Scaling Up Deliberation for Multilingual ASR
This work addresses accuracy improvements for multilingual ASR systems, though it is incremental as it builds on existing deliberation methods by scaling them to multilingual contexts.
The paper tackles the problem of improving multilingual automatic speech recognition by investigating second-pass deliberation, showing that it reduces the average word error rate on 9 languages by 4% relative compared to single-pass models, with improvements up to 9% when scaled to 1B parameters.
Multilingual end-to-end automatic speech recognition models are attractive due to its simplicity in training and deployment. Recent work on large-scale training of such models has shown promising results compared to monolingual models. However, the work often focuses on multilingual models themselves in a single-pass setup. In this work, we investigate second-pass deliberation for multilingual speech recognition. Our proposed deliberation is multilingual, i.e., the text encoder encodes hypothesis text from multiple languages, and the decoder attends to multilingual text and audio. We investigate scaling the deliberation text encoder and decoder, and compare scaling the deliberation decoder and the first-pass cascaded encoder. We show that deliberation improves the average WER on 9 languages by 4% relative compared to the single-pass model. By increasing the size of the deliberation up to 1B parameters, the average WER improvement increases to 9%, with up to 14% for certain languages. Our deliberation rescorer is based on transformer layers and can be parallelized during rescoring.