Research and processing of the kazakh children’s acoustic corpus
Diana Rakhimova, Zh. Duisenbekkyzy, Eşref Adalı, Sh. Toleugali · Bulletin of the National Engineering Academy of the Republic of Kazakhstan · 2024
Recent advancements in speech recognition technology have significantly enhanced accessibility and functionality across various sectors. Nonetheless, the task of recognizing kid’s speech presents considerable challenges. Children from different age groups exhibit distinct speech characteristics, including variations in intonation, articulation, and vocabulary. These variabilities necessitate the development of a robust speech recognition system capable of adapting to the evolving speech patterns of children. This paper describes a methodology for the compilation and processing of a kid’s speech corpus in the Kazakh language, informed by analyses of acoustic corpus from other languages. To construct this corpus, a comprehensive dataset was assembled. A novel hybrid error correction strategy, employing the Levenshtein distance algorithm, was developed to address orthographic discrepancies. This method ena- bles the practical evaluation of speech recognition efficacy, quantified through the computational steps required for word correction.