wav2vec2-base-timit-demo-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: 0.5255
• Wer: 0.3330

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
3.5942	1.0	500	2.3849	1.0011
0.9765	2.01	1000	0.5907	0.5202
0.4424	3.01	1500	0.4547	0.4661
0.3008	4.02	2000	0.4194	0.4228
0.2316	5.02	2500	0.3933	0.4099
0.1921	6.02	3000	0.4532	0.3965
0.1561	7.03	3500	0.4315	0.3777
0.1378	8.03	4000	0.4463	0.3847
0.1222	9.04	4500	0.4402	0.3784
0.1076	10.04	5000	0.4253	0.3735
0.0924	11.04	5500	0.4844	0.3732
0.0866	12.05	6000	0.4758	0.3646
0.086	13.05	6500	0.6395	0.4594
0.0763	14.06	7000	0.4951	0.3647
0.0684	15.06	7500	0.4870	0.3577
0.0616	16.06	8000	0.5442	0.3591
0.0594	17.07	8500	0.5305	0.3606
0.0613	18.07	9000	0.5434	0.3546
0.0473	19.08	9500	0.4818	0.3532
0.0463	20.08	10000	0.5086	0.3514
0.042	21.08	10500	0.5017	0.3484
0.0365	22.09	11000	0.5129	0.3536
0.0336	23.09	11500	0.5411	0.3433
0.0325	24.1	12000	0.5307	0.3424
0.0282	25.1	12500	0.5261	0.3404
0.0245	26.1	13000	0.5306	0.3388
0.0257	27.11	13500	0.5242	0.3369
0.0234	28.11	14000	0.5216	0.3359
0.0221	29.12	14500	0.5255	0.3330

Framework versions

• Transformers 4.17.0
• Pytorch 1.11.0+cu113
• Datasets 1.18.3
• Tokenizers 0.12.1