io.spokestack.spokestack.nlu.tensorflow

Class TensorflowNLU

java.lang.Object

io.spokestack.spokestack.nlu.tensorflow.TensorflowNLU

All Implemented Interfaces:

NLUService, AutoCloseable

public final class TensorflowNLU
extends Object
implements NLUService

On-device natural language understanding powered by DistilBERT and TensorFlow Lite.

This component uses a pre-trained TensorFlow Lite model to classify user utterances into intents and extract slot values. Metadata supplied alongside the model is used to determine the types of any slot values detected. Classification is performed on a background thread.

This component supports the following configuration properties:

• nlu-model-path (string, required): file system path to the NLU TensorFlow Lite model.
• nlu-metadata-path (string, required): file system path to the model's metadata, used to decode intent and slot names and types.
• wordpiece-vocab-path (string, required): file system path to the wordpiece vocabulary file used by the wordpiece token encoder.
• slot-<slotType> (string, optional): class name of a slot parser capable of parsing slots with the slotType type. For example, a custom slot parser used to parse slots listed as user in the NLU metadata should be provided under the key slot-user.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	TensorflowNLU.Builder Fluent builder interface for initializing a TensorFlow NLU model.

Constructor Summary

Constructors

Constructor and Description
TensorflowNLU(SpeechConfig speechConfig, NLUContext nluContext) Public constructor for NLUManager participation.

Method Summary

Modifier and Type	Method and Description
void	addListener(TraceListener listener) Add a new listener to receive trace events from the NLU subsystem.
AsyncResult<NLUResult>	classify(String utterance) Classify a user utterance, returning a wrapper that can either block until the classification is complete or call a registered callback when the result is ready.
AsyncResult<NLUResult>	classify(String utterance, NLUContext nluContext) Classifies a user utterance.
void	close()
void	removeListener(TraceListener listener) Remove a trace listener, allowing it to be garbage collected.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

TensorflowNLU

public TensorflowNLU(SpeechConfig speechConfig,
                     NLUContext nluContext)

Public constructor for NLUManager participation. Uses a WordpieceTextEncoder and a default TensorFlow model loader.

The model and text encoder are loaded on a background thread.

Parameters:

speechConfig - configuration properties

nluContext - The context used to register listeners and deliver trace and error events.

Method Detail

close

public void close()
           throws Exception

Specified by:

close in interface AutoCloseable

Throws:

Exception

classify

public AsyncResult<NLUResult> classify(String utterance)

Classify a user utterance, returning a wrapper that can either block until the classification is complete or call a registered callback when the result is ready.

Parameters:

utterance - The utterance to classify.

Returns:

An object representing the result of the asynchronous classification.

classify

public AsyncResult<NLUResult> classify(String utterance,
                                       NLUContext nluContext)

Description copied from interface: NLUService

Classifies a user utterance. Classification should be performed on a background thread, but the use of an AsyncResult allows the caller to either block while waiting for a result or register a callback to be executed when the result is available.

Specified by:

classify in interface NLUService

Parameters:

utterance - The user utterance to be classified.

nluContext - The current NLU context, containing request metadata and the ability to fire trace events.

Returns:

An AsyncResult representing the result of the classification task.

See Also:

AsyncResult.registerCallback(io.spokestack.spokestack.util.Callback)

addListener

public void addListener(TraceListener listener)

Add a new listener to receive trace events from the NLU subsystem.

Parameters:

listener - The listener to add.

removeListener

public void removeListener(TraceListener listener)

Remove a trace listener, allowing it to be garbage collected.

Parameters:

listener - The listener to remove.