File size: 3,189 Bytes
29a8690 eae2655 da20fc1 6ee1e1a da20fc1 29a8690 399c972 29a8690 399c972 29a8690 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 |
---
license: apache-2.0
tags:
- generated_from_trainer
- whisper_sprint
- Arabic
- multilingual
- STT
metrics:
- wer
model-index:
- name: KalemaTech-Arabic-STT-ASR-based-on-Whisper-Small
results: []
---
<!-- 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. -->
# KalemaTech-Arabic-STT-ASR-based-on-Whisper-Small
This model is a fine-tuned version of [openai/whisper-small](https://huggingface.co/openai/whisper-small) on Common_Voice_Arabic_12.0_Augmented.
It achieves the following results on the evaluation set:
- Loss: 0.5362
- Wer: 58.5848
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
Common_Voice_Arabic_12.0 and I made some augmentations to it as follows:
- 25% of the data TimeMasking
- 25% of the data SpecAugmentation
- 25% of the data WavAugmentation (AddGaussianNoise)
- The final dataset is the original common voice plus the augmented files
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 64
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 25
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|:-------------:|:-----:|:-----:|:---------------:|:-------:|
| 0.2728 | 1.01 | 1000 | 0.3063 | 60.4733 |
| 0.1442 | 2.01 | 2000 | 0.2878 | 55.6935 |
| 0.0648 | 3.02 | 3000 | 0.3009 | 59.2568 |
| 0.0318 | 4.03 | 4000 | 0.3278 | 59.2993 |
| 0.0148 | 5.04 | 5000 | 0.3539 | 61.0364 |
| 0.0088 | 6.04 | 6000 | 0.3714 | 56.9154 |
| 0.0061 | 7.05 | 7000 | 0.3920 | 57.5515 |
| 0.0041 | 8.06 | 8000 | 0.4149 | 61.6328 |
| 0.0033 | 9.06 | 9000 | 0.4217 | 58.0310 |
| 0.0033 | 10.07 | 10000 | 0.4376 | 59.9594 |
| 0.0021 | 11.08 | 11000 | 0.4485 | 56.7812 |
| 0.0015 | 12.08 | 12000 | 0.4577 | 57.6936 |
| 0.0013 | 13.09 | 13000 | 0.4671 | 60.6606 |
| 0.0011 | 14.1 | 14000 | 0.4686 | 59.8159 |
| 0.0008 | 15.11 | 15000 | 0.4856 | 60.7111 |
| 0.0011 | 16.11 | 16000 | 0.4851 | 59.5198 |
| 0.0005 | 17.12 | 17000 | 0.4936 | 59.2608 |
| 0.0004 | 18.13 | 18000 | 0.4995 | 57.9619 |
| 0.0003 | 19.13 | 19000 | 0.5085 | 58.3630 |
| 0.0002 | 20.14 | 20000 | 0.5155 | 58.0987 |
| 0.0001 | 21.15 | 21000 | 0.5251 | 58.8504 |
| 0.0001 | 22.16 | 22000 | 0.5268 | 58.4228 |
| 0.0001 | 23.16 | 23000 | 0.5317 | 59.0881 |
| 0.0001 | 24.17 | 24000 | 0.5362 | 58.5848 |
### Framework versions
- Transformers 4.25.1
- Pytorch 1.13.1+cu117
- Datasets 2.8.0
- Tokenizers 0.13.2
|