Skip to content
DatasetSpeech and audio

SloPalSpeech

SloPalSpeech is an open Slovak speech dataset for training automatic speech recognition models.

Open on Hugging FacePaper

Can I use it?

The license also allows commercial use if you credit the authors and specify the license.

About this resource

SloPalSpeech is an open Slovak speech dataset for training automatic speech recognition models. It contains 2,806 hours of recordings from sessions of the National Council of the Slovak Republic, split into roughly 403k segments of 20 to 30 s, paired with the official session transcripts. Transcripts were aligned to audio automatically, and segments with WER above 40% against a Whisper transcription were removed, so some label noise remains. Only a train split is published, with no official test set. The domain is formal parliamentary speech, so models fine-tuned on it alone tend to drift toward parliamentary vocabulary.

See a mistake or a missing detail? Write to us.

More from NaiveNeuron FMPI CU

NaiveNeuron FMPI CU