Voice Based Retrieval using Convolution Neural Network in Deep Learning

Abstract

In this paper, we propose a Content Based Voice Retrieval(CBVR) is used to search a specific audio files from a large data base. Using Deep learning features are learned automatically in the training phase. Convolution Neural Network is used in this research for CBVR. On this paper, we recommend a novel technique to key word search(KWS) in low-resource languages, which presents an replacement method for retrieving the phrases of curiosity, in particular for the out of vocabulary (OOV) ones. Our procedure contains the approaches of question-by using-illustration retrieval tasks into KWS and conducts the hunt by the use of the subsequence dynamic time warping (sDTW) algorithm. For this, text queries are modeled as sequences of function vectors and used as templates within the search. A Convolution neural network-headquartered model is informed to gain knowledge of a frame-degree distance metric to be used in sDTW and the right question model frame representations for this realized distance. This new procedure can be used as a substitute to traditional LVCSR-situated KWS programs, or in combination with them, to attain the intention of filling the gap between OOV and in-vocabulary (IV) retrieval performances.