SeedTTSOptions

Options & Configuration · AiDotNet.TextToSpeech.CodecBased

Options for the disclosed Seed-TTS autoregressive pipeline.

Inheritance: ModelOptions → TtsModelOptions → CodecTtsOptions → SeedTTSOptions

Known derived types: SeedTTSCloneOptions

How It Works

The Seed-TTS technical report discloses the four model blocks and their objectives, but intentionally does not publish widths, depths, tokenizer configuration, source code, or weights. Consequently, the values below are resource-conscious defaults for AiDotNet’s native implementation, not purported ByteDance hyperparameters. Dimensions and the native diffusion schedule are configurable; undisclosed private block details cannot be certified by these options or by loading an AiDotNet-format checkpoint.

Constructors

Constructor Summary
SeedTTSOptions() Creates the native reference configuration.
SeedTTSOptions(SeedTTSOptions other) Creates a copy.

Properties

Property Summary
SeedTTSTrainingComponent TrainingComponent { get; set; } The component optimized by Train(TtsTrainingSample).
double DiffusionBetaEnd { get; set; } Last native linear VP beta; inference uses deterministic DDIM with clean-feature prediction.
double DiffusionBetaStart { get; set; } First native linear VP beta, following the standard DDPM configuration, not a reported Seed value.
double Temperature { get; set; } Autoregressive sampling temperature.
int AcousticDimension { get; set; } Continuous width produced by the tokenizer and consumed by diffusion and the vocoder.
int CodecFrameRate { get; set; } The speech-token frame rate, paired with SampleRate and CodecRates.
int DiffusionLayers { get; set; } Number of bidirectional transformer blocks in the token diffusion model.
int DiffusionSteps { get; set; } Number of coarse-to-fine denoising iterations at inference.
int DiffusionTrainingSteps { get; set; } Native VP diffusion training schedule length; the Seed report does not disclose a schedule.
int DiscriminatorChannels { get; set; } Gets or sets the discriminator channels.
int DiscriminatorWidthDivisor { get; set; } Gets or sets the discriminator width divisor.
int FramesPerTextToken { get; set; } Generation budget per target text token when an end token has not yet been emitted.
int GenerationSeed { get; set; } Seed used by the speech-token sampler and diffusion initial noise.
int MaxPromptCodecFrames { get; set; } Maximum number of prompt speech tokens retained for in-context continuation.
int MinimumGeneratedCodecFrames { get; set; } Minimum speech-token count before the autoregressive decoder may accept its end token.
int NumCodebooks { get; set; } The native pipeline’s single speech-token stream.
int NumEncoderLayers { get; set; } The causal token language model’s depth, forwarded to NumLLMLayers.
int SampleRate { get; set; } The waveform sample rate, paired with CodecFrameRate and CodecRates.
int SamplesPerCodecFrame { get; } Samples represented by one speech token in this native configuration.
int TokenizerChannels { get; set; } Configurable DAC-family tokenizer encoder width; Seed-TTS does not disclose its tokenizer widths.
int TokenizerCodeDimension { get; set; } Gets or sets the feature width of the tokenizer code.
int TopK { get; set; } Top-k speech-token sampling; zero uses the entire vocabulary.
int VocoderHiddenDimension { get; set; } Hidden width of the separately trained waveform decoder.
int[] CodecRates { get; set; } Gets or sets the codec rates.
int[] DiscriminatorWindows { get; set; } Gets or sets the discriminator windows.
int[] TrainingMelBins { get; set; } Gets or sets the training mel bins.
int[] TrainingMelWindows { get; set; } Gets or sets the training mel windows.