Automatic Speech Recognition
Transformers
PyTorch
TensorBoard
Bengali
whisper
hf-asr-leaderboard
Generated from Trainer
Eval Results (legacy)
# Load model directly
from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq
processor = AutoProcessor.from_pretrained("Rounak28/bengaliAI-finetuned-0-10000-50-percent-new")
model = AutoModelForSpeechSeq2Seq.from_pretrained("Rounak28/bengaliAI-finetuned-0-10000-50-percent-new", device_map="auto")Quick Links
whisper-small fintuned-0-10000-50%
This model is a fine-tuned version of openai/whisper-small on the bengaliAI-kaggle dataset. It achieves the following results on the evaluation set:
- Loss: 0.6193
- Wer: 90.4418
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 4
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- 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: 50
- training_steps: 200
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.8424 | 0.4 | 100 | 0.7538 | 103.2021 |
| 0.6195 | 0.8 | 200 | 0.6193 | 90.4418 |
Framework versions
- Transformers 4.31.0
- Pytorch 2.0.0
- Datasets 2.14.4
- Tokenizers 0.13.3
- Downloads last month
- 5
Model tree for Rounak28/bengaliAI-finetuned-0-10000-50-percent-new
Base model
openai/whisper-smallEvaluation results
- Wer on bengaliAI-kaggleself-reported90.442
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("automatic-speech-recognition", model="Rounak28/bengaliAI-finetuned-0-10000-50-percent-new")