Tusom2021: A Phonetically Transcribed Speech Dataset from an Endangered Language for Universal Phone Recognition Experiments

TL;DR

This paper introduces a publicly available, phonetically transcribed speech dataset from the endangered Tangkhulic language East Tusom, designed to facilitate universal phone recognition research in low-resource language settings.

Contribution

It provides a novel, phonetic transcription dataset for an endangered language, along with benchmarks for universal phone recognition systems, addressing data scarcity in low-resource language technologies.

Findings

Dataset contains 2255 utterances in East Tusom language.

Baseline performance metrics for universal phone recognition systems are established.

The dataset is more suitable for phone recognition than phoneme-based datasets.

Abstract

There is growing interest in ASR systems that can recognize phones in a language-independent fashion. There is additionally interest in building language technologies for low-resource and endangered languages. However, there is a paucity of realistic data that can be used to test such systems and technologies. This paper presents a publicly available, phonetically transcribed corpus of 2255 utterances (words and short phrases) in the endangered Tangkhulic language East Tusom (no ISO 639-3 code), a Tibeto-Burman language variety spoken mostly in India. Because the dataset is transcribed in terms of phones, rather than phonemes, it is a better match for universal phone recognition systems than many larger (phonemically transcribed) datasets. This paper describes the dataset and the methodology used to produce it. It further presents basic benchmarks of state-of-the-art universal phone recognition systems on the dataset as baselines for future experiments.