Publication Details
Eat: Enhanced ASR-TTS for Self-Supervised Speech Recognition
Burget Lukáš, doc. Ing., Ph.D. (DCGM)
Watanabe Shinji (FIT)
ASTUDILLO, R.
Černocký Jan, prof. Dr. Ing. (DCGM)
cycle-consistency, self-supervision, sequence-tosequence, speech recognition
Self-supervised ASR-TTS models suffer in out-of-domain data conditions. Here we propose an enhanced ASR-TTS (EAT) model that incorporates two main features: 1) The ASR!TTS direction is equipped with a language model reward to penalize the ASR hypotheses before forwarding it to TTS. 2) In the TTS!ASR direction, a hyper-parameter is introduced to scale the attention context from synthesized speech before sending it to ASR to handle out-ofdomain data. Training strategies and the effectiveness of the EAT model are explored under out-of-domain data conditions. The results show that EAT reduces the performance gap between supervised and self-supervised training significantly by absolute 2.6% and 2.7% on Librispeech and BABEL respectively.
@inproceedings{BUT175793,
author="BASKAR, M. and BURGET, L. and WATANABE, S. and ASTUDILLO, R. and ČERNOCKÝ, J.",
title="Eat: Enhanced ASR-TTS for Self-Supervised Speech Recognition",
booktitle="ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)",
year="2021",
pages="6753--6757",
publisher="IEEE Signal Processing Society",
address="Toronto, Ontario",
doi="10.1109/ICASSP39728.2021.9413375",
isbn="978-1-7281-7605-5",
url="https://ieeexplore.ieee.org/document/9413375"
}