---
license: apache-2.0
tags:
- generated_from_trainer
base_model: facebook/wav2vec2-xls-r-300m
datasets:
- common_voice_15_0
metrics:
- wer
model-index:
- name: wav2vec2-xls-r-300m-br
  results:
  - task:
      type: automatic-speech-recognition
      name: Automatic Speech Recognition
    dataset:
      name: common_voice_15_0
      type: common_voice_15_0
      config: br
      split: None
      args: br
    metrics:
    - type: wer
      value: 41
      name: WER
    - type: cer
      value: 14.7
      name: CER
language:
- br
pipeline_tag: automatic-speech-recognition
---

# wav2vec2-xls-r-300m-br

This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on Mozilla Common Voice 15 Breton dataset and [Roadennoù](https://github.com/gweltou/roadennou) dataset. It achieves the following results on the MCV15-br test set:
- Wer: 41.0
- Cer: 14.7

## Model description

This model was trained to assess the performance wav2vec2-xls-r-300m for fine-tuning a Breton ASR model. 

## Intended uses & limitations

This model is a research model. Usage for production is not recommended.

## Training and evaluation data

The training dataset consists of MCV15-br train dataset and 90% of the Roadennoù dataset.
The validation dataset consists of  MCV15-br validation dataset and the remaining 10% of the Roadennoù dataset.
The final test dataset consists of MCV15-br test dataset.

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 6e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 40
- mixed_precision_training: Native AMP


### Framework versions

- Transformers 4.39.1
- Pytorch 2.0.1+cu117
- Datasets 2.18.0
- Tokenizers 0.15.2