Starting Speech Recognition

To initiate speech recognition, your code should call StartContinuousRecognition after providing one or more grammars and one or more handlers for Recognition events, such as ContinuousRecognitionResultAvailable. The SpeechRecognizer will then attempt to match utterances from the user against the provided grammars, delivering SpeechRecognitionResult objects to the consuming code via the provided handlers on successful recognition, or raising errors through other events such as ContinuousRecognitionStatusChanged.

Instead of following an eventing model, call RecognizeAsync to listen for the next single phrase. Each call to RecognizeAsync attempts to match one utterance from a user to an enabled grammar in a speech recognizer’s grammar set. If any of the following occur, the asynchronous recognition operation will be finalized. The result of the asynchronous operation is a SpeechRecognitionResult object.

• Recognition was successful. The user’s utterance was matched to a phrase in an enabled grammar, and the SpeechRecognitionResult object has details.
• Recognition was not successful. The user’s utterance could not be matched to a phrase in an enabled grammar.
• A recognition timeout interval expired. If the interval for one of the InitialSilenceTimeout, BabbleTimeout, or EndSilenceTimeout expires, the recognition operation is finalized without a match to an enabled grammar. For more information about timeouts, see Customizing speech recognizer settings in NUI Speech.
• A recognition error occurred, such as a failure to load the provided grammar. See Speech recognition errors for information about recognition error codes.

When recognition is started, either through StartContinuousRecognition or RecognizeAsync, the recognizer will automatically load all not-yet-loaded grammars. To avoid delays when starting recognition, it is usually advisable to request a grammar load operation before you need to begin Recognition. Preloading the grammars gives the advantage of less perceived delay to start a recognition operation.

You may instantiate only a single recognizer at one time. Subsequent attempts to instantiate another speech recognizer will result in HRESULT_FROM_WIN32(ERROR_NO_SYSTEM_RESOURCES).

Speech and Application Focus

Speech is a focus-sensitive input. When a title goes out of focus, the following occurs:

• Audio processing is suspended until focus is regained.
• The ContinuousRecognition_Stopped event will be sent to the title if there was a continuous recognition in progress.
• RecognizeAsync and PreloadGrammarsAsync operations are cancelled.
• If the user stops speaking, the audio queue is emptied.

When the title regains focus, the following occurs:

• If there was a continuous recognition in progress before the recognizer went out of focus, and the recognizer was not stopped by the title while out of focus, a ContinuousRecognition_InProgress event will be sent, and audio processing/recognition will be automatically resumed.
• The state loaded is from when the title went out of focus.
• If speech was paused, the title must call ResumeContinousRecognition to receive recognition events from the audio stream.