FPETS speeds up TTS by 600X and reduces errors.
problem High latency and errors in end-to-end TTS systems.
method Non-autoregressive, fully parallel approach with UFANS and trainable position encoding.
result Significant speed up and better quality audios with fewer errors.