Fundamental frequency (f₀)

In Japanese, a mora basically corresponds to a syllable, the exceptions are prolonged syllables with two moras. As is well known, a Japanese word of M mora length has M+1 accent types expressed by mora position of pitch-drop. Fujisaki et al. proposed a practical f₀ estimation method based on these accent types [5]. Conventional synthesis methods usually estimate f₀ contours of target words using some method like this, and select waveforms with estimated f₀ values or manipulate them to yield the required values. We use the mora position of the syllable within the word directly to represent the pitch of the syllable because this eliminates the need to consider the f₀ value itself. A careful investigation of the f₀ contours of Japanese city names and town names clarified that most city and town names with the same mora length have the same accent type, especially among words with four to six mora. Figure 1 shows the mean f₀ contour and standard deviations obtained from 2,800 of Japanese city and town names with five moras pronounced by a single narrator. Standard deviations are small enough to neglect accent types. Consequently, the f₀ of a syllable within a word of mora length M can be represented by its mora position m and M.

  

Figure 1: Means and standard deviations of f₀