self-supervised-audio

Code for self-supervised learning of audio (waveform, but also maybe spectrograms)

• [ x ] implement VicReg
• [ x ] implement audio dataloaders
• [ x ] visualization & analysis

current version

• all segments 10ms (~441 samples) in length
• vicreg v1: no overlap between audio segments
• vicreg v2: 50% overlap between audio segments

next versions?

• longer segments (30ms? 40?)
• CPC (with AR context?)
• vicreg with context?