cminja's picture
init
7eb4e5d verified
|
raw
history blame
410 Bytes
metadata
language:
  - sr
tags:
  - audio
  - automatic-speech-recognition
license: mit
datasets:
  - juzne-vesti-srpski
  - google/fleurs
  - mozilla-foundation/common_voice_16_1
  - espnet/yodas

Model

Fine-tune of openAI's whisper-medium on the multiple datasets.

It achieves the following results on the evaluation set:

  • Loss: 0.1358
  • Wer Ortho: 0.1814
  • Wer: 0.0901