@JJitsevLAION goes audio - LAION-CLAP.
Using same idea as CLIP did when using contrastive InfoNCE loss for image-text pairs for audio-text pairs.
First publication out of MILA-LAION collaboration. Good to see cooperations working again so well.
t.co/Vw3kXGcZGH
@kawamuramasaharCLAP : contrastive language-audio pretraining to develop audio representation by combining audio data with natural language descriptions using LAION-Audio-630K, large collection of 633526 audio-text pairs from different data sources
t.co/lzo41ZVfz4t.co/k8DrxKvdgd