Personalised Hearing for Savox-Otos
The Challenge
Professional hearing protection headsets are largely one-size-fits-all in terms of audio tuning. Workers with noise-induced hearing loss struggle to understand speech, reducing safety and increasing fatigue. The challenge was to apply worker-specific hearing profiles to the audio signal without modifying the certified headset hardware, while ensuring ultra-low latency (<3ms) and preserving high speech intelligibility avoiding low-frequency masking.
My Role & Collaborators
As Tech Lead (Acoustics and Audio Tech), I engineered the real-time audio pipeline and DSP for embedded hardware (Teensy 3.2/4.0). I implemented the NAL-R prescription algorithm with 8-band WDRC compression and managed the overall technical architecture.
Our Solution
We engineered an inline NFC-enabled DSP module that reads individual audiometric data from NFC-encoded cards and applies frequency-specific amplification using the NAL-R prescription formula to maximize speech intelligibility. The hardware employs an 8-band parametric equalisation architecture, which is being migrated to a multiband WDRC compression pipeline with soft-knee per-band compression.
Key Techniques
- NAL-R Prescription Formula Implementation
- 8-Band Parametric EQ & WDRC Compression Pipeline
- Embedded DSP (Teensy 3.2/4.0, ARM Cortex-M4/M7, PJRC Audio Shield)
- NFC-based Audiogram Loading (NTAG213, HIMSA Noah format XMLs)
- Real-time Audio Processing (under 3ms latency)
- Psychoacoustic Testing with KEMAR mannequin
Project Links
Results
Validated a functional prototype that applies NAL-R correction in real-time. Conducted A/B comparisons with simulated hearing loss showing profound intelligibility improvements, shifting towards multi-band WDRC for optimal loudness growth.
System Architecture
Project Gallery
