Comparative Evaluation of Speech-to- Text Models for Lithuanian Transcription: Effects of Audio Formats and Recording Environments

Santrauka

This study evaluates the performance of various speech-to-text models for Lithuanian transcription, focusing on how audio formats and recording environments affect their accuracy. Among the models tested, Google’s Chirp-2 demonstrated the highest accuracy under optimal conditions. However, its performance declined with increased playback speeds and in environments with significant background noise, highlighting the importance of controlled recording conditions for effective deployment of STT systems in realworld applications.