0%
Jivascope
Hear what the ear can't.
Cardiopulmonary audio intelligence
Scroll to begin ↓
01 See it in action
Product overview

AI-powered cardiopulmonary audio intelligence.

Jivascope turns raw stethoscope recordings into structured diagnostic insight using three Audio Spectrogram Transformers in parallel — no manual interpretation, no ambiguity.

  • Heart sound classification at 97.3% F1-score
  • Multi-view lung analysis catching crackles & wheezes
  • Automatic audio segmentation & routing
  • AI validation rejects non-medical audio
02 What it does

Three neural networks decode what your stethoscope heard.

Upload a recording. Audio Spectrogram Transformers filter, segment, and classify it in seconds — confidence scores, spectrograms, and actionable results. No manual interpretation. No ambiguity.

  • Heart

    Murmurs, irregular rhythms and abnormalities at 97.3% F1.

  • Lungs

    Multi-view analysis catching crackles and wheezes.

  • Validation

    Non-medical audio rejected before any model runs.

03 The numbers don't lie
0 Cardiac F1-score Murmur & arrhythmia detection across validation datasets
0 AST models Cardiac classifier, Multi-View lung analyzer, segmentation
0 Full pipeline Upload to structured diagnostic report with spectrograms
04 The raw tech under the hood
  1. AST

    Audio Spectrogram Transformer

    Built on MIT's AST, fine-tuned on AudioSet. Audio is bandpass-filtered, converted to 128-bin mel spectrograms, normalized, and fed into transformer encoders with 768-dim hidden states.

  2. x3

    Triple model architecture

    One AST for cardiac events, one Multi-View AST for respiratory patterns, one tiny AST for segmentation. Each purpose-built, loaded on-demand with automatic idle unloading.

  3. MEL

    Spectrogram intelligence

    Every upload generates a 1024-frame mel spectrogram with 128 frequency bins. Log-power scaling, z-score normalization and adaptive padding keep inputs consistent regardless of length.

  4. SIG

    Sigmoid confidence scoring

    No vague predictions. Each output passes through sigmoid activation producing independent probability scores. Every number is actionable.

05 The minds behind it
  • 01

    Tunir Sahoo

    Co-founder & HealthTech Lead

    James Dyson Award winner, 60+ hackathon victories. MBA from IIM Kashipur, B.Pharm Tech.

  • 02

    Avijit Bhuin

    Co-founder & AI/ML Engineer

    Data scientist, patent co-inventor at Cognizant. Built and trained all the AST models powering Jivascope.

  • 03

    Rishab Lal

    Co-founder & Backend Engineer

    Engineered the FastAPI backend, model serving pipeline and deployment processing audio in under three seconds.

06 How it works
  1. 01

    Record body audio

    Capture 5–10 seconds of heart or lung sounds. WAV, MP3, OGG, FLAC — all accepted.

  2. 02

    AI validates

    Your upload is compared against reference medical audio. Non-medical recordings get rejected at the gate.

  3. 03

    AST models process

    Segmentation routes the audio. Bandpass filtering, mel extraction, transformer inference and sigmoid scoring — all automated.

  4. 04

    Read your diagnosis

    Labels, confidence percentages, spectrograms, waveforms and BPM estimates appear on screen. Structured. Shareable.

07 In the news

Jivascope, as covered by leading publications

Get started

Upload. Analyze. Know.

Open the dashboard, upload an audio file, and let three AST neural networks decode it in seconds.

Launch Jivascope →
Supports WAV, MP3, OGG, FLAC · Automated segmentation · Structured output