raw-waveform

WavTTS presents the first raw waveform generative text-to-speech model using flow matching and Diffusion Transformer, achieving performance comparable to latent-space diffusion models while avoiding information loss from compressed representations.