Fast transcription of speech in low-resource languages

TL;DR

This paper introduces a rapid and resource-efficient speech transcription method for low-resource languages, leveraging minimal text data, a pretrained acoustic model, and a G2P table to produce accurate transcriptions within hours.

Contribution

It presents a novel low-resource speech transcription pipeline that requires only a few hours and minimal data, enabling quick transcription of diverse languages.

Findings

Successfully transcribed 40 hours of speech in multiple languages

Operates with only tens of megabytes of noisy text data

Uses a combination of pretrained models and G2P tables

Abstract

We present software that, in only a few hours, transcribes forty hours of recorded speech in a surprise language, using only a few tens of megabytes of noisy text in that language, and a zero-resource grapheme to phoneme (G2P) table. A pretrained acoustic model maps acoustic features to phonemes; a reversed G2P maps these to graphemes; then a language model maps these to a most-likely grapheme sequence, i.e., a transcription. This software has worked successfully with corpora in Arabic, Assam, Kinyarwanda, Russian, Sinhalese, Swahili, Tagalog, and Tamil.