In recent years, people have become more interested in using acoustics to locate equipment in intelligent human-computer interfaces (HCI). However, existing acoustic locating systems have poor compatibility with smart devices. To achieve millimeter-level accuracy, most systems require large microphone spacing, complex algorithms, and a high cost. In this paper, we propose a new HCI system, MilliLoc, to achieve millimeter-level real-time acoustic locating. Our first contribution is to design a high-performance and convenient microphone array. It has a high sampling rate of 96 kHz, real-time complete dataflow, and is compatible with any smart device. The second contribution is to propose an improved generalized cross-correlation phase transform method with simplicity and real-time capability. Even if the signal-to-noise ratio is not high enough, it can still estimate the time delay accurately. We combine the microphone array with the algorithm to illustrate the implementation of MilliLoc. Experiments show that MilliLoc has a median 2D accuracy of 7.2 mm in a 60 cm sector area and achieves millimeter-level locating.