carlacastedo's picture
Update README.md
c1b14b1 verified
metadata
license: apache-2.0
base_model: facebook/wav2vec2-xls-r-300m
datasets:
  - openslr/openslr
  - mozilla-foundation/common_voice_17_0
  - GTM-UVigo/FalAI
  - google/fleurs
  - proxectonos/Nos_Parlaspeech-GL
language:
  - gl
metrics:
  - wer
  - cer
tags:
  - audio
  - automatic-speech-recognition
  - gl
model-index:
  - name: Wav2Vec2-XLS-R-300M-GL
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Common Voice 17.0
          type: mozilla-foundation/common_voice_17_0
          args: gl
        metrics:
          - name: WER
            type: wer
            value: 12.04
          - name: CER
            type: cer
            value: 3.82
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: OpenSLR
          type: openslr
          args: gl
        metrics:
          - name: WER
            type: wer
            value: 7.85
          - name: CER
            type: cer
            value: 1.66
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: FalAI
          type: falai
          args: validated
        metrics:
          - name: WER
            type: wer
            value: 4.39
          - name: CER
            type: cer
            value: 1.17
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: FLEURS
          type: fleurs
          args: gl_es
        metrics:
          - name: WER
            type: wer
            value: 15.83
          - name: CER
            type: cer
            value: 5.08
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Nos_Parlaspeech-GL
          type: nos_parlaspeech-GL
          args: clean
        metrics:
          - name: WER
            type: wer
            value: 8.92
          - name: CER
            type: cer
            value: 2.65

Wav2Vec2-XLS-R-300M-GL

This model is a finetuned version of Facebook's Wav2Vec2 XLS-R 300M for Galician on the datasets Common Voice Corpus 17.0, Open SLR77, FalAI, Fleurs and Nos_ParlaSpeech-GL.

Test

This model has been tested in the test splits of the Galician OpenSLR dataset, the Galician Common Voice 17.0 dataset, the FalAI dataset, the Galician FLEURS dataset and Nos_Parlaspeech-GL. The results are shown in the following tables:

Corpus WER CER RTF
Common Voice 17.0 7.85 1.66 0.0085
Open SLR77 12.04 3.82 0.0087
FalAI 4.39 1.17 0.0260
FLEURS 15.83 5.08 0.0091
Nos_Parlaspeech-GL 8.92 2.65 0.0114

Citation information

If you use this model, please cite as follows:

Moscoso Sánchez, Antonio; Magariños, Carmen; Castedo, Carla. 2025. Nos_ASR-wav2vec2-xls-r-300m-gl. URL: https://huggingface.co/proxectonos/Nos_ASR-wav2vec2-xls-r-300m-gl