PhD Researcher · C4DM · QMUL

Ilias Ibnyahya

Building differentiable audio systems at the intersection of signal processing and machine learning. Specialising in artificial reverberation, Feedback Delay Networks, and intelligent audio effects — based at the Centre for Digital Music.

I'm a PhD researcher at Queen Mary University of London, working within the Centre for Digital Music (C4DM) under the supervision of Prof. Joshua D. Reiss. Prior to my PhD, I spent 8 years at DiGiCo, developing professional audio systems and bridging engineering with live sound production.

My research focuses on making reverberation algorithms fully differentiable — enabling gradient-based optimisation of parameters that were previously hand-tuned or inaccessible to learning frameworks. The goal is to close the gap between the perceptual richness of convolution reverb and the parametric control of algorithmic approaches.

I'm currently finalising work on a differentiable Feedback Delay Network that clones target room impulse responses using gradient descent, with a novel parametric EQ approach that reduces filter complexity by over two-thirds without sacrificing accuracy.

Affiliation C4DM, QMUL
Ex-Industry DiGiCo 8 yrs
Focus Differentiable DSP
Publications 3 papers
Venues DAFx · AES · ISMIR
Location London, UK
Artificial Reverberation

Designing and optimising Feedback Delay Networks (FDNs) to synthesise realistic room acoustics with parametric control over frequency-dependent decay.

Differentiable Audio

Making classical DSP building blocks — IIR filters, attenuation stages, delay lines — fully differentiable and compatible with gradient-based learning frameworks.

Audio Effects Modelling

Applying inference-time optimisation and style transfer to control audio effects. Collaborating on differentiable compression, EQ, and reverberation for production contexts.

2025
Differentiable Attenuation Filters for Feedback Delay Networks
Ilias Ibnyahya, Joshua D. Reiss
DAFx 2025

A scalable parametric EQ approach for FDN attenuation filters using shared SOS biquads — fully differentiable and trainable via gradient descent. Reduces filter count by over two-thirds vs. graphic EQ designs while matching perceptual accuracy.

2024
ST-ITO: Controlling Audio Effects for Style Transfer with Inference-Time Optimization
Christian J. Steinmetz, Shubhr Singh, Marco Comunità, Ilias Ibnyahya, Shanxin Yuan, Emmanouil Benetos, Joshua D. Reiss
ISMIR 2024Best Paper

Style transfer across arbitrary audio effects using inference-time optimisation against a pretrained audio similarity metric — no retraining required. Applied to compression, EQ, delay, and reverberation across speech, voice, and full music.

2022
A Method for Matching Room Impulse Responses with Feedback Delay Networks
Ilias Ibnyahya, Joshua D. Reiss
AES Convention 153

Multi-stage pipeline fitting FDN coefficients to a target room impulse response, combining IR feature extraction with a genetic algorithm using an MFCC cost function. A truncated IR hybrid enables better early reflection accuracy.

CPU / GPU Value Finder
Find the best-value desktop & workstation hardware at real market prices
Launch Tool ↗

Tired of paying MSRP? This tool scrapes live eBay pricing across GPUs and CPUs — both consumer and professional-grade — and ranks them by performance-per-pound. Whether you're building a desktop rig or sourcing a workstation, it cuts through the noise and shows you where the genuine value is right now.

GPUs CPUs Desktop Workstation / Pro Live eBay Pricing Performance / £ Ranking
Let's talk
audio.

Open to collaborations in differentiable DSP, audio effects research, and industry applications of intelligent audio processing.