Modelpublic

cutten/wav2vec2-large-multilang-cv-ru-night

sourceHugging Faceapache-2.0updated 4y agoView on Hugging Face

0likes14downloads

Model Card

wav2vec2-large-multilang-cv-ru-night

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

—Loss: 0.6617
—Wer: 0.5097

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: 16
—evalbatchsize: 8
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_steps: 1000
—num_epochs: 100
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
8.725	1.58	500	3.2788	1.0
3.1184	3.15	1000	2.4018	1.0015
1.2393	4.73	1500	0.6213	0.7655
0.6899	6.31	2000	0.5518	0.6811
0.5532	7.89	2500	0.5102	0.6467
0.4604	9.46	3000	0.4887	0.6213
0.4095	11.04	3500	0.4874	0.6042
0.3565	12.62	4000	0.4810	0.5893
0.3238	14.2	4500	0.5028	0.5890
0.3011	15.77	5000	0.5475	0.5808
0.2827	17.35	5500	0.5289	0.5720
0.2659	18.93	6000	0.5496	0.5733
0.2445	20.5	6500	0.5354	0.5737
0.2366	22.08	7000	0.5357	0.5686
0.2181	23.66	7500	0.5491	0.5611
0.2146	25.24	8000	0.5591	0.5597
0.2006	26.81	8500	0.5625	0.5631
0.1912	28.39	9000	0.5577	0.5647
0.1821	29.97	9500	0.5684	0.5519
0.1744	31.55	10000	0.5639	0.5551
0.1691	33.12	10500	0.5596	0.5425
0.1577	34.7	11000	0.5770	0.5551
0.1522	36.28	11500	0.5634	0.5560
0.1468	37.85	12000	0.5815	0.5453
0.1508	39.43	12500	0.6053	0.5490
0.1394	41.01	13000	0.6193	0.5504
0.1291	42.59	13500	0.5930	0.5424
0.1345	44.16	14000	0.6283	0.5442
0.1296	45.74	14500	0.6063	0.5560
0.1286	47.32	15000	0.6248	0.5378
0.1231	48.9	15500	0.6106	0.5405
0.1189	50.47	16000	0.6164	0.5342
0.1127	52.05	16500	0.6269	0.5359
0.112	53.63	17000	0.6170	0.5390
0.1113	55.21	17500	0.6489	0.5385
0.1023	56.78	18000	0.6826	0.5490
0.1069	58.36	18500	0.6147	0.5296
0.1008	59.94	19000	0.6414	0.5332
0.1018	61.51	19500	0.6454	0.5288
0.0989	63.09	20000	0.6603	0.5303
0.0944	64.67	20500	0.6350	0.5288
0.0905	66.25	21000	0.6386	0.5247
0.0837	67.82	21500	0.6563	0.5298
0.0868	69.4	22000	0.6375	0.5208
0.0827	70.98	22500	0.6401	0.5271
0.0797	72.56	23000	0.6723	0.5191
0.0847	74.13	23500	0.6610	0.5213
0.0818	75.71	24000	0.6774	0.5254
0.0793	77.29	24500	0.6543	0.5250
0.0758	78.86	25000	0.6607	0.5218
0.0755	80.44	25500	0.6599	0.5160
0.0722	82.02	26000	0.6683	0.5196
0.0714	83.6	26500	0.6941	0.5180
0.0684	85.17	27000	0.6581	0.5167
0.0686	86.75	27500	0.6651	0.5172
0.0712	88.33	28000	0.6547	0.5208
0.0697	89.91	28500	0.6555	0.5162
0.0696	91.48	29000	0.6678	0.5107
0.0686	93.06	29500	0.6630	0.5124
0.0671	94.64	30000	0.6675	0.5143
0.0668	96.21	30500	0.6602	0.5107
0.0666	97.79	31000	0.6611	0.5097
0.0664	99.37	31500	0.6617	0.5097

Framework versions

—Transformers 4.19.2
—Pytorch 1.11.0
—Datasets 2.2.2
—Tokenizers 0.12.1