Main facade for the KeenASR speech recognition plugin. More...

Public Member Functions
delegate void	InitializedHandler (bool status)
	Called when the ASR engine has finished initializing.

delegate void	PartialASRResultHandler (string text)
	Called during recognition with intermediate (non-final) text.

delegate void	FinalASRResultHandler (ASRResponse response)
	Called when recognition is complete with the final response. The caller receives ownership of the ASRResponse and must call Dispose() on it when done to release the native handle.

delegate void	UnwindAppAudioBeforeAudioInterruptHandler ()
	Called immediately before KeenASR unwinds its audio stack due to an audio interrupt (e.g. the app going to background). Stop any app-controlled audio playback in this handler. Do not modify the audio session or interact with the recognizer.

delegate void	RecognizerReadyToListenAfterInterruptHandler ()
	Called when the recognizer is ready to listen again after an audio interrupt (e.g. the app returning to foreground). Use this to re-enable UI elements.

Decoding Graphs
bool	DecodingGraphWithNameExists (string dgName)
	Checks whether a decoding graph with the given name exists on the device.

bool	CreateDecodingGraphFromPhrases (string dgName, string[] phrases, SpeakingTask task=SpeakingTask.Default, WordPronunciation[] alternativePronunciations=null, float spokenNoiseProbability=0.5f)
	Creates a decoding graph from the given list of phrases. The decoding graph defines the set of utterances the recognizer can detect. The graph is saved on the device and can be referenced later by name.

bool	CreateContextualDecodingGraphFromPhrases (string dgName, string[][] contextualPhrases, SpeakingTask task=SpeakingTask.Default, WordPronunciation[] alternativePronunciations=null, float spokenNoiseProbability=0.5f)
	Creates a contextual decoding graph from multiple groups of phrases. Each group represents a context that can be switched at runtime using PrepareForListeningWithContextualDecodingGraph. This is more efficient than creating separate decoding graphs when you have many contexts (e.g. pages in a book).

Listening
bool	PrepareForListeningWithDecodingGraph (string dgName, bool computeGoP=false)
	Loads a previously created decoding graph and prepares the recognizer for listening. After this call succeeds, the recognizer state transitions to ReadyToListen.

bool	PrepareForListeningWithDecodingGraphAtPath (string path, bool computeGoP=false)
	Loads a decoding graph from the specified filesystem path and prepares the recognizer for listening. After this call succeeds, the recognizer state transitions to ReadyToListen.

bool	PrepareForListeningWithContextualDecodingGraph (string dgName, int contextId, bool computeGoP=false)
	Loads a contextual decoding graph and prepares the recognizer for listening with the specified context. Use this with decoding graphs created via CreateContextualDecodingGraphFromPhrases. After this call succeeds, the recognizer state transitions to ReadyToListen.

bool	PrepareForListeningWithContextualDecodingGraphAtPath (string path, int contextId, bool computeGoP=false)
	Loads a contextual decoding graph from the specified filesystem path and prepares the recognizer for listening with the specified context. After this call succeeds, the recognizer state transitions to ReadyToListen.

bool	StartListening ()
	Starts listening for speech. The recognizer must be in ReadyToListen state (i.e. PrepareForListeningWithDecodingGraph must have been called first). Recognition results are delivered via onPartialASRResultReceived and onFinalResponseReceived events.

void	StopListening ()
	Stops the recognizer from processing incoming audio. Note that calling this method will NOT trigger the onFinalResponseReceived callback. If you need the final result, set VAD timeout thresholds (e.g. TimeoutEndSilenceForGoodMatch) to short values so that the recognizer stops naturally via VAD and delivers the result through the callback.

RecognizerState	GetRecognizerState ()
	Returns the current state of the recognizer.

float	InputLevel ()
	Returns the current input audio level in dB. Poll this method periodically to drive a UI element that gives the user visual feedback that the system is hearing them. Only meaningful while the recognizer is in Listening state.

