Litcius/Paper detail

Gender and Age Group Predictions from Speech Features using Multi-Layer Perceptron Model

Saksham Goyal, Vinay Vasanth Patage, Sourabh Tiwari

202016 citationsDOI

Abstract

Gender and Age estimation from speech is a challenging research area for voice technology domains for use in personal voice assistants, user-profiling, targeted marketing etc. In this paper we discuss the prediction of gender and age group from speech features using Multi-Layer Perceptron architecture. Features like Mel Frequency Cepstral Coefficients, formants, pitch, chroma values etc. are extracted from every speech sample and are then selected using Principal Component Analysis and Redundant Feature Elimination techniques. The paper explains 2 approaches named sequential model and single model approach for gender and age group prediction. The Common Voice repository from the Mozilla Speech Database is used for training and testing. The implemented model gave us an accuracy of 89.58%. We have explained the approaches used in existing solutions and their drawbacks, and proved that our approach is giving better accuracy than state of the art solutions.

Topics & Concepts

Computer scienceFormantMel-frequency cepstrumPerceptronSpeech recognitionMultilayer perceptronPrincipal component analysisProfiling (computer programming)Artificial neural networkClassifier (UML)Feature (linguistics)Artificial intelligenceFeature extractionSpeech processingLinguisticsOperating systemPhilosophyVowelMusic and Audio ProcessingSpeech and Audio ProcessingSpeech Recognition and Synthesis