Skip to content
Partner With Us
Technology

Intelligence that understands the sounds around you.

Shravix is designed to run directly on wearable hardware — keeping detection local, responsive and independent of cloud connectivity. Here's how the system is designed to work, end to end.

Technical architecture

From ambient sound to a timely response.

A high-level view of the processing pipeline. Proprietary implementation details, model weights and dataset specifics are not disclosed.

01

Audio capture

Always-on microphones continuously capture the surrounding acoustic environment alongside playback.

02

Audio segmentation

Incoming audio is windowed and preprocessed into segments suitable for real-time classification.

03

Embedded model

A compact MLP-based classifier runs directly on the wearable, without cloud or smartphone dependency.

04

Hazard classification

The model evaluates each segment for relevant acoustic events — currently tuned to vehicle horns.

05

Alert generation

When a relevant event is recognized, playback is briefly interrupted with a short spoken warning.

06

Playback control

After the alert, the listening experience resumes automatically where it left off.

  1. Ambient audio
  2. Microphone input
  3. Audio preprocessing
  4. MLP-based classification
  5. Hazard decision
  6. Interruptive voice alert
  7. Playback restoration
Performance

Prototype validation results.

MetricPrototype result
Detection accuracy94% under stated validation conditions
Inference latency~30 ms
False-positive rate~2 / 30 min
InferenceOn-device
ConnectivityNo cloud dependency
Smartphone dependencyNone for classification
Technology readinessTRL4

Results reflect prototype validation on development-board-class embedded hardware under the stated metropolitan-noise conditions. This is not a finished consumer wearable, and production performance may vary by hardware, environment and model configuration.

Hardware integration

Designed to fit the hardware you already build.

Shravix is designed for programmable Bluetooth audio SoCs, reusing existing microphones where compatible and integrating at the firmware level — with no mandatory cloud service.

  • Microphone reuse where compatible
  • Runs on programmable Bluetooth audio SoCs
  • Uses available DSP headroom
  • Modest RAM / Flash footprint
  • SDK / reference firmware pathway

Technical eligibility requirements

  • Always-on microphones
  • Programmable DSP
  • Firmware upgradability
  • Sufficient RAM / Flash
  • Multi-microphone processing

These are technical eligibility requirements for evaluation, not universal guarantees of compatibility.

Indian Patent No. 592435 · Granted
Intellectual property

Protected technology, built for deployment.

The granted patent covers the stated combination of microphone input, MLP-based classification, interruptive voice alert and associated touch-panel control, with a 20-year term.

The patent is one component of a broader defensibility story — not every element below is legally protected, but together they form a practical moat around real-world deployment.

Granted patent
Embedded AI expertise
Training pipeline
Real-world data
Model optimization
Firmware integration know-how
OEM SDK
Deployment experience

Evaluating Shravix for your platform?

Talk to us about SoC compatibility, integration effort and running a pilot.

Request a technical brief