|
--- |
|
license: apache-2.0 |
|
tags: |
|
- generated_from_trainer |
|
datasets: |
|
- mozilla-foundation/common_voice_12_0 |
|
metrics: |
|
- wer |
|
model-index: |
|
- name: wav2vec2-large-xls-r-1b-frisian |
|
results: |
|
- task: |
|
name: Automatic Speech Recognition |
|
type: automatic-speech-recognition |
|
dataset: |
|
name: common_voice_12_0 |
|
type: common_voice_12_0 |
|
config: fy-NL |
|
split: test |
|
args: fy-NL |
|
metrics: |
|
- name: Wer |
|
type: wer |
|
value: 0.15990775235054105 |
|
language: |
|
- fy |
|
--- |
|
|
|
<!-- This model card has been generated automatically according to the information the Trainer had access to. You |
|
should probably proofread and complete it, then remove this comment. --> |
|
|
|
# wav2vec2-large-xls-r-1b-frisian |
|
|
|
This model is a fine-tuned version of [facebook/wav2vec2-xls-r-1b](https://huggingface.co/facebook/wav2vec2-xls-r-1b) on the common_voice_12_0 dataset. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 0.2634 |
|
- WER: 0.1599 |
|
|
|
This model was developed together with [golesheed](https://huggingface.co/golesheed) for the course "Speech Recognition II" of the "MSc Voice Technology" program at Rijksuniversiteit Groningen - Campus Fryslân. |
|
|
|
## Intended uses & limitations |
|
|
|
Intended use is for recognizing Frisian speech. |
|
|
|
Limitations include not enough hyperparameter tuning, no LM rescoring, and using v12 of Common Voice instead of v13. |
|
|
|
## Training and evaluation data |
|
|
|
Training and evaluation splits used are the ones available in the Common Voice dataset. |
|
|
|
## Training procedure |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 8e-05 |
|
- train_batch_size: 16 |
|
- eval_batch_size: 8 |
|
- seed: 42 |
|
- gradient_accumulation_steps: 2 |
|
- total_train_batch_size: 32 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: linear |
|
- lr_scheduler_warmup_steps: 500 |
|
- num_epochs: 50 |
|
- mixed_precision_training: Native AMP |
|
|
|
### Training results |
|
|
|
| Training Loss | Epoch | Step | Validation Loss | Wer | |
|
|:-------------:|:-----:|:----:|:---------------:|:------:| |
|
| 4.7284 | 2.1 | 250 | 2.9453 | 1.0 | |
|
| 1.7496 | 4.2 | 500 | 0.5141 | 0.4771 | |
|
| 0.8168 | 6.3 | 750 | 0.3220 | 0.3148 | |
|
| 0.7403 | 8.4 | 1000 | 0.2988 | 0.2573 | |
|
| 0.7298 | 10.5 | 1250 | 0.2794 | 0.2347 | |
|
| 0.6303 | 12.61 | 1500 | 0.2577 | 0.2164 | |
|
| 0.5201 | 14.71 | 1750 | 0.2746 | 0.2162 | |
|
| 0.5189 | 16.81 | 2000 | 0.2543 | 0.2034 | |
|
| 0.5054 | 18.91 | 2250 | 0.2847 | 0.2071 | |
|
| 0.5112 | 21.01 | 2500 | 0.2772 | 0.1979 | |
|
| 0.5105 | 23.11 | 2750 | 0.2633 | 0.1920 | |
|
| 0.5032 | 25.21 | 3000 | 0.2667 | 0.1856 | |
|
| 0.46 | 27.31 | 3250 | 0.2730 | 0.1852 | |
|
| 0.4992 | 29.41 | 3500 | 0.2626 | 0.1782 | |
|
| 0.4535 | 31.51 | 3750 | 0.2778 | 0.1749 | |
|
| 0.4036 | 33.61 | 4000 | 0.2825 | 0.1747 | |
|
| 0.3347 | 35.71 | 4250 | 0.2797 | 0.1708 | |
|
| 0.2708 | 37.82 | 4500 | 0.2662 | 0.1712 | |
|
| 0.1825 | 39.92 | 4750 | 0.2652 | 0.1648 | |
|
| 0.1654 | 42.02 | 5000 | 0.2719 | 0.1628 | |
|
| 0.1387 | 44.12 | 5250 | 0.2552 | 0.1607 | |
|
| 0.1367 | 46.22 | 5500 | 0.2641 | 0.1591 | |
|
| 0.1218 | 48.32 | 5750 | 0.2634 | 0.1598 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.27.3 |
|
- Pytorch 2.0.0+cu117 |
|
- Datasets 2.10.1 |
|
- Tokenizers 0.13.2 |
