daniel0321forever/personalize-whiper-model

Personalized Whisper Small - Wei Fang

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

• Loss: 0.2946
• Wer: 8.3723

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: 32
• eval_batch_size: 8
• 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: 4000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.9814	0.32	100	0.8164	13.2172
0.3013	0.64	200	0.2578	11.7722
0.2074	0.96	300	0.2192	10.4972
0.1429	1.28	400	0.2245	11.0072
0.1565	1.6	500	0.2102	10.6247
0.1554	1.92	600	0.2137	11.2197
0.0684	2.24	700	0.2139	8.8823
0.0717	2.56	800	0.2142	9.6898
0.0795	2.88	900	0.2128	9.2223
0.0329	3.21	1000	0.2341	9.3073
0.03	3.53	1100	0.2324	8.9673
0.0319	3.85	1200	0.2365	9.0948
0.0137	4.17	1300	0.2403	9.0523
0.0145	4.49	1400	0.2470	8.3723
0.0145	4.81	1500	0.2596	9.4348
0.0067	5.13	1600	0.2544	8.9248
0.0088	5.45	1700	0.2553	8.4573
0.0065	5.77	1800	0.2729	8.8823
0.0018	6.09	1900	0.2680	8.7973
0.0023	6.41	2000	0.2710	9.0948
0.0018	6.73	2100	0.2762	8.8398
0.002	7.05	2200	0.2717	8.5848
0.0011	7.37	2300	0.2784	8.5423
0.0012	7.69	2400	0.2797	8.4573
0.0011	8.01	2500	0.2782	8.3723
0.0007	8.33	2600	0.2838	8.1598
0.0007	8.65	2700	0.2826	8.2448
0.0013	8.97	2800	0.2835	8.4148
0.0006	9.29	2900	0.2913	8.2448
0.0006	9.62	3000	0.2906	8.4148
0.001	9.94	3100	0.2886	8.6273
0.0005	10.26	3200	0.2890	8.3723
0.0005	10.58	3300	0.2905	8.3723
0.0005	10.9	3400	0.2917	8.4573
0.0008	11.22	3500	0.2927	8.3723
0.0019	11.54	3600	0.2932	8.3723
0.0004	11.86	3700	0.2939	8.3723
0.0004	12.18	3800	0.2941	8.3723
0.0005	12.5	3900	0.2944	8.3723
0.0005	12.82	4000	0.2946	8.3723

Framework versions

• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2