The First Slovak ASR Benchmark Is Now Available on SLAIH

The first dedicated automatic speech recognition (ASR/STT) benchmark for Slovak on SLAIH. The benchmark contains 354 hours of Slovak speech recordings across multiple domains and provides a systematic comparison of 22 ASR models evaluated exclusively on Slovak speech data.

Its goal is to establish transparent and reproducible evaluation standards for Slovak speech technologies while supporting the growth of the Slovak ASR ecosystem.

Why did we create the Slovak ASR Benchmark?

Existing benchmarks and publicly available resources have primarily focused on English or multilingual datasets where Slovak represents only a very small fraction of the data. As a result, objectively comparing ASR systems for Slovak has been difficult.

The new Slovak ASR Benchmark addresses this issue by evaluating models directly on Slovak speech data across diverse domains and recording conditions, enabling a more realistic assessment of model performance in real-world applications.

The current benchmark includes

• 354 hours of Slovak speech recordings
• 22 evaluated ASR models
• comparisons across multiple domains and data sources
• transparent metrics and reproducible evaluation

We believe this benchmark will help researchers, companies, and public institutions better navigate the landscape of Slovak speech technologies while also encouraging the development of stronger open datasets and ASR models for Slovak.

🌐 Benchmark