The article outlines a research initiative that investigates the use of machine learning to enhance acoustic eavesdropping attacks, particularly focusing on smartphones. Conducted by an intern and a newly hired engineer, the study pivots on the capabilities of gyroscopes and accelerometers—sensors that are standard in mobile devices but often overlooked for security vulnerabilities. Traditional microphone-based eavesdropping requires user consent, while these inertial sensors can be accessed without explicit permission, making them a target for side-channel attacks. The researchers highlight two significant projects: Gyrophone and AccEar. Gyrophone successfully extracted small amounts of speech from gyroscope data using signal processing techniques, while AccEar advanced this work by utilizing a Generative Adversarial Network (GAN) to restore about 85% of the original speech from accelerometer readings. The team is replicating these experiments using a Google Pixel 7a to test the ongoing viability of these exploits. They aim to preprocess recorded sensor data and apply extensive AI modeling to further improve speech recovery from the sensors. Upcoming parts of the series will detail the preliminary testing phase and the results of their attempts to exploit these vulnerabilities. The research raises significant concerns about user privacy, given the potential for unauthorized access to sensitive audio information through commonly used smartphone sensors.