SpeakerRecognitionOptions
Options & Configuration · AiDotNet.Models.Options
Shared hyperparameters for the speaker recognition models — the embedding extractor and the verifier built on top of it.
Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → SpeakerRecognitionOptions
Known derived types: SpeakerEmbeddingOptions, SpeakerVerifierOptions
For Beginners
These models turn a recording of someone speaking into a short list of numbers — a “voiceprint” — that is close to other recordings of the same person and far from everyone else’s. The settings here describe the audio they expect and the network that produces the voiceprint.
How It Works
Both models follow the x-vector design of Snyder et al. (ICASSP 2018): a stack of time-delay layers over frame-level features, statistics pooling across time, then dense layers producing the embedding. There is no attention in that design, which is why there is no head count here.
Properties
| Property | Summary |
|---|---|
int EmbeddingDimension { get; set; } |
Gets or sets the length of the voiceprint the model produces. |
int HiddenDim { get; set; } |
Gets or sets the width of the encoder layers. |
int NumEncoderLayers { get; set; } |
Gets or sets how many encoder layers the model builds. |
int SampleRate { get; set; } |
Gets or sets the sample rate, in hertz, that input audio is expected to have. |