wav2vec2-age-gender

This model is a fine-tuned version of audeering/wav2vec2-large-robust-6-ft-age-gender on the arrow dataset. It achieves the following results on the evaluation set:

• Loss: 2.5537
• Accuracy: 0.4728
• F1 Score: 0.2652
• Mse: 1.3505
• Mae: 0.7527
• Mae^m: 1.2759

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: 0.0003
• train_batch_size: 9
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 18
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1 Score	Mse	Mae	Mae^m
1.6537	0.6116	100	1.5436	0.3951	0.1176	1.8719	0.9510	1.8697
1.444	1.2232	200	1.3586	0.4142	0.2591	1.1662	0.7629	1.0623
1.4129	1.8349	300	1.4103	0.3351	0.1886	1.1853	0.8256	1.2259
1.4089	2.4465	400	1.5204	0.3188	0.1708	1.2425	0.8556	1.1245
1.3624	3.0581	500	1.4430	0.3924	0.1947	1.2507	0.7984	1.1701
1.2793	3.6697	600	1.4646	0.4060	0.3666	1.3678	0.8174	0.8894
1.0964	4.2813	700	1.5239	0.4441	0.2978	1.1090	0.7275	0.9776
1.1362	4.8930	800	1.6786	0.4496	0.3366	1.2561	0.7548	0.8756
0.9191	5.5046	900	1.5732	0.4441	0.3106	1.0845	0.7139	0.8963
0.6949	6.1162	1000	1.8249	0.4469	0.2538	1.1008	0.7139	1.0736
0.8272	6.7278	1100	1.8225	0.4550	0.3115	1.0599	0.6948	1.0455
0.7242	7.3394	1200	2.1688	0.3951	0.2593	1.0599	0.7493	1.0673
0.7385	7.9511	1300	2.3316	0.4114	0.2894	1.2234	0.7711	1.0547
0.6214	8.5627	1400	2.2313	0.4768	0.2632	0.9455	0.6567	1.0143
0.3996	9.1743	1500	2.3696	0.4714	0.2979	1.0763	0.6948	1.0522
0.467	9.7859	1600	2.4082	0.4005	0.2902	1.1144	0.7493	0.8743
0.3559	10.3976	1700	2.9747	0.4632	0.3154	1.1281	0.7084	1.1496
0.5321	11.0092	1800	2.8344	0.4578	0.3201	0.9973	0.6812	0.9616
0.3078	11.6208	1900	3.0384	0.4251	0.3050	1.0981	0.7330	1.0609
0.216	12.2324	2000	3.1374	0.4578	0.3000	0.9564	0.6676	0.9610
0.3241	12.8440	2100	3.6605	0.4142	0.2530	1.0245	0.7193	1.1348
0.2642	13.4557	2200	3.4759	0.4278	0.3076	1.0354	0.7139	0.9519
0.1692	14.0673	2300	3.9204	0.4223	0.2506	1.0763	0.7275	1.1008
0.1729	14.6789	2400	3.8804	0.4578	0.3171	1.0545	0.6948	1.0291
0.2636	15.2905	2500	4.1746	0.4523	0.3043	1.0845	0.7084	0.9463
0.1523	15.9021	2600	4.1583	0.4169	0.2919	1.1144	0.7384	0.9452
0.101	16.5138	2700	4.2574	0.4496	0.2650	0.9837	0.6839	1.0895
0.1593	17.1254	2800	4.3649	0.4387	0.2716	1.1253	0.7275	1.1297
0.0707	17.7370	2900	4.6972	0.4060	0.2930	1.1907	0.7657	1.0690
0.1647	18.3486	3000	4.9096	0.4360	0.2576	1.0518	0.7139	1.0402
0.2	18.9602	3100	4.9426	0.4142	0.2755	1.0409	0.7248	1.0775
0.0381	19.5719	3200	4.7156	0.4441	0.2706	1.0136	0.6975	0.9823
0.1236	20.1835	3300	5.1736	0.4305	0.2666	1.0136	0.7084	0.9442
0.0415	20.7951	3400	5.2558	0.4305	0.2625	1.0681	0.7248	0.9997
0.0307	21.4067	3500	5.4287	0.3924	0.2482	1.1635	0.7711	1.0877
0.0079	22.0183	3600	5.3503	0.4169	0.2505	1.1035	0.7439	1.0535
0.0588	22.6300	3700	5.4146	0.4060	0.2740	1.1172	0.7520	0.9442
0.1119	23.2416	3800	5.5098	0.4278	0.2842	1.0899	0.7248	1.0109
0.0135	23.8532	3900	6.0306	0.4196	0.2634	1.0381	0.7221	1.0756
0.0457	24.4648	4000	5.8127	0.4469	0.2854	1.0327	0.7003	1.1027
0.1018	25.0765	4100	5.6878	0.4469	0.2939	1.1117	0.7193	1.0586
0.0082	25.6881	4200	5.6874	0.4332	0.2795	1.1090	0.7330	1.0770
0.0308	26.2997	4300	5.8580	0.4305	0.2743	1.0899	0.7302	1.0835
0.0081	26.9113	4400	6.2890	0.4332	0.2487	1.0599	0.7221	1.1730
0.0185	27.5229	4500	6.1682	0.4441	0.2758	1.0463	0.7084	1.0905
0.016	28.1346	4600	6.0424	0.4414	0.2930	1.0954	0.7248	1.1141
0.0156	28.7462	4700	6.1145	0.4414	0.2847	1.0926	0.7221	1.1320
0.0035	29.3578	4800	6.0960	0.4441	0.2912	1.1035	0.7221	1.1203

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.0
• Tokenizers 0.19.1