Whisper Small ar - Mohammed Bakheet
ูู ูุฐุฌ ููุงู ููุชุนุฑู ุนูู ุงูุตูุชุ ูุฐุง ุงููู ูุฐุฌ ูุชู ูุฒ ุจุฏูุฉ ุนุงููุฉ ูู ุงูุชุนุฑู ุนูู ุงูุตูุช ุจุงููุบุฉ ุงูุนุฑุจูุฉ.
This model is a fine-tuned version of openai/whisper-large on the Common Voice 11.0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.1921
- Wer: 12.6150
Model description
This model is a fine-tuned version of openai/whisper-large on the Common Voice 11.0 dataset. It achieves 12.61 WER. Data augmentation can be implemented to further improve the model performance.
Intended uses & limitations
from datasets import load_dataset
from transformers import WhisperProcessor, WhisperForConditionalGeneration
from datasets import Audio
# load the dataset
test_dataset = load_dataset("mozilla-foundation/common_voice_11_0", "ar", split="test", use_auth_token=True, trust_remote_code=True)
# get the processor and model from mohammed/whisper-small-arabic-cv-11
processor = WhisperProcessor.from_pretrained("mohammed/whisper-large-arabic-cv-11")
model = WhisperForConditionalGeneration.from_pretrained("mohammed/whisper-large-arabic-cv-11")
model.config.forced_decoder_ids = None
# resample the audio files to 16000
test_dataset = test_dataset.cast_column("audio", Audio(sampling_rate=16000))
# get 10 exmaples of model transcription
for i in range(10):
sample = test_dataset[i]["audio"]
input_features = processor(sample["array"], sampling_rate=sample["sampling_rate"], return_tensors="pt").input_features
predicted_ids = model.generate(input_features)
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=False)
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
print(f"{i} Reference Sentence: {test_dataset[i]['sentence']}")
print(f"{i} Predicted Sentence: {transcription[0]}")
0 Reference Sentence: ุฒุงุฑูู ูู ุฃูุงุฆู ุงูุดูุฑ ุจุฏุฑู
0 Predicted Sentence: ุฒุงุฑูู ูู ุฃูุงุฆู ุงูุดูุฑ ุจุฏุฑู
1 Reference Sentence: ุฅุจูู ุจุทู.
1 Predicted Sentence: ุงุจูู ุจุทู
2 Reference Sentence: ุงููุงุนุธ ุงูุฃู
ุฑุฏ ูุฐุง ุงูุฐู
2 Predicted Sentence: ุฃูุงุนุฒ ุงูุฃู
ุฑุฌ ูุฐุง ุงูุฐู
3 Reference Sentence: ุณู
ุญ ูู ูุฐุง ุจุงูุชุฎุตุต ูู ุงูุจุฑููุฒ ุงูุตุบูุฑุ ุงูุฐู ูุชู
ุฅูุชุงุฌู ุจุดูู ุฑุฆูุณู ูู
ุฑุจุญ ููุชุตุฏูุฑ.
3 Predicted Sentence: ุณู
ุญ ูู ูุฐุง ุจุงูุชุฎุตุต ูู ุงูุจูููุฒ ุงูุตุบูุฑ ุงูุฐู ุงุนุชู
ุฏ ู
ูุชุงุฌู ุจุดูู ุฑุฆูุณู ูุบุฑุจุญ ููุชุตุฏูุฑ
4 Reference Sentence: ุฃูุฏูู ููู
ุ
4 Predicted Sentence: ุฃูุฏูู ููู
5 Reference Sentence: ูุง ูุฏูู
ู ูุณู
ุจู ุงูู ุงูุตูุจุงุก
5 Predicted Sentence: ูุง ูุฏูู
ู ูุณู
ุจู ุฅูู ุงูุตุญุจุงุก
6 Reference Sentence: ุฅูู ุชูุจุฑ ุงูู
ุดููุฉ.
6 Predicted Sentence: ุฅูู ุชูุจุฑ ุงูู
ุดููุฉ
7 Reference Sentence: ูุฑุบุจ ุฃู ููุชูู ุจู.
7 Predicted Sentence: ูุฑุบุจ ุฃู ููุชูู ุจู
8 Reference Sentence: ุฅููู
ูุง ูุนุฑููู ูู
ุงุฐุง ุญุชู.
8 Predicted Sentence: ุฅููู
ูุง ูุนุฑููู ูู
ุงุฐุง ุญุชู
9 Reference Sentence: ุณูุณุนุฏูู ู
ุณุงุนุฏุชู ุฃู ููุช ุชุญุจ.
9 Predicted Sentence: ุณูุณุนุฏูู ู
ุณุงุนุฏุชู ุฃู ููุช ุชุญุจ
Training and evaluation data
This model is trained on the Common Voice 11.0 dataset.
Training procedure
The model is trained on 64 cores CPU, Nvidia A100 GPU with 48 VRAM, and 100GB Disk space. The GPU utilization reached 100%. Please check the training hyperparameters below.
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 4
- eval_batch_size: 4
- seed: 42
- gradient_accumulation_steps: 16
- total_train_batch_size: 64
- 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
Training results
Training Loss | Epoch | Step | Validation Loss | Wer |
---|---|---|---|---|
0.1952 | 1.6630 | 1000 | 0.1843 | 14.0098 |
0.0339 | 3.3261 | 2000 | 0.1921 | 12.6150 |
Framework versions
- Transformers 4.43.3
- Pytorch 2.2.0
- Datasets 2.20.0
- Tokenizers 0.19.1
- Downloads last month
- 27
Model tree for mohammed/whisper-large-arabic-cv-11
Base model
openai/whisper-large