Che-Yu HsuPo-Han ChenTing-Yi ChenShang-Yu LinChing-Jen WangChingwei YehTay-Jyi LinPEI-ZEN CHANGWEI-CHANG LI2025-03-212025-03-212024-10-20https://scholars.lib.ntu.edu.tw/handle/123456789/726004This study demonstrates the development of miniaturized throat microphones using MEMS-based vibration sensors and a voice conversion algorithm. A MEMS-based accelerometer, attached to the throat along with an audio microphone, records the throat vibrations and sound wave data from the TMHINT Mandarin Chinese script, consisting of 320 sentences. The recorded vibration signals are processed to match the microphone signals by an artificial neural network and evaluated using objective metrics such as the Mean Opinion Score (MOS) and Short-Time Objective Intelligibility (STOI). The converted results show that a MEMS vibration sensor with a bandwidth of 2.4 kHz is sufficient for the purpose of voice recovery.MEMS accelerometersspeech quality assessmentthroat microphonesvibration sensorsvoice conversion[SDGs]SDG3[SDGs]SDG7conference paper10.1109/SENSORS60989.2024.10785086