JPH0762800B2 - Pitch conversion method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a pitch conversion method for converting a pitch of a so-called karaoke accompaniment to a pitch higher or lower to a pitch that a singer can easily sing.

Background Art Pitch conversion is proportional movement of the frequency component of the signal,
To do this, it is necessary to expand or contract the time axis. However, the expansion and contraction of the time axis inevitably causes the tempo of accompaniment to slow down. To prevent this, the tempo can be kept constant by repeating decimation or repetition of a short-time signal that is hard to hear by the human ear from the acoustic signal at regular intervals. The corrugated connection must be very smooth. Waveform connection methods include a connection method at a zero crossing point, a weight function that smoothly changes from 1 to 0, and a connection method that multiplies each signal by a weight function that smoothly changes from 0 to 1 and adds the signals. However, in any of the methods, if the pitch phases of the signals are not aligned, the waveform envelope has a dent, and if this is repeated, noise such as tremolo occurs. There is a technique of compressing or expanding the time axis of the audio signal as an inside out of the technique of converting the pitch height. For this technology,
Nikkei Electronics July 26, 1976 issue and June 1978
It is listed in the 12th issue of the month.

The time-axis compression / decompression technology for audio signals and the pitch conversion technology are in a conflicting relationship, whereas the former speeds up or slows down the playback speed without changing the pitch, whereas the latter changes the playback speed. It does not increase or decrease the pitch frequency, but is the same technology in principle. In this prior art, it is desired that the waveforms cannot be connected by a smooth weighting function, and the place where the pitch synchronization connection is possible is improved. As another prior art, BBD (Backet
Although there is also a method using Brigade Device), this prior art uses the method of connecting at the zero crossing point of the waveform, and the phase of the pitch is also adjusted, but that process requires a great deal of processing time. Also, the waveforms can be continuously connected at the zero point, but the first-order differential waveform becomes discontinuous and high-frequency noise may occur.

OBJECT It is an object of the present invention to provide a pitch conversion method capable of smoothly performing pitch-phase-synchronized waveform connection and preventing noise due to amplitude fluctuation.

To achieve the above object, the present invention converts the pitch frequency by making the ratio of the sampling frequency of the input signal and the sampling frequency of the output signal equal to the ratio of the pitch to be converted, and the sampling frequency of the input and output. In the method of converting the pitch frequency of the signal by correcting the expansion and contraction of the time axis caused by the difference of the signal by periodically thinning out or repeating a part of the signal and returning to the original time axis, the thinning processing of the partial signal waveform / A weighting function W1 (t) (0≤t≤T) that monotonically decreases from one logical value to the other logical value as the connection processing of the signal waveform accompanying the iterative processing,
A monotonically increasing weight function W2 in which the time of the weight function is reversed.
(T) = W1 (T−t) (0 ≦ t ≦ T) is prepared, the former weight function W1 (t) is applied to the preceding signal waveform, and the latter weight function W2 (t) is applied to the subsequent signal waveform. Then, when adding both, the signal waveform is smoothed by matching the phase of the pitch waveform of the preceding signal and the phase of the pitch waveform of the subsequent signal based on the phase signal detected in real time by the peak tracking method. It is a pitch conversion method characterized by being connected.

Embodiment An embodiment of the present invention will be described below with reference to the drawings.

(A) Pitch conversion amount The pitch conversion ratio ρ is represented by the ratio between the original pitch frequency PI and the converted pitch frequency PO as shown in the first equation.

ρ = PO / PI (1) The conversion amount in the pitch of the natural scale is shown in Table 1.

Here O = 24kHz.

Next, the pitch conversion method will be described.

(B) Sampling of Input Signal The relationship between the sampling frequency I of the input signal and the sampling frequency O of the output signal is set as in the second equation in order to perform frequency conversion for the key shift.

1 (1) is a graph showing an input signal waveform at a pitch conversion rate ρ = 1.25, FIG. 1 (2) is a graph showing an output signal waveform, and FIG. 2 (1) is a spectrum of the input waveform. 2 is a graph showing the spectrum of the output waveform.

By changing the sampling frequency of the input waveform and the output waveform, the pitch of the whole song can be changed, but at the same time the time axis expands and contracts. That is, when the pitch is raised, the time is compressed and the tempo becomes faster, and when the pitch is lowered, the time is extended and the tempo becomes slower. In order to restore the expansion and contraction of this time axis, insertion or thinning processing is repeated.

(C) Repeated insertion and thinning-out process FIG. 3 is a diagram for explaining the state of the repeated insertion process, and FIG. 4 is a diagram showing a method of connecting repeated insertion parts. By contracting the time axis of the input / output waveforms shown in FIGS. 3 (1) and 3 (2) by inserting repetitive insertion portions R1, R2 ... As shown in FIG. 3 (3). It was corrected and returned to the original time axis. The connecting method of the repeated insertion portion is performed by weighting the crossing portion so that the old waveform is gradually attenuated and the new waveform is gradually raised as shown in FIG. Repeated insertion part Ri length M and input waveform part
The relation between the length L of Si + Ri and the length L is shown in the third equation.

The crossover length N must be less than L + M. That is, it is necessary that N <L + M. The repetition cycle L + M of the insertion process is shown in the fourth equation.

FIG. 5 is a diagram for explaining the state of thinning processing,
FIG. 6 is a diagram showing a method of connecting thinned-out portions. As shown in FIG. 5 (1) and FIG. 5 (2), the time-axis extension of the input / output waveform is thinned out as shown in FIG. 5 (3) by the thinned-out portions C1, C2 ,. It is corrected by and is returned to the original time axis. The connection method of the thinned-out portions Ci is, as shown in FIG. 6, connected by weighting as in the case of the repeated insertion described above. The relationship between the length M of the thinned-out portion Ci and the length L of the input waveform portion Si is shown in the fifth equation.

The length N of the intersection must be shorter than L. That is, it is necessary that N <L. The repeating period L of the thinning-out process is shown in the sixth equation.

(D) Weighting Function in Crossover Area Various functions are conceivable as the weighting function, and the representative one is the Hanning Function. As shown in FIG. 7, the weighting function is a monotonically decreasing weighting function W1 (t) (0≤t≤T), and the weighting function W2 (t) is monotonically increasing with the time of the weighting function reversed. = W1 (T-
When t) is l2, when both the weighting functions W1 (t) and W2 (t) are added, it is represented by a straight line l3. Hanning function W1
(N) is shown by the seventh equation, and W2 (n) is shown by the eighth equation.

Here, N is the number of sampling points in the intersection area.
The output signal in the crossover area is the two output signals to be connected.
Assuming y1 (n) and y2 (n), the connection waveform y0 (n) is expressed by the equation 9.

y0 (n) = y1 (n) × W1 (n) + y2 (n) × W2 (n) ...
(9) (e) Phase matching in waveform connection (1) Effect of phase matching When the waveforms are connected by a weighting function, if they are connected without matching the phase, a large amplitude groove appears as shown in FIG. When it occurs and this is repeated, a module like a tremolo is heard and becomes very annoying. This is especially true for sounds that are clearly pitched.

In order to prevent this, it is possible to prevent this by matching the phases of the pitch waveforms at the points where the two weighting functions cross each other. FIG. 9 shows a waveform when the phases are matched and connected. Waveforms can be connected very smoothly by matching the phases. Figures 10 and 11 show the effect of this phase matching with the simulation.
It can be seen that phasing allows a very smooth connection without amplitude changes.

(F) Phase Detection FIG. 12 is a graph for explaining the peak tracking method. As described above, the effect of phase matching has been clarified, but in order to realize this, it is a problem to detect the phase. Here, we propose a peak tracking method as a method to detect the phase with less hardware. This method
A peak envelope of the waveform is created, and a phase signal is output when the peak envelope is in contact with the waveform. Corresponding to the input signal shown in Fig. 12 (2), the plus side of Fig. 12 (1) is shown. And the minus side phase signal shown in FIG. 12 (2). For phase matching when connecting waveforms, it is of course necessary to perform phase matching with the plus peak for the plus peak and with the minus peak for the minus peak. As for whether the phase should be matched with the plus peak or the minus peak, it seems quite natural to match the peak with the larger absolute value of the amplitude. However, more hardware is required to match the peak with the larger absolute value, and the phase matching algorithm is also complicated.

FIG. 13 is a phase signal generation circuit diagram by the peak tracking method. This circuit consists of diodes D20, D21, resistors R21 to R24,
It is composed of capacitors C20 to C22, comparators K20 and K21, and an amplifier A20. As a method of detecting the peak of the signal waveform, a signal that tracks the peak envelope of the signal waveform by charging the signal waveform with capacitors C20, C21 through diodes D20, D21 and gradually discharging it with resistors R21, R22. Then, the peak position is detected by comparing the voltage value of this signal with the voltage value of the original waveform signal by the comparators K20 and K21. The time constant T = RC for tracking the peak seems to be several tens of ms, but it is necessary to select the optimum value through experiments.

(G) Selection of parameters (1) Selection of sampling frequency The frequency band required for karaoke was set to 100 to 10 kHz, and the sampling frequency of the output waveform was set to 24 kHz in consideration of an anti-alias filter that can be realized at low cost. The sampling frequency of the input waveform in this case is shown in Table 1.
As shown in.

(2) Selection of repeat width or thinning width and crossing width To select repeat width or thinning width and crossing width, set the values of M and N shown in FIGS. 4 and 5. There is a need. The values of M and N were set to M = 1024 and N = 512 from the viewpoint of hardware feasibility. The reason for this is that by performing the phase matching, the deterioration of the sound quality due to the removal of M and N from the optimum values becomes extremely small. The value of N is 1024-N for phase matching.
It is necessary to carry out in the section of, and N = 512.

(3) Selection of weighting function As for the weighting function, we initially considered 12 kinds of weighting functions and investigated which one was better by experiments, but we excluded the ones with particularly bad sound quality by performing phase matching. Therefore, the difference due to the weight function is almost eliminated. In the end, we chose the most common Hanning function.

(4) Time constant for peak tracking The time constant for peak tracking depends on the time change of the amplitude envelope. It seems that there is an optimum value depending on the type of music, but it is an average for karaoke. It is necessary to find the optimum value for. When 20 ms was the best when tested for 5 ms, 10 ms, and 20 ms, it was set to 20 ms.

(H) Hardware (1) Hardware configuration FIG. 14 is a circuit diagram showing a hardware data system configuration, and FIG. 15 is a circuit diagram showing a hardware control system configuration. The hardware configuration is a low pass filter,
It is composed of a buffer for converting the sampling frequency of the waveform, a waveform connection circuit, a phase detection circuit, a frequency divider for generating the sampling clock, and a microcomputer for controlling the whole. The function of each part will be described below.

The low-pass filters LPF1 to LPF6 exclude folding noise, and the cutoff frequency c1 = 6.0kHz, c2 = 6.8kHz, c3 = 7.6kHz, c based on the sampling frequency of the input waveform.
It is necessary to set 6 = 12.0kHz. Lowpass Filter LP
