excalibur12/pic-20s_asr-scr_w2v2-base_002

pic-20s_asr-scr_w2v2-base_003

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: 1.4215
• Per: 0.1497
• Pcc: 0.6339
• Ctc Loss: 0.5259
• Mse Loss: 0.8821

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: 3333
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2247
• training_steps: 22470
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Per	Pcc	Ctc Loss	Mse Loss
17.0464	3.0	2247	5.0097	0.9979	0.6184	3.7876	1.2860
4.3019	6.0	4494	4.2229	0.9979	0.7055	3.7328	0.6675
3.9383	9.0	6741	4.1717	0.9979	0.7012	3.7059	0.7331
3.3316	12.0	8988	2.9515	0.6216	0.6761	2.3269	0.7956
1.5694	15.0	11235	1.8634	0.2235	0.6674	0.8822	0.9706
0.8929	18.0	13482	1.5733	0.1742	0.6392	0.6657	0.8867
0.6847	21.0	15729	1.6522	0.1613	0.6497	0.5817	1.0250
0.5739	24.0	17976	1.4394	0.1534	0.6165	0.5482	0.8750
0.5063	27.0	20223	1.4105	0.1510	0.6296	0.5322	0.8668
0.4701	30.0	22470	1.4215	0.1497	0.6339	0.5259	0.8821

Framework versions

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