Configuration
bool	SetVADParameter (VadParameter vadParameter, float value)
	Sets a Voice Activity Detection parameter. VAD parameters control when the recognizer automatically stops listening (e.g. after a period of silence or after reaching a maximum duration).

bool	IsEchoCancellationAvaialable ()
	Checks whether the device supports echo cancellation.

void	PerformEchoCancellation (bool value)
	Enables or disables echo cancellation. When enabled, audio played by the app through the device speaker is removed from the microphone signal, allowing recognition to work during audio playback.

Speaker Adaptation
void	AdaptToSpeakerWithName (string speakerName)
	Starts speaker adaptation for the named user. If a saved adaptation profile exists for this user, it will be loaded. Use this when you know with high confidence which user will be speaking. In shared-device environments, it is better not to use adaptation.

void	ResetSpeakerAdaptation ()
	Resets speaker adaptation to the baseline model, discarding any in-session adaptation data. Does not remove saved adaptation profiles.

void	SaveSpeakerAdaptationProfile ()
	Persists the current speaker adaptation profile so it can be reused in future sessions when AdaptToSpeakerWithName is called with the same name.

bool	RemoveAllSpeakerAdaptationProfiles ()
	Removes all saved speaker adaptation profiles from the device.

bool	RemoveSpeakerAdaptationProfiles (string speakerName)
	Removes the saved adaptation profile for the specified speaker.

Static Public Member Functions
Initialization
static bool	Initialize (string bundleName)
	Initializes the ASR engine with the named ASR bundle from StreamingAssets.

static bool	InitializeWithASRBundleAtPath (string asrBundlePath)
	Initializes the ASR engine with the ASR bundle at the given filesystem path.

static bool	Teardown ()
	Tears down the recognizer and releases all associated resources. All audio playback should be stopped before calling this method.

static void	SetLogLevel (LogLevel logLevel)
	Sets the framework logging verbosity. Can be called before Initialize().

Dashboard
static bool	StartDataUploader (string appKey)
	Starts a background uploader thread that uploads queued responses to Dashboard. Call this once during setup; responses are queued via ASRResponse.QueueForUpload().

static bool	ResumeUploader ()
	Resumes the data uploader after it has been paused.

static void	PauseUploader ()
	Pauses the data uploader. Queued files are retained and will be uploaded when the uploader is resumed.

static bool	StopUploader ()
	Stops the data uploader thread. After stopping, StartDataUploader() must be called again to resume uploads.

Properties
static KeenASR	Instance `[get]`
	Gets the shared KeenASR instance. Returns null if the SDK has not been initialized.

Events
static InitializedHandler	onInitializedReceived
	Raised when the ASR engine has finished initializing. Subscribe to this event before calling Initialize().

Events
PartialASRResultHandler	onPartialASRResultReceived
	Raised during recognition with intermediate (non-final) text as it becomes available.

FinalASRResultHandler	onFinalResponseReceived
	Raised when recognition is complete with the final response. The handler receives an ASRResponse that it must dispose when done.

UnwindAppAudioBeforeAudioInterruptHandler	onUnwindAppAudioBeforeAudioInterruptReceived
	Raised immediately before KeenASR unwinds its audio stack due to an audio interrupt. See UnwindAppAudioBeforeAudioInterruptHandler for details.

RecognizerReadyToListenAfterInterruptHandler	onRecognizerReadyToListenAfterInterruptReceived
	Raised when the recognizer is ready to listen again after an audio interrupt.

Detailed Description

Main facade for the KeenASR speech recognition plugin.

Provides a singleton interface for initializing the ASR engine, managing decoding graphs, controlling the recognizer lifecycle (listening, VAD, echo cancellation), and handling recognition results. Works on iOS and Android; uses a stub in the Unity Editor.

Typical usage:

KeenASR.onInitializedReceived += OnInit;
KeenASR.Initialize("keenA1m-nnet3chain-en-us");
 
void OnInit(bool success) {
    KeenASR.Instance.CreateDecodingGraphFromPhrases("myDG", phrases);
    KeenASR.Instance.PrepareForListeningWithDecodingGraph("myDG");
    KeenASR.Instance.onFinalResponseReceived += OnResponse;
    KeenASR.Instance.StartListening();
}
 
void OnResponse(ASRResponse response) {
    Debug.Log(response.result.cleanText);
    response.Dispose();
}

Member Function Documentation

◆ AdaptToSpeakerWithName()

void KeenResearch.KeenASR.AdaptToSpeakerWithName ( string speakerName )

inline

Starts speaker adaptation for the named user. If a saved adaptation profile exists for this user, it will be loaded. Use this when you know with high confidence which user will be speaking. In shared-device environments, it is better not to use adaptation.

