AI Smart Hearing Amplifier based on BES2600

Client

A hearable technology startup developing an over-the-counter hearing amplifier that automatically adapts to acoustic environments without user intervention.

Challenge

Running real-time environmental sound classification concurrently with low-latency audio processing (under 10ms) on a battery-powered Bluetooth audio SoC.

Product Concept & Architecture

The BES2600 dual-core Bluetooth audio SoC handles both audio DSP and BLE profiles. Three Knowles SPH0645 MEMS microphones form a beamforming array. A Knowles balanced armature receiver delivers high-fidelity amplification.

Engineering Solution

We allocated one BES2600 core to the audio pipeline (dual-mic beamforming + WDRC amplification + feedback cancellation) and the second core to environmental classification. Every 3 seconds, a 300ms audio snippet is classified by a TinyML CNN trained on 20 acoustic scenes. On scene change, the amplification profile transitions smoothly over 500ms.

Result

Sub-8ms audio latency. 92% scene classification accuracy. 15dB SNR improvement in restaurant environments. 12-hour continuous battery life.