wav2vec2-base-ami_multi-nithin4

This model is a fine-tuned version of facebook/wav2vec2-base on the AMI-IHM dataset. It achieves the following results on the evaluation set:

  • Loss: 2.0790
  • Wer: 0.4478

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
  • train_batch_size: 8
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 32
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 500
  • num_epochs: 30.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
2.8893 1.07 2500 3.7944 1.0000
2.0331 2.13 5000 2.0323 0.5840
1.9009 3.2 7500 1.8876 0.5173
1.8367 4.27 10000 2.1239 0.4847
1.8007 5.33 12500 1.9126 0.4684
1.743 6.4 15000 2.0750 0.4570
1.7329 7.47 17500 1.9226 0.4460
1.7013 8.53 20000 1.9677 0.4392
1.6674 9.6 22500 1.9064 0.4360
1.6568 10.67 25000 1.8144 0.4304
1.6507 11.73 27500 1.8881 0.4248
1.5973 12.8 30000 1.7907 0.4267
1.6316 13.87 32500 1.7567 0.4207
1.6053 14.93 35000 1.7838 0.4192
1.599 16.0 37500 1.8054 0.4181
1.5629 17.06 40000 1.7739 0.4135
1.6124 18.13 42500 2.0690 0.4138
1.5623 19.2 45000 1.9308 0.4144
1.5524 20.26 47500 1.8130 0.4121
1.5654 21.33 50000 1.8344 0.4131
1.5552 22.4 52500 1.9365 0.4116
1.5357 23.46 55000 1.9330 0.4114
1.534 24.53 57500 1.8155 0.4079
1.5333 25.6 60000 1.7895 0.4069
1.5315 26.66 62500 1.7903 0.4082
1.5174 27.73 65000 1.8356 0.4080
1.5209 28.8 67500 1.8147 0.4077
1.5696 29.86 70000 1.8219 0.4076

Framework versions

  • Transformers 4.12.0.dev0
  • Pytorch 1.9.1
  • Datasets 1.12.2.dev0
  • Tokenizers 0.10.3
Downloads last month
27
Hosted inference API
Automatic Speech Recognition
or
This model can be loaded on the Inference API on-demand.