ASR Models
Collection
Automatic Speech Recognition models • 5 items • Updated
Automatic Speech Recognition
Transformers
PyTorch
JAX
Safetensors
Galician
wav2vec2
audio
Eval Results (legacy)
# Load model directly
from transformers import AutoProcessor, AutoModelForCTC
processor = AutoProcessor.from_pretrained("proxectonos/Nos_ASR-wav2vec2-large-xlsr-53-gl-with-lm")
model = AutoModelForCTC.from_pretrained("proxectonos/Nos_ASR-wav2vec2-large-xlsr-53-gl-with-lm")Quick Links
Wav2Vec2-Large-XLSR-53-Galician-With-LM
This is a copy of the model diego-fustes/wav2vec2-large-xlsr-gl with an integrated language model.
Improvement This model has been compared with the baseline (diego-fustes/wav2vec2-large-xlsr-gl) on:
- The test subset of the Galician OpenSLR dataset (837 speech samples)
- The test subset of the Galician Common Voice 7.0 dataset (1716 speech samples)
The results are shown in the following tables:
OpenSLR77:
| Model | WER | CER |
|---|---|---|
| diego-fustes/wav2vec2-large-xlsr-gl | 9.10% | 3.94% |
| cmagui/wav2vec2-large-xlsr-53-galician-with-lm | 6.86% | 2.20% |
Common_voice-gl:
| Model | WER | CER |
|---|---|---|
| diego-fustes/wav2vec2-large-xlsr-gl | 22.12% | 5.09% |
| cmagui/wav2vec2-large-xlsr-53-galician-with-lm | 15.20% | 3.87% |
Citation information
If you use this model, please cite as follows:
Magariños, Carmen; Gamallo, Pablo. 2023. Nos_ASR-wav2vec2-large-xlsr-53-gl-with-lm. URL: https://huggingface.co/proxectonos/Nos_ASR-wav2vec2-large-xlsr-53-gl-with-lm
- Downloads last month
- 64
Dataset used to train proxectonos/Nos_ASR-wav2vec2-large-xlsr-53-gl-with-lm
Space using proxectonos/Nos_ASR-wav2vec2-large-xlsr-53-gl-with-lm 1
Collection including proxectonos/Nos_ASR-wav2vec2-large-xlsr-53-gl-with-lm
Evaluation results
- Test WER on OpenSLRself-reported9.100
- Test CER on OpenSLRself-reported3.940
- Test WER (+LM) on OpenSLRself-reported6.860
- Test CER (+LM) on OpenSLRself-reported2.200
- Test WER on Common Voice 7.0self-reported22.120
- Test CER on Common Voice 7.0self-reported5.090
- Test WER (+LM) on Common Voice 7.0self-reported15.200
- Test CER (+LM) on Common Voice 7.0self-reported3.870
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("automatic-speech-recognition", model="proxectonos/Nos_ASR-wav2vec2-large-xlsr-53-gl-with-lm")