Introduction and Problem Space
Acoustic sensing plays a growing role in smart infrastructure for detecting occupancy, appliance usage, and safety-critical anomalies where latency is key and event timing is unpredictable. These scenarios demand always-on operation, but fixed high-rate microphone sampling incurs high power and storage costs (see figure below), limiting use in low-cost, embedded systems. Prior adaptive sampling approaches often rely on narrow-band speech, analog filters, multi-microphone arrays, or task-specific heuristics, each increasing hardware complexity or reducing generalizability. We present EfficientMic, a digital, aliasing-aware framework for adaptive acoustic sensing using only a single microphone.
Contributions
We introduce EfficientMic, a lightweight, aliasing-aware adaptive sampling framework that dynamically adjusts microphone sampling rates in real time using a single microphone. EfficientMic detects aliasing directly from Short-time Fourier Transform (STFT) features using a compact XGBoost classifier [1] and increases the sampling rate only when necessary to preserve signal fidelity. This approach reduces computational and storage overhead without requiring task-specific tuning or additional hardware. EfficientMic is designed for deployment on low-power, always-on embedded systems that lack analog filtering or multi-channel support, making it suitable for a wide range of smart infrastructure applications.
The contributions of this work are:
1) The first aliasing-aware adaptive sampling framework that operates fully in the digital domain using only a single microphone.
2) We develop a lightweight aliasing detection pipeline using STFT features and an XGBoost classifier, enabling real-time adaptation on low-power embedded platforms.
3) We build and release a curated dataset of 587,350 unique 1-second aliased and non-aliased STFT and MFCC audio representations at varying frequencies derived from public datasets to support reproducibility future research:
https://anonymous.4open.science/r/Aliasing-Audio-STFT-Dataset-5156