I had a presentation at the speaker and language recognition workshop 'Odyssey 2010' held in Brno, Czech. This is a collaboration with Phillip from NMSU and Michael from ftw on speaker verification systems. We reported that synthesised speech generated from speaker-adaptive HMM-based speech synthesis systems is enough high to allow these synthesised voices to pass for true human claimants despite the good performance of the speaker verification systems and emphasised that we need to develop new features or strategies to discriminate synthetic speech from real speech because the conventional methods to detect synthetic speech is no longer robust enough.

Slides (pdf format)
odyssey10-ver3