Language-Queried Audio Source Separation (LASS) dataset for text-to-audio retrieval. Retrieve audio clips corresponding to natural language text descriptions/captions.The original dataset is based on the AudioCaps dataset.The source audio has been synthesized by mixing two audio with their labelled snr ratio as indicated in the dataset.
Reference… See the full description on the dataset page:
https://huggingface.co/datasets/mteb/lass-synth-t2a.
