metadata

license: cc-by-nc-4.0
datasets:
  - mozilla-foundation/common_voice_11_0
language:
  - fr
  - es
  - pt
  - da
  - de
  - nl
  - fy
  - zh
  - ja
  - ar
  - sw
  - gn
library_name: fairseq

HUTTER-12: H(uBERT) UTTER model covering 12 languages.

Total training hours: 1,622 from Romance (French: 300h, Spanish: 300h, Portuguese: 102.3h), West-Germanic (Danish: 3.5h, German: 300h, Dutch: 72.1h, Frisian: 41.2h) and other languages (Chinese (zh-CN): 104.6h, Japanese: 37h, Arabic: 61h, Swahili 300h, Guaraní: 0.4h)
Number of updates: 400K
Number of iterations: 3
Clustering approach: mini-batch K-means (100% of the data)
Dataset: CommonVoice v13

Funding

This is an output of the European Project UTTER (Unified Transcription and Translation for Extended Reality) under grant number 101070631. For more information go to https://he-utter.eu/