Medical AI Dictation

Zyemed Sonus

Voice. Intelligence. Care. A secure, centralized medical dictation service built exclusively for the Zyemed ecosystem. Deliver instant, structured, and clinically-validated reports.

ON-PREMISE BOUNDARY (ZDPA)

WHISPER-POWERED SPEECH AI

Absolute Privacy & Local Residency

Unlike consumer cloud voice utilities, Zyemed Sonus runs completely self-hosted inside your hospital's security boundary. Your audio recordings never leave the deployment network, ensuring strict compliance with local regulations, including the Zimbabwe Data Protection Act (ZDPA).

Acoustic Pre-Processing

5-Stage Audio Pipeline

Before passing spoken clips to the speech-to-text models, Sonus processes and normalizes the recording to maximize speech quality and eliminate errors.

1. Format Standardisation

FFmpeg seamlessly transcodes WebM, MP4, and OGG formats into lossless 16 kHz Mono WAV streams.

2. Voice Activity Detection (VAD)

Silero-VAD strips silent frames and environment noise, preventing model hallucinations.

3. Perceived Loudness Normalisation

Applies EBU R128 protocols to target -23 LUFS, compensating for varied microphone levels.

4. DC Offset Elimination

Removes spectral bias to eliminate waveform clipping and clean acoustic artifacts.

5. SNR Quality Validation

Computes a live signal-to-noise ratio score (0–1) to flag low-quality audio or hardware faults.

Natural Language Engine

7-Stage Medical NLP

Raw transcriptions are automatically post-processed through a clinical NLP pipeline to standardize typography and format diagnostic templates.

1. Punctuation Restoration

Applies deep punctuation restoration models to intelligently structure sentences.

2. Sentence Casing

Validates and standardizes capitalization patterns across clinical documentation.

3. Terminology Normalisation

Resolves phonetic errors using a local clinical lexicon with ~100 key terminology mappings.

4. Unit Standardisation

Normalises values instantly (e.g., 'point five centimetres' becomes '0.5 cm').

5. Anatomy Capitalisation

Capitalises anatomical descriptors (e.g., Left, Right, Bilateral) for clear reports.

6. Section Header Identification

Detects spoken headers to automatically format sections (e.g., 'HISTORY:', 'IMPRESSION:').

7. Structured Field Extraction

Extracts sections into clean JSON objects, enabling structured population of EHR tables.

Seamless Developer SDK

Integrating clinical voice recording into web modules takes only minutes with our pre-built TypeScript client and React hook.

npm Package

Direct access to raw API client hooks and WebSocket abstractions.

npm install @zyenova/sonus-client

React Native Capture

Built-in audio capture utilizing browser MediaRecorder API with noise and echo cancellation.

DICOM Mapping

Transcriptions map cleanly to DICOM Study Instance UIDs for automated archiving.

useSonusDictation.tsx

import { useSonusDictation } from '@zyenova/sonus-client';

function DictationWidget({ studyUid }) {
  const {
    isRecording,
    isProcessing,
    transcript,
    startRecording,
    stopRecording,
  } = useSonusDictation({
    sonusUrl: 'https://sonus.zyemed.com',
    token: currentUserToken,
    studyUid,
    onComplete: (result) => {
      // Structured medical transcript sections en route
      console.log(result.transcript.sections.impression);
    }
  });

  return (
    <button onClick={isRecording ? stopRecording : startRecording}>
      {isRecording ? "Listening..." : "Dictate Report"}
    </button>
  );
}

Supercharge Reporting Workflows

Bring self-hosted, highly-accurate clinical dictation into your diagnostics environment.

Schedule Demo