Voiced-Unvoiced Classification of Speech Using a Neural Network Trained with LPC Coefficients

摘要

Voiced-Unvoiced classification (V-UV) is a well understood but still not perfectly solved problem. It tackles the problem of determining whether a signal frame contains harmonic content or not. This paper presents a new approach to this problem using a conventional multi-layer perceptron neural network trained with linear predictive coding (LPC) coefficients. LPC is a method that results in a number of coefficients that can be transformed to the envelope of the spectrum of the input frame. As a spectrum is suitable for determining the harmonic content, so are the LPC-coefficients. The proposed neural network works reasonably well compared to other approaches and has been evaluated on a small dataset of 4 different speakers.