Siyong/MT

wav2vec-base-Millad_TIMIT

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: 1.3772
• Wer: 0.6859
• Cer: 0.3217

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: 5000
• num_epochs: 60
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
No log	2.36	2000	2.6233	1.0130	0.6241
No log	4.73	4000	2.2206	0.9535	0.5032
No log	7.09	6000	2.3036	0.9368	0.5063
1.235	9.46	8000	1.9932	0.9275	0.5032
1.235	11.82	10000	2.0207	0.8922	0.4498
1.235	14.18	12000	1.6171	0.7993	0.3976
1.235	16.55	14000	1.6729	0.8309	0.4209
0.2779	18.91	16000	1.7043	0.8141	0.4340
0.2779	21.28	18000	1.7426	0.7658	0.3960
0.2779	23.64	20000	1.5230	0.7361	0.3830
0.2779	26.0	22000	1.4286	0.7658	0.3794
0.1929	28.37	24000	1.4450	0.7379	0.3644
0.1929	30.73	26000	1.5922	0.7491	0.3826
0.1929	33.1	28000	1.4443	0.7454	0.3617
0.1929	35.46	30000	1.5450	0.7268	0.3621
0.1394	37.83	32000	1.9268	0.7491	0.3763
0.1394	40.19	34000	1.7094	0.7342	0.3783
0.1394	42.55	36000	1.4024	0.7082	0.3494
0.1394	44.92	38000	1.4467	0.6840	0.3395
0.104	47.28	40000	1.4145	0.6933	0.3407
0.104	49.65	42000	1.3901	0.6970	0.3403
0.104	52.01	44000	1.3589	0.6636	0.3348
0.104	54.37	46000	1.3716	0.6952	0.3340
0.0781	56.74	48000	1.4025	0.6896	0.3312
0.0781	59.1	50000	1.3772	0.6859	0.3217

Framework versions

• Transformers 4.17.0
• Pytorch 1.12.0+cu113
• Datasets 1.18.3
• Tokenizers 0.12.1