Support CVVP & fix for major bug in API

update do_tts

update api constants

update

implement clip-guided generation (and never use it...)

Remove intelligibility refinement

Update sweep & eval_multiple with new voices

support latents into the diffusion decoder

Updates

support presets for generation

Clip diffusion inputs

Updates

integrate new autoregressive model and fix new diffusion bug

Integrate new diffusion network

port do_tts to use the API

Update API to have more expressive interface for controlling various generation knobs

Upgrade CLIP model and add eval_multiple