartCoroutine(audioConversionJob.WaitFor()));
if (audioConversionJob.ErrorMessage != null)
{
if (m_OnError != null)
{
m_OnError(audioConversionJob.ErrorMessage);
}
yield break;
}
m_StreamingSpeechToTextProcess.StandardInput.WriteLine(rawAudioFilePath);
}
/// <summary>
/// Translates speech to text by making a request to the speech-to-text API.
/// </summary>
protected override IEnumerator TranslateRecordingToText()
{
m_TempAudioComponent.ClearTempAudioFiles();
// Save recorded audio to a WAV file and convert it to FLAC format.
string wavAudioFilePath = SavWav.Save(m_TempAudioComponent.TempAudioRelativePath(), AudioRecordingManager.Instance.RecordedAudio);
string flacAudioFilePath = IOUtilities.MakeFilePathUnique(Path.ChangeExtension(wavAudioFilePath, "flac"));
SmartLogger.Log(DebugFlags.GoogleNonStreamingSpeechToText, "converting audio");
var audioConversionJob = new SoXAudioConversionJob(wavAudioFilePath, flacAudioFilePath, 16000);
audioConversionJob.Start();
yield return(StartCoroutine(audioConversionJob.WaitFor()));
if (audioConversionJob.ErrorMessage != null)
{
if (m_OnError != null)
{
m_OnError(audioConversionJob.ErrorMessage);
}
yield break;
}
var request = new Request("POST", Constants.GoogleNonStreamingSpeechToTextURL +
"?" + Constants.GoogleAPIKeyParameterName + "=" + m_APIKey);
request.headers.Add("Content-Type", "application/json");
// Construct JSON request body.
JSONObject requestJSON = new JSONObject();
JSONObject requestConfig = new JSONObject();
requestConfig.AddField(Constants.GoogleRequestJSONConfigEncodingFieldKey, "FLAC");
requestConfig.AddField(Constants.GoogleRequestJSONConfigSampleRateFieldKey, "16000");
JSONObject requestAudio = new JSONObject();
requestAudio.AddField(Constants.GoogleRequestJSONAudioContentFieldKey, Convert.ToBase64String(File.ReadAllBytes(flacAudioFilePath)));
requestJSON.AddField(Constants.GoogleRequestJSONConfigFieldKey, requestConfig);
requestJSON.AddField(Constants.GoogleRequestJSONAudioFieldKey, requestAudio);
request.Text = requestJSON.ToString();
request.Send();
SmartLogger.Log(DebugFlags.GoogleNonStreamingSpeechToText, "sent request");
while (!request.isDone)
{
yield return(null);
}
// Grab the response JSON once the request is done and parse it.
var responseJSON = new JSONObject(request.response.Text, int.MaxValue);
SmartLogger.Log(DebugFlags.GoogleNonStreamingSpeechToText, responseJSON.ToString());
string errorText = GoogleSpeechToTextResponseJSONParser.GetErrorFromResponseJSON(responseJSON);
if (errorText != null)
{
if (m_OnError != null)
{
m_OnError(errorText);
}
}
SpeechToTextResult textResult;
JSONObject resultsJSON = responseJSON.GetField(Constants.GoogleResponseJSONResultsFieldKey);
if (resultsJSON != null && resultsJSON.Count > 0)
{
JSONObject resultJSON = resultsJSON[0];
textResult = GoogleSpeechToTextResponseJSONParser.GetTextResultFromResultJSON(resultJSON);
}
else
{
textResult = GoogleSpeechToTextResponseJSONParser.GetDefaultGoogleSpeechToTextResult();
}
if (m_OnTextResult != null)
{
m_OnTextResult(textResult);
}
m_TempAudioComponent.ClearTempAudioFiles();
}
