Kunnafonidilaw ka Cadeau: an ASR dataset of present-day Bambara

TL;DR

This paper introduces Kunkado, a comprehensive Bambara ASR dataset from radio archives, and demonstrates how finetuning models on this data improves recognition accuracy in real-world scenarios.

Contribution

The creation of Kunkado, a large-scale, realistic Bambara speech dataset, and the demonstration of improved ASR performance through finetuning on this data.

Findings

Finetuning reduces WER from 44.47% to 37.12%.

Finetuning reduces WER from 36.07% to 32.33%.

Models trained on Kunkado outperform those trained on cleaner speech.

Abstract

We present Kunkado, a 160-hour Bambara ASR dataset compiled from Malian radio archives to capture present-day spontaneous speech across a wide range of topics. It includes code-switching, disfluencies, background noise, and overlapping speakers that practical ASR systems encounter in real-world use. We finetuned Parakeet-based models on a 33.47-hour human-reviewed subset and apply pragmatic transcript normalization to reduce variability in number formatting, tags, and code-switching annotations. Evaluated on two real-world test sets, finetuning with Kunkado reduces WER from 44.47\% to 37.12\% on one and from 36.07\% to 32.33\% on the other. In human evaluation, the resulting model also outperforms a comparable system with the same architecture trained on 98 hours of cleaner, less realistic speech. We release the data and models to support robust ASR for predominantly oral languages.