audiophile

Create audio datasets for ML/DL projects by scrapping FreeSound! (don't tell anyone...) 🤫

What is it?

audiophile let you scrape audio files from FreeSound directly without asking for API keys or manually browsing files. It is intended to be used for building audio datasets for deep learning projects with minimum effort.

How to use

• Install dependencies by running the requirements.txt file with pip: pip install -r requirements.txt. Doing this in a virtualenv will minimize disruption to other projects.

• Set your scraping parameters in the config.json file:

  • query_type: Choose how to query the audio store, either by setting as search (regular search) or tags (search by a tag).
  • query: The value for the query (cars, birds, conversations, etc.).
  • format: Filter the audio formats to be downloaded (mp3, wav, etc.).
  • auth: Cookies values from an active session in FreeSound, I recommend that you use the Cookies extension to get them, these are used for requests validation only.

• Run the scraper.py file.

• Your data and cached pages will be downloaded at the data folder.

Coming soon (maybe not)

• Maximum audio file size filter
• Maximum audio duration filter
• Pack download

We are open for pull requests!

Licence

MIT, attribution is appreciated but not required.