File size: 3,988 Bytes
9d44a45 cbe5f88 9d44a45 cbe5f88 9d44a45 cbe5f88 9d44a45 cbe5f88 9d44a45 cbe5f88 9d44a45 cbe5f88 9d44a45 cbe5f88 9d44a45 cbe5f88 |
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 94 95 96 97 98 99 100 101 |
---
license: apache-2.0
base_model: openai/whisper-tiny
tags:
- generated_from_trainer
metrics:
- wer
model-index:
- name: whisper-tiny-myanmar
results: []
datasets:
- chuuhtetnaing/myanmar-speech-dataset-openslr-80
language:
- my
pipeline_tag: automatic-speech-recognition
library_name: transformers
---
<!-- 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. -->
# whisper-tiny-myanmar
This model is a fine-tuned version of [openai/whisper-tiny](https://huggingface.co/openai/whisper-tiny) on the [chuuhtetnaing/myanmar-speech-dataset-openslr-80](https://huggingface.co/datasets/chuuhtetnaing/myanmar-speech-dataset-openslr-80) dataset.
It achieves the following results on the evaluation set:
- Loss: 0.2353
- Wer: 61.8878
## Usage
```python
from datasets import Audio, load_dataset
from transformers import pipeline
# Load a sample audio
dataset = load_dataset("chuuhtetnaing/myanmar-speech-dataset-openslr-80")
dataset = dataset.cast_column("audio", Audio(sampling_rate=16000))
test_dataset = dataset['test']
input_speech = test_dataset[42]['audio']
pipe = pipeline(model='chuuhtetnaing/whisper-tiny-myanmar')
output = pipe(input_speech, generate_kwargs={"language": "myanmar", "task": "transcribe"})
print(output['text']) # ကျွန်မ ပြည်ပ မှာ ပညာ သင် တော့ စာမြီးပွဲ ကို တပတ်တခါ စစ်တယ်
```
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0003
- train_batch_size: 128
- eval_batch_size: 128
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 200
- num_epochs: 30
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|:-------------:|:-----:|:----:|:---------------:|:--------:|
| No log | 1.0 | 18 | 1.2679 | 357.6135 |
| 1.483 | 2.0 | 36 | 1.0660 | 102.5378 |
| 1.0703 | 3.0 | 54 | 0.9530 | 106.3669 |
| 1.0703 | 4.0 | 72 | 0.8399 | 100.5343 |
| 0.8951 | 5.0 | 90 | 0.7728 | 107.6581 |
| 0.7857 | 6.0 | 108 | 0.7143 | 107.5245 |
| 0.6614 | 7.0 | 126 | 0.5174 | 104.4078 |
| 0.6614 | 8.0 | 144 | 0.3004 | 90.3384 |
| 0.3519 | 9.0 | 162 | 0.2447 | 82.4577 |
| 0.2165 | 10.0 | 180 | 0.2333 | 83.8825 |
| 0.2165 | 11.0 | 198 | 0.2022 | 77.0258 |
| 0.1532 | 12.0 | 216 | 0.1759 | 73.0632 |
| 0.1039 | 13.0 | 234 | 0.1852 | 72.0837 |
| 0.0675 | 14.0 | 252 | 0.1902 | 71.2823 |
| 0.0675 | 15.0 | 270 | 0.1882 | 70.5254 |
| 0.0517 | 16.0 | 288 | 0.2002 | 69.7240 |
| 0.0522 | 17.0 | 306 | 0.1965 | 67.7649 |
| 0.0522 | 18.0 | 324 | 0.1935 | 68.2102 |
| 0.0404 | 19.0 | 342 | 0.2132 | 67.9430 |
| 0.0308 | 20.0 | 360 | 0.2110 | 66.6963 |
| 0.0236 | 21.0 | 378 | 0.2141 | 65.9394 |
| 0.0236 | 22.0 | 396 | 0.2200 | 64.4702 |
| 0.0116 | 23.0 | 414 | 0.2227 | 63.4016 |
| 0.0055 | 24.0 | 432 | 0.2244 | 64.1585 |
| 0.0025 | 25.0 | 450 | 0.2254 | 62.4666 |
| 0.0025 | 26.0 | 468 | 0.2282 | 63.1790 |
| 0.0006 | 27.0 | 486 | 0.2320 | 61.7097 |
| 0.0002 | 28.0 | 504 | 0.2342 | 62.0659 |
| 0.0002 | 29.0 | 522 | 0.2350 | 62.0214 |
| 0.0001 | 30.0 | 540 | 0.2353 | 61.8878 |
### Framework versions
- Transformers 4.35.2
- Pytorch 2.1.1+cu121
- Datasets 2.14.5
- Tokenizers 0.15.1 |