Source-linked AI summary
A Frequency-Domain Artificial Reverberator Plug-In
Jonas Roth, Nishanth Kumar, Silvan Krebs, David Wieland, Christoph Studer
TL;DR
Artificial reverberation needs models that capture both early reflections and stochastic late tails while also supporting creative effects. FDverb addresses this with an STFT-based noise-carrier reverberator, early-reflection processing, nonlinear decay, and pitch shifting. The result is an open-source DAW plug-in and Python reference implementation for hybrid reverberation and sound design.
Problem
Artificial reverberation must represent distinguishable early reflections and stochastic late reverberation tails, which have traditionally been modeled from physical-room responses.
Method
FDverb uses an STFT to shape spectral noise with input-derived envelopes for the late tail, combines it with sparse early reflections, and adds nonlinear decay and pitch shifting.
Results
FDverb is provided as an open-source DAW plug-in and Python reference implementation, with nonlinear decay and pitch shifting supporting creative sound-design applications.
Takeaways & Limitations
FDverb provides a hybrid reverberation algorithm whose tail and creative controls can be used for sound-design applications.
Takeaways & Limitations
The linear-frequency pitch shift does not preserve harmonic intervals, producing an inharmonic reverb tail.
Abstract
from arXiv · showhide
We present FDverb, a frequency-domain artificial reverberator, based on the idea of a vocoder with a noise carrier signal. Using a short-time Fourier transform (STFT) for analysis and synthesis, FDverb generates late reverberation by weighting spectral noise components with envelopes. We extend FDverb with early reflections, nonlinear decay, and pitch shifting. These extensions enable creative sound-design applications. We provide FDverb as an open-source DAW plug-in, using the JUCE framework.
1. INTRODUCTION
FDverb treats late reverberation as frequency-domain noise shaped by spectral envelopes, while combining it with early reflections into a hybrid reverberator. The system adds creative controls and is released as an open-source DAW plug-in with a Python reference implementation.
- Artificial reverberation models physical-room responses through distinguishable early reflections and a stochastic late reverberation tail.The late tail is motivated by observations that some concert-hall impulse responses resemble exponentially decaying white noise.
- FDverb generates its late tail in the frequency domain using an STFT and noise-carrier spectral envelopes, then combines it with early reflections.The early-reflection unit and tail reverberator together form the hybrid FDverb algorithm.
- Nonlinear decay and pitch shifting extend FDverb beyond basic reverberation toward creative sound-design applications.
- FDverb is provided as an open-source DAW plug-in together with a Python reference implementation.
2. THE FDVERB ALGORITHM
FDverb combines time-domain early reflections with an STFT-based tail reverberator that shapes spectral noise using input envelopes. Its processing balances frequency resolution, temporal character, and latency while adding creative controls.
- Hybrid architecture: FDverb outputs a weighted combination of the dry input, sparse early reflections, and a delayed frequency-domain reverb tail.The early-reflection signal is produced by sparse time-domain convolution, while the tail is generated by spectrally shaping noise.
- Latency and timing: Early reflections allow FDverb to preserve a short perceived pre-delay even when block-based tail processing introduces longer latency.Time-domain early-reflection convolution avoids the block latency inherent to frequency-domain processing, while the tail can be independently delayed.
- Tail processing: The tail algorithm applies STFT analysis, extracts per-bin magnitude envelopes, modulates spectral noise with those envelopes, and reconstructs the signal using inverse STFT.Noise is generated directly in the spectral domain, then weighted by the AR-averaged envelopes before overlap-add synthesis.
- Envelope following: The envelope follower detects spectral magnitudes and smooths them with an attack-release averager whose attack and release coefficients control temporal response.The default detector exponent γ = 1 corresponds to amplitude detection, while the release behavior is tied to frequency-dependent decay settings.
- STFT design trade-offs: A larger STFT block provides finer frequency resolution and a more tonal tail but increases temporal smearing, whereas a shorter block produces a sharper, more noise-like tail.The block size also determines algorithmic latency, creating a trade-off between spectral detail, temporal tracking, and responsiveness.
- Creative extensions: The paper extends FDverb with nonlinear decay and pitch-shifting controls for creative sound-design applications.The section introduces enhancements beyond the basic tail algorithm, including additional creative controls and pitch-shifting features.
3. DAW PLUG-IN IMPLEMENTATION
FDverb is implemented as a JUCE-based DAW plug-in in VST3 and AU formats, with an envelope visualization and measured low per-track CPU use.
- 3. DAW PLUG-IN IMPLEMENTATION: FDverb uses JUCE and is distributed as VST3 and Audio Unit plug-ins for digital audio workstations.Its early-reflection unit uses a delay line supporting up to 32 echoes with maximum delay of 500 ms.
- 3. DAW PLUG-IN IMPLEMENTATION: 0.011–3.81 ms average per-block processing time was measured across the reported block sizes.The measurements span N = 25 to N = 214.
- 3. DAW PLUG-IN IMPLEMENTATION: Less than 2.6 % per-track CPU usage was reported for one FDverb instance across all supported block sizes.Measurements used REAPER with the FDverb VST3 plug-in and a 48 kHz sample rate.
- 3. DAW PLUG-IN IMPLEMENTATION: The interface adds a real-time envelope-gram that visualizes the envelopes as a rolling-spectrogram-like display.Figure 4 places the real-time envelope visualization at the top right of the plug-in interface.
4. VISITOR EXPERIENCE
Visitors can explore FDverb hands-on by adjusting parameters in real time with presets, prerecorded material, music, or their own audio device.
- 4. VISITOR EXPERIENCE: Visitors are invited to operate the DAW plug-in through a laptop, audio interface, and headphones.They can begin with provided parameter presets and tweak FDverb in real time.
- 4. VISITOR EXPERIENCE: Audio input can come from prerecorded instrument samples, music, or a visitor’s own connected device.