metadata
model-index:
- name: whisper-large-v2-uk
results:
- task:
type: automatic-speech-recognition
name: Automatic Speech Recognition
dataset:
name: mozilla-foundation/common_voice_11_0
type: mozilla-foundation/common_voice_11_0
config: uk
split: test
metrics:
- type: wer
value: 13.01
name: WER
tags:
- whisper-event
Whisper model finetuned using audio data from CommonVoice Ukrainian v10 train and dev set with additional data via semi-supervised data.
There is a differences in tokenization of source data (in our data normalization process, we replace punctucation with "" rather than Whisper's " "). This mismatch leads to a slight degradation on CommonVoice.