las-20s_asr-scr_w2v2-base_001

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

• Loss: 1.7692
• Per: 0.1618
• Pcc: 0.6714
• Ctc Loss: 0.5287
• Mse Loss: 1.1764

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: 16
• eval_batch_size: 1
• seed: 1111
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2226
• training_steps: 22260
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Per	Pcc	Ctc Loss	Mse Loss
39.2618	1.0	742	18.2143	0.9897	0.3509	5.3212	12.9207
9.6068	2.0	1484	4.9042	0.9897	0.6173	3.8345	1.0954
4.8526	3.0	2226	5.0768	0.9897	0.6396	3.7815	1.3610
4.5882	4.0	2968	4.7560	0.9897	0.6884	3.7659	1.0929
4.4015	5.0	3710	4.6983	0.9897	0.6865	3.7593	1.0752
4.2302	6.0	4452	4.9552	0.9897	0.6803	3.7220	1.3942
4.0589	7.0	5194	4.8305	0.9897	0.6951	3.6546	1.3568
3.8724	8.0	5936	4.4403	0.9860	0.6980	3.5953	1.0528
3.6873	9.0	6678	4.1864	0.9719	0.6821	3.4283	0.9776
3.3602	10.0	7420	3.8269	0.9710	0.6819	2.9748	1.0437
2.7948	11.0	8162	3.3716	0.8033	0.6600	2.2695	1.2205
2.1454	12.0	8904	2.5213	0.5346	0.6626	1.5874	0.9981
1.6658	13.0	9646	2.3257	0.3574	0.6619	1.2086	1.1291
1.3504	14.0	10388	2.1408	0.2585	0.6690	0.9703	1.1523
1.1362	15.0	11130	2.1374	0.2231	0.6696	0.8224	1.2663
1.01	16.0	11872	1.9479	0.2058	0.6627	0.7386	1.1614
0.9113	17.0	12614	2.5121	0.1947	0.6641	0.6834	1.7065
0.8491	18.0	13356	2.1075	0.1857	0.6662	0.6464	1.3748
0.7892	19.0	14098	2.1469	0.1795	0.6725	0.6196	1.4298
0.7354	20.0	14840	2.0128	0.1745	0.6604	0.5952	1.3312
0.6968	21.0	15582	1.9962	0.1719	0.6711	0.5793	1.3293
0.6684	22.0	16324	1.7687	0.1689	0.6746	0.5668	1.1428
0.6384	23.0	17066	1.6411	0.1670	0.6702	0.5582	1.0406
0.6081	24.0	17808	1.7926	0.1649	0.6720	0.5454	1.1815
0.6029	25.0	18550	1.5970	0.1638	0.6694	0.5426	1.0173
0.5845	26.0	19292	1.6099	0.1633	0.6718	0.5400	1.0310
0.5701	27.0	20034	1.6468	0.1631	0.6725	0.5370	1.0654
0.5512	28.0	20776	1.6962	0.1621	0.6696	0.5322	1.1116
0.5411	29.0	21518	1.8220	0.1616	0.6689	0.5303	1.2200
0.5394	30.0	22260	1.7692	0.1618	0.6714	0.5287	1.1764

Framework versions

• Transformers 4.38.1
• Pytorch 2.0.1
• Datasets 2.18.0
• Tokenizers 0.15.2