Search papers, labs, and topics across Lattice.
1
0
3
Ditching mel-spectrograms unlocks SOTA text-to-speech with a surprisingly simple diffusion model operating directly on waveform latents.
