Skip to content

2021-04-08

Pre-release
Pre-release
Compare
Choose a tag to compare
@comodoro comodoro released this 08 Apr 19:24
· 6 commits to master since this release
e0251d1

Results on some test sets (without a language model):

  • Czech Commonvoice test dataset: WER: 0.446331, CER: 0.112392, loss: 14.529331
  • Vystadial 2016 test set: WER: 0.569942, CER: 0.226371, loss: 20.126215
  • Large Corpus of Czech Parliament Plenary Hearings test set: WER 0.209104, CER: 0.048405, loss: 17.649645

While not quite SOTA, it may be a sufficient basis for recognition with limited vocabulary.