Acoustic detection in confined spaces is vital for security surveillance, structural diagnostics and biological behavior monitoring. However, the complex acoustic environment in confined spaces—characterized by reflection, reverberation, and attenuation, makes it difficult to accurately acquire specific frequency bands signals. Therefore, it is essential to develop acoustic sensors for confined spaces with high precision, enhanced sensitivity, and strong frequency selectivity, particularly for a widespread and common mid-frequency acoustic signals. In this work, we propose a mid-frequency triboelectric acoustic sensor (MTAS) engineered for 200–3000   Hz signal acquisition in confined spaces. A SrTiO₃-doped polyvinylidene fluoride (PVDF) composite film is developed to enhance triboelectric charge density. This enhancement enables the MTAS to exhibit high sensitivity at 60-90   dB, along with excellent mid-frequency selectivity. In a cylindrical confined space, the MTAS successfully detected acoustic signals generated from 17 distinct units on a hemispherical surface using a single sensing channel. When combined with a one-dimensional probability density convolutional neural network, the system achieves a 97.86% average recognition rate for sound source localization. This research offers an efficient approach to improve the sensitivity of triboelectric acoustic sensors and expands the potential for sound detection in confined spaces.