SpeakerEmbeddingOptions

Options & Configuration · AiDotNet.Audio.Speaker

Configuration options for the SpeakerEmbeddingExtractor.

Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → SpeakerRecognitionOptions → SpeakerEmbeddingOptions

For Beginners

The embedding extractor turns a recording into a “voiceprint” — a short list of numbers that identifies a speaker. The values here are the ones it ships with.

How It Works

#2090 note.FftSize, HopLength and NumMfcc were declared here before but never reached the MFCC front end, which always built itself from its own defaults. They are now passed through, so changing them changes what the model computes.

Constructors

Constructor Summary
SpeakerEmbeddingOptions() Initializes a new instance carrying the model’s published defaults.
SpeakerEmbeddingOptions(SpeakerEmbeddingOptions other) Initializes a new instance by copying another instance.

Properties

Property Summary
OnnxModelOptions OnnxOptions { get; set; } Gets or sets the ONNX runtime options used when loading the model from an ONNX file.
double MinimumDurationSeconds { get; set; } Gets or sets the shortest recording, in seconds, the model will accept.
int FftSize { get; set; } Gets or sets the FFT window size, in samples, used by the MFCC front end.
int HopLength { get; set; } Gets or sets the stride between analysis frames, in samples.
int NormalizationWindowFrames { get; set; } Frames of the sliding window over which features are mean-normalized (section 3.2: up to 3 seconds).
int NumMfcc { get; set; } Gets or sets the number of MFCC coefficients per frame.
int NumTrainingSpeakers { get; set; } Speakers of the training set, the width of the softmax layer that trains the x-vector network (section 2.2); zero builds no classifier.
int PoolingDim { get; set; } Units of the last frame-level layer, whose statistics are pooled (Table 1: 1500).

Methods

Method Summary
void Validate() Throws if a value this model requires has been left unset or is not positive.