Speaker Adaptive Real-Time Korean Single Vowel Recognition for an Animation Producing

Abstract

Voice Recognition technique has been developed and it has been actively applied to various information devices in Korea such as smart phones and car navigation systems. Since the basic research technique related the speech recognition has been based on research results of other languages such as English and Japanese, it is possible to meet a sort of difficulties or some problems in point of view from the recognition. It should check once at least or a margin for applying the Korean vocal sound system to improve the recognition of Korean speech, 44 since Korean phonemes always have a same phonetic value. However, the scope of this study is the recognition of single vowels for a digital contents producing, particularly lip sync animation, since the lip sync producing generally requires tedious hand work of animators and it seriously affects the animation producing cost and development period to get a high quality of lip animation. In this research, a real time processed automatic lip sync algorithm for virtual characters as the animation key in digital contents is studied by considering Korean vocal sound system. The proposed algorithm contributes to produce a natural condonable lip animation with the lower producing cost and the shorter development period. The system of real time vowel recognition for producing digital contents focusing on formants frequencies is proposed. The recognition process consists of speech signal as the input, filtering, Fast Fourier Transform and identification. The algorithm based on the formant frequency using F1 and F2 was proposed, whose output was applied to the autonomic natural animating of the character’ s mouth shape for small and medium sized animation productions or e-learning contents productions. The result shows the proposed speaker dependent single vowel recognition system is able to distinguish Korean single vowels from dialogue of a dubbing artist with real-time. The average of the recognition ratio was 97.3 % in the laboratory environment. It gives a possibility that the more condonable lip sync produces automatically without any animator involved.