---
license: mit
language:
- br
base_model:
- openai/whisper-small
pipeline_tag: automatic-speech-recognition
---

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