A 300 million parameter Wav2Vec2 model fine-tuned for Czech speech recognition using 250 hours of training data.