Parameters

speakerName A name or pseudonym identifying the speaker.

◆ CreateContextualDecodingGraphFromPhrases()

bool KeenResearch.KeenASR.CreateContextualDecodingGraphFromPhrases	(	string	dgName,
		string	contextualPhrases[][],
		SpeakingTask	task = `SpeakingTask::Default`,
		WordPronunciation[]	alternativePronunciations = `null`,
		float	spokenNoiseProbability = `0::5f`
	)

inline

Creates a contextual decoding graph from multiple groups of phrases. Each group represents a context that can be switched at runtime using PrepareForListeningWithContextualDecodingGraph. This is more efficient than creating separate decoding graphs when you have many contexts (e.g. pages in a book).

Parameters

dgName	Name for the decoding graph (used to reference it later).
contextualPhrases	Array of phrase arrays, where each inner array is a context. The index of each inner array becomes its contextId.
task	Speaking task type (affects language model optimization).
alternativePronunciations	Optional array of alternative pronunciations for words in the phrases. Pass null if not needed.
spokenNoiseProbability	Probability of <SPOKEN_NOISE> appearing in results (0.0–1.0, default 0.5).

Returns: true if the decoding graph was created successfully, false otherwise.

◆ CreateDecodingGraphFromPhrases()

bool KeenResearch.KeenASR.CreateDecodingGraphFromPhrases	(	string	dgName,
		string[]	phrases,
		SpeakingTask	task = `SpeakingTask::Default`,
		WordPronunciation[]	alternativePronunciations = `null`,
		float	spokenNoiseProbability = `0::5f`
	)

inline

Creates a decoding graph from the given list of phrases. The decoding graph defines the set of utterances the recognizer can detect. The graph is saved on the device and can be referenced later by name.

Parameters

dgName	Name for the decoding graph (used to reference it later).
phrases	Array of phrases the user is likely to say.
task	Speaking task type (affects language model optimization).
alternativePronunciations	Optional array of alternative pronunciations for words in the phrases. Pass null if not needed.
spokenNoiseProbability	Probability of <SPOKEN_NOISE> appearing in results (0.0–1.0, default 0.5). Lower values reduce its frequency; higher values increase it.

Returns: true if the decoding graph was created successfully, false otherwise.

◆ DecodingGraphWithNameExists()

bool KeenResearch.KeenASR.DecodingGraphWithNameExists ( string dgName )

inline

Checks whether a decoding graph with the given name exists on the device.

Parameters

dgName Name of the decoding graph.

Returns: true if the decoding graph exists, false otherwise.

◆ FinalASRResultHandler()

delegate void KeenResearch.KeenASR.FinalASRResultHandler ( ASRResponse response )

Called when recognition is complete with the final response. The caller receives ownership of the ASRResponse and must call Dispose() on it when done to release the native handle.

Parameters

response The final recognition response. Caller must dispose.

◆ GetRecognizerState()

RecognizerState KeenResearch.KeenASR.GetRecognizerState ( )

inline

Returns the current state of the recognizer.

Returns: The current RecognizerState.

◆ Initialize()

static bool KeenResearch.KeenASR.Initialize ( string bundleName )

inlinestatic

Initializes the ASR engine with the named ASR bundle from StreamingAssets.

On Android this call runs asynchronously; on iOS it runs synchronously. On both platforms, the onInitializedReceived event is raised when initialization completes. Subscribe to this event before calling Initialize().

Calling this method when the SDK is already initialized returns false.

Parameters

bundleName Name of the ASR bundle directory in StreamingAssets (e.g. "keenA1m-nnet3chain-en-us").

Returns: true if initialization was started successfully, false if already initialized.

