JPH1020886A - System for detecting harmonic waveform component existing in waveform data - Google Patents

Abstract

PROBLEM TO BE SOLVED: To precisely detect a pitch period of a harmonic waveform by using short waveform data and to make possible specifying the harmonic waveform by obtaining the amplitude of a sine waveform component and a cosine waveform component related to plural values from the waveform data and detecting the period which gives a maximum for the sum of their squares as the pitch period. SOLUTION: The values obtained by multiplying respectively the waveform data by a sine waveform and a cosine waveform of a period T/n and adding them ranging over a sectional length rT are used as the waveform data, and the amplitude of the sine waveform component and cosine waveform component are determined for plural values of an integer n. Then, the period that the sum of the squares of the amplitude becomes a maximum when the period T is made a variable is detected as the pitch period. Here n, r are arbitrary integers. Then, by transmitting a waveform parameter (the pitch period and amplitude information of respective harmonics) obtained from an output terminal (b) and a residual waveform (low bit waveform information) obtained from an output terminal (c), reduction (band compression) of a bit rate in transmission of a voice signal becomes possible.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION The present invention relates to a method for detecting a harmonic waveform contained in a voice signal, a music signal, or the like, and can be used for detecting a pitch period and compressing a band of an audio signal.

[0002]

2. Description of the Related Art It is well known that when a fluctuation of a pitch period is small and waveform data sufficiently longer than the pitch period is given, the pitch period can be detected from Fourier analysis (DFT, FFT) or measurement of an autocorrelation function. It is a matter. However, when the pitch period fluctuates greatly and it is necessary to detect the period using waveform data of the same length as the pitch period, the conventional method has a large error and is not practical.

[0003]

An object of the present invention is to detect a pitch period of a harmonic waveform with high accuracy using short waveform data. Specifying a harmonic waveform using the detected pitch period.

[0004]

In order to solve the above problems, the present invention multiplies waveform data by a sine waveform and a cosine waveform having a period T / n, where n and r are arbitrary integers and T is a period. Using at least a value obtained by adding over the section length rT, the amplitudes of the sine waveform component and the cosine waveform component are converted from the waveform data to the integer n.
, And a cycle in which the sum of the squares of the amplitude becomes a maximum when the cycle T is a variable is detected as a pitch cycle, or the pitch cycle and 1 / n of the pitch cycle are detected. A means for detecting a harmonic waveform component present in the waveform data, wherein a harmonic waveform included in the waveform data is specified using the amplitudes of the sine waveform component and the cosine waveform component.

[0005]

The operation of the present invention will be described below using mathematical expressions.
When the waveform data is represented by W (m) (m = 1, 2,..., M), an arbitrary integer is represented by n (> 0), and the cycle is represented by T, the waveform data has a sine waveform and a cosine waveform having a cycle T / n. The value X multiplied by each and added over the section length rT (≦ M)
(T / n) and Y (T / n)

(Equation 1) It is expressed as Using this, the amplitudes A (T / n) and B (T / n) of the sine and cosine waveform components are respectively

(Equation 2) Given by Next, a plurality of values of n, for example, n = 1, 2, 2,
.., K, and the sum of the squares of the amplitudes is defined as P (T).

(Equation 3) Becomes T when P (T) is maximized using T as a variable
Is T ₁ ,

(Equation 4) It is expressed as Pitch period of the harmonic wave included the T ₁ to the waveform data _{W (m), A (T} 1 / n) of the sine wave component of the n harmonic amplitude, B _(T 1 / n) of the n-th harmonic Assuming that the amplitude of the cosine waveform component is the harmonic waveform H (m,
T ₁ )

(Equation 5) Is represented by

Here, W (m) is generally expressed by the following equation.

(Equation 6) The operation of the present invention will be described for the case represented by In the above (Equation 1) to (Equation 3), T = T ₁
+ D (d << T ₁ ), the amplitudes of the sine and cosine waveform components are respectively

(Equation 7) Becomes Therefore, P (T) at d = 0 because the maximum, T = _{T 1} is obtained as the pitch period. At this time, the amplitudes of the sine and cosine waveform components of the k-th harmonic are respectively

(Equation 8) And the detected harmonic waveform H (m, T ₁ ) is

(Equation 9) Becomes

[0007]

FIG. 1 (a) shows a harmonic waveform W composed of five sine waves having a basic period (pitch period) of T ₁ = 120.
(M) is shown, and (d) is its spectrum. FIG. 1B shows a waveform in a case where a harmonic waveform component is detected by the present method using the section from m = 1 to 512 of W (m), the waveform is synthesized, and extrapolated (m> 512). (E) is the spectrum. FIG. 1 (c)
Similarly, a section of W (m) from m = 1 to 512 is analyzed by FFT, and shows a waveform subjected to IFFT (inverse Fourier transform), and (f) shows its spectrum. From these results, it can be seen that FFT does not detect the sine wave that actually constitutes the harmonic waveform. On the other hand, it is shown that the method according to the present invention can detect a sine wave with high accuracy.

In the method of detecting a harmonic waveform component of the present invention used in FIG. 1, the sum (summation) range given by (Equation 1) is set to m = 513-4T to 512, and T is set to 10 128. As described above, an arbitrary section of the waveform data can be used for analysis, so that the amplitude A (T /
For n) and B (T / n), an average value of values obtained from a plurality of sections may be used.

FIG. 2 is a block diagram of an embodiment in which the present invention is used for band compression of an audio signal. In FIG. 2, 1 is a storage unit for temporarily storing one frame of input waveform data, 2 is a waveform data analysis unit for detecting a harmonic waveform component according to the present invention, 3 is a pitch period and amplitude of a harmonic component. Represents a waveform subtraction unit that removes a harmonic waveform component from input data and outputs a residual waveform, a represents a waveform data input terminal, b represents a waveform parameter output terminal, and c represents a waveform parameter output terminal. Indicates a residual waveform output terminal.

If the input waveform data is an audio waveform quantized with an 8-bit amplitude and sampled at a sampling frequency of 8 kHz, it takes 64000 bits / sec to transmit (or record) this waveform. On the other hand, the harmonic waveform component (vowel component) of the speech waveform is converted into 16 ms (1
If the pitch period information is represented by 8 bits and the amplitude information of the sine waveform component and the cosine waveform component up to the eighth harmonic is represented by 6 bits for every 28 data), the transmission of the harmonic waveform can be performed at 6500 bits / sec. . Since the harmonic waveform component is the main component in the audio waveform, the residual waveform from which the harmonic waveform component has been removed can be transmitted at a low bit rate. Therefore, waveform parameters (pitch period and amplitude information of each harmonic) obtained from the output terminal b in FIG.
And transmitting the residual waveform (low bit waveform information) obtained from the output terminal c, it is possible to reduce the bit rate (band compression) in the transmission of the audio signal.

In the waveform synthesizing unit 3 shown in FIG. 2, the pitch period obtained by the analysis is T1, and the amplitudes of the sine waveform component and the cosine waveform component of the n-th harmonic are A (T ₁ /
n) and B (T ₁ / n), a composite waveform H (m, T ₁ ) is obtained by the operation of (Equation 5). Further, in the waveform subtracting section 4 in FIG. 2, when the input waveform data is W (m), the residual waveform R (m, T ₁ ) is obtained by the following equation.

(Equation 10)

FIG. 3 is a block diagram of an embodiment in which the present invention is used for band compression of a music signal. In this embodiment, a plurality of harmonic waveform components are generally detected from waveform data by repeating the operation of detecting the harmonic waveform components 1 to 4 in FIG. 2, and the harmonic waveform components are again detected from the residual waveform. Is detected. That is, the comparison control unit 5 in FIG. 3 stores the input residual waveform in the storage unit 1 and detects and removes the harmonic waveform component again, and compares the residual waveform with the previous residual waveform data. If the change is small (when the power ratio of the residual waveform is close to 1), it is determined that all the harmonic waveform components have been detected, the previous residual waveform is output to the output terminal c, and the 1 storage unit is stored. Issue a command to input new waveform data. Since a music signal generally contains sounds of a plurality of pitches, band compression (low bit rate transmission) becomes possible by detecting a plurality of harmonic waveform components and transmitting the waveform parameters and the residual waveform.

[0013]

According to the present invention, the pitch period of the harmonic waveform included in the waveform data can be detected using the short waveform data, and the harmonic waveform component can be specified. It can be used for period detection and detection of harmonic waveform components. By expressing a harmonic waveform by waveform parameters, there is an effect that a low bit rate transmission (band compression) of an audio signal or a music signal becomes possible.

[Brief description of the drawings]

FIG. 1 is a diagram showing an analysis result of a harmonic waveform according to the present invention. (A) is a harmonic waveform W (m), (b) is an analytically synthesized waveform of W (m) (m ≦ 512) and extrapolated waveform (m> 512),
(C) is a waveform obtained by performing an inverse Fourier transform (IFFT) on the analysis result of W (m) by FFT, and (d), (e), and (f) are spectra of waveforms (a), (b), and (c), respectively. It is shown.

FIG. 2 is a block diagram of an embodiment in which the present invention is used for band compression of an audio signal.

FIG. 3 is a block diagram of an embodiment in which the present invention is used for band compression of a music signal.

[Explanation of symbols]

 Reference Signs List 1 storage unit 2 waveform analysis unit 3 waveform synthesis unit 4 waveform subtraction unit 5 comparison control unit a waveform data input terminal b waveform parameter output terminal c residual waveform output terminal

Claims (1)

[Claims] 1. Using at least values obtained by multiplying waveform data by a sine waveform and a cosine waveform having a period T / n and adding them over a section length rT, where n and r are arbitrary integers and T is a period, From the waveform data, the amplitudes of the sine waveform component and the cosine waveform component are converted to the integer n
, And a cycle in which the sum of the squares of the amplitude becomes a maximum when the cycle T is a variable is detected as a pitch cycle, or the pitch cycle and 1 / n of the pitch cycle are detected. A method for detecting a harmonic waveform component present in waveform data, wherein a harmonic waveform included in the waveform data is specified using amplitudes of a sine waveform component and a cosine waveform component.