KeywordSpotting<T>

Models & Types · AiDotNet.SpeechRecognition.Specialized

Keyword Spotting: lightweight wake-word and command detection

Inheritance: NeuralNetworkBase<T> → AudioNeuralNetworkBase<T> → KeywordSpotting<T>

Implements: ISpeechRecognizer<T>

For Beginners

Keyword Spotting provides lightweight, always-on detection of specific keywords and voice commands using a small feed-forward neural network.

How It Works

References:

  • Paper: “Small-footprint Keyword Spotting Using Deep Neural Networks” (2014)

The cited Deep KWS model stacks log-filterbank frames, passes them through fully connected ReLU layers, and produces frame-level keyword-label posteriors with a softmax output layer.

Example

// Create a Keyword Spotting model for wake-word detection
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.OneDimensional,
    taskType: NeuralNetworkTaskType.SpeechRecognition,
    inputHeight: 16000, inputWidth: 1, inputDepth: 1, outputSize: 32);
var model = new KeywordSpotting<double>(architecture);
            
// Or load a pre-trained ONNX model for edge-device keyword detection
var onnxModel = new KeywordSpotting<double>(architecture, "keywordspotting.onnx");

Constructors

Constructor Summary
KeywordSpotting(NeuralNetworkArchitecture<T> architecture, KeywordSpottingOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer)
KeywordSpotting(NeuralNetworkArchitecture<T> architecture, string modelPath, KeywordSpottingOptions options)

Properties

Property Summary
IReadOnlyList<string> SupportedLanguages { get; }
bool SupportsStreaming { get; }
bool SupportsWordTimestamps { get; }

Methods

Method Summary
IReadOnlyDictionary<string, T> DetectLanguageProbabilities(Tensor<T> audio)
IStreamingTranscriptionSession<T> StartStreamingSession(string language)
ModelMetadata<T> GetModelMetadata()
Task<TranscriptionResult<T>> TranscribeAsync(Tensor<T> audio, string language, bool? includeTimestamps, CancellationToken cancellationToken)
TranscriptionResult<T> Transcribe(Tensor<T> audio, string language, bool? includeTimestamps) Detects keywords using the paper’s compact feed-forward Deep KWS network.
string DetectLanguage(Tensor<T> audio)