SoundStream

Creator

Creator

Created

Created

2024 Oct 24 9:31

Editor

Editor

Edited

Edited

2024 Nov 1 1:1

Refs

Refs

Audio → Audio token with
Prosody,
Timbre

Hierarchical Token Structure

긴 대화 생성 작업을 위해 600 bps 속도로 음향 토큰

토큰들이 시간 프레임에 따라 위계적으로 그룹화

google-research.github.io

Authors: Neil Zeghidour, Alejandro Luebs, Ahmed Omran, Jan Skoglund, Marco Tagliasacchi

https://google-research.github.io/seanet/soundstream/examples/

https://arxiv.org/pdf/2107.03312

Recommendations

////////