We just evaluated our trained Wav2Vec models (1 epoch), and they gave a small but noticeable improvement over our baseline.

Even more importantly — we now have fully working training and evaluation scripts that anyone can use to keep improving these models. This opens the door for more experimentation, hyperparameter tuning, and base model changes to push the score higher.

Replay: https://youtu.be/kRRJdAlZrig

📺 Past streams: https://www.youtube.com/@koleshjr

🎮 Watch live: https://www.twitch.tv/koleshjr/schedule