This Wav2Vec2 model was first pretrained on 500 hours Kalmyk TV recordings and 1000 hours Mongolian speech recognition dataset. After that, the model was finetuned on a 300 hours Kalmyk synthetic STT dataset created by a voice conversion model.
50% WER on a private test set created from Kalmyk TV recordnings
on clean voice recordings, the model should have much lower WER
