Table of Contents

Class SpectralFeatureExtractor<T>

Namespace
AiDotNet.Audio.Features
Assembly
AiDotNet.dll

Extracts spectral features from audio signals including centroid, bandwidth, rolloff, and flux.

public class SpectralFeatureExtractor<T> : AudioFeatureExtractorBase<T>, IAudioFeatureExtractor<T>

Type Parameters

T

The numeric type used for calculations.

Inheritance
SpectralFeatureExtractor<T>
Implements
Inherited Members

Remarks

Spectral features describe the shape and characteristics of an audio signal's frequency content. They are widely used for audio classification, music analysis, and speech processing.

For Beginners: These features describe "what the sound looks like" in terms of frequency:

  • Spectral Centroid: The "center of mass" of the spectrum - high for bright sounds (cymbals), low for dull sounds (bass drum). Think of it as the "brightness" of the sound.
  • Spectral Bandwidth: How spread out the frequencies are. Wide for rich sounds (orchestra), narrow for pure tones (flute).
  • Spectral Rolloff: The frequency below which most (e.g., 85%) of the energy is concentrated. Useful for distinguishing voiced from unvoiced speech.
  • Spectral Flux: How much the spectrum changes between frames. High during transients (drum hits), low during sustained sounds.
  • Spectral Flatness: How "noisy" vs "tonal" the sound is. 1.0 = pure noise, 0.0 = pure tone.

Usage:

var extractor = new SpectralFeatureExtractor<float>();
var features = extractor.Extract(audioTensor);
// features.Shape = [numFrames, numFeatures]

Constructors

SpectralFeatureExtractor(SpectralFeatureOptions?)

Initializes a new spectral feature extractor.

public SpectralFeatureExtractor(SpectralFeatureOptions? options = null)

Parameters

options SpectralFeatureOptions

Spectral feature extraction options.

Properties

FeatureDimension

Gets the number of features produced per frame.

public override int FeatureDimension { get; }

Property Value

int

Name

Gets the name of this feature extractor.

public override string Name { get; }

Property Value

string

Methods

Extract(Tensor<T>)

Extracts features from an audio waveform.

public override Tensor<T> Extract(Tensor<T> audio)

Parameters

audio Tensor<T>

The audio waveform as a 1D tensor [samples].

Returns

Tensor<T>

Features as a 2D tensor [frames, features].

GetFeatureIndex(SpectralFeatureType)

Gets the column index for a specific feature type in the extracted feature tensor.

public int GetFeatureIndex(SpectralFeatureType featureType)

Parameters

featureType SpectralFeatureType

The feature type to find the index for.

Returns

int

The column index of the feature in the output tensor, or -1 if the feature is not enabled.

Remarks

Feature indices depend on which features are enabled. For example, if only Flux and Flatness are enabled, Flux will be at index 0 and Flatness at index 1. With Basic features (Centroid, Bandwidth, Rolloff, Flux, Flatness), Flux will be at index 3.

Edit this page