CosyVoiceConformerOptions

Options & Configuration · AiDotNet.TextToSpeech.CodecBased

Native Conformer/U-Net flow and S3 encoder settings used by CosyVoice v1/v2 and Chatterbox.

Inheritance: ModelOptions → CosyVoiceFamilyOptions → CosyVoiceConformerOptions

Known derived types: ChatterboxOptions, CosyVoiceAutoregressiveOptions

Properties

Property Summary
CosyVoiceFlowLoss FlowLoss { get; set; } The flow-matching loss: the paper’s absolute error (eq.
double FlowAttentionDropoutRate { get; set; } The flow encoder’s attention dropout.
double InferenceCfgRate { get; set; } The flow’s guidance weight at inference.
double TrainingCfgRate { get; set; } The probability a flow training step drops its conditions (classifier-free guidance training).
int DecoderBlocks { get; set; } The estimator’s transformer blocks per U-Net level.
int DecoderHeadDim { get; set; } The estimator’s attention head width.
int DecoderHeads { get; set; } The estimator’s attention heads.
int DecoderMidBlocks { get; set; } The estimator’s middle blocks.
int FlowDim { get; set; } The flow’s width (input_size, its encoder’s output_size).
int FlowEncoderLayers { get; set; } The flow encoder’s Conformer blocks.
int FlowFeedForwardDim { get; set; } The flow encoder’s feed-forward width.
int FlowHeads { get; set; } The flow encoder’s attention heads.
int FlowTimesteps { get; set; } Euler steps of the flow’s sampler.
int SpeechTokenizerHeads { get; set; } The S3 speech tokenizer’s attention heads.
int SpeechTokenizerLayers { get; set; } The S3 speech tokenizer’s blocks.
int SpeechTokenizerState { get; set; } The S3 speech tokenizer’s width (1,280 in every release).
int[] DecoderChannels { get; set; } The flow estimator’s U-Net channels.