RESTORING PUNCTUATION IN AUTOMATIC SPEECH RECOGNITION FOR KAZAKH LANGUAGE

Abstract

This paper considers the problem of punctuation restoration in Kazakh automatic speech recognition (ASR) systems. The proposed approach demonstrates a solution to the punctuation restoration problem in Kazakh ASR for four main punctuation classes (COMMA, PERIOD, QUESTION, and EXCLAMATION) and the unpunctuated class (O). Experiments were conducted on both monolingual Kazakh texts and mixed Kazakh-Russian data, covering the practical parameters of multilingual communication in modern ASR applications. To reduce the imbalance between the exclamation and question mark classes, a specially labeled corpus was created based on the augmentation method for restoring Kazakh punctuation. Experiments showed that the punctuation restoration model achieved a macro-F1 score of 93.55% for four punctuation classes (COMMA, PERIOD, QUESTION, and EXCLAMATION). The model demonstrated high performance, showing F1 scores of 97.96% for period, 97.78% for question mark, 88.57% for comma, and 89.89% for exclamation mark, which is a significant improvement over previous works and confirms the robustness of the approach for automatic speech recognition results in Kazakh.