The govori.si platform was developed with the primary goal of creating a speech recognition tool optimized for the Slovenian language. This innovative solution facilitates the transcription of field recordings, interviews and manual dictations in professional environments and also supports the creation of subtitles.
To address the lack of open speech transcription solutions for low-resource languages, we developed and implemented govori.si specifically for the Slovenian language. We trained an advanced automatic speech recognition model, that is among the best current Slovenian ASR (Automatic Speech Recognition) models, and used state-of-the-art techniques to overcome transcription challenges of transcription, such as diarization, capitalization, punctuation, user-defined substitution dictionaries and numerical notation parsing.
The platform is freely available for research and non-commercial use, with access via registration details provided by the authors on request. User feedback has been overwhelmingly positive, highlighting the value of the platform for applications such as legislative procedures, journalism and research. In the future, we plan to enhance both the performance and usability of the tool by refining the speaker segmentation model, integrating large language models for text summarization and correction, and continuing the development of a user-friendly interface tailored to different applications.
Publications: