bnriiitb/whisper-small-te-4k

Whisper Small Telugu - Naga Budigam

This model is a fine-tuned version of openai/whisper-small on the Chai_Bisket_Stories_16-08-2021_14-17 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2875
• Wer: 38.1492

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: 8
• 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: 15000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2064	0.66	500	0.2053	60.1707
0.1399	1.33	1000	0.1535	49.3269
0.1093	1.99	1500	0.1365	44.5516
0.0771	2.66	2000	0.1316	42.1136
0.0508	3.32	2500	0.1395	41.1384
0.0498	3.99	3000	0.1386	40.5395
0.0302	4.65	3500	0.1529	40.9529
0.0157	5.32	4000	0.1719	40.6667
0.0183	5.98	4500	0.1723	40.3646
0.0083	6.65	5000	0.1911	40.4335
0.0061	7.31	5500	0.2109	40.4176
0.0055	7.98	6000	0.2075	39.7021
0.0039	8.64	6500	0.2186	40.2639
0.0026	9.31	7000	0.2254	39.1032
0.0035	9.97	7500	0.2289	39.2834
0.0016	10.64	8000	0.2332	39.1456
0.0016	11.3	8500	0.2395	39.4371
0.0016	11.97	9000	0.2447	39.2410
0.0009	12.63	9500	0.2548	38.7799
0.0008	13.3	10000	0.2551	38.7481
0.0008	13.96	10500	0.2621	38.8276
0.0007	14.63	11000	0.2633	38.6686
0.0003	15.29	11500	0.2711	38.4566
0.0005	15.96	12000	0.2772	38.7852
0.0001	16.62	12500	0.2771	38.2658
0.0001	17.29	13000	0.2808	38.2393
0.0001	17.95	13500	0.2815	38.1810
0.0	18.62	14000	0.2854	38.2022
0.0	19.28	14500	0.2872	38.1333
0.0	19.95	15000	0.2875	38.1492

Framework versions

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