Search is a problem ubiquitous in diverse domains and modalities. Speech search engine plays a vital role in numerous speech communication applications, including audio/video search and retrieval, audio broadcast monitoring, and voice-based surveillance. The performance of the speech search system depends critically on the representation of the speech signal. A good feature representation should be speech-specific, and at the same time, it should be robust to the speaker and channel variability. In this project, we are developing two kinds of speech search systems – (i) language-dependent (works for a specific language), and (ii) language-agnostic (works for any language). The system will be used for NPTEL audio/video search for Indian languages.