Mandarin Tones Recognition by Segments of Fundamental Frequency Contours

Mandarin is one of the tonal languages. In Mandarin tones, there are four lexical tones (tone 1 to tone 4) with four different fundamental frequency (f0), such as flat and high, rising, falling and then rising, and falling, respectively. In order to process signal on lexical tone, at first we have to identify which tone is. We would like to find out an efficient approach to identify Mandarin tones by the segments of fundamental frequency contours. In this study, 3 male and 3 female participants engaged in recording speech materials. All participants are native Mandarin speakers, no history of any speech or hearing disorder, and all passed articulation and voice assessment. There are two target syllables (/ti/ and /tu/) of four lexical tones used for the materials. In our experiment, we analysed the signal features and the acoustics characteristics that included the range of f0, the average of f0 and so on. Then we could predict which tone is from what the segments told. The result of this study revealed that the segments of the contours could not target the corresponding tone correctly. The approach of this study may not provide a way for hearing-devices to predict Mandarin tone before signal processing. The further study for prediction by segments of f0 contours is required.