Date of Graduation

5-2020

Document Type

Thesis

Degree Name

Bachelor of Science

Degree Level

Undergraduate

Department

Computer Science and Computer Engineering

Advisor/Mentor

Luu, Khoa

Committee Member/Reader

Gauch, John

Committee Member/Second Reader

Li, Qinghua

Abstract

Deep learning has been recently proven to be a viable asset in determining features in the field of Speech Analysis. Deep learning methods like Convolutional Neural Networks facilitate the expansion of specific feature information in waveforms, allowing networks to create more feature dense representations of data. Our work attempts to address the problem of re-creating a face given a speaker's voice and speaker identification using deep learning methods. In this work, we first review the fundamental background in speech processing and its related applications. Then we introduce novel deep learning-based methods to speech feature analysis. Finally, we will present our deep learning approaches to speaker identification and speech to face synthesis. The presented method can convert a speaker audio sample to an image of their predicted face. This framework is composed of several chained together networks, each with an essential step in the conversion process. These include Audio embedding, encoding, and face generation networks, respectively. Our experiments show that certain features can map to the face and that with a speaker's voice, DNNs can create their face and that a GUI could be used in conjunction to display a speaker recognition network's data.

Keywords

Speaker Recognition, Speech to Face, MFCC, Convolution Neural Networks, Service Learning

Download

Share

COinS