kattojuprashanth238/whisper-small-te-v5

Whisper Small Te - Prashanth Kattoju

This model is a fine-tuned version of openai/whisper-small-v5 on the Common Voice 17 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0948
• Wer Ortho: 32.4176
• Wer: 10.6227

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: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 50
• training_steps: 1500

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
1.4434	0.4338	50	0.9813	101.6484	70.6960
0.6517	0.8677	100	0.4498	81.8681	31.6850
0.3973	1.2950	150	0.3015	75.2747	28.3883
0.2864	1.7289	200	0.1763	65.9341	23.4432
0.1998	2.1562	250	0.1545	59.3407	21.2454
0.1631	2.5900	300	0.1231	53.2967	19.4139
0.1368	3.0174	350	0.1128	50.0	20.3297
0.0823	3.4512	400	0.1249	48.3516	16.6667
0.0795	3.8850	450	0.1094	48.3516	18.6813
0.0486	4.3124	500	0.1131	43.9560	17.7656
0.0386	4.7462	550	0.1044	43.9560	15.7509
0.0275	5.1735	600	0.1020	35.1648	16.3004
0.0323	5.6074	650	0.1050	43.4066	15.9341
0.0257	6.0347	700	0.1067	39.0110	14.6520
0.0196	6.4685	750	0.1096	39.0110	13.9194
0.0231	6.9024	800	0.1158	41.2088	14.8352
0.0213	7.3297	850	0.0915	39.0110	15.3846
0.0142	7.7636	900	0.1107	41.2088	15.3846
0.0129	8.1909	950	0.1132	37.9121	14.4689
0.01	8.6247	1000	0.1190	35.7143	13.7363
0.0123	9.0521	1050	0.1081	37.9121	14.4689
0.0144	9.4859	1100	0.1086	39.0110	14.2857
0.0115	9.9197	1150	0.1021	37.9121	16.6667
0.0097	10.3471	1200	0.0998	34.0659	14.1026
0.0075	10.7809	1250	0.0976	28.5714	10.8059
0.0062	11.2082	1300	0.1252	34.6154	12.6374
0.0081	11.6421	1350	0.1029	31.3187	13.7363
0.0058	12.0694	1400	0.0986	34.6154	11.9048
0.0092	12.5033	1450	0.1020	32.9670	13.7363
0.0047	12.9371	1500	0.0948	32.4176	10.6227

Framework versions

• Transformers 4.48.1
• Pytorch 2.5.1
• Datasets 3.2.0
• Tokenizers 0.21.0