Agent Atlas

The Builders

Holly Herndon

Composer who released Holly+, an instrument other people can perform in her voice.

Train the voice on the artist's own stems, and decide in the open who may use it commercially.

Steal this

  • Train on your own material. Holly+ learned from hours of her isolated vocal stems.
  • Decide who may use the voice, and how the profits are shared, before you release it.

The lesson and the takeaways are our summary. Everything below is theirs, with where they said it.

The setup

Factual
Machines

The model runs on a GPU backend and can be played live.

Software

Holly+, a DAO for approval and profits, and the singing networks on her 2019 album PROTO.

Models

A voice model trained on her vocal stems. In 2026 she says it was trained with Voctro Labs.

As of March 3, 2026. Setups change; follow the links.

How they work

Factual
  1. She released Holly+ so others could make work with her voice, and planned to decentralize decisions and profits. (Holly Herndon)
  2. She says PROTO, on 4AD in 2019, was the first album to use singing neural networks. (Holly Herndon)
  3. Holly+ was trained on multiple hours of her isolated vocal stems and runs on a GPU backend. (Holly Herndon)
  4. On 3 Mar 2026 she says her partner's British accent came through in her cloned voice, and that seamless is five to 10 years off. (Scientific American)