wav2vec2-large-xlsr-korean-demo-test

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.9829
• Wer: 0.5580

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• 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
8.1603	0.4	400	5.0560	1.0
3.0513	0.79	800	2.1226	0.9984
1.7673	1.19	1200	1.2358	0.9273
1.4577	1.59	1600	1.0198	0.8512
1.3308	1.98	2000	0.9258	0.8325
1.1798	2.38	2400	0.8587	0.7933
1.1268	2.77	2800	0.8166	0.7677
1.0664	3.17	3200	0.7911	0.7428
0.9923	3.57	3600	0.7964	0.7481
1.0059	3.96	4000	0.7617	0.7163
0.9141	4.36	4400	0.7854	0.7280
0.8939	4.76	4800	0.7364	0.7160
0.8689	5.15	5200	0.7895	0.6996
0.8236	5.55	5600	0.7756	0.7100
0.8409	5.95	6000	0.7433	0.6915
0.7643	6.34	6400	0.7566	0.6993
0.7601	6.74	6800	0.7873	0.6836
0.7367	7.14	7200	0.7353	0.6640
0.7099	7.53	7600	0.7421	0.6766
0.7084	7.93	8000	0.7396	0.6740
0.6837	8.32	8400	0.7717	0.6647
0.6513	8.72	8800	0.7763	0.6798
0.6458	9.12	9200	0.7659	0.6494
0.6132	9.51	9600	0.7693	0.6511
0.6287	9.91	10000	0.7555	0.6469
0.6008	10.31	10400	0.7606	0.6408
0.5796	10.7	10800	0.7622	0.6397
0.5753	11.1	11200	0.7816	0.6510
0.5531	11.5	11600	0.8351	0.6658
0.5215	11.89	12000	0.7843	0.6416
0.5205	12.29	12400	0.7674	0.6256
0.5219	12.69	12800	0.7594	0.6287
0.5186	13.08	13200	0.7863	0.6243
0.473	13.48	13600	0.8209	0.6469
0.4938	13.87	14000	0.8002	0.6241
0.474	14.27	14400	0.8008	0.6122
0.442	14.67	14800	0.8047	0.6089
0.4521	15.06	15200	0.8341	0.6123
0.4289	15.46	15600	0.8217	0.6122
0.4278	15.86	16000	0.8400	0.6152
0.4051	16.25	16400	0.8634	0.6182
0.4063	16.65	16800	0.8486	0.6097
0.4101	17.05	17200	0.8825	0.6002
0.3896	17.44	17600	0.9575	0.6205
0.3833	17.84	18000	0.8946	0.6216
0.3678	18.24	18400	0.8905	0.5952
0.3715	18.63	18800	0.8918	0.5994
0.3748	19.03	19200	0.8856	0.5953
0.3485	19.42	19600	0.9326	0.5906
0.3522	19.82	20000	0.9237	0.5932
0.3551	20.22	20400	0.9274	0.5932
0.3339	20.61	20800	0.9075	0.5883
0.3354	21.01	21200	0.9306	0.5861
0.318	21.41	21600	0.8994	0.5854
0.3235	21.8	22000	0.9114	0.5831
0.3201	22.2	22400	0.9415	0.5867
0.308	22.6	22800	0.9695	0.5807
0.3049	22.99	23200	0.9166	0.5765
0.2858	23.39	23600	0.9643	0.5746
0.2938	23.79	24000	0.9461	0.5724
0.2856	24.18	24400	0.9658	0.5710
0.2827	24.58	24800	0.9534	0.5693
0.2745	24.97	25200	0.9436	0.5675
0.2705	25.37	25600	0.9849	0.5701
0.2656	25.77	26000	0.9854	0.5662
0.2645	26.16	26400	0.9795	0.5662
0.262	26.56	26800	0.9496	0.5626
0.2553	26.96	27200	0.9787	0.5659
0.2602	27.35	27600	0.9814	0.5640
0.2519	27.75	28000	0.9816	0.5631
0.2386	28.15	28400	1.0012	0.5580
0.2398	28.54	28800	0.9892	0.5567
0.2368	28.94	29200	0.9909	0.5590
0.2366	29.34	29600	0.9827	0.5567
0.2347	29.73	30000	0.9829	0.5580

Framework versions

• Transformers 4.21.1
• Pytorch 1.12.1+cu113
• Datasets 2.4.0
• Tokenizers 0.12.1