◆ InitializedHandler()

delegate void KeenResearch.KeenASR.InitializedHandler ( bool status )

Called when the ASR engine has finished initializing.

Parameters

status true if initialization succeeded, false otherwise.

◆ InitializeWithASRBundleAtPath()

static bool KeenResearch.KeenASR.InitializeWithASRBundleAtPath ( string asrBundlePath )

inlinestatic

Initializes the ASR engine with the ASR bundle at the given filesystem path.

On Android this call runs asynchronously; on iOS it runs synchronously. On both platforms, the onInitializedReceived event is raised when initialization completes. Subscribe to this event before calling this method.

Calling this method when the SDK is already initialized returns false.

Parameters

asrBundlePath Full filesystem path to the ASR bundle directory.

Returns: true if initialization was started successfully, false if already initialized.

◆ InputLevel()

float KeenResearch.KeenASR.InputLevel ( )

inline

Returns the current input audio level in dB. Poll this method periodically to drive a UI element that gives the user visual feedback that the system is hearing them. Only meaningful while the recognizer is in Listening state.

Returns: Input signal level in decibels.

◆ IsEchoCancellationAvaialable()

bool KeenResearch.KeenASR.IsEchoCancellationAvaialable ( )

inline

Checks whether the device supports echo cancellation.

Returns: true if echo cancellation is available, false otherwise.

◆ PartialASRResultHandler()

delegate void KeenResearch.KeenASR.PartialASRResultHandler ( string text )

Called during recognition with intermediate (non-final) text.

Parameters

text	The partial recognition text so far.

◆ PauseUploader()

static void KeenResearch.KeenASR.PauseUploader ( )

inlinestatic

Pauses the data uploader. Queued files are retained and will be uploaded when the uploader is resumed.

◆ PerformEchoCancellation()

void KeenResearch.KeenASR.PerformEchoCancellation ( bool value )

inline

Enables or disables echo cancellation. When enabled, audio played by the app through the device speaker is removed from the microphone signal, allowing recognition to work during audio playback.

Parameters

value true to enable echo cancellation, false to disable.

◆ PrepareForListeningWithContextualDecodingGraph()

bool KeenResearch.KeenASR.PrepareForListeningWithContextualDecodingGraph	(	string	dgName,
		int	contextId,
		bool	computeGoP = `false`
	)

inline

Loads a contextual decoding graph and prepares the recognizer for listening with the specified context. Use this with decoding graphs created via CreateContextualDecodingGraphFromPhrases. After this call succeeds, the recognizer state transitions to ReadyToListen.

Parameters

dgName	Name of a contextual decoding graph.
contextId	Index of the context to use (corresponds to the index in the contextualPhrases array used during creation).
computeGoP	If true, compute goodness of pronunciation scores at the phoneme level in the final result.

Returns: true if successful, false otherwise.

◆ PrepareForListeningWithContextualDecodingGraphAtPath()

bool KeenResearch.KeenASR.PrepareForListeningWithContextualDecodingGraphAtPath	(	string	path,
		int	contextId,
		bool	computeGoP = `false`
	)

inline

Loads a contextual decoding graph from the specified filesystem path and prepares the recognizer for listening with the specified context. After this call succeeds, the recognizer state transitions to ReadyToListen.

Parameters

path	Full filesystem path to the contextual decoding graph directory.
contextId	Index of the context to use.
computeGoP	If true, compute goodness of pronunciation scores at the phoneme level in the final result.

Returns: true if successful, false otherwise.

◆ PrepareForListeningWithDecodingGraph()

bool KeenResearch.KeenASR.PrepareForListeningWithDecodingGraph	(	string	dgName,
		bool	computeGoP = `false`
	)

inline

Loads a previously created decoding graph and prepares the recognizer for listening. After this call succeeds, the recognizer state transitions to ReadyToListen.

Parameters

dgName	Name of a decoding graph created with CreateDecodingGraphFromPhrases.
computeGoP	If true, compute goodness of pronunciation scores at the phoneme level in the final result.

Returns: true if successful, false otherwise.

◆ PrepareForListeningWithDecodingGraphAtPath()

bool KeenResearch.KeenASR.PrepareForListeningWithDecodingGraphAtPath	(	string	path,
		bool	computeGoP = `false`
	)

inline

Loads a decoding graph from the specified filesystem path and prepares the recognizer for listening. After this call succeeds, the recognizer state transitions to ReadyToListen.

Parameters

path	Full filesystem path to the decoding graph directory.
computeGoP	If true, compute goodness of pronunciation scores at the phoneme level in the final result.

Returns: true if successful, false otherwise.

◆ RecognizerReadyToListenAfterInterruptHandler()

delegate void KeenResearch.KeenASR.RecognizerReadyToListenAfterInterruptHandler ( )

Called when the recognizer is ready to listen again after an audio interrupt (e.g. the app returning to foreground). Use this to re-enable UI elements.

◆ RemoveAllSpeakerAdaptationProfiles()

bool KeenResearch.KeenASR.RemoveAllSpeakerAdaptationProfiles ( )

inline

Removes all saved speaker adaptation profiles from the device.

Returns: true if all profiles were removed, false otherwise.

◆ RemoveSpeakerAdaptationProfiles()

bool KeenResearch.KeenASR.RemoveSpeakerAdaptationProfiles ( string speakerName )

inline

Removes the saved adaptation profile for the specified speaker.

Parameters

speakerName Name of the speaker whose profile should be removed.

Returns: true if the profile was removed, false otherwise.

◆ ResetSpeakerAdaptation()

void KeenResearch.KeenASR.ResetSpeakerAdaptation ( )

inline

Resets speaker adaptation to the baseline model, discarding any in-session adaptation data. Does not remove saved adaptation profiles.

◆ ResumeUploader()

static bool KeenResearch.KeenASR.ResumeUploader ( )

inlinestatic

Resumes the data uploader after it has been paused.

Returns: true if the uploader was resumed successfully, false otherwise.

◆ SaveSpeakerAdaptationProfile()

void KeenResearch.KeenASR.SaveSpeakerAdaptationProfile ( )

inline

Persists the current speaker adaptation profile so it can be reused in future sessions when AdaptToSpeakerWithName is called with the same name.

◆ SetLogLevel()

static void KeenResearch.KeenASR.SetLogLevel ( LogLevel logLevel )

inlinestatic

Sets the framework logging verbosity. Can be called before Initialize().

Parameters

logLevel The desired log level.

◆ SetVADParameter()

bool KeenResearch.KeenASR.SetVADParameter	(	VadParameter	vadParameter,
		float	value
	)

inline

Sets a Voice Activity Detection parameter. VAD parameters control when the recognizer automatically stops listening (e.g. after a period of silence or after reaching a maximum duration).

Parameters

vadParameter	The VAD parameter to set.
value	Timeout value in seconds.

Returns: true if the parameter was set successfully, false otherwise.

◆ StartDataUploader()

static bool KeenResearch.KeenASR.StartDataUploader ( string appKey )

inlinestatic

Starts a background uploader thread that uploads queued responses to Dashboard. Call this once during setup; responses are queued via ASRResponse.QueueForUpload().

Parameters

appKey The APP_KEY for your app as shown in Dashboard.

Returns: true if the uploader was started successfully, false otherwise.

◆ StartListening()

bool KeenResearch.KeenASR.StartListening ( )

inline

Starts listening for speech. The recognizer must be in ReadyToListen state (i.e. PrepareForListeningWithDecodingGraph must have been called first). Recognition results are delivered via onPartialASRResultReceived and onFinalResponseReceived events.

Returns: true if listening was started successfully, false otherwise.

◆ StopListening()

void KeenResearch.KeenASR.StopListening ( )

inline

Stops the recognizer from processing incoming audio. Note that calling this method will NOT trigger the onFinalResponseReceived callback. If you need the final result, set VAD timeout thresholds (e.g. TimeoutEndSilenceForGoodMatch) to short values so that the recognizer stops naturally via VAD and delivers the result through the callback.

◆ StopUploader()

static bool KeenResearch.KeenASR.StopUploader ( )

inlinestatic

Stops the data uploader thread. After stopping, StartDataUploader() must be called again to resume uploads.

Returns: true if the uploader was stopped successfully, false otherwise.

◆ Teardown()

static bool KeenResearch.KeenASR.Teardown ( )

inlinestatic

Tears down the recognizer and releases all associated resources. All audio playback should be stopped before calling this method.

After teardown, Instance will return null. The SDK can be re-initialized by calling Initialize() again.

Returns: true if teardown succeeded, false if the SDK was not initialized.

◆ UnwindAppAudioBeforeAudioInterruptHandler()

delegate void KeenResearch.KeenASR.UnwindAppAudioBeforeAudioInterruptHandler ( )

Called immediately before KeenASR unwinds its audio stack due to an audio interrupt (e.g. the app going to background). Stop any app-controlled audio playback in this handler. Do not modify the audio session or interact with the recognizer.

This handler must return quickly, otherwise the audio stack may not unwind before the app is suspended.

Property Documentation

◆ Instance

KeenASR KeenResearch.KeenASR.Instance

staticget

Gets the shared KeenASR instance. Returns null if the SDK has not been initialized.

Event Documentation

◆ onFinalResponseReceived

FinalASRResultHandler KeenResearch.KeenASR.onFinalResponseReceived

Raised when recognition is complete with the final response. The handler receives an ASRResponse that it must dispose when done.

◆ onInitializedReceived

InitializedHandler KeenResearch.KeenASR.onInitializedReceived

static

Raised when the ASR engine has finished initializing. Subscribe to this event before calling Initialize().

◆ onPartialASRResultReceived

PartialASRResultHandler KeenResearch.KeenASR.onPartialASRResultReceived

Raised during recognition with intermediate (non-final) text as it becomes available.

◆ onRecognizerReadyToListenAfterInterruptReceived

RecognizerReadyToListenAfterInterruptHandler KeenResearch.KeenASR.onRecognizerReadyToListenAfterInterruptReceived

Raised when the recognizer is ready to listen again after an audio interrupt.

◆ onUnwindAppAudioBeforeAudioInterruptReceived

UnwindAppAudioBeforeAudioInterruptHandler KeenResearch.KeenASR.onUnwindAppAudioBeforeAudioInterruptReceived

Raised immediately before KeenASR unwinds its audio stack due to an audio interrupt. See UnwindAppAudioBeforeAudioInterruptHandler for details.

Public Member Functions

Static Public Member Functions

Properties

Events

Detailed Description

Member Function Documentation

◆ AdaptToSpeakerWithName()

◆ CreateContextualDecodingGraphFromPhrases()

◆ CreateDecodingGraphFromPhrases()

◆ DecodingGraphWithNameExists()

◆ FinalASRResultHandler()

◆ GetRecognizerState()

◆ Initialize()

◆ InitializedHandler()

◆ InitializeWithASRBundleAtPath()

◆ InputLevel()

◆ IsEchoCancellationAvaialable()

◆ PartialASRResultHandler()

◆ PauseUploader()

◆ PerformEchoCancellation()

◆ PrepareForListeningWithContextualDecodingGraph()

◆ PrepareForListeningWithContextualDecodingGraphAtPath()

◆ PrepareForListeningWithDecodingGraph()

◆ PrepareForListeningWithDecodingGraphAtPath()

◆ RecognizerReadyToListenAfterInterruptHandler()

◆ RemoveAllSpeakerAdaptationProfiles()

◆ RemoveSpeakerAdaptationProfiles()

◆ ResetSpeakerAdaptation()

◆ ResumeUploader()

◆ SaveSpeakerAdaptationProfile()

◆ SetLogLevel()

◆ SetVADParameter()

◆ StartDataUploader()

◆ StartListening()

◆ StopListening()

◆ StopUploader()

◆ Teardown()

◆ UnwindAppAudioBeforeAudioInterruptHandler()

Property Documentation

◆ Instance

Event Documentation

◆ onFinalResponseReceived

◆ onInitializedReceived

◆ onPartialASRResultReceived

◆ onRecognizerReadyToListenAfterInterruptReceived

◆ onUnwindAppAudioBeforeAudioInterruptReceived