ALM/whisper-it-small-augmented

Whisper Small Italian - Robust

This model is a fine-tuned version of openai/whisper-small on the mozilla-foundation/common_voice_11_0 it dataset. It achieves the following results on the evaluation set:

• Loss: 0.1980
• Wer: 8.7457

IMPORTANT The model has been trained using data augmentation to improve its generalization capabilities and robustness. The results on the eval set during training are biased towards data augmentation applied to evaluation data.

Results on eval set

• Mozilla CV 11.0 - Italian: 8.00 wer (using official script)
• TODO

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: 5e-05
• train_batch_size: 64
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 25000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1927	1.0	2500	0.2506	14.9991
0.0736	2.01	5000	0.2258	12.7864
0.0413	3.01	7500	0.2144	11.4508
0.0201	4.02	10000	0.2146	10.8774
0.0129	5.02	12500	0.2127	10.6920
0.0091	6.03	15000	0.2117	10.2867
0.0043	7.03	17500	0.2076	9.6860
0.0018	8.04	20000	0.2065	9.4235
0.0013	9.04	22500	0.2003	8.9105
0.0009	10.05	25000	0.1978	8.7497

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1
• Tokenizers 0.13.2