wav2vec2-base-timit-google-colab

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: 0.4659
• Wer: 0.3080

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
• train_batch_size: 8
• 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: 1000
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.5787	0.87	500	1.7648	1.0305
0.8692	1.73	1000	0.5136	0.5103
0.4346	2.6	1500	0.4364	0.4515
0.31	3.46	2000	0.3889	0.4070
0.234	4.33	2500	0.4161	0.3863
0.2054	5.19	3000	0.3845	0.3722
0.165	6.06	3500	0.4035	0.3643
0.1436	6.92	4000	0.4090	0.3623
0.1381	7.79	4500	0.4007	0.3673
0.1175	8.65	5000	0.4588	0.3632
0.1052	9.52	5500	0.4441	0.3588
0.0988	10.38	6000	0.4133	0.3489
0.0877	11.25	6500	0.4758	0.3510
0.0856	12.11	7000	0.4454	0.3425
0.0731	12.98	7500	0.4252	0.3351
0.0712	13.84	8000	0.4163	0.3370
0.0711	14.71	8500	0.4166	0.3367
0.06	15.57	9000	0.4195	0.3347
0.0588	16.44	9500	0.4697	0.3367
0.0497	17.3	10000	0.4255	0.3314
0.0523	18.17	10500	0.4676	0.3307
0.0444	19.03	11000	0.4570	0.3244
0.0435	19.9	11500	0.4307	0.3243
0.0348	20.76	12000	0.4763	0.3245
0.036	21.63	12500	0.4635	0.3238
0.0347	22.49	13000	0.4602	0.3212
0.0333	23.36	13500	0.4472	0.3195
0.0311	24.22	14000	0.4449	0.3183
0.0294	25.09	14500	0.4631	0.3175
0.025	25.95	15000	0.4466	0.3164
0.023	26.82	15500	0.4581	0.3138
0.0216	27.68	16000	0.4665	0.3114
0.0198	28.55	16500	0.4590	0.3092
0.0181	29.41	17000	0.4659	0.3080

Framework versions

• Transformers 4.20.0.dev0
• Pytorch 1.11.0+cu113
• Datasets 1.18.3
• Tokenizers 0.12.1