OravontOravont Systems LLP

Track 01 · Underwater domain awareness

Underwater domain awareness: detect, classify and re-identify vessels

Passive-sonar AI services for defence organisations and research institutions, from Oravont Systems in Noida, India, led by its founder, Capt (Dr) Sunil Tyagi: systems that detect, classify and re-identify vessels from the sound they radiate, DEMON (Detection of Envelope Modulation on Noise) analysis of vessel recordings, and deployment that runs entirely on the user's own infrastructure. Underwater domain awareness (UDA) is knowing what is present and moving beneath the sea surface, and passive acoustics is its natural sensing modality: sound propagates efficiently under water, and a vessel under way cannot avoid radiating the noise of its own machinery.

Discuss your recordings Capability overview (PDF)
UDA services

Passive-sonar AI services

01Detection and classification pipelines

End-to-end systems that take raw hydrophone recordings and return detected, classified acoustic events: detection that adapts to the local ambient conditions of each recording, calibrated per-class probabilities, and evidence aggregated within a recording before anything is reported.

02Vessel re-identification against a gallery

Acoustic fingerprints that match a new contact to a known hull or flag it as new. Enrolling a vessel is a single forward pass, with no retraining, and every comparison returns a score that can be ranked and thresholded.

03DEMON analysis of vessel recordings

Shaft-rate and blade-rate estimation from cavitation noise, with the open-source DEMON-Analysis toolkit: interpretable, briefing-ready propulsion features that run alongside learned models and show the physical reason behind a match.

04Offline, in-house deployment

Systems built to run entirely on the user organisation's own infrastructure and data: a frozen model with a growing gallery, no data egress, no retraining, GPU acceleration with CPU fallback, and open-source dependencies.

Recordings you already hold can be sent for analysis: ship-radiated noise analysis and vessel re-identification →

Engagements are scoped on enquiry through the Contact page. All services across the three tracks are on the Services page; the sections below describe the work behind them.

UDA models are not sold as off-the-shelf software: SKANN is published research, and Oravont builds SKANN-based models for an organisation's own recordings as a service, including offline deployment with no data egress. See Using SKANN on your own recordings.

Three questions

One signal, three questions

Detection

Is something there?

Event detection that adapts to the local ambient conditions of each recording, rather than relying on a fixed threshold, before anything is passed on for classification.

Classification

What kind of source?

Seven acoustic classes — large and small vessels, whales, dolphins, other marine life, natural sound and other man-made sound — scored with calibrated per-class probabilities, and evidence aggregated within a recording before anything is reported.

Re-identification

Which hull?

A compact acoustic fingerprint for each recording, compared by cosine similarity against an enrolled gallery: match to a known hull, or flag as new. Every comparison returns a number, so matches can be ranked and thresholded rather than re-judged by eye.

Physics first

DEMON: the propulsion signature, read in the open

A vessel's propeller amplitude-modulates its own cavitation noise, and that modulation carries the propulsion signature. DEMON — Detection of Envelope Modulation on Noise — extracts it: bandpass the cavitation-dominated band, take the Hilbert envelope, and transform the envelope into a low-frequency modulation spectrum. Shaft-rate and blade-rate lines and their harmonics appear as a comb. Because blade rate equals shaft rate times blade count, the comb reads out shaft speed and an integer hull attribute — the number of blades — directly from the sound.

These interpretable, briefing-ready features run alongside the learned representations described on the Deep Learning track: two views of the same acoustics, one transparent, one propagation-robust. When an embedding model separates two recordings, DEMON often shows the physical reason — different blade rates, a different comb — in a form a review board can read.

The toolkit is open: DEMON-Analysis on GitHub (code under MIT; documentation, figures and samples under CC BY 4.0). The accompanying working note is on the Research page.

DEMON modulation spectrum of a vessel recording, with blade-rate harmonics marked Blade-rate harmonics (1×–5× BR) extracted from the cavitation envelope of a vessel recording, using Oravont's DEMON tool.
Engagements

Programmes

National-security AI challenge · NCIIPC / Startup India

PS12 — underwater acoustic event detection and classification Concluded May 2026

Under Problem Statement 12 of the AI Grand Challenge, as technical partner to an industry participant, Oravont designed an end-to-end system that takes raw hydrophone recordings and returns detected, classified acoustic events across seven target classes. Detection adapts to local ambient conditions; classification runs on SKANN (Selective-Kernel Acoustic Neural Network) with calibrated per-class probabilities; a runtime clustering stage aggregates evidence within each recording.

The pipeline was developed entirely on open-source acoustic corpora, runs offline on GPU or CPU, and produced strong, well-calibrated classification across acoustically diverse classes.

Class territory map: UMAP of the learned embedding, Voronoi regions for four sound classes Class territory map: a UMAP projection of the learned embedding space, with Voronoi territories drawn from class centroids (backbone stage, four acoustic superclasses).

Oravont Systems · technical partner · 2025 – May 2026 · Concluded

iDEX · Ministry of Defence · Indian Navy, Underwater Domain Awareness

DISC5 Challenge 10 — vessel re-identification from passive sonar Active · winding down

Classification answers what kind of source is this. Re-identification answers a harder question: which specific vessel is this? As technical partner to an industry participant in DISC5, Oravont built the deep-learning core that distils each recording into a compact acoustic fingerprint — a 512-number embedding trained with a metric-learning objective so that recordings of the same hull cluster together while different hulls separate — and delivered the enrolment-and-inference application to the participant.

A new contact is either matched to a known hull in the gallery or flagged as new — an open-set decision — and enrolling a vessel is a single forward pass, with no retraining and no new class.

Built for sovereign deployment.

  • Frozen model, growing gallery — the user organisation enrols and scores on its own data, entirely in-house.
  • No data egress, no retraining — fully offline operation on the user's infrastructure.
  • Robust by design — training augmented with synthetic channel colouring, measured ambient noise, multipath and time shifts, so the fingerprint keys on the vessel, not the recording chain.
  • Licence-clean — developed on open-source acoustic corpora with open-source dependencies; GPU-accelerated with CPU fallback.
Three 512-dimensional acoustic fingerprints: vessel A twice (cosine 0.92), vessel B (0.16) Acoustic fingerprints compared by cosine similarity: two recordings of the same hull agree (cos 0.92); a different hull does not (cos 0.16).

Oravont Systems · technical partner to an industry participant · Programme active; Oravont's part winding down. The enrolment-and-inference application is public on GitHub; the training code and its documentation are held privately under the programme. Enquiries through the Contact page.

Measurement

Numbers an analyst can act on

Re-identification is only useful if its score means what it says. Oravont's evaluation protocol, published in the September 2026 preprint (Tyagi, arXiv:2609.07399), closes the easiest routes to a flattering result: train and test splits that are disjoint by hull, keyed to vessel identity; galleries and queries drawn from separate passages of each hull; source-pure galleries, so that cross-corpus hardware signatures cannot act as a ranking shortcut; and an audio-adjudicated gate that merges recordings of the same physical transit before any score is computed.

On public data, transit deduplication alone removes a 16–21 point apparent rank-1 advantage (rank-1: the share of queries whose correct hull is ranked first) — larger than the difference between the methods compared. On a 40-hull public gallery, cross-passage rank-1 is 0.25 for the SKANN embedding and 0.35 when it is fused with a narrowband-tonal comparator. The results support analyst triage over a ranked shortlist, not identification of record, and the practice says so in print. Details and figures are on the Deep Learning track and the Research page.

Further reading: Re-identifying vessels from radiated noise, a working note by Capt (Dr) Sunil Tyagi on embeddings, galleries and enrolling new vessels without retraining.

In brief

Questions this track is asked

What is underwater domain awareness?
Underwater domain awareness (UDA) is knowing what is present and moving in the underwater environment: vessels on or below the surface, marine life, and the conditions that shape how sound travels. Sound carries far under water, and a vessel under way cannot avoid radiating the noise of its machinery and propeller. Passive acoustics, with hydrophones listening continuously, is therefore its natural sensing modality. Oravont's work concentrates on vessels: detecting, classifying and re-identifying them from their radiated noise.
How is passive sonar different from active sonar?
Active sonar transmits a pulse and listens for the echo. That gives range, but the transmission can also disclose the platform that sent it. Passive sonar only listens, to the sound vessels radiate themselves. It emits no probing signal that could disclose the listener's presence, and it can run continuously. With an array or directional sensors it gives bearing; range has to be estimated over time or from several sensors. The sound it records carries a vessel's machinery and propeller signatures, the information that detection, classification and re-identification work from.
What is DEMON analysis?
DEMON (Detection of Envelope Modulation on Noise) recovers a vessel's propulsion signature from its cavitation noise. The cavitation band is bandpassed, its Hilbert envelope taken, and the envelope transformed into a low-frequency modulation spectrum, in which shaft-rate and blade-rate lines appear as a comb. Blade rate is shaft rate times blade count. When the shaft-rate and blade-rate components can be resolved and correctly assigned, their relationship gives estimates of shaft speed and blade count. Oravont's DEMON-Analysis toolkit is open source.
Can a vessel be identified from its radiated noise?
Partly. Recordings of the same hull share machinery tonals, blade-rate lines and cavitation character. A learned acoustic fingerprint can therefore rank a new recording against a gallery of known hulls, or flag it as new. On public hydrophone data, Oravont's evaluation supports analyst triage over a ranked shortlist, not identification of record.
Contact

Starting a UDA engagement

Write to info@oravontsystems.com with the problem, the platform or data involved, and the timeframe. Enquiries reach a person, not a form.

Discuss your recordings
How this connects

The other two tracks

Work with Oravont. To discuss a project, write to info@oravontsystems.com, call +91 99222 42303 or use the contact page. Oravont Systems LLP, Noida, India.