Inclusive & Assistive Products
ISDN2001/2002: Second Year Design Project
Microphone
Placement
Research Context
While extensive research has been conducted on microphone placement for optimizing audio quality, particularly in contexts such as singing and voice projection (Brixen, n.d.), the specific impact of microphone placement on capturing infant sounds for machine learning applications remains largely unexplored. This gap is particularly pronounced in diverse acoustic environments, where the nuances of infant vocalizations must be accurately detected and classified.
Research Questions
1. Whether a wearable microphone or a detached microphone captures the most accurate infant vocalisations in diverse acoustic environments.
2. How the microphone placement affects the accuracy of machine learning models designed to detect and classify infant sounds.
Method
Distances measured
(control variable:
-
40cm directly above mouth
-
15cm directly above mouth
-
10cm directly above mouth
-
below the chin
This investigation will be conducted by comprehensively analysing existing literature studies on microphone placements for infants, focusing on its impact on audio quality and machine learning model performance. Following this, potential microphone placements will be identified and narrowed down. Subsequently, experiments will be conducted using various microphone placements, such as on the arm or detached at different distances. MATLAB will be utilized to assess the effects of these placements on frequency response and signal-to-noise ratio (SNR). Finally, the performance of machine learning models trained with data from each microphone placement will be compared to determine the most effective setup.
Results
Machine Learning Results Analysis
The analysis showed that when audio recordings from various microphone positions were used, the predictions for the recordings taken 10 cm above the mouth were the most accurate. As the distance between the microphone and the sound source increased, the prediction accuracy decreased significantly, often resulting in incorrect or low-confidence predictions. This confirms that microphone distance plays a crucial role in the quality of audio data for analysis.
Signal-to-Noise Ratio (SNR) Analysis Findings
Analysis of SNR using MATLAB revealed that microphone proximity to the infant correlated with higher signal-to-noise ratios. However, the submental position (directly below the chin) exhibited suboptimal SNR performance, likely due to acoustic wave obstruction – a phenomenon consistent with hypotheses in prior literature.
Conclusion
Current baby monitors offer real-time video and audio feeds, yet many caregivers, especially mothers, experience anxiety about missing critical moments when not actively monitoring. My research focused on identifying the most suitable microphone placement for infants, guided by both SNR analysis and machine learning model accuracy. The findings confirmed that microphone proximity significantly impacts audio quality and prediction accuracy, with optimal results achieved closest to the infants, specifically, within a range of less than 30 cm from the baby.
Based on these insights, our group determined that the final baby monitor design could feature an adjustable microphone mount to allow for portability and convenience. However, it is essential to inform parents that for best performance, the microphone should be placed at a distance of less than 30 cm from the infant. This informed design approach will enhance the reliability of automated alerts, reduce caregiver anxiety, and directly address the core problem statement.