The research focused on understanding different synthetic speech production techniques and developing signal processing based features for distinction of synthetic and natural speech.

(July, 2013 – June, 2014)

Thesis title: Analysis of Voice Biometric Attacks (Automatic speaker recognition systems
Author: Adarsa
Supervisor: Dr. Hemant A.Patil, DAIICT
Download

The improvement in text-to-speech (TTS) synthesis poses the problem of biometric attack on speaker verification systems(SVS). In this context, the study analyses the performance of TTS systems using false acceptance rate to impostor using artificial speech.
The research identifies different features for distinction of natural and synthetic speech. Three specific naturalness components are identified considering prosody, nonlinearity. Further method of integrating these into GMM-UBM system is proposed.