Wav2Small<T>

Models & Types · AiDotNet.Audio.Emotion

Wav2Small lightweight speech emotion recognition model (Gomez-Alanis et al., 2024).

Inheritance: NeuralNetworkBase<T> → AudioNeuralNetworkBase<T> → AudioClassifierBase<T> → Wav2Small<T>

Implements: IEmotionRecognizer<T>

For Beginners

Wav2Small is a small, fast emotion detector that learned its skills from a much larger model. It can tell if someone sounds happy, sad, angry, or neutral—even on a phone or embedded device that can’t run large AI models.

Usage:

How It Works

Wav2Small distills knowledge from large wav2vec 2.0 models into a compact architecture suitable for edge deployment. It achieves competitive accuracy on IEMOCAP and RAVDESS benchmarks while requiring significantly fewer parameters and computation.

Constructors

Constructor Summary
Wav2Small(NeuralNetworkArchitecture<T> architecture, Wav2SmallOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer)
Wav2Small(NeuralNetworkArchitecture<T> architecture, string modelPath, Wav2SmallOptions options)

Properties

Property Summary
IReadOnlyList<string> SupportedEmotions { get; }

Methods

Method Summary
EmotionResult<T> RecognizeEmotion(Tensor<T> audio)
IReadOnlyDictionary<string, T> GetEmotionProbabilities(Tensor<T> audio)
IReadOnlyList<TimedEmotionResult<T>> RecognizeEmotionTimeSeries(Tensor<T> audio, int windowSizeMs, int hopSizeMs)
ModelMetadata<T> GetModelMetadata()
T GetArousal(Tensor<T> audio)
T GetValence(Tensor<T> audio)
Vector<T> ExtractEmotionFeatures(Tensor<T> audio)