metadata

license: apache-2.0
tags:
  - automatic-speech-recognition
  - timit_asr
  - generated_from_trainer
datasets:
  - timit_asr
model-index:
  - name: sew-d-small-100k-ft-timit
    results: []

sew-d-small-100k-ft-timit

This model is a fine-tuned version of asapp/sew-d-small-100k on the TIMIT_ASR - NA dataset. It achieves the following results on the evaluation set:

Loss: 1.7430
Wer: 0.8090

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: 32
eval_batch_size: 1
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: 20.0
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.2068	0.69	100	4.0802	1.0
2.9806	1.38	200	2.9792	1.0
2.9781	2.07	300	2.9408	1.0
2.9656	2.76	400	2.9143	1.0
2.8953	3.45	500	2.8774	1.0
2.7714	4.14	600	2.7774	0.9999
2.6883	4.83	700	2.6552	0.9854
2.6477	5.52	800	2.5424	1.0114
2.3607	6.21	900	2.4291	1.1012
2.1048	6.9	1000	2.2526	0.9744
2.2448	7.59	1100	2.1697	0.9332
2.3525	8.28	1200	2.0429	0.8721
2.099	8.97	1300	2.0076	0.9183
1.8236	9.66	1400	1.9695	0.8263
1.6602	10.34	1500	1.9328	0.8537
2.2146	11.03	1600	1.8795	0.8443
1.9278	11.72	1700	1.8749	0.8075
1.6324	12.41	1800	1.8504	0.8091
1.6517	13.1	1900	1.8228	0.8233
2.0463	13.79	2000	1.8194	0.8564
1.8736	14.48	2100	1.7950	0.8225
1.6233	15.17	2200	1.7896	0.8429
1.4982	15.86	2300	1.7793	0.8441
1.8955	16.55	2400	1.7640	0.8058
1.8253	17.24	2500	1.7668	0.8134
1.5332	17.93	2600	1.7611	0.8015
1.452	18.62	2700	1.7497	0.8072
1.7609	19.31	2800	1.7464	0.8062
1.7743	20.0	2900	1.7430	0.8090

Framework versions

Transformers 4.12.0.dev0
Pytorch 1.8.1
Datasets 1.14.1.dev0
Tokenizers 0.10.3