A Comparison of Semi-Supervised Learning Techniques for Streaming ASR at Scale. (arXiv:2304.11053v1 [cs.CL])

Unpaired text and audio injection have emerged as dominant methods for
improving ASR performance in the absence of a large labeled corpus. However,
little guidance exists on deploying these methods to improve production ASR
systems that are trained on very large supervised corpora and with realistic
requirements like a constrained model size and CPU budget, streaming
capability, and a rich lattice for rescoring and for downstream NLU tasks. In
this work, we compare three state-of-the-art semi-supervised methods
encompassing both unpaired …

art arxiv asr audio budget comparison cpu guidance nlu performance production requirements scale semi-supervised semi-supervised learning state streaming supervised learning systems text work