Towards a System for Automatic Media Transcription in Macedonian
Filip Mitrovski, Branislav Gerazov, Zoran A. Ivanovski, Dimitar Taškovski · 2020
This paper presents the design and evaluation of a system for Automatic Speech Recognition used for automatic speech transcription of media in Macedonian. The system is based on the powerful Kaldi Speech Recognition Toolkit. We use several methods and algorithms to train, align and implement different models. A comparison between the models is made to analyze their performance and discuss their pros and cons in the specific situation where a relatively small dataset is used. The results show that the delta model outperforms even the more complex models. Finally, we compare the impact of the language model on the overall system performance.