MaskClip: Detachable Clip-on Piezoelectric Sensing of Mask Surface Vibrations for Real-time Noise-Robust Speech Input

TL;DR

MaskClip addresses speech recognition degradation when wearing masks in noisy settings by introducing a hardware-based, detachable clip-on microphone that senses mask-surface vibrations via a piezoelectric sensor. The method selectively records speech-induced mask vibrations while attenuating ambient noise, enabling robust recognition without heavy GPU processing. Key results show a CER of 5.1% in quiet and 6.1% in noisy conditions, outperforming conventional pin microphones (9.4% and 19.7%), with high subjective audio quality (MOS) in normal speech. The approach promises practical voice interfaces in medical, cleanroom, and industrial environments and outlines future work toward multimodal integration and mobility studies to broaden deployment.

Abstract

Masks are essential in medical settings and during infectious outbreaks but significantly impair speech communication, especially in environments with background noise. Existing solutions often require substantial computational resources or compromise hygiene and comfort. We propose a novel sensing approach that captures only the wearer's voice by detecting mask surface vibrations using a piezoelectric sensor. Our developed device, MaskClip, employs a stainless steel clip with an optimally positioned piezoelectric sensor to selectively capture speech vibrations while inherently filtering out ambient noise. Evaluation experiments demonstrated superior performance with a low Character Error Rate of 6.1\% in noisy environments compared to conventional microphones. Subjective evaluations by 102 participants also showed high satisfaction scores. This approach shows promise for applications in settings where clear voice communication must be maintained while wearing protective equipment, such as medical facilities, cleanrooms, and industrial environments.

Figures (8)

• Figure 1: Comparison of speech recognition performance between MaskClip and a unidirectional microphone. The upper spectrogram shows the recording from MaskClip, demonstrating clear speech articulation. The lower spectrogram shows the recording from a conventional unidirectional microphone, revealing significant background noise interference. The contrast illustrates MaskClip's superior noise resistance capabilities.
• Figure 2: Detailed hardware implementation of the MaskClip device. The left image shows the physical device implementation, while the right diagram illustrates the electrical circuit configuration. The core components consist of a piezoelectric sensor, preamplifier circuit, codec, and ESP32 xiao microcontroller. The signal processing path operates at 3.3V, where signals from the piezoelectric sensor are amplified by the preamplifier, digitized by the codec, and processed by the microcontroller. The stainless-steel clip design ensures secure attachment and stable signal detection. This architecture enables low-power operation while maintaining high-performance voice detection capabilities. The compact form factor and integrated design demonstrate the system's practical viability for real-world applications.
• Figure 3: Overview of experimental conditions used for systematic evaluation of MaskClip's sensor configuration parameters. From left to right: clip materials (stainless steel versus plastic), sensor directions (inward-facing versus outward-facing), and positioning (non-woven mask versus fabric mask). Performance impacts were evaluated across combinations of these parameters.
• Figure 4: shows the results of testing different sensor configuration parameters for the MaskClip device. The graph compares performance between stainless steel and plastic clips, sensor orientations (inward-facing versus outward-facing), and positions ranging from 0mm to 70mm from the left edge of the mask. The stainless steel clip with an inward-facing sensor achieved the best performance, with peak performance of approximately 70 dB at the 10mm position. Even at distances of 30-60mm, the stainless steel configuration maintained stable signal detection around 45-50 dB. These results demonstrate that the stainless steel clip provides superior vibration transmission characteristics, making it the optimal choice for the final design.
• Figure 5: Comprehensive overview of the speech recognition evaluation process. (a) Comparison between software-based and hardware-based evaluation methodologies, (b) structure of the utilized datasets, and (c) experimental setup configuration. The evaluation employed wTIMIT and LibriMix datasets to simulate real-world conditions, enabling robust performance assessment.