Whisper Small English

This model is a fine-tuned version of openai/whisper-small.en on the google/fleurs en_us dataset. It achieves the following results on the evaluation set:

• Loss: 0.6007
• Wer: 7.9908

Model description

This model was created as part of the Whisper Fine-Tune Event. This is my first attempt at fine-tuning the Whisper neural network. Honestly, it's my second time ever trying anything related to training a neural network, and my first time was pretty bad (but I did get a lot of rather funny images out of it, so perhaps it wasn't entirely fruitless?), and it seems like the WER only went up after step 2000, so... I'm not sure if I did a good job or if I just wasted GPU cycles, but maybe I can try again and get a better score?

I'm learning.

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: 1e-05
• train_batch_size: 64
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2