Reduce and Reconstruct: ASR for Low-Resource Phonetic Languages

TL;DR

This paper introduces a technique to improve low-resource phonetic language ASR by reducing the output alphabet based on phonetic similarity and then reconstructing the original sequence, leading to better performance with limited data.

Contribution

It proposes a novel reduction and reconstruction method for low-resource ASR systems, utilizing linguistically meaningful reductions and a finite state transducer-based reconstruction module.

Findings

Up to 7% relative WER reduction on Gujarati and Telugu with 10 hours of data

Effective alphabet reduction simplifies ASR training in low-resource settings

Reconstruction module accurately recovers original sequences from reduced alphabet predictions

Abstract

This work presents a seemingly simple but effective technique to improve low-resource ASR systems for phonetic languages. By identifying sets of acoustically similar graphemes in these languages, we first reduce the output alphabet of the ASR system using linguistically meaningful reductions and then reconstruct the original alphabet using a standalone module. We demonstrate that this lessens the burden and improves the performance of low-resource end-to-end ASR systems (because only reduced-alphabet predictions are needed) and that it is possible to design a very simple but effective reconstruction module that recovers sequences in the original alphabet from sequences in the reduced alphabet. We present a finite state transducer-based reconstruction module that operates on the 1-best ASR hypothesis in the reduced alphabet. We demonstrate the efficacy of our proposed technique using ASR systems for two Indian languages, Gujarati and Telugu. With access to only 10 hrs of speech data, we obtain relative WER reductions of up to 7% compared to systems that do not use any reduction.