ivanlau's picture
added eval.sh ad eval results; added robust speech event tag
d3c1869
|
raw
history blame
2.15 kB
metadata
language:
  - zh-HK
license: apache-2.0
tags:
  - automatic-speech-recognition
  - mozilla-foundation/common_voice_8_0
  - generated_from_trainer
  - zh-HK
  - robust-speech-event
datasets:
  - common_voice
model-index:
  - name: ''
    results: []

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - ZH-HK dataset. It achieves the following results on the evaluation set:

  • Loss: 2.6726
  • Wer: 0.9815

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: 0.0003
  • train_batch_size: 32
  • eval_batch_size: 16
  • seed: 42
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 64
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 500
  • num_epochs: 10.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
No log 1.0 183 47.8442 1.0
No log 2.0 366 6.3109 1.0
41.8902 3.0 549 6.2392 1.0
41.8902 4.0 732 5.9739 1.1123
41.8902 5.0 915 4.9014 1.9474
5.5817 6.0 1098 3.9892 1.0188
5.5817 7.0 1281 3.5080 1.0104
5.5817 8.0 1464 3.0797 0.9905
3.5579 9.0 1647 2.8111 0.9836
3.5579 10.0 1830 2.6726 0.9815

Framework versions

  • Transformers 4.17.0.dev0
  • Pytorch 1.10.2+cu102
  • Datasets 1.18.3
  • Tokenizers 0.11.0