Gujarati-English Code-Switching Speech Recognition using ensemble prediction of spoken language

TL;DR

This work proposes two methods of introducing language specific parameters and explainability in the multi-head attention mechanism, and implements a Temporal Loss that helps maintain continuity in input alignment in end-to-end Automatic Speech Recognition models.

Abstract

An important and difficult task in code-switched speech recognition is to recognize the language, as lots of words in two languages can sound similar, especially in some accents. We focus on improving performance of end-to-end Automatic Speech Recognition models by conditioning transformer layers on language ID of words and character in the output in an per layer supervised manner. To this end, we propose two methods of introducing language specific parameters and explainability in the multi-head attention mechanism, and implement a Temporal Loss that helps maintain continuity in input alignment. Despite being unable to reduce WER significantly, our method shows promise in predicting the correct language from just spoken data. We introduce regularization in the language prediction by dropping LID in the sequence, which helps align long repeated output sequences.

Figures (4)

• Figure 1: Language conditioned Neural Machine Translation with parameter sharing
• Figure 2: Average Gating Loss vs Layer #
• Figure 3: Gating weights alignment to reference LID
• Figure :