| 摘要 |
PROBLEM TO BE SOLVED: To provide a method for frequency analysis by two-stage processing which can efficiently perform the processing while maintaining the capability of increasing the reduction effect of a side lobe by a window without increasing the total throughput. SOLUTION: Time-series data of a sound signal are cut while overlapped if necessary and a window for reducing the side lobe of the frequency-division processing is multiplied by a window of a degree longer than the degree of discrete Fourier transform performed based on a frequency-division processing for the cut time-series data; and the time-series data having been multiplied by the window are divided by degrees of discrete Fourier transform performed on a time-division processing and added by sampling, and when the frequency- division processing is carried out on the basis of an arbitrary center frequency, the frequency shift that the discrete Fourier transform result has is compensated to take a detailed frequency analysis of the result of the frequency shift compensation with finer analysis width. |
