Founded in 2020 to make cutting-edge technology expressive and useful for musicians. We build machine learning instruments and create powerful audio tools.

Projects

Holly+

We collaborated with Holly Herndon to create Holly+: a first-of-its-kind website that lets anyone transform their audio files through a custom model trained on Holly Herndon’s voice.

Real-time Hardware

Low-latency processing is the key to making machine learning a musical instrument. Never Before Heard Sounds is building a real-time neural-net audio processor, probably the world's first!

gan.style

gan.style resynthesizes audio from YouTube using a neural net trained on vocal and instrumental recordings.

Technology

Sound Style-Transfer

Our web services are built on a scalable GPU backend for speedy audio inference with any of our models.

SoundPipe

Generating audio files just-in-time, SoundPipe allows you to stretch, trim, filter, transcode, sequence, and much more through our streaming backend.

Models

The core of our technology are the models which learn musical features in a generative algorithm. We focus on ethically sourced data and full attribution in addition to unique and high quality sound.