| Summary: | 碩士 === 國立中興大學 === 統計學研究所 === 104 === Chinese is a tonal language. The difference vowel phoneme will produce different formant frequencies. This thesis proposes an algorithm which extracts the first and second formant frequencies from the mandarin vowel phoneme’s frequency domain and observes the formants distribution under different vowel phonemes. The algorithm is as follows: First of all, using the Fast Fourier Transform (FFT) to transform sound waves from the time domain to frequency domain; second, find the relative extremum from the frequency domain waveforms. Third, remove the extremum if the energy is too minor; fourth, calculate the frequency difference from the relative extremum and use the maximum as the fundamental frequency of the sound waves. Finally, calculate the fundamental frequency and the ratio of high frequency energy to low frequency energy to find the possible frequency position of the highest formant energy. Then calculate the largest and second largest frequency of extremum from this range, use it as the first and second formant frequency position respectively. In this study, a total of 19 subjects and 209,760 voice data are used to do the test and observe the distribution of the different phonemes.
|
|---|
