Keyword extraction is widely used for information indexing, compressing, summarizing, etc. Existing keyword extraction techniques apply various text-based algorithms and metrics to locate the keywords. At the same time, some types of audio and audiovisual content, e. g. lectures, talks, interviews and other speech-oriented information, allow to perform keyword search by prosodic accents made by a speaker. This paper presents PitchKeywordExtractor — an algorithm with its software prototype for prosody-based automatic keyword extraction in speech content. It operates together with a third-party automatic speech recognition system, handles speech prosody by a pitch detection algorithm and locates the keywords using pitch contour cross-correlation with four tone units taken from D. Brazil discourse intonation model.