Artificial Speech Corpus in Japanese for Anti-spoofing Automatic Speaker Recognition

Abstract

In this paper, we introduce an artificial speech corpus in Japanese designed for anti-spoofing automatic speaker recognition. This speech corpus contains the speech data which are generated by a voice conversion method and a text-to-speech synthesizer. Using this speech corpus, we conduct speaker recognition experiments. Experimental results show that it is possible to spoof the automatic speaker recognizer using the artificial speech. Especially, it is difficult to distinguish the synthesis speech from the genuine speech compared with the voice conversion speech.