/// <summary>
/// Builds and runs a speech recognition pipeline using the Azure speech recognizer. Requires a valid Cognitive Services
/// subscription key. See https://docs.microsoft.com/en-us/azure/cognitive-services/cognitive-services-apis-create-account.
/// </summary>
/// <remarks>
/// If you are getting a <see cref="System.InvalidOperationException"/> with the message 'AzureSpeechRecognizer returned
/// OnConversationError with error code: LoginFailed. Original error text: Transport error', this most likely is due to
/// an invalid subscription key. Please check your Azure portal at https://portal.azure.com and ensure that you have
/// added a subscription to the Azure Speech API on your account.
/// </remarks>
public static void RunAzureSpeech()
{
// Create the pipeline object.
using (Pipeline pipeline = Pipeline.Create())
{
// Create the AudioSource component to capture audio from the default device in 16 kHz 1-channel
// PCM format as required by both the voice activity detector and speech recognition components.
IProducer <AudioBuffer> audioInput = new AudioCapture(pipeline, new AudioCaptureConfiguration()
{
DeviceName = "plughw:0,0", Format = WaveFormat.Create16kHz1Channel16BitPcm()
});
// Perform voice activity detection using the voice activity detector component
var vad = new SimpleVoiceActivityDetector(pipeline);
audioInput.PipeTo(vad);
// Create Azure speech recognizer component
var recognizer = new AzureSpeechRecognizer(pipeline, new AzureSpeechRecognizerConfiguration()
{
SubscriptionKey = Program.azureSubscriptionKey, Region = Program.azureRegion
});
// The input audio to the Azure speech recognizer needs to be annotated with a voice activity flag.
// This can be constructed by using the Psi Join() operator to combine the audio and VAD streams.
var annotatedAudio = audioInput.Join(vad);
// Subscribe the recognizer to the annotated audio
annotatedAudio.PipeTo(recognizer);
// Partial and final speech recognition results are posted on the same stream. Here
// we use Psi's Where() operator to filter out only the final recognition results.
var finalResults = recognizer.Out.Where(result => result.IsFinal);
// Print the recognized text of the final recognition result to the console.
finalResults.Do(result => Console.WriteLine(result.Text));
// Register an event handler to catch pipeline errors
pipeline.PipelineExceptionNotHandled += Pipeline_PipelineException;
// Register an event handler to be notified when the pipeline completes
pipeline.PipelineCompleted += Pipeline_PipelineCompleted;
// Run the pipeline
pipeline.RunAsync();
// Azure speech transcribes speech to text
Console.WriteLine("Say anything");
Console.WriteLine("Press any key to exit...");
Console.ReadKey(true);
}
}
public void VoiceActivity_DetectFromFile()
{
// Initialize components and wire up pipeline.
using (var pipeline = Pipeline.Create(nameof(this.VoiceActivity_DetectFromFile)))
{
var vad = new SimpleVoiceActivityDetector(pipeline, new SimpleVoiceActivityDetectorConfiguration());
var audioInput = new WaveFileAudioSource(pipeline, "16kHz1chan.wav", null, 20);
audioInput.PipeTo(vad);
// Add results from outputs.
var results = new List <bool>();
vad.Out.Do(r => results.Add(r));
// Run pipeline and wait for completion.
pipeline.Run(null, false);
Assert.IsTrue(results.Count > 0, "No results!");
CollectionAssert.AreEqual(Enumerable.Repeat(false, 29).ToList(), results.GetRange(0, 29), "Initial silence detection failed!");
CollectionAssert.AreEqual(Enumerable.Repeat(true, 182).ToList(), results.GetRange(29, 182), "Voice activity detection failed!");
CollectionAssert.AreEqual(Enumerable.Repeat(false, 62).ToList(), results.GetRange(211, 62), "Trailing silence detection failed!");
}
}
