A Cookbook for Community-driven Data Collection of Impaired Speech in LowResource Languages

TL;DR

This paper introduces a community-driven approach and a practical 'cookbook' for collecting impaired speech data in low-resource languages, demonstrated through creating an Akan speech dataset and fine-tuning ASR models.

Contribution

It provides the first open-source impaired speech dataset for Akan and develops a reproducible methodology for community-based data collection and model training.

Findings

Successful creation of the first Akan impaired speech dataset

Open-source tools and practices for community-driven data collection

Initial improvements in ASR accuracy for impaired speech in Akan

Abstract

This study presents an approach for collecting speech samples to build Automatic Speech Recognition (ASR) models for impaired speech, particularly, low-resource languages. It aims to democratize ASR technology and data collection by developing a "cookbook" of best practices and training for community-driven data collection and ASR model building. As a proof-of-concept, this study curated the first open-source dataset of impaired speech in Akan: a widely spoken indigenous language in Ghana. The study involved participants from diverse backgrounds with speech impairments. The resulting dataset, along with the cookbook and open-source tools, are publicly available to enable researchers and practitioners to create inclusive ASR technologies tailored to the unique needs of speech impaired individuals. In addition, this study presents the initial results of fine-tuning open-source ASR models to better recognize impaired speech in Akan.