The invention discloses a multimedia resource retrieval method based on audio content retrieval, comprising the following steps: 1) a pretreatment server converts video and audio into standard voice to be recognized; 2) a voice recognition server trains training corpus into an acoustical model, and matches the voice to be recognized with the acoustical model to obtain semantic text indexes; (3) an index server stores and recognizes keyword indexes, and matches the indexes with a retrieval condition to get a retrieval result. By using keyword retrieval technology in audio, the internal semantic information of audio / video resource can be obtained, the text type semantic information is indexed, therefore, the method provides more comprehensive and reliable audio / video resource information indexes, such that a retrieval system can get matched multimedia resource more accurately, and locate the precise position of retrieval keywords in audio / video.