Analysis of Speech Signal Data of Mising Vowels using Logistic Regression and K-Means Clustering

Abstract

In this paper, an attempt has been made to study and analyze speech signal data. Here, the sound or speech data has different attributes like time, pitch, formant frequencies, speaker type, Vowel No etc. The dataset used here is speech signal data which are analog in nature and has been converted to digital format. After converting the data into digital format we want to establish a Logit model to predict the speaker gender on the basis of the pitch signal values which is also considered as fundamental formant frequency. That is our objective is to predict whether a speaker is male or female by looking at the pitch value by using logistic regression. We have applied clustering techniques to visualize and interpret how it works in speech signal data. The logistic model gives us 91% accuracy rate with low and efficient AIC value where as in case of the clustering algorithm we get a 93% accuracy for the whole sample.