|
--- |
|
base_model: masoudmzb/wav2vec2-xlsr-multilingual-53-fa |
|
metrics: |
|
- wer |
|
widget: |
|
- example_title: M22N20 |
|
src: >- |
|
https://huggingface.co/lnxdx/Wav2Vec2-Large-XLSR-Persian-ShEMO/blob/main/M16A01.wav |
|
- example_title: Common Voice sample 2978 |
|
src: >- |
|
https://huggingface.co/m3hrdadfi/wav2vec2-large-xlsr-persian-v3/resolve/main/sample2978.flac |
|
- example_title: Common Voice sample 5168 |
|
src: >- |
|
https://huggingface.co/m3hrdadfi/wav2vec2-large-xlsr-persian-v3/resolve/main/sample5168.flac |
|
model-index: |
|
- name: wav2vec2-large-xlsr-persian-shemo |
|
results: |
|
- task: |
|
name: Speech Recognition |
|
type: automatic-speech-recognition |
|
dataset: |
|
name: Common Voice 13.0 fa |
|
type: common_voice_13_0 |
|
args: fa |
|
metrics: |
|
- name: Test WER |
|
type: wer |
|
value: 19.21 |
|
- task: |
|
name: Speech Recognition |
|
type: automatic-speech-recognition |
|
dataset: |
|
name: ShEMO |
|
type: shemo |
|
args: fa |
|
metrics: |
|
- name: Test WER |
|
type: wer |
|
value: 32.85 |
|
language: |
|
- fa |
|
pipeline_tag: automatic-speech-recognition |
|
tags: |
|
- audio |
|
- speech |
|
- automatic-speech-recognition |
|
- asr |
|
--- |
|
|
|
<!-- 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 XLSR Persian ShEMO |
|
|
|
This model is a fine-tuned version of [masoudmzb/wav2vec2-xlsr-multilingual-53-fa](https://huggingface.co/masoudmzb/wav2vec2-xlsr-multilingual-53-fa) |
|
on the [ShEMO](https://github.com/pariajm/sharif-emotional-speech-dataset) dataset for speech recognition in Persian (Farsi). |
|
When using this model, make sure that your speech input is sampled at 16 kHz. |
|
|
|
It achieves the following results: |
|
- Loss on ShEMO train set: 0.7618 |
|
- Loss on ShEMO dev set: 0.6728 |
|
- WER on ShEMO train set: 30.47 |
|
- WER on ShEMO dev set: 32.85 |
|
- WER on Common Voice 13 test set: 19.21 |
|
|
|
## Evaluation results 🧪 |
|
| Checkpoint Name | WER on ShEMO dev set | WER on Common Voice 13 test set | Max :) | |
|
| :---------------------------------------------------------------------------------------------------------------: | :------: | :-------: | :---: | |
|
| [m3hrdadfi/wav2vec2-large-xlsr-persian-v3](https://huggingface.co/m3hrdadfi/wav2vec2-large-xlsr-persian-v3) | 46.55 | **17.43** | 46.55 | |
|
| [m3hrdadfi/wav2vec2-large-xlsr-persian-shemo](https://huggingface.co/m3hrdadfi/wav2vec2-large-xlsr-persian-shemo) | **7.42** | 33.88 | 33.88 | |
|
| [masoudmzb/wav2vec2-xlsr-multilingual-53-fa](https://huggingface.co/masoudmzb/wav2vec2-xlsr-multilingual-53-fa) | 56.54 | 24.68 | 56.54 | |
|
| This checkpoint | 32.85 | 19.21 | **32.85** | |
|
|
|
As you can see, my model performs better in maximum case :D |
|
|
|
## Training procedure 🏋️ |
|
|
|
#### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 1e-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 |
|
- training_steps: 2000 |
|
- mixed_precision_training: Native AMP |
|
|
|
You may need *gradient_accumulation* because you need more batch size. |
|
|
|
#### Training log 📉 |
|
|
|
| Training Loss | Epoch | Step | Validation Loss | Wer | |
|
|:-------------:|:-----:|:----:|:---------------:|:------:| |
|
| 1.8553 | 0.62 | 100 | 1.4126 | 0.4866 | |
|
| 1.4083 | 1.25 | 200 | 1.0428 | 0.4366 | |
|
| 1.1718 | 1.88 | 300 | 0.8683 | 0.4127 | |
|
| 0.9919 | 2.5 | 400 | 0.7921 | 0.3919 | |
|
| 0.9493 | 3.12 | 500 | 0.7676 | 0.3744 | |
|
| 0.9414 | 3.75 | 600 | 0.7247 | 0.3695 | |
|
| 0.8897 | 4.38 | 700 | 0.7202 | 0.3598 | |
|
| 0.8716 | 5.0 | 800 | 0.7096 | 0.3546 | |
|
| 0.8467 | 5.62 | 900 | 0.7023 | 0.3499 | |
|
| 0.8227 | 6.25 | 1000 | 0.6994 | 0.3411 | |
|
| 0.855 | 6.88 | 1100 | 0.6883 | 0.3432 | |
|
| 0.8457 | 7.5 | 1200 | 0.6773 | 0.3426 | |
|
| 0.7614 | 8.12 | 1300 | 0.6913 | 0.3344 | |
|
| 0.8127 | 8.75 | 1400 | 0.6827 | 0.3335 | |
|
| 0.8443 | 9.38 | 1500 | 0.6725 | 0.3356 | |
|
| 0.7548 | 10.0 | 1600 | 0.6759 | 0.3318 | |
|
| 0.7839 | 10.62 | 1700 | 0.6773 | 0.3286 | |
|
| 0.7912 | 11.25 | 1800 | 0.6748 | 0.3286 | |
|
| 0.8238 | 11.88 | 1900 | 0.6735 | 0.3297 | |
|
| 0.7618 | 12.5 | 2000 | 0.6728 | 0.3286 | |
|
|
|
#### Hyperparameter tuning 🔧 |
|
Several models with differet hyperparameters were trained. The following figures show the training process for three of them. |
|
 |
|
 |
|
**20_2000_1e-5_hp-mehrdad** is the current model (lnxdx/Wav2Vec2-Large-XLSR-Persian-ShEMO) and it's hyperparameters are: |
|
```python |
|
model = Wav2Vec2ForCTC.from_pretrained( |
|
model_name_or_path if not last_checkpoint else last_checkpoint, |
|
# hp-mehrdad: Hyperparams of 'm3hrdadfi/wav2vec2-large-xlsr-persian-v3' |
|
attention_dropout = 0.05316, |
|
hidden_dropout = 0.01941, |
|
feat_proj_dropout = 0.01249, |
|
mask_time_prob = 0.04529, |
|
layerdrop = 0.01377, |
|
ctc_loss_reduction = 'mean', |
|
ctc_zero_infinity = True, |
|
) |
|
``` |
|
The hyperparameters of **19_2000_1e-5_hp-base** are: |
|
```python |
|
model = Wav2Vec2ForCTC.from_pretrained( |
|
model_name_or_path if not last_checkpoint else last_checkpoint, |
|
# hp-base: Hyperparams simmilar to ('facebook/wav2vec2-large-xlsr-53' or 'facebook/wav2vec2-xls-r-300m') |
|
attention_dropout = 0.1, |
|
hidden_dropout = 0.1, |
|
feat_proj_dropout = 0.1, |
|
mask_time_prob = 0.075, |
|
layerdrop = 0.1, |
|
ctc_loss_reduction = 'mean', |
|
ctc_zero_infinity = True, |
|
) |
|
``` |
|
|
|
And the hyperparameters of **22_2000_1e-5_hp-masoud** are: |
|
```python |
|
model = Wav2Vec2ForCTC.from_pretrained( |
|
model_name_or_path if not last_checkpoint else last_checkpoint, |
|
# hp-masoud: Hyperparams of 'masoudmzb/wav2vec2-xlsr-multilingual-53-fa' |
|
attention_dropout = 0.2, |
|
hidden_dropout = 0.2, |
|
feat_proj_dropout = 0.1, |
|
mask_time_prob = 0.2, |
|
layerdrop = 0.2, |
|
ctc_loss_reduction = 'mean', |
|
ctc_zero_infinity = True, |
|
) |
|
``` |
|
Learning rate is 1e-5 for all three models. |
|
|
|
As you can see this model performs better with WER metric on validation(evaluation) set. |
|
|
|
The script used for training can be found [here](https://colab.research.google.com/github/m3hrdadfi/notebooks/blob/main/Fine_Tune_XLSR_Wav2Vec2_on_Persian_ShEMO_ASR_with_%F0%9F%A4%97_Transformers_ipynb.ipynb). |
|
|
|
Check out [this blog](https://huggingface.co/blog/fine-tune-xlsr-wav2vec2) for more information. |
|
|
|
#### Framework versions |
|
|
|
- Transformers 4.35.2 |
|
- Pytorch 2.1.0+cu118 |
|
- Datasets 2.15.0 |
|
- Tokenizers 0.15.0 |
|
|
|
## Contact us 🛎 |
|
If you have any technical question regarding the model, pretraining, code or publication, please create an issue in the repository. This is the *best* way to reach us. |
|
|
|
## Citation ↩️ |
|
*TO DO!* |
|
|
|
**Fine-tuned with ❤️ without ☕︎** |
