The embodiment of the invention provides a multi-person scene speaker-voice matching method which comprises the following steps of dividing an audio to be matched into a plurality of sound fragments;performing voice recognition on the sound fragments to obtain voice fragments in the sound fragments; obtaining video fragments corresponding to the voice fragments; performing face detection on the video fragments to obtain all predicated speakers in the voice fragments; according to the pixel difference values of adjacent gray level frames in the video fragments, obtaining the hitting information of each predicated speaker in the adjacent gray level frames; and counting the hitting times of each predicated speaker in the video fragments according to the hitting information, wherein the predicated speaker with the maximum number of the hitting times is a target speaker of the voice fragments. The multi-person scene speaker-voice matching method achieves the goal of automatically binding voice to the subordinate target speaker; the workload of manually matching the voice with the target speaker in a subsequent process can be greatly reduced; and the practicability of a visual auditorycognitive technology is favorably promoted.