Skip to content

Commit

Permalink
large-v3 (#1761)
Browse files Browse the repository at this point in the history
* mel_filters() loads 128 mel bins

* can load 100-language models

* large-v3 checkpoint and evals

* add mandarin alias

* remove unused path

* flake8 fix

* formatting fix
  • Loading branch information
jongwook authored Nov 6, 2023
1 parent f6f01c5 commit c5d4256
Show file tree
Hide file tree
Showing 11 changed files with 7,145 additions and 2,235 deletions.
4 changes: 2 additions & 2 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -69,9 +69,9 @@ There are five model sizes, four with English-only versions, offering speed and

The `.en` models for English-only applications tend to perform better, especially for the `tiny.en` and `base.en` models. We observed that the difference becomes less significant for the `small.en` and `medium.en` models.

Whisper's performance varies widely depending on the language. The figure below shows a WER (Word Error Rate) breakdown by languages of the Fleurs dataset using the `large-v2` model (The smaller the numbers, the better the performance). Additional WER scores corresponding to the other models and datasets can be found in Appendix D.1, D.2, and D.4. Meanwhile, more BLEU (Bilingual Evaluation Understudy) scores can be found in Appendix D.3. Both are found in [the paper](https://arxiv.org/abs/2212.04356).
Whisper's performance varies widely depending on the language. The figure below shows a performance breakdown of `large-v3` and `large-v2` models by language, using WERs (word error rates) or CER (character error rates, shown in *Italic*) evaluated on the Common Voice 15 and Fleurs datasets. Additional WER/CER metrics corresponding to the other models and datasets can be found in Appendix D.1, D.2, and D.4 of [the paper](https://arxiv.org/abs/2212.04356), as well as the BLEU (Bilingual Evaluation Understudy) scores for translation in Appendix D.3.

![WER breakdown by language](https://raw.githubusercontent.com/openai/whisper/main/language-breakdown.svg)
![WER breakdown by language](https://github.com/openai/whisper/assets/266841/f4619d66-1058-4005-8f67-a9d811b77c62)



Expand Down
Loading

0 comments on commit c5d4256

Please sign in to comment.