As the demand for machine learning and data analysis increases with the global proliferation of online services and AI, its enormous computational load is becoming a challenge. This research aims to realize reservoir computing (RC) technology that enables in-situ sequential analysis of speech and other signal waveforms obtained from microphones at high speed with low computational cost by utilizing the nonlinear characteristics of Micro Electro-Mechanical Systems (MEMS) oscillators. Here, we propose a MEMS-RC that electrically connects multiple oscillators with a modulation mechanism using a thermal expansion actuator, and the strength of the coupled operation can be easily adjusted. we optimized the driving conditions for one and two coupled oscillators and compared their performance and found that the learning performance was improved with the two-coupled oscillators. Furthermore, combinations of coupling strengths that improve the learning performance of the two resonators were plotted on a heat map to identify regions of consistently high performance.