Cross-modal retravel has been used in many fields, such as business and search engines. Most search engines for business are text-based, but text-based search engines are limited by equipment and the strict requirement for knowledge. Text-based search needs keyboards to finish the search process, which requires users to have the knowledge of using keyboards. Compared to the text-based search, audio-based search has advantages. First, it avoids the traditional ways of inputting information. And it gets rid of the gap in time between inputting information for searching and getting useful information. In this paper, we propose a way to use audio to search images for business applications. We use deep learning to implement cross-modal retrieval systems between images and audio. We first extract features from images and audio respectively. And then we implement a neural network with two identical networks to learn the correspondence between images and audio. The first network extracts the features from images and audio further for calculation, and the second network learns whether two features from different modalities are related. This research provides a new way for business applications to search for information more instantly. © Published under licence by IOP Publishing Ltd.