A method for retrieving an audio file includes: collecting an audio segment in real time; and for every two chronologically adjacent audio frames in a plurality of audio frames of the audio segment, acquiring a difference value between spectral centroids of a sub-band corresponding to the two audio frames, to obtain a plurality of difference values; and obtaining an audio fingerprint corresponding to the two audio frames based on the plurality of difference values. A quantity of the plurality of difference values equaling a quantity of sub-bands of one of the two audio frames. Each bit of the audio fingerprint being determined based on a comparison between a difference value corresponding to the bit and a preset difference value threshold. The method also includes retrieving, in an audio file library based on audio fingerprints of the plurality of audio frames, a target audio file matching the audio segment.