I'm setting up an application for Android using Unity, where I want to apply speech to text functionalities, for that purpose I'm using the SDK of Azure's Speech-Service. The problem is that when I say a word, the recognizer tries to guess the word instead of even considering what I was saying, for example when saying pseudowords.
That's why I wanted to implement the Phrase List, which it's suppose to help you improving the accuracy of speech recognition.
So I used the PhraseListGrammar Class, and it was working well when using "en-US" as the SpeechRecognitionLanguage but when I try to do the same when using "en-ES" it just omits all the phrases added and the problem persist for all the other languages, perhaps there's something I'm doing wrong.
Yesterday I was trying this and it was working like a charm when using "en-US", now it's not working so there must be something missing.
var config = SpeechConfig.FromSubscription("mysubscriptionkey", "westus2");
// config.SpeechRecognitionLanguage = "en-US";
config.SpeechRecognitionLanguage = "es-ES";
config.OutputFormat = OutputFormat.Detailed;
using (var recognizer = new SpeechRecognizer(config))
{
lock (threadLocker)
{
waitingForReco = true;
}
PhraseListGrammar phraseList = PhraseListGrammar.FromRecognizer(recognizer);
phraseList.AddPhrase("Mover un badodo");
phraseList.AddPhrase("lolololalala");
phraseList.AddPhrase("badodo");
var result = await recognizer.RecognizeOnceAsync().ConfigureAwait(false);
if (result.Reason == ResultReason.RecognizedSpeech)
{
consoleMessage = JsonConvert.SerializeObject(result.Best());
newMessage = JsonConvert.SerializeObject(result);
}
else if (result.Reason == ResultReason.NoMatch)
{
newMessage = "NOMATCH: Speech could not be recognized.";
}
else if (result.Reason == ResultReason.Canceled)
{
var cancellation = CancellationDetails.FromResult(result);
newMessage = $"CANCELED: Reason={cancellation.Reason} ErrorDetails={cancellation.ErrorDetails}";
}
lock (threadLocker)
{
cmessage = consoleMessage;
message = newMessage;
waitingForReco = false;
}
}
I expected it to improve the recognition of the added keywords, but It's not working.