<p>Obstructive sleep apnea (OSA) is a chronic respiratory disorder that is frequently underdiagnosed. The primary clinical methods for OSA detection are polysomnography (PSG) and polygraphy (PG). While these tests provide an accurate assessment of the patient’s condition, they remain impractical for large-scale population screening. Consequently, considerable research has been dedicated to developing audio-based OSA assessment techniques based on data from synchronized audio and PSG/PG recordings. Given that most PSG/PG equipment used in sleep centers lacks built-in support for synchronizing with audio recordings, this works presents a novel method for synchronizing audio with PG signals to enhance audio-based analysis of OSA. The proposed method introduces an iterative time-alignment algorithm based on the cross-correlation between an estimated respiratory sound signal and the nasal flow signal from PG. To estimate the respiratory sound from the captured sound signal, constrained non-negative matrix factorization (NMF) is applied using a combination of orthogonality and sparsity constraints to identify the component that best models the respiratory pattern of each analyzed audio segment. The synchronization performance of the proposed method was tested on a newly developed dataset comprising full-night measurements from 32 subjects, achieving a mean absolute error of less than one second. Finally, the developed dataset demonstrated its validity for OSA severity assessment by providing comparable results to those obtained from other datasets when evaluated using a state-of-the-art method.</p>