This paper details the design and implementation of an embedded, real-time localization system founded on a MEMS microphone array to address the demand for real-time detection and localization of transient sound events within complex acoustic environments. A novel transient sound detection algorithm is proposed, which integrates a dynamic threshold with a four-state finite state machine (FSM). This algorithm dynamically updates the noise baseline via an exponentially weighted moving average model and employs the FSM's state transition mechanism for characterizing the waveform evolution of transient sounds, aiming to enhance detection robustness under conditions of low signal-to-noise ratios and non-stationary noise. The sound source localization method utilizes the Generalized Cross-Correlation with Phase Transform algorithm for estimating the time difference of arrival. Through acceleration strategies, including the selection of optimal microphone pairs and the pre-computation of a constant matrix, the computational complexity is significantly diminished while preserving localization accuracy. The system is architected around an STM32H743 microcontroller, integrates an 8-channel MEMS microphone array for synchronous data acquisition, and facilitates wireless communication with a host computer using an ESP8266 module. Experimental results indicate that the proposed transient sound detection algorithm sustains excellent performance in non-stationary noise environments. The accelerated sound source localization method achieves effective localization, with an average single-localization time of just 26 ms, thereby satisfying the real-time processing demands of the embedded system. This research provides a low-cost, high-efficiency technical solution for monitoring transient sound events in scenarios such as security surveillance and industrial equipment anomaly detection.