wav2vec2-base-nsc-demo-4

This model is a fine-tuned version of facebook/wav2vec2-base-960h on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3016
• Wer: 0.1720

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: 5.9591386586384804e-05
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 51

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7542	2.27	50	0.3351	0.1948
0.3912	4.55	100	0.3016	0.1720
0.2497	6.82	150	0.3247	0.1757
0.201	9.09	200	0.3111	0.1728
0.1602	11.36	250	0.3259	0.1723
0.1334	13.64	300	0.3431	0.1765
0.1083	15.91	350	0.3413	0.1726
0.1114	18.18	400	0.4089	0.1768
0.0828	20.45	450	0.3531	0.1765
0.0926	22.73	500	0.3481	0.1755
0.093	25.0	550	0.3379	0.1742
0.0772	27.27	600	0.3628	0.1779
0.0701	29.55	650	0.3747	0.1773
0.0736	31.82	700	0.3834	0.1808
0.0607	34.09	750	0.3747	0.1742
0.0629	36.36	800	0.3683	0.1734
0.0713	38.64	850	0.3671	0.1744
0.0728	40.91	900	0.3632	0.1749
0.0696	43.18	950	0.3615	0.1731
0.0638	45.45	1000	0.3591	0.1755
0.0552	47.73	1050	0.3608	0.1779
0.0578	50.0	1100	0.3630	0.1752

Framework versions

• Transformers 4.33.0
• Pytorch 2.0.0
• Datasets 2.1.0
• Tokenizers 0.13.3