/asr_project_template

Template for ASR project

Primary LanguagePythonMIT LicenseMIT

ASR project barebones

Installation guide

< Write your installation guide here >

pip install -r ./requirements.txt

Recommended implementation order

You might be a little intimidated by the number of folders and classes. Try to follow this steps to gradually undestand the workflow.

  1. Test hw_asr/tests/test_dataset.py and hw_asr/tests/test_config.py and make sure everythin works for you
  2. Implement missing functions to fix tests in hw_asr\tests\test_text_encoder.py
  3. Implement missing functions to fix tests in hw_asr\tests\test_dataloader.py
  4. Implement functions in hw_asr\metric\utils.py
  5. Implement missing function to run train.py with a baseline model
  6. Write your own model and try to overfit it on a single batch
  7. Pain and suffering Implement your own models and train them. You've mastered this template when you can tune your experimental setup just by tuning configs.json file and running train.py
  8. Don't forget to write a report about your work
  9. Get hired by Google the next day

Before submitting

  1. Make sure your projects run on a new machine after complemeting installation guide
  2. Search project for # TODO: your code here and implement missing functionality
  3. Make sure all tests work without errors
    python -m unittest discover hw_asr/tests
  4. Make sure test.py works fine and works as expected. You should create files default_test_config.json and your installation guide should download your model checpoint and configs in default_test_model/checkpoint.pth and default_test_model/config.json.
    python test.py \
       -c default_test_config.json \
       -r default_test_model/checkpoint.pth \
       -t test_data \
       -o test_result.json
  5. Use train.py for training

Credits

this repository is based on a heavily modified fork of pytorch-template repository.

TODO

These barebones can use more tests. We highly encourage students to create pull requests to add more tests / new functionality. Current demands:

  • Tests for beam search
  • W&B logger backend
  • README section to describe folders
  • Notebook to show how to work with ConfigParser and config_parser.init_obj(...)