A Piezoelectric Micromachined Ultrasonic Transducer-Based Bone Conduction Microphone System for Enhancing Speech...

Area of Science:

• Acoustics
• Signal Processing
• Biomedical Engineering

Background:

• Speech recognition in noisy environments is challenging due to limitations of air conduction microphones (ACMs).
• Environmental noise significantly degrades the quality of speech captured by ACMs.
• Existing methods struggle to achieve robust speech recognition under high noise conditions.

Purpose of the Study:

• To develop a novel system for enhanced speech capture and recognition in noisy environments.
• To leverage bone conduction (BC) for more robust speech signal acquisition.
• To improve speech enhancement by fusing BC and AC speech signals.

Main Methods:

• A customized bone conduction microphone (BCM) system using a piezoelectric micromachined ultrasonic transducer was developed.
• Simultaneous capture of speech via real-time bone conduction (BC) and air conduction (AC) using integrated microphones.
• A speech enhancement module fused BC and AC speech, using BC speech as a reference for noise-insensitive adaptation.

Main Results:

• The BCM system achieved a signal-to-noise amplitude ratio over five times greater than ACMs in a 68 dB noise environment.
• Fused speech enhancement reduced character error rate by over 20% compared to noisy speech.
• Enhanced speech accuracy approached that of clean speech recognition.

Conclusions:

• The developed system efficiently captures BC speech, offering a simpler and more robust alternative.
• Fusion of BC and AC speech signals significantly enhances speech recognition in noisy conditions.
• This innovative approach improves speech clarity and recognition accuracy in challenging acoustic environments.