Skip to content

A platform for the collaborative creation of open audio collections labeled by humans and based on Freesound content.


Notifications You must be signed in to change notification settings


Repository files navigation

Freesound Annotator

The platform changed its name from Freesound Datasets to Freesound Annotator.

Freesound Annotator is a platform for the collaborative creation of open audio collections labeled by humans and based on Freesound content. Freesound Annotator allows the following functionalites:

  • explore the contents of datasets
  • contribute to the creation of the datasets by providing annotations
  • it will allow to download different timestamped releases of the datasets
  • it also promotes discussions around both platform and datasets

This repository serves the following main purposes:

  • development and maintenance of the Freesound Annotator
  • allow people to see the ongoing progress in a transparent manner
  • concentrate discussion from the community

We would like the community to get involved and to share comments and suggestions with us and other users. Feel free to take a look at the issues and join ongoing discussions, or create a new issue. We encourage discussion about several aspects of the datasets and the platform, including but not limited to: faulty audio samples, wrong annotations, annotation tasks protocol, etc. You can check the Discussion page on the Freesound Annotator for some more ideas for discussion.

The first dataset created through the Freesound Annotator is FSD: a large-scale, general-purpose dataset composed of Freesound content annotated with labels from Google’s AudioSet Ontology. All datasets collected through the platform will be openly available under Creative Commons licenses.

You can find more information about the platform and the creation of FSD in our paper:

E. Fonseca, J. Pons, X. Favory, F. Font, D. Bogdanov, A. Ferraro, S. Oramas, A. Porter and X. Serra. “Freesound Datasets: A Platform for the Creation of Open Audio Datasets” In Proceedings of the 18th International Society for Music Information Retrieval Conference, Suzhou, China, 2017.

Getting Started

You'll need to have docker and docker-compose installed.


Copy freesound_datasets/ to freesound_datasets/ and follow the instructions in the file to fill in services credentials for your project.

To allow downloads, you need to fill in


If you want to log in with an external service fill in the relevant SOCIAL_AUTH_ keys.

Otherwise, to create a user using Django's models you can run

docker-compose run --rm web python createsuperuser

You will need to install the PostgreSQL pg_trgm extension in order to enable the text-search in the sound curation task. After having started the containers (docker-compose up), from an other terminal, you can run

docker-compose run --rm db psql -h db -U postgres


The first time you load the application you will need to perform migrations:

docker-compose run --rm web python migrate

Run Freesound Annotator using docker-compose:

docker-compose up

And point your browser at http://localhost:8000.

Dummy data

You will need test data to develop. To load some data, first load the dataset fixtures by running

docker-compose run --rm web python loaddata datasets/fixtures/initial.json

This will create an empty dataset with a loaded taxonomy.

Once you have the dataset, you can generate fake data (sounds, annotations and votes), by running

docker-compose run --rm web python generate_fake_data fsd 100 5 1000 1000

This will create 100 fake sounds, 5 fake users, 1000 fake annotations and 1000 fake annotation votes. You can run this command again to generate more data.

Add a new dataset with a validation task

  1. Create dataset object from the admin

  2. Create a taxonomy.json file which looks like the example below and load using command (get the dataset ID from the admin):

python load_taxonomy 'DATASET_ID PATH/TO/TAOXNOMY_FILE.json

        "name":"Main class",
        "description":"A description for this cateogry.",
        "name":"Sub class",
        "description":"A description for this cateogry.",
  1. Create objects for the taxonomy nodes using command (get the taxonomy ID from the admin):

python create_taxonomy_node_instances TAXONOMY_ID

  1. Load candidate sound annotations from a JSON file using the command (algorithm_name is ):

python load_sounds_for_dataset short_ds_name filepath.json algorithm_name

      "description":"Cat sniffing: sniffing microphone. Recorded with a ZOOM H6 recorder and X/Y capsule. The sound was post-processed to remove any background noise or room tone.",
      "name":"Cat Sniffing.wav"
      "description":"dog bark ",