wav2vec2-base-timit-demo-google-colab

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 0.6342
• Wer: 0.5808

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
9.1358	1.19	500	3.2710	1.0
3.0499	2.38	1000	1.8976	1.0
1.279	3.56	1500	0.7502	0.8228
0.7953	4.75	2000	0.5914	0.7343
0.6451	5.94	2500	0.6152	0.7280
0.5351	7.13	3000	0.5948	0.7041
0.4633	8.31	3500	0.5585	0.6712
0.4272	9.5	4000	0.5372	0.6457
0.3803	10.69	4500	0.5404	0.6402
0.3462	11.88	5000	0.5862	0.6484
0.3302	13.06	5500	0.5991	0.6426
0.3096	14.25	6000	0.5687	0.6287
0.2839	15.44	6500	0.5798	0.6384
0.2701	16.63	7000	0.5775	0.6047
0.2507	17.81	7500	0.5638	0.6065
0.2376	19.0	8000	0.5937	0.6094
0.2264	20.19	8500	0.5944	0.6065
0.2146	21.38	9000	0.6050	0.6122
0.1947	22.57	9500	0.6283	0.5992
0.1982	23.75	10000	0.6126	0.6018
0.1924	24.94	10500	0.6075	0.5962
0.1855	26.13	11000	0.6344	0.5938
0.1839	27.32	11500	0.6118	0.5880
0.1741	28.5	12000	0.6381	0.5878
0.1726	29.69	12500	0.6342	0.5808

Framework versions

• Transformers 4.19.2
• Pytorch 1.11.0
• Datasets 2.2.2
• Tokenizers 0.12.1