Modelpublic

jimregan/wav2vec2-xls-r-300m-phoneme-timit

sourceHugging Faceapache-2.0updated 2y agoView on Hugging Face

0likes26downloads

Model Card

working

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:

—Loss: 0.3630
—Wer: 0.6243
—Cer: 0.1316

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: 3e-05
—trainbatchsize: 8
—evalbatchsize: 8
—seed: 42
—gradientaccumulationsteps: 4
—totaltrainbatch_size: 32
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_steps: 2000
—training_steps: 10000
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.5325	11.9	1000	3.4897	1.0	0.9266
2.1973	23.81	2000	1.1350	0.8396	0.2403
1.4762	35.71	3000	0.5270	0.6845	0.1563
1.2409	47.62	4000	0.4195	0.6331	0.1403
1.1241	59.52	5000	0.3845	0.6362	0.1379
1.024	71.43	6000	0.3716	0.6321	0.1355
0.9922	83.33	7000	0.3728	0.6290	0.1331
0.9432	95.24	8000	0.3648	0.6170	0.1321
0.9279	107.14	9000	0.3643	0.6248	0.1325
0.9268	119.05	10000	0.3630	0.6243	0.1316

Framework versions

—Transformers 4.38.2
—Pytorch 2.0.1+cu118
—Datasets 2.18.0
—Tokenizers 0.15.2