Mandarin Tone Recognition Based on Pre-Classification
Lin Xin Tang, Junxun Yin · 2006
Mandarin tone behavior varies greatly from speaker to speaker and it is this that presents the greatest challenge to any speaker-independent tone recognition system. This paper presents a speaker normalizing algorithm which is designed to reduce this influence. In addition to this, the other tone variations are pre-classified into four categories: the neutral tone, and the isolated syllable, the disyllable and the polysyllable. The neutral tone is a special tonal sandhi, and the recognition rate of the neutral tone is poor in many current tone recognition systems. So, this paper proposes a special treatment for the neutral tone. The experimental results show that by using these new methods, an accuracy rate of 96.64% can be achieved in Mandarin tone recognition.