TODM: Train Once Deploy Many Efficient Supernet-Based RNN-T Compression For On-Device ASR Models.
Yuan ShangguanHaichuan YangDanni LiChunyang WuYassir FathullahDilin WangAyushi DalmiaRaghuraman KrishnamoorthiOzlem KalinliJunteng JiaJay MahadeokarXin LeiMike SeltzerVikas ChandraPublished in: ICASSP (2024)