Speaker: Md Sahidullah (Visiting Researcher)
Date: August 31, 2017
Automatic speaker verification (ASV) technology is recently finding its way to end-user applications. This voice-based authentication technology shows promising recognition performance in the controlled conditions. However, ASV technology is highly vulnerable to spoofing attacks where an intruder uses a synthetic or recorded voice to get illegitimate access. Over the years, several interesting studies were carried out to develop spoofing countermeasures, but experiments were conducted with in-house datasets consisting of a limited number of spoofing attacks. The automatic speaker verification spoofing and countermeasures challenges (ASVspoof) have been organized recently for systematic benchmarking and evaluation of anti-spoofing algorithms. With a participation of more than 100 research groups, this challenge series has gained considerable attention from the academia and industry. In both editions of the challenge, we observe a significant improvement over baseline state-of-the-art systems.
This talk will have two parts. First, a brief overview of speaker recognition system will be presented. In the next part, spoofing problem and anti-spoofing measures will be discussed. This will contain the summary of the findings and progress made in recent years. We will conclude by discussing some open research problems, such as the development of generalized spoofing countermeasures, joint recognition of speaker and spoofed speech, and recognition of spoofed speech in noisy conditions.