Mapping Acoustic Vector Space and Document Vector Space by RNN-LSTM

Abstract

In this research, we propose a method to search between different media(cross media mapping)using Recurrent Neural Network(RNN)which is a kind of Neural Network(NN). By using the proposed method, it becomes possible to correlate music and lyrics, it is possible to search music using documents. Application of this model realizes a system that provides appropriate BGM from conversation contents by monitoring human interactions. In this paper, we constructed a proposal model, conducted an evaluation experiment, and confirmed the possibility of cross media mapping.