Adversarial Masking Approach for Robust Target Source Localization in the SH Domain
Gyanajyoti Routray, Priyadarshini Dwivedi, Rahul Peter, Rajesh M. Hegde
Asilomar 2025 (Paper #1523)
COMPUTATIONAL ACOUSTICS · DSP · MACHINE LEARNING
.....
You are a function of what the whole universe is doing in the same way that a wave is a function of what the whole ocean is doing. — Alan Watts
Visiting Researcher @ MATINE Lab, KAIST · simulating the recorder's mystery tonesM.Sc. Acoustics & Audio Technology @ Aalto University
SIGNAL PATH
About My Journey
I started out making music in my bedroom when I was 15, and I got deep into production during my later years. During my undergrad, coursework in DSP and math brought me closer to my interests. I couldn't go for my higher studies right after my bachelor's (to save up), during which I worked as a software engineer. During this time, I was able to beef up my coding skills and craft a path for me to pursue my research in acoustics and signal processing (which I'm now invested in at Aalto). I'm continuing to do this by participating in challenges with ML/Audio/Acoustics and even winning one this year. I'll continue to push these boundaries.
SOUL & CIRCUITRY
Background
Following 2 years at PineLabs building scalable fintech microservices, I transitioned into specialized acoustics and audio research. Currently, I'm an RnD Intern at Trinnov Audio in Paris, working on advanced ML/DSP methods for loudspeaker characterization in collaboration with Inria. Living and working in France has been an enriching experience, allowing me to contribute to cutting-edge audio technologies while immersing myself in a new cultural and professional environment that bridges industry and academia.
Focus Areas
- ML/DSP · CALIBRATIONModels that extract source characteristics from room measurements — current focus at Trinnov
- SPATIAL AUDIO · SELDSound event localization with Conformer models trained on synthetic stereo via SpatialScaper
- ACOUSTICS · TDOAICASSP research on beamforming and adversarial masking for robust source localization
- EMBEDDED · HEARINGReal-time embedded DSP pipelines for hearing protection (Savox/Otos)
Approach
I try to deconstruct complex technical logic (which I used to abstract in my early days) and apply the same to real-world scenarios. Now I'm learning more at my uni!
FREQUENCY BANDS
Technical Skills
As you can see, I've worked across a lot of domains but code's common to all of it (80% acoustics and audio signal-processing).
Computational Acoustics
- Finite-difference / FDTD wave solvers
- Incompressible Navier–Stokes & aeroacoustics (Lighthill)
- Physical modelling (strings, hammers, modal synthesis)
- Inverse problems & system identification
- GPU / HPC simulation
Signal Processing
- Spectral & partial tracking (STFT, inharmonicity)
- Room modes & loudspeaker–room calibration
- TDOA, beamforming, spherical harmonics
- Real-time DSP (C++, embedded)
Machine Learning
- PyTorch / TensorFlow
- Physics-guided & hybrid models
- Conformers, SELD, audio encoders
- Data augmentation (SpatialScaper, RIRs)
Acoustic Measurement
- Impulse responses & reverberation time
- In-room loudspeaker measurements
- KEMAR / psychoacoustic testing
- MATLAB / Python analysis
Audio & Music
- JUCE / C++ audio plugins
- Session piano & composition
- Production, mixing, mastering (Ableton, Pro Tools)
- Max4Live (Gen~), MIDI / MPE
Software & Systems
- Python, C++, Rust, Kotlin
- React / Next.js, Three.js, Web Audio
- Docker, Kubernetes, CI/CD
- AWS / GCP, ParaView
OUTPUT STAGE
Featured Projects
Some projects are proprietary but I've shared what's possible.
Mystery Tones in the Recorder
Coupled flow–acoustic simulations of a capped recorder head to track down a faint note that sounds one octave below the main tone. Ongoing research at MATINE Lab, KAIST.
- the mystery tone
- f₀/2
- the mystery tone
- coupled flow + sound
- 3D
- coupled flow + sound
- FD solvers
- GPU
- FD solvers
Computational AeroacousticsNavier–StokesFinite DifferencesGPU / HPCMusical Acoustics
PianoID: Inverse Physical Modelling of the Piano
Calibrating a real-time physical piano model to one specific acoustic instrument from a sparse set of recorded notes: analytical string identification, 88-key compass interpolation, and a nonlinear hammer inverse. JUCE/C++ prototype running; research ongoing.
- stage inverse
- 4
- stage inverse
- keys from ~10
- 88
- keys from ~10
- modal bandwidth
- 12 kHz
- modal bandwidth
Computational AcousticsInverse ProblemsSystem IdentificationPhysical ModellingJUCE / C++DSP
OrbKord: Piano Analysis & Transcription
A browser app for studying harmony: real-time chord recognition, piano audio-to-MIDI transcription, a harmoniser and microtonal tunings. Live, with paying users.
- monthly users
- 700+
- monthly users
- paid licences
- 60+
- paid licences
- chord shapes
- 63
- chord shapes
Web AudioWeb MIDITransKunMusic TheoryMicrotonalProduct
Personalised Hearing for Savox-Otos
Developing an inline audio module for professional hearing protection headsets to apply personalized hearing profiles.
Embedded SystemsDSPNFCProduct DevelopmentUser ResearchC++Python
PEER REVIEWED
Selected Papers
Titles, authors, and links to the work. No abstracts here.
Class-Aware Hybrid Ensemble for Query-by-Vocal Imitation
Rahul Peter, Vivek Mohan
AES International Conference on Artificial Intelligence and Machine Learning for Audio 2025
INTRUSIVE LYRIC INTELLIGIBILITY VIA HYBRID ATTENTION OVER ASR EMBEDDINGS AND FEATURE BLOCKS
Rahul Peter, Vivek Mohan, Seralathan Subramanian, Lauri Juvela
ICASSP 2026 (Published)
WET SIGNAL
The Creatives
Side quests in music production and interactive audio tools
Electronic Music by me
IMPULSE RESPONSE
Experience & Education
Moving places quite a lot.
Visiting Research Student – MATINE Lab, KAIST
Sept 2026 – PresentMusical Acoustics & Transducer Instrument Engineering Lab, Graduate School of Culture Technology, KAIST, Daejeon
Running coupled flow–acoustic simulations of a recorder head to work out where its “mystery tones” (half-harmonics) come from: incompressible Navier–Stokes with a hydrodynamic/acoustic splitting method, Lighthill-type sources and GPU finite-difference solvers. Working with Jin Woo Lee. See the project.
Computational aeroacousticsGPU / HPC simulationInverse modelling of the recorder head
R&D Intern – Trinnov Audio
Feb 2026 – Aug 2026Trinnov Audio, Paris (co-supervised by Inria)
ML/DSP research on low-frequency loudspeaker–room interaction (modal behaviour, source and room characterisation, calibration), using data-driven and physics-guided methods validated on real measurements. Supervised by Romain Couderc and Théophile Dupré. Co-supervised by Antoine Deleforge (Inria). Working within the research team headed by Miguel Moleron and Axel Coulon.
Headed for production ahead of WaveformingPatent and publication in progressCo-supervised with Inria
M.Sc. Acoustics & Audio Technology
Sept 2025 – PresentAalto University
Focus on computational acoustics, physical modelling, audio signal processing and audio machine learning.
GPA 4.43 / 5.00Awarded the Aalto Excellence scholarship
Backend Developer – PineLabs (Fintech)
2023 – 2025PineLabs, Bengaluru
Designed and maintained Kotlin/Ktor microservices, built Rust DLQ processors for webhook reliability, and led CI/CD pipelines on Kubernetes.
Cut webhook latency by 34% with custom Rust serialisationAutomated 10+ deployment pipelines (Docker, Helm)
Researcher (part time) – Multimodal Information Processing Lab
Jun 2024 – Sep 2024IIT Kanpur
Built the Adversarial Masking (Adv-M) framework for robust source localisation in the spherical-harmonic domain: a GAN-based real-time mask generator feeding a CNN DOA estimator, with TDOA-informed design choices.
~30% better localisation accuracy44% lower RMSEPaper at IEEE Asilomar 2025
B.E. Electronics & Communication
2019 – 2023Birla Institute of Technology and Science, Pilani, K.K. Birla Goa Campus
Specializing in Signal Processing, Programming, Applied Mathematics and Algorithms.
CGPA: 8.14/10.00Research in Speech Signal Processing
OPEN CHANNEL
Let's Connect
Whether you’re hiring for audio research, backend engineering, or creative collaborations, let’s connect!