Music – Instruments – Electrical musical tone generation
Reexamination Certificate
2002-01-03
2003-12-09
Donels, Jeffrey (Department: 2837)
Music
Instruments
Electrical musical tone generation
C704S209000
Reexamination Certificate
active
06660923
ABSTRACT:
CROSS-REFERENCE TO RELATED APPLICATIONS
The present application claims priorities under 35 U.S.C. §119 to Japanese Patent Application No. 2001-001409, filed on Jan. 9, 2001 and entitled “Method for extracting formants of a musical tone, recording medium and apparatus for extracting formants of a musical tone”, Japanese Patent Application No. 2001-375423, filed on Dec. 10, 2001 and entitled “Method for extracting formants of a musical tone, recording medium and apparatus for extracting formants of a musical tone”, and Japanese Patent Application No. 2001-392305, filed on Dec. 25, 2001 and entitled “Method for extracting formants of a musical tone, recording medium and apparatus for extracting formants of a musical tone”. The contents of these applications are incorporated herein by reference in their entirety.
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a method for extracting formants of waveform data of a sampled musical tone, a recording medium and an apparatus for extracting formants of a musical tone.
2. Discussion of Background
Frequency characteristics are ones that show the characteristics of a musical tone waveform. Usually, line spectra are found by FFT (Fast Fourier Transformation) and are evaluated as the frequency characteristics. However, it is difficult to grasp the entire characteristics since too much detailed information is contained.
When the line spectra found by the FFT are smoothed to obtain formants, and when the formants are evaluated as the frequency characteristics of the musical tone waveform, it is easier to grasp the entire characteristics, and, e.g., treatment of the waveform becomes easier.
As a method for obtaining formants, it has been proposed to find formants by performing cepstral analysis.
The cepstrum is one that is obtained by performing FFT on an input signal, taking logarithms of the amplitude spectra of the transformed input signal and then performing Inverse FFT on the logarithms. The dimension is called quefrency. The quefrency has the same dimension as time. The fine structure of spectra appears at a higher quefrency, and the spectral envelope (formants) appears at a lower quefrency.
The cepstral analysis means that only parts having a lower quefrency at the dimension of quefrency are extracted (Hereinbelow, the maximum quefrency on extraction will be called the coefficient of the cepstral analysis.), and that FFT is performed on the extracted parts to extract formants of an input signal.
FIG. 12
shows a case wherein the coefficient of the cepstral analysis is 80, and
FIG. 13
shows a case wherein the coefficient of the cepstral analysis is 40.
However, even after the cepstral analysis, minute fluctuations due to harmonic components remain, and the positions and the levels of peaks have changed in comparison with the original data as in the case shown in FIG.
12
. When the coefficient is decreased to reduce fluctuations due to harmonic components, wide fluctuations are also lost, failing to show the characteristics of the original data, as shown in FIG.
13
.
In the cepstral analysis, peaks are lowered under the influence of valleys between line spectra, and not only minute fluctuations at harmonic component levels but also wide fluctuations in the entirety are lost.
In the case of a normal musical tone, there is occurred a case wherein although the fundamental tone level is the greatest and harmonic component levels become smaller as the frequency increase, levels in the vicinity of the fundamental tone level (in particular, frequency components not higher than the fundamental tone) become smaller under the influence of valleys that are not higher than the fundamental tone.
SUMMARY OF THE INVENTION
The present invention is provided in consideration of these problems and proposes a method for extracting formants that reflect the entire characteristics of the waveform data of an original musical tone with fidelity, a recording medium with a program capable of performing the extracting method saved thereto, and an apparatus for extracting formants of a musical stone.
From the viewpoints, the method for extracting formants of a musical tone according to a first aspect of the present invention is basically characterized in that the method carries out the steps of:
finding power line spectra of a waveform to be processed;
performing level interpolation control on the power line spectra at every unit of a certain frequency, which is up to and including half a sampling frequency;
performing Fast Fourier Transformation or Inverse Fast Fourier Transformation on ones obtained by connecting peaks of harmonic components by the level interpolation control;
performing level setting with a specified coefficient to smooth a spectral envelope to be obtained as formants later on; and
obtaining the spectral envelope by performing Inverse Fast Fourier Transformation or Fast Fourier Transformation on values that are found by performing the level setting with the specified coefficient.
The arrangement according to the first aspect can be free from minute fluctuations due to harmonic components, can prevent peaks from lowering by eliminating valleys between line spectra with the level interpolation control before the cepstral analysis, and can prevent levels in the vicinity of the fundamental tone (in particular, frequency components not higher than the fundamental tone) from becoming smaller under the influence of valleys that are not higher than the fundamental tone, thereby obtaining formants that represent the characteristics of the original data in terms of all respects, such as the positions and the levels of the peaks.
The reason why the level interpolation control is limited to the frequencies that are up to and including half a sampling frequency is that the frequency equal to half a sampling frequency is an upper limit according to the sampling theorem. The certain frequency for the level interpolation control may be arbitrarily set as long as the cycle is not beyond the upper limit.
The arrangement according to a second aspect of the present invention is directed to one of the ways of carrying out the level interpolation control in the level interpolation control step, which specifically comprises the steps of finding a frequency F
1
and a level L
1
of a spectrum having a maximum level before and after a fundamental tone of the waveform; bringing all levels of the power line spectra at 0 up to the frequency F
1
into L
1
; repeating processing wherein, at every frequency that is an integral multiply of a frequency of the fundamental tone and is up to and including half the sampling frequency, a frequency Fn and a level Ln of a spectrum having a maximum level are found before and after the respective integral multiple frequencies, and the levels from the frequency having subjected to the level control at the previous stage to the frequency Fn are controlled to have values interpolated from a level Ln−1 to the level Ln; finding a frequency FN and a level LN of a last harmonic component; and bringing all levels from the frequency FN up to the frequency of the last harmonic component set for the level interpolation control to LN, thereby performing level interpolation control to connect peaks of harmonic components with the result that valleys not higher than the fundamental tone or valleys between line spectra are eliminated.
The arrangement according to a third aspect of the present invention is directed to another example of the ways of carrying out the level interpolation control in the level interpolation control step, which specifically comprises the steps of finding a frequency F
1
and a level L
1
of a spectrum having a maximum level before and after a fundamental tone of the waveform; bringing all levels of the power line spectra at 0 up to the frequency F
1
into L
1
; repeating processing wherein a frequency Fn and a level Ln of a spectrum having a maximum level are found before and after respective frequencies that are obtained by adding the frequency of the fundamental tone to a fr
Donels Jeffrey
Kabushiki Kaisha Kawai Gakki Seisakusho
Oblon & Spivak, McClelland, Maier & Neustadt P.C.
LandOfFree
Method for extracting the formant of a musical tone,... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method for extracting the formant of a musical tone,..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for extracting the formant of a musical tone,... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3183465