DDSP
Differentiable Digital Signal Processing: DSP modules (oscillators, filters) usable inside neural nets.
| Language | Python |
|---|---|
| Category | ML & generative |
| License | Apache-2.0 |
| Platforms | macOS Windows Linux |
| Install | pip install ddsp |
| First released | 2020 |
| Maintained | Yes |
| Links | Website · Source |
Strengths
- Combines interpretable DSP with deep learning
- Enables timbre transfer with small, data-efficient models
Limitations
- Built on older TensorFlow tooling, finicky install
- Niche/research API with a learning curve
Best for
Neural audio synthesis and timbre transfer with interpretable, controllable DSP.
Minimal example
import ddsp
harmonic = ddsp.synths.Harmonic()
audio = harmonic(amplitudes, harmonic_distribution, f0_hz)