thennal/whisper-large-v2-ml

Whisper Large V2 Malayalam

This model is a fine-tuned version of openai/whisper-large-v2 on the ICFOSS Malayalam Speech Corpus dataset. It achieves the following results on the evaluation set:

• Loss: 0.0617
• Wer: 44.1379
• Cer: 9.6895

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: 8
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• 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: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.1071	0.13	500	0.1274	62.9885	15.0225
0.0693	0.26	1000	0.1052	57.4713	13.0696
0.054	0.39	1500	0.0902	48.0460	11.5173
0.0494	0.51	2000	0.0774	46.4368	10.7912
0.0446	0.64	2500	0.0722	46.8966	10.7161
0.0463	0.77	3000	0.0699	46.2069	10.3405
0.0347	0.9	3500	0.0662	43.6782	10.2404
0.0233	1.03	4000	0.0688	45.7471	10.4407
0.0226	1.16	4500	0.0642	44.5977	10.1152
0.0194	1.28	5000	0.0617	44.1379	9.6895

Framework versions

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