Freesound Forums
Freesound Datasets: help us to annotate Freesound!
Started January 12th, 2018 · 8 replies · Latest reply by Calvindavid 1 month, 2 weeks ago
Dear Freesound community,
At the Music Technology Group, Universitat Pompeu Fabra, we are developing the Freesound Datasets platform for the collaborative creation of open audio datasets based on Freesound content. Our goal is to create datasets (collections of audio data) that can be used for research, for example enabling machines to hear and interpret sound similarly to humans. This would also allow us to implement innovative ways of searching and browsing content in Freesound from which you’ll eventually benefit 
The first dataset we want to build is called FSD. FSD is a general-purpose dataset composed of Freesound content organised by the AudioSet taxonomy. The AudioSet taxonomy is a collection of sound categories that aims to hierarchically organize a wide range of everyday sounds. We have automatically pre-filled AudioSet categories with thousands of audio samples from Freesound. Now we need your help to confirm that we have assigned sounds to the right categories!
The usefulness of FSD for research depends on the accuracy of these assignments (also called “annotations”). Therefore, we need to validate these annotations in order to keep only the most accurate ones. We plan to achieve this via crowdsourcing, and you guys, the Freesound community, are our first stop . For you to be able to help us, we have set up an online platform at https://datasets.freesound.org in which you can listen to the sounds we’ve assigned to AudioSet categories and validate whether the assignments are correct or not. Our platform is in beta version now, so we would appreciate it if you took 15-20 minutes to use the platform, validate some annotations and leave some feedback for improvements before the official launch.
If you want to participate, you simply need to visit the following URL:
https://datasets.freesound.org/fsd/annotate/
You can either choose the “beginners” or the “advanced” task, then you’ll be asked to login (you can use your Freesound account). Follow the guidelines on screen and validate one sound category. Then, please fill in the following form to give us some feedback:
If you did the beginners task, please fill in this form.
If you did the advanced task, please fill in this other form.
Thanks a lot in advance! We’ll keep you updated with more information about this project in the coming months.
Cheers,
the Freesound team
edfonseca wrote:Dear Freesound community,
At the Music Technology Group, Universitat Pompeu Fabra, we are developing the Freesound Datasets platform for the collaborative creation of open audio datasets based on Freesound content. Our goal is to create datasets (collections of audio data) that can be used for research, for example enabling machines to hear and interpret sound similarly to humans. This would also allow us to implement innovative ways of searching and browsing content in Freesound from which you’ll eventually benefit
The first dataset we want to build is called FSD. FSD is a general-purpose dataset composed of Freesound content organised by the AudioSet taxonomy. The AudioSet taxonomy is a collection of sound categories that aims to hierarchically organize a wide range of everyday sounds. We have automatically pre-filled AudioSet categories with thousands of audio samples from Freesound. Now we need your help to confirm that we have assigned sounds to the right categories!
https://drift-boss.proThe usefulness of FSD for research depends on the accuracy of these assignments (also called “annotations”). Therefore, we need to validate these annotations in order to keep only the most accurate ones. We plan to achieve this via crowdsourcing, and you guys, the Freesound community, are our first stop
If you want to participate, you simply need to visit the following URL:
https://datasets.freesound.org/fsd/annotate/
You can either choose the “beginners” or the “advanced” task, then you’ll be asked to login (you can use your Freesound account). Follow the guidelines on screen and validate one sound category. Then, please fill in the following form to give us some feedback:
If you did the beginners task, please fill in this form.
If you did the advanced task, please fill in this other form.Thanks a lot in advance! We’ll keep you updated with more information about this project in the coming months.
Cheers,
the Freesound team
Thank so much.
Thank you for sharing this exciting initiative with the community! The Freesound Datasets platform sounds like a fantastic step forward in leveraging collaborative efforts to improve audio dataset annotations and advance sound research. It’s inspiring to see how this project could enhance both machine understanding of audio and the overall user experience on Freesound.
I’ve taken some time to explore the platform and validate annotations. The interface is intuitive, and the task instructions are clear, making it easy to jump in and contribute. I especially appreciate the choice between “beginners” and “advanced” tasks, which ensures inclusivity for participants of varying expertise levels.
While using the platform, I noticed a few areas where improvements could make the experience even smoother. For example, adding a brief tutorial or examples for first-time users could help them better understand what constitutes a "correct" or "incorrect" annotation. Additionally, providing an option to revisit or edit previous annotations within the same session would be helpful for those who might want to double-check their work.
Overall, this is an exciting project, and I’m thrilled to see the progress you’ll make with FSD and other datasets. I’m happy to participate and look forward to seeing how these datasets contribute to innovative sound research and new features in Freesound.