I've just released a demo of the latest Whisper Large V3 Turbo model for Albanian on Huggingface Spaces. Trained on over 400 hours of carefully curated speech data, it now reaches around 16% word error rate. My goal is to bring it closer to 10%. Please try it yourself. 👇
huggingface.co
Whisper Turbo Shqip - a Hugging Face Space by Kushtrim
This application transcribes Albanian audio into text. You can upload an audio file, record your voice, or provide a YouTube URL, and it will generate a transcript for you.
