Indexingand retrieval of speechusing Power Normalised Cepstral Coefficients and Sub Band Coding

Audio which includes speech, music and other kinds of sounds is a part of multimedia and wide variety of research is being carried out in managing and organizing online audio databases. The goal of speech indexing and retrieval is used to facilitate the user to retrieve relevant speech in efficient manner. In this paper a novel method for speech indexing and retrieval is proposed using Power NormalisedCepstral Coefficients (PNCC), and Sub Band Coding (SBC). Thus each complete speech dialogue is separated into individual words by marking the Root Means Square (RMS) energy envelope using voice activity detection method. A query speech clip retrieval is done using Gaussian mixture model (GMM) based on features extracted. The probability that the feature vector belongs to the Gaussian is computed and retrieval is based on highest probability.