Syllabic component collecting stage

1.

Generation of syllables

• All target words $\{W\}$ are classified according to mora length. $\{W_M\}$ is the word set whose mora length is M. Thus, $\{W\}=\cup_{M=1}^{K}\{W_M\}$ where K is the longest mora length among the targeted words.
• Develop each of the classified words into a series of syllables. $W_{M,i}=\{Sy(P,N)_{m,M}\}$ $(i=1~{\rm to}~L,~m=1~{\rm to}~M)$ where L is the number of words of mora length M. As an example, the 5 mora word ``yokohamashi'', a Japanese city, can be expressed as $\{{\rm yo}(,/k/)_{1,5},$ ${\rm ko}(/o/,/h/)_{2,5},$ ${\rm ha}(/o/,/m/)_{3,5},$ ${\rm ma}(/a/,/s/)_{4,5},$ ${\rm shi}(/a/,)_{5,5}\}$.

2.

Selection of a word set and recording

• Collect all the different syllables
• Select the minimum set of words that includes all the above syllables.
• Record the selected words using a designated narrator.

3.

Database creation of syllabic components

• Divide the recorded words into a set of syllabic waveforms.
• Pair each syllabic component with its waveform, and store them in a database. Thus syllabic component Sy(P,N)_m,M is expressed as $\{Sy(P,N)_{m,M},{\rm index~to~the~waveform}\}$.

This procedure is shown in Fig.2.

  

Figure 2: Database creation process