基於微機電加速度計之機器學習喉部振動語音轉換可行性研究

Abstract

利用高規格壓電加速度計，可以很好的接收喉部表面的振動訊號，並且具備很好的環境雜訊免疫力。結合機器學習的語音轉換技術，可以補償喉部振動訊號在高頻訊號調音缺失的問題。為了推進喉部振動麥克風的商業應用普及化，本文針對喉部振動的感測器進行探討，分析不同感測器對於語音訊號的影響，提出運用微機電加速度計來改善原本壓電式體積龐大以及造價昂貴的問題。同時，探討運用傳統的語音增強方法來改善以微機電加速度計為感測基礎的振動訊號，進一步提升語音轉換的品質。

By utilizing high-quality piezoelectric accelerometers, the vibrations on the neck surface can be effectively captured, which is exhibiting the excellent immunity to environmental noise. With machine-learning-based speech conversion techniques, it becomes possible to compensate for the loss of high-frequency components in the vibration signal. To promote the widespread commercial application of neck vibration microphones, this study dedicates to investigate the different sensors' impact on speech signals. And proposes the use of MEMS accelerometers as a replacement for piezoelectric types, in order to address the issues of size and cost. In addition, the study explores the application of conventional speech enhancement methods to improve the vibration signals captured by MEMS accelerometers, for enhancing the quality of speech conversion.