DDSP-VST
Key Features
- Real-time DDSP neural synthesis as both an audio effect and a MIDI instrument
- Transforms incoming audio into trained instrument-like timbres while retaining pitch and dynamics
- Tone-shaping controls for moving between more realistic model ranges and unstable experimental colors
- DDSP Effect mode for processing voice, instruments, percussion, nature recordings, and other audio inputs
- DDSP Synth mode for driving embedded models from MIDI sources without pitch-detection errors
- Support for custom trained models through Magenta's Colab-based DDSP training workflow
- Open-source project with macOS and Windows plugin binaries linked from the official product page
Description
DDSP-VST is a neural synthesis plugin from Magenta that can run as both an audio effect and a MIDI-driven instrument inside a DAW. It analyzes pitch and loudness from an input signal, then uses a DDSP model to resynthesize that performance as another trained timbre while preserving expressive contours.
The effect version is the most distinctive workflow: feed it a monophonic voice, instrument, percussion hit, or field recording and shape how the trained model responds. Magenta's own guide notes that monophonic input gives the cleanest musical result, while unusual sources can create more experimental textures.
DDSP-VST also includes a synth version for playing the models from MIDI, which avoids pitch-tracking errors and adds envelope control for generated audio. The plugin ships with embedded models, and users can train additional models through Magenta's Colab workflow using a short monophonic recording.
Community discussion frames DDSP-VST as an early but inspiring neural resynthesis tool rather than a polished conventional synth replacement. It is strongest for strange timbre transfer, expressive sound design, ambient layers, and research-minded production workflows where the artifacts are part of the appeal.
Video Preview
Frequently Asked Questions
Is DDSP-VST an effect or an instrument?
It is both. Magenta provides DDSP Effect for transforming incoming audio and DDSP Synth for playing embedded models from MIDI inside a DAW.
What kind of audio works best with DDSP Effect?
Magenta's getting-started guide says monophonic input works best for clean, musical results. More complex sources can still be useful when the goal is unusual textures rather than realistic instrument replacement.
Can DDSP-VST use custom models?
Yes. The GitHub guide explains that the plugins include 11 embedded models and can load trained .tflite models made through Magenta's DDSP training Colab.
Is the project still actively maintained?
The official GitHub repository is archived and read-only as of October 9, 2024. The product page and download buttons are still live, so this should be treated as an available research-era plugin rather than an actively updated commercial instrument.
