May, 2023
SoundStorm: 高效并行音频生成
SoundStorm: Efficient Parallel Audio Generation
Zalán Borsos, Matt Sharifi, Damien Vincent, Eugene Kharitonov, Neil Zeghidour...
TL;DRSoundStorm is a non-autoregressive audio generation model that uses semantic tokens and bidirectional attention to efficiently generate high-quality audio with consistency, comparable with autoregressive generation while being two orders of magnitude faster.