Fine-Tuning the Wav2vec2 Model for Kazakh Speech: A Study on a Limited Corpus

Bauyrzhan Kairatuly, Мадина Мансурова, Assel Ospan

2023

Closed access Metadata (CSL-JSON)
DOI 10.1109/sist58284.2023.10223504
Data confidence 85 high
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

In this study, we developed a model for automatic recognition of Kazakh speech by fine-tuning the XLSR-Wav2Vec2 pre-trained model to a corpus of Kazakh speech. Our results show that fine-tuning the wav2vec2 model on a small corpus of Kazakh speech allows a significant increase in recognition accuracy. However, larger datasets are needed to further evaluate the effectiveness of this approach. The results of this study contribute to ongoing efforts to improve speech recognition technology for low-resource languages such as Kazakh.