EAT: Enhanced ASR-TTS for Self-supervised Speech Recognition
This work addresses performance degradation in speech recognition for out-of-domain conditions, representing an incremental improvement in self-supervised learning methods.
The paper tackles the problem of self-supervised ASR-TTS models performing poorly on out-of-domain data by proposing an enhanced model (EAT) with language model rewards and attention scaling, resulting in absolute reductions of 2.6% and 2.7% in performance gaps on Librispeech and BABEL datasets.
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$\rightarrow$TTS direction is equipped with a language model reward to penalize the ASR hypotheses before forwarding it to TTS. 2) In the TTS$\rightarrow$ASR direction, a hyper-parameter is introduced to scale the attention context from synthesized speech before sending it to ASR to handle out-of-domain 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.