CosyVoiceTrainingStage

Enums · AiDotNet.TextToSpeech.CodecBased

Which CosyVoice network a training step trains (the recipes train llm, flow and hifigan in turn).

Fields

Field Summary
CosyVoiceTrainingStage Flow The speech-token-to-mel flow: conditional flow matching.
CosyVoiceTrainingStage LanguageModel The text-to-speech-token language model: next-token cross-entropy.
CosyVoiceTrainingStage Vocoder HiFT against its discriminators (HiFiGan): per utterance a discriminator step, then a generator step.
int value__