DCASE 2024 Task 9: Language-Queried Audio Source Separation | Development Set

== Description == 

The development set is composed of audio samples from FSD50K [1] and Clotho v2 [2] datasets. FSD50K contains over 51k audio clips (~100 hours) manually labeled using 200 classes drawn from the AudioSet Ontology. For each audio clip in the FSD50K dataset, we generated one automatic caption for each audio clip by prompting ChatGPT (GPT-4) with its sound event tags. All audio files should be converted to mono 16 kHz audio for training LASS models. 

Clotho v2: https://zenodo.org/records/4783391

FSD50K: https://zenodo.org/records/4060432

Automatic captions generated for FSD50K:

• fsd50k_dev_auto_caption.json
• fsd50k_eval_auto_caption.json

Prompt for generating captions:

I will give you a number of lists containing sound events. Please write an one-sentence audio caption to describe these sounds.

Make sure you are using grammatical subject-verb-object sentences. Directly describe the sounds and avoid using the word “heard”. Please don't describe the temporal order of these sound events. The caption should be less than 20 words.

In addition to the development set, participants are free to use any external data (including private data) but are not allowed to use audio in Freesound uploaded between April and October 2023. Participants must specify all external resources utilized in their submission in the technical report.

== References ==

[1] Fonseca E, Favory X, Pons J, et al. FSD50k: an open dataset of human-labeled sound events. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2021, 30: 829-852.

[2] Drossos K, Lipping S, Virtanen T. Clotho: An audio captioning dataset. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). 2020: 736-740.

== Contact ==

Xubo Liu, xubo.liu@surrey.ac.uk