Improving Contextual Biasing with Text Injection
Tara N. Sainath, Rohit Prabhavalkar, Diamantino Caseiro, Pat Rondon, Cyril Allauzen
Abstract
In this work, we present a model-based approach to improving contextual biasing that improves quality without drastically increasing model computation during inference. Specifically, we look at injecting text data during training which is representative of contextually-relevant context that will be seen at inference, using a modality-matching text injection method known as JOIST. As JOIST injects text data directly into the E2E model, there is no additional model computation during inference, which is a big difference compared to most model-based biasing techniques. We find that our proposed approach, when combined with an FST-based context model, improves recognition of contacts between 5–15% relative.
BibTeX
@inproceedings{icassp2023_improvingcontext,
title = {Improving Contextual Biasing with Text Injection},
author = {Tara N. Sainath and Rohit Prabhavalkar and Diamantino Caseiro and Pat Rondon and Cyril Allauzen},
booktitle = {ICASSP 2023},
year = {2023}
}