Hubert-base-ASR-eu

This model is a fine-tuned version of facebook/hubert-base-ls960 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.1922
• Wer: 0.3154
• Cer: 0.0583

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: 64
• eval_batch_size: 4
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.98) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
5.5577	0.1652	1000	3.5579	0.9999	0.9883
3.3467	0.3304	2000	2.9818	0.9999	0.9883
2.934	0.4955	3000	2.8354	0.9999	0.9883
2.8012	0.6607	4000	2.8179	0.9999	0.9883
2.7896	0.8259	5000	2.8299	0.9999	0.9883
1.4043	0.9911	6000	0.9904	0.9859	0.2788
0.903	1.1563	7000	0.5862	0.8110	0.1676
0.6955	1.3214	8000	0.4828	0.7192	0.1411
0.5764	1.4866	9000	0.4330	0.6535	0.1257
0.5195	1.6518	10000	0.3948	0.6083	0.1151
0.4949	1.8170	11000	0.3698	0.5676	0.1073
0.4481	1.9822	12000	0.3475	0.5343	0.1007
0.4251	2.1473	13000	0.3286	0.5162	0.0962
0.4257	2.3125	14000	0.3139	0.4933	0.0918
0.3982	2.4777	15000	0.3016	0.4780	0.0890
0.3672	2.6429	16000	0.2925	0.4616	0.0852
0.3741	2.8081	17000	0.2823	0.4504	0.0830
0.3991	2.9732	18000	0.2748	0.4372	0.0804
0.3624	3.1384	19000	0.2684	0.4293	0.0786
0.3362	3.3036	20000	0.2638	0.4180	0.0772
0.3209	3.4688	21000	0.2577	0.4072	0.0752
0.314	3.6340	22000	0.2498	0.4042	0.0744
0.3066	3.7991	23000	0.2452	0.3957	0.0730
0.3206	3.9643	24000	0.2466	0.3877	0.0717
0.3009	4.1295	25000	0.2396	0.3826	0.0708
0.3803	4.2947	26000	0.2391	0.3814	0.0705
0.2932	4.4599	27000	0.2334	0.3748	0.0692
0.2825	4.6250	28000	0.2292	0.3679	0.0679
0.2882	4.7902	29000	0.2253	0.3655	0.0675
0.2792	4.9554	30000	0.2233	0.3612	0.0667
0.2829	5.1206	31000	0.2224	0.3584	0.0664
0.2778	5.2858	32000	0.2221	0.3524	0.0652
0.2736	5.4509	33000	0.2160	0.3516	0.0647
0.2999	5.6161	34000	0.2156	0.3483	0.0646
0.2586	5.7813	35000	0.2149	0.3445	0.0634
0.2704	5.9465	36000	0.2132	0.3452	0.0635
0.2607	6.1117	37000	0.2112	0.3377	0.0626
0.2577	6.2768	38000	0.2074	0.3366	0.0620
0.2477	6.4420	39000	0.2064	0.3355	0.0617
0.2611	6.6072	40000	0.2062	0.3339	0.0615
0.2428	6.7724	41000	0.2044	0.3316	0.0611
0.246	6.9376	42000	0.2028	0.3314	0.0609
0.2664	7.1027	43000	0.2006	0.3295	0.0605
0.25	7.2679	44000	0.2027	0.3263	0.0601
0.2458	7.4331	45000	0.2009	0.3241	0.0599
0.2446	7.5983	46000	0.1995	0.3238	0.0598
0.2377	7.7635	47000	0.1997	0.3226	0.0594
0.2428	7.9286	48000	0.1981	0.3210	0.0592
0.2426	8.0938	49000	0.1964	0.3202	0.0591
0.253	8.2590	50000	0.1950	0.3191	0.0590
0.2635	8.4242	51000	0.1943	0.3191	0.0590
0.2789	8.5894	52000	0.1940	0.3175	0.0589
0.2608	8.7545	53000	0.1957	0.3173	0.0587
0.2231	8.9197	54000	0.1932	0.3167	0.0584
0.2406	9.0849	55000	0.1933	0.3164	0.0586
0.2347	9.2501	56000	0.1915	0.3161	0.0585
0.2502	9.4153	57000	0.1935	0.3153	0.0583
0.2353	9.5804	58000	0.1926	0.3154	0.0583
0.2358	9.7456	59000	0.1918	0.3156	0.0582
0.2395	9.9108	60000	0.1922	0.3154	0.0583

Framework versions

• Transformers 4.49.0
• Pytorch 2.6.0+cu124
• Datasets 3.3.2
• Tokenizers 0.21.0