<p>Sound control in noisy or reverberant spaces is important for applications ranging from communication to immersive audio. However, existing methods often struggle to deliver sound selectively to specific listeners without interference. Here we show that an active acoustic metasurface, composed of programmable elements that both sense and re-emit sound, enables precise targeting of audio in complex environments. Each element processes signals in real time using convolution filtering, allowing us to exploit reciprocity and time-reversal symmetry in wave propagation. Experiments with audible sound in reverberant rooms demonstrate that this approach creates clear, individualized sound channels while suppressing unwanted noise. This research opens new possibilities for adaptive sound delivery in crowded or dynamic settings, with potential applications in conferencing, entertainment, and assistive listening technologies.</p>