Encountering natural fractures or unbalanced formation pressure during oil drilling can result in economic loss and environmental pollution due to well leakage. Existing detection methods encounter challenges such as high costs, complex downhole environments, and difficult data acquisition. To address these issues, we propose a well leakage detection method using cepstrum for analyzing transient pressure waves. Cepstrum is a signal Fourier transform after logarithmic operation and then Fourier inverse spectrum obtained. By studying the propagation of transient pressure waves in the wellbore, we identify drilling fluid leakage location and amount based on time-dependent and amplitude changes of pressure wave signal characteristic peaks. To handle noise in the pressure wave signal, we employ adaptive noise-complete ensemble empirical modal decomposition (CEEMDAN) and wavelet threshold (WT) joint denoising. Correlation coefficient (CCF) with the Hilbert joint spectrum (HJS) is used to extract main frequency components, achieving denoising. Experimental results confirm: ① Noise interference in transient pressure waves is effectively suppressed using the CEEMDAN-WT-CCF-HJS denoising method. ② Cepstrum analysis of the pressure wave signal during wellbore annulus system leakage reveals distinct reflected wave characteristic peaks, aiding in locating different leakage points, with the amplitude of these peaks reflecting the size of the leakage. ③ This method efficiently utilizes time-frequency information from the excitation pressure wave signal, offering advantages over traditional time-domain and frequency-domain analysis. Experiments covering various leakage scenarios, amounts, and borehole sizes yielded controlled experimental errors (2.25%–9.10%), within a reasonable range. The method's validity and reliability were confirmed, providing theoretical support and technical guidance for well leakage detection in oil drilling.