Natural Perspective Spatial Audio

A model designs a custom 7.1 spatial mix for every recording. Most recordings are made with one or two microphones and played back over two speakers, and there's almost no content for the surround systems in cars and home theaters. Natural Perspective creates a genuine spatial mix from those same stereo sources.

How it works

A recording is separated into its instrument stems. A language model then studies what it can learn (the artist, the song, whether it was cut live or in a studio, and the measured level of every stem) and designs a mix: it invents a scene, describes where you are sitting, and routes each instrument across the eight channels of a 7.1 system.

The examples are real outputs from my own music library, each with its scene, stage map and the model's reasoning. The code is open source on GitHub.

How to read a stage map

You are the white marker at the center, facing the stage at the top. Each instrument is a field placed where it sits in the mix: width shows its stereo spread, front-to-back position shows depth.

In a live mix the crowd wraps wide and behind you; in a studio mix the players gather close around the front.