The acoustic signature is the thread
A ship under way cannot help making noise: machinery tonals at shaft and generator lines, propeller cavitation modulated at the blade rate, flow over the hull. That radiated signature is what a stealth designer works to suppress, what a passive-sonar system listens for, and what a neural network learns to read.
- Stealth makes a platform's own signature quieter: acoustic modelling, underwater radiated noise, target strength, shock and vibration control.
- Underwater domain awareness detects, classifies and re-identifies other vessels by their signatures, from raw hydrophone recordings.
- Deep learning is the engine that reads signatures: learned filterbanks on the raw waveform, metric-learned acoustic fingerprints, and evaluation protocols built to resist flattering scores.
Each track improves the others. The stealth models say what a detector should expect to hear; the detectors show which features of a signature survive the ocean; and the same understanding of how a hull makes noise sits behind all three.
Three tracks of work
Detect, classify, re-identify
Vessel detection, classification and re-identification from passive-sonar recordings. DEMON analysis for interpretable propulsion features; SKANN classification across seven acoustic classes under the NCIIPC AI Grand Challenge (PS12, concluded May 2026); acoustic fingerprints for open-set re-identification under iDEX DISC5 Challenge 10 (Indian Navy, active).
Learning from the raw waveform
SKANN, a selective-kernel acoustic neural network that learns its own multi-scale filterbank from the raw hydrophone waveform; metric-learned 512-dimensional embeddings; physics-grounded synthetic data for pretraining; and a cross-passage evaluation protocol, published on arXiv and filed as an Indian provisional patent in September 2026.
Designing platforms that stay quiet
Acoustic modelling and underwater radiated noise estimation, sonar signal reflection and target strength, acoustic interference to own sonar, airborne noise for habitability, shock and vibration control, and mast vibration — with three in-house engineering tools: SSR-Sim, ABNContour and SMS-Suite.
Founder & Principal: Capt (Dr) Sunil Tyagi, Indian Navy (Retd)
Profile, CV and the full publication record are on suniltyagi.in. The figures above are taken from the founder's CV.
From the practice
Open-set vessel re-identification from ship-radiated noise, with SKANN
A raw-waveform selective-kernel encoder, a cross-passage evaluation protocol that removes the easiest routes to a flattering score, and honest numbers on public hydrophone data. Companion artefacts on Zenodo under CC BY 4.0; aspects of the method covered by an Indian provisional patent application.
Physics-grounded synthetic underwater acoustic dataset
12,000 five-second clips across four vessel classes and ambient ocean noise, generated from first-principle acoustic models in physical units and released for self-supervised pretraining.
DEMON Analysis: reading a vessel's propulsion from its own noise
The physics, the Hilbert envelope and the harmonic comb — and a case where the classical method caught a defect in a synthetic training dataset that spectrograms had hidden. The toolkit is public under MIT and CC BY 4.0.
Work with Oravont
Collaborative R&D, consultancy in platform acoustics, academic collaboration, and enquiries about the engineering tools — for defence organisations, research laboratories, shipyards and industry. Enquiries reach a person, not a form.
Start a conversation