Estimating Social Background Profiling of Indian Speakers by Acoustic Speech Features

Social background profiling of speakers refers to estimating the geographical origin by their speech features. Methods for accent that use linguistic features, require phoneme alignment and transcription samples. This paper proposes a purely acoustic model, composed multiple convolutional networks with global average-pooling layers, classify temporal sequence The bottleneck representations networks, trained original signals low-pass filtered copies, are fed Support Vector Machine classifier final prediction. model has been analysed dataset Indian from social backgrounds spread across India. It shown up 85% accuracy is achievable classifying geographic corresponding regional languages; 17% higher than benchmark deep learning using same Results have also indicated classification accents easier second language speakers, as compared native language.