Modelpublic

alexziweiwang/exp21-uaspeech-foundation

sourceHugging Faceupdated 4y agoView on Hugging Face

0likes6downloads

Model Card

exp21-uaspeech-foundation

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

—Loss: 2.4644
—Wer: 1.2633

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: 0.0001
—trainbatchsize: 4
—evalbatchsize: 8
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_steps: 1000
—num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
27.1608	0.67	500	3.0053	1.0
3.2225	1.35	1000	2.8180	1.0
2.7882	2.02	1500	2.5589	1.3125
2.4911	2.7	2000	2.1110	1.9129
2.1921	3.37	2500	1.8404	1.8504
1.833	4.04	3000	1.5873	1.8030
1.51	4.72	3500	1.3015	1.6951
1.2721	5.39	4000	1.2160	1.6136
1.1451	6.06	4500	1.4102	1.6212
0.9631	6.74	5000	1.1730	1.5587
0.9608	7.41	5500	1.4232	1.5473
0.8454	8.09	6000	1.3427	1.4830
0.7377	8.76	6500	1.3691	1.4716
0.6465	9.43	7000	1.4927	1.4489
0.6538	10.11	7500	1.3825	1.4072
0.591	10.78	8000	1.3166	1.4186
0.5476	11.46	8500	1.2636	1.3826
0.515	12.13	9000	1.4861	1.4072
0.5027	12.8	9500	1.7483	1.3826
0.4714	13.48	10000	1.4656	1.3598
0.4506	14.15	10500	1.4430	1.3693
0.3944	14.82	11000	2.0191	1.3352
0.3604	15.5	11500	1.6739	1.3428
0.3533	16.17	12000	1.7428	1.3314
0.3345	16.85	12500	1.6851	1.3277
0.3133	17.52	13000	1.8589	1.3106
0.2772	18.19	13500	2.0114	1.2955
0.2789	18.87	14000	2.0333	1.2879
0.2361	19.54	14500	2.6048	1.3030
0.2406	20.22	15000	2.8444	1.2841
0.2673	20.89	15500	1.9573	1.2746
0.2067	21.56	16000	2.1780	1.2936
0.2281	22.24	16500	2.2629	1.2765
0.1735	22.91	17000	2.2492	1.2822
0.1902	23.58	17500	2.0317	1.2803
0.1659	24.26	18000	2.0658	1.2652
0.1746	24.93	18500	2.2515	1.2708
0.1409	25.61	19000	2.1689	1.2595
0.1461	26.28	19500	2.2441	1.2614
0.1699	26.95	20000	2.1327	1.2633
0.1378	27.63	20500	2.1795	1.2670
0.1108	28.3	21000	2.4057	1.2576
0.1505	28.98	21500	2.4389	1.2595
0.1492	29.65	22000	2.4644	1.2633

Framework versions

—Transformers 4.23.1
—Pytorch 1.12.1+cu113
—Datasets 1.18.3
—Tokenizers 0.13.2