---
id: magenta
title: "Neural Synthesis"
year: 2018
category: "AI & Code"
lang: en
canonical: https://nodemusic.cc/node/magenta
source: node.music
---

# Neural Synthesis (2018)

AI & Code · https://nodemusic.cc/node/magenta

## Summary

AI learned to hear timbre. Projects like Google Magenta allowed transferring the sound of a violin onto a dog's bark (Tone Transfer). Synthesis ceased to be mathematical and became neural.

## Full text

For decades, synthesis relied on mathematical approximations of sound waves. Google Magenta changed the paradigm by teaching computers to 'hallucinate' sound based on physics. Using Differentiable Digital Signal Processing (DDSP), neural networks analyze thousands of hours of audio to understand not just the notes, but the physical cause of the sound—the friction of a bow, the breath in a flute, the resonance of wood.

This enables 'Tone Transfer': you can tap a rhythm on a table, and the AI will re-render it as a professional violin solo, preserving your human timing but completely replacing the timbre. It is not sampling; it is a reconstruction of reality. The AI acts as a 'timbral prosthesis,' allowing a musician to play instruments they have never touched with a nuance that MIDI could never achieve.

## Connections

### Influenced by

- [Generative Music](https://nodemusic.cc/node/generative.md) - EVOLVED
- [The Voder (Bell Labs)](https://nodemusic.cc/node/voder.md) - EVOLVED

### Influenced

- [Generative AI](https://nodemusic.cc/node/gen_ai.md) - EVOLVED
- [PROTO / Spawn](https://nodemusic.cc/node/proto_spawn.md) - EVOLVED
- [Dadabots](https://nodemusic.cc/node/dadabots.md) - TECHNIQUE

Wikipedia: https://en.wikipedia.org/wiki/Google_Magenta