wav2vec2-large-xlsr-mn_en_v1

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.9588
• Wer: 0.3868

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: 6
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 12
• 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	Wer
3.3016	0.31	400	1.8974	0.9524
1.4532	0.62	800	0.9980	0.7182
1.0379	0.93	1200	0.7873	0.6193
0.836	1.24	1600	0.7248	0.5773
0.7698	1.55	2000	0.6732	0.5495
0.7235	1.86	2400	0.6309	0.5178
0.6405	2.17	2800	0.6529	0.5105
0.6158	2.48	3200	0.6189	0.4981
0.5668	2.79	3600	0.6242	0.4865
0.5359	3.1	4000	0.6143	0.4703
0.5	3.41	4400	0.6288	0.4708
0.5006	3.72	4800	0.5983	0.4630
0.4874	4.03	5200	0.6253	0.4632
0.4425	4.34	5600	0.6035	0.4477
0.4364	4.65	6000	0.5864	0.4451
0.4342	4.96	6400	0.5920	0.4467
0.3932	5.27	6800	0.6110	0.4483
0.3919	5.58	7200	0.6228	0.4411
0.3996	5.89	7600	0.6215	0.4401
0.3724	6.2	8000	0.6375	0.4463
0.3541	6.51	8400	0.5837	0.4332
0.3573	6.82	8800	0.6098	0.4482
0.3495	7.13	9200	0.5830	0.4364
0.3238	7.44	9600	0.5945	0.4343
0.3244	7.75	10000	0.6099	0.4375
0.3296	8.06	10400	0.6191	0.4339
0.2991	8.37	10800	0.6058	0.4273
0.3043	8.68	11200	0.6060	0.4289
0.3034	8.99	11600	0.6172	0.4328
0.2711	9.3	12000	0.6674	0.4241
0.2869	9.6	12400	0.6451	0.4186
0.2828	9.91	12800	0.6250	0.4198
0.2586	10.22	13200	0.6477	0.4177
0.2625	10.53	13600	0.6563	0.4173
0.2588	10.84	14000	0.6566	0.4277
0.2431	11.15	14400	0.6822	0.4230
0.246	11.46	14800	0.6835	0.4188
0.2372	11.77	15200	0.6824	0.4319
0.2427	12.08	15600	0.7079	0.4188
0.2232	12.39	16000	0.6642	0.4131
0.2186	12.7	16400	0.6705	0.4158
0.2268	13.01	16800	0.6710	0.4179
0.2037	13.32	17200	0.6766	0.4152
0.2082	13.63	17600	0.6658	0.4155
0.2122	13.94	18000	0.6845	0.4248
0.1933	14.25	18400	0.7068	0.4107
0.1933	14.56	18800	0.7056	0.4149
0.1868	14.87	19200	0.6932	0.4152
0.1891	15.18	19600	0.7369	0.4104
0.1741	15.49	20000	0.7334	0.4095
0.1785	15.8	20400	0.7364	0.4142
0.1794	16.11	20800	0.7588	0.4146
0.167	16.42	21200	0.7607	0.4115
0.1666	16.73	21600	0.7603	0.4137
0.1715	17.04	22000	0.7678	0.4087
0.1641	17.35	22400	0.7855	0.4092
0.1514	17.66	22800	0.7811	0.4116
0.1648	17.97	23200	0.7801	0.4111
0.1421	18.28	23600	0.8001	0.4123
0.1528	18.59	24000	0.7936	0.4043
0.1516	18.9	24400	0.8146	0.4115
0.145	19.21	24800	0.7926	0.4093
0.1405	19.52	25200	0.8165	0.4083
0.1406	19.83	25600	0.8154	0.4060
0.1354	20.14	26000	0.8114	0.4026
0.1307	20.45	26400	0.8073	0.4023
0.1314	20.76	26800	0.8122	0.4018
0.1236	21.07	27200	0.8162	0.4043
0.1245	21.38	27600	0.8339	0.4002
0.1207	21.69	28000	0.8461	0.4059
0.1228	22.0	28400	0.8508	0.3984
0.1127	22.31	28800	0.8681	0.4026
0.1178	22.62	29200	0.8610	0.4006
0.1169	22.93	29600	0.8775	0.3996
0.1115	23.24	30000	0.8591	0.3999
0.105	23.55	30400	0.8945	0.3958
0.1085	23.86	30800	0.8954	0.3933
0.1054	24.17	31200	0.8982	0.3940
0.1007	24.48	31600	0.9025	0.3962
0.1018	24.79	32000	0.8998	0.3962
0.0973	25.1	32400	0.9150	0.3917
0.0969	25.41	32800	0.9348	0.3915
0.0934	25.72	33200	0.8974	0.3930
0.0983	26.03	33600	0.9129	0.3960
0.0962	26.34	34000	0.9215	0.3901
0.092	26.65	34400	0.9314	0.3881
0.0894	26.96	34800	0.9286	0.3885
0.087	27.27	35200	0.9277	0.3882
0.0906	27.58	35600	0.9400	0.3887
0.0886	27.89	36000	0.9285	0.3890
0.086	28.2	36400	0.9455	0.3890
0.0821	28.51	36800	0.9504	0.3874
0.0843	28.81	37200	0.9445	0.3875
0.0802	29.12	37600	0.9565	0.3872
0.0821	29.43	38000	0.9587	0.3871
0.0852	29.74	38400	0.9588	0.3868

Framework versions

• Transformers 4.24.0
• Pytorch 1.12.1+cu102
• Datasets 1.18.3
• Tokenizers 0.13.2