DJ Tagger

This will be a reimagined DJ Tagging system. I'm going to build it for my needs first, but will gradually make it more universal. It think tagging, clustering and visualization is core to DJing and I'm excited to build something that innovates here.

DJ Tagger (working title) will incorporate some of what makes Lexicon great, be open source like One Tagger and make use of SOTA audio processing and analysis.

This will be the first step. Next I will take on reverse engineering a radical new approach to DJing itself that incorporates "skills" that I'd like to be able perform. I think current DJ software is incredibly clunky. I will build it for my deck first and again make it more universal over time. I'll build it as "plug-in" first so that you can assemble your own DJ visualizations, skills, etc.. and it will use DJ Tagger as the active track index underneath.

Excited just writing this, but I know this will be a lot of work.

• MERT (Acoustic Music Understanding Model): Developed with large-scale self-supervised training using acoustic and musical teacher models, yielding superior universal representations for downstream music tasks than standard tag-trained CNNs.
• CLAP (Contrastive Language-Audio Pretraining): Models like CLAP HTS-AT link text and music directly into a shared embedding space, allowing for zero-shot music tagging and flexible text-based queries without relying on a fixed set of closed-vocabulary labels.
• MusicFM and Music2Vec: Newer transformer-based backbones designed specifically to capture long-term musical structure, harmony, and rhythm rather than short 3-second spectrogram chunks.
• PANNs (Pretrained Audio Neural Networks): While slightly broader, architectures like CNN14 provide robust general-purpose audio and music event detection baselines.

https://vibesdj.io/features/auto-tag https://cyanite.ai/