Breathe Module: Acoustic Monitoring
10 min read Peer Verified
Primary Input Architecture
The Breathe Module utilizes high-gain omnidirectional microphones configured as dual-channel arrays for spatial sound localization and noise rejection.
| Parameter | Specification | Clinical Rationale |
|---|---|---|
| Microphone Type | Electret condenser, omni | Captures respiratory sounds from multiple angles |
| Sample Rate | 48 kHz (configurable 24-96 kHz) | Nyquist frequency >20 kHz for high-frequency wheezes |
| Bit Depth | 24-bit | Dynamic range >110 dB for quiet breathing detection |
| Frequency Response | 20 Hz - 20 kHz ±3 dB | Full spectrum respiratory sound capture |
Analysis Core: FFT + TRNN
The Breathe Module employs a dual-stage analysis pipeline combining classical signal processing with deep learning:
Stage 1: Fast Fourier Transform (FFT)
Time-domain audio is transformed into frequency-domain representation using a 512-point Hann-windowed FFT with 75% overlap.
Stage 2: Temporal Rhythmic Neural Network (TRNN)
The TRNN is a bidirectional LSTM architecture trained on 150,000+ annotated respiratory cycles.
Output Specifications
1. Respiratory Rate (RR)Primary Metric
Calculated using autocorrelation of amplitude envelope. Accuracy: Sensitivity >92% for 10-60 breaths/min.
2. Cough FrequencyEvent Detection
Detected using spectral flux thresholding combined with duration filtering.
3. Wheeze DetectionPathology
Identified using spectral centroid elevation combined with harmonic structure analysis. Confidence scoring >0.85 triggers urgent alert.
>92%
Detection vs. gold-standard auscultation
±1.2 BPM
Mean absolute error vs. manual count
<8%
In clinical noise environments
Integration Architecture
AiVet enables seamless integration with practice management systems through secure webhooks and standardized FHIR resources.