xlsr-sg-lm / README.md
anton-l's picture
anton-l HF staff
Upload README.md
d1fc6dc
metadata
language: gsw
tags:
  - hf-asr-leaderboard
  - robust-speech-event
widget:
  - example_title: swiss parliament sample 1
    src: >-
      https://huggingface.co/manifoldix/xlsr-sg-lm/resolve/main/07e73bcaa2ab192aea9524d72db45f34f274d1b3d5672434c462d32d44d792be.mp3
  - example_title: swiss parliament sample 2
    src: >-
      https://huggingface.co/manifoldix/xlsr-sg-lm/resolve/main/14a2f855363920f111c7b30e8632c19e5f340ab5031e1ed2621db39baf452ae0.mp3
model-index:
  - name: XLS-R-1b Wav2Vec2 Swiss German
    results:
      - task:
          name: Speech Recognition
          type: automatic-speech-recognition
        metrics:
          - name: Test WER on Swiss parliament
            type: wer
            value: 34.6%
          - name: Test WER on Swiss dialect test set
            type: wer
            value: 40%

XLSR-1b Swiss German

Fine-tuned on the Swiss parliament dataset from FHNW v1 (70h).

Tested on the Swiss parliament test set with a WER of 34.6%

Tested on the "Swiss German Dialects" with a WER of 40%

Both test sets can be accessed here: fhnw_datasets

The Swiss German dialect private test set has been uploaded on huggingface: huggingface_swiss_dialects