CosyVoiceOptions

Options & Configuration · AiDotNet.TextToSpeech.CodecBased

Options for CosyVoice.

Inheritance: ModelOptions → CosyVoiceFamilyOptions → CosyVoiceConformerOptions → CosyVoiceAutoregressiveOptions → CosyVoiceOptions

Known derived types: CosyVoiceCloneOptions

For Beginners

These options configure the CosyVoice model. Default values follow the released model.

How It Works

The defaults are CosyVoice-300M’s (FunAudioLLM/CosyVoice examples/libritts/cosyvoice/conf/cosyvoice.yaml): Whisper’s 100-language text tokens (51,866) and the 50 Hz S3 speech tokens (4,096); the LM’s 6-block Conformer text encoder (512 in, 1,024 wide, 16 heads, 4,096 units) and 14-block Transformer (1,024 wide); the flow’s 6-block Conformer (512 wide, 8 heads, 2,048 units) and U-Net estimator (channels 256, 256; 4 blocks; 12 mid blocks; 8 heads of 64), 10 Euler steps with guidance 0.7; HiFT at 22.05 kHz (base 512, 8 harmonics, upsampling 8·8, iSTFT 16/4); a 192-wide CAM++ x-vector.

Constructors

Constructor Summary
CosyVoiceOptions() Initializes a new instance with default values.
CosyVoiceOptions(CosyVoiceOptions other) Initializes a new instance by copying all CosyVoice settings.

Properties

Property Summary
int LmFeedForwardDim { get; set; } The LM’s feed-forward width (linear_units), in the text encoder and the Transformer alike.
int NumEncoderLayers { get; set; } The v1 Conformer text encoder’s block count.
int TextEncoderDim { get; set; } The v1 Conformer text encoder’s input width.
int TextLanguages { get; set; } Language tokens in the text tokenizer: 100 (Whisper large-v3’s) for CosyVoice-300M.