August 14, 2025Applied SciencesOpen Access

Investigation of ASR Models for Low-Resource Kazakh Child Speech: Corpus Development, Model Adaptation, and Evaluation

View Full Paper

Ask AI

Bookmark

ESPnet achieved the highest accuracy at 32% for sentences, while Vosk excelled with 68% accuracy on short words.
The study evaluated four ASR models, with whisper and espnet showing notable performance improvements through corpus adaptation.
Data were collected using a custom Telegram bot and Dictaphone recordings, focusing on children aged 2 to 8 years.
Findings highlight the need for specialized acoustic corpora in low-resource environments supporting child speech technology.

Discussion

No takes yet. Share an insight, caveat, or question.

Analysis reveals ASR models' varying accuracy for kazakh child speech, suggesting tailored data is essential.

Rakhimova et al. (2025) studied this question.

View Full Paper

Ask AI

Bookmark