In this study, a piezoelectric MEMS (Micro-Electro-Mechanical-System) speaker with thoughtfully selected resonant frequencies of dual cantilever units is designed and implemented. Based on the lead zirconium titanate (PZT) thin film with superior piezoelectric coefficient, the cantilever diaphragm is designed through the modal analysis. In the 1.5mm by 1.5mm cantilever array, a high-frequency actuation unit with resonance of 13.1kHz and a low-frequency actuation unit with resonance of 6.35kHz constitute the proposed piezoelectric MEMS speaker. The boost of sound pressure level (SPL) due to the resonant mode of low-frequency actuation unit reduces the total harmonic distortion (THD) around the subharmonic frequency of high-frequency actuation unit. Meanwhile, the asymmetric triangle diaphragm reduces the maximum stress in the structure when operating at the resonant frequency, thereby mitigating the THD spikes resulting from the nonlinear structural behavior. In pressure-field measurements, the proposed design can reach SPL ≥ 70.0dB from 2.7kHz to 15.0kHz with only 0.179 Vrms input. At the same time, THD remains below 3.0% from 3.2kHz to 20kHz. The outstanding performance under larger input signal and bias voltage is also verified. Actuated by 0.354 Vrms and 2 VDC bias, the maximum SPL achieves 111.9dB and the SPL is higher than 80.0dB from 3.3kHz to 14.1kHz. Furthermore, THD is lower than 3.0% from 1.0kHz to 8.9kHz. A promising solution of piezoelectric MEMS speaker for in-ear applications is demonstrated in this study.