gweltou/whisper-small-br_260315

A Whisper-small model finetuned on ~80h of data.

Training data:

• —51.5 % - Brezhoweb (up to September 2024)
• —11.6 % - Banque Sonore des Dialectes Bretons
• —10.2 % - Dizale
• —5.0 % - #Brezhoneg (vol. 11 to vol. 44)
• —3.9 % - Mozilla Common Voice 22 (train split)
• —2.5 % - Mélanie Jouitteau
• —2.5 % - RKB
• —1.1 % - Ofis Public ar Brezhoneg
• —1.1 % - Becedia
• —10.6 % - Other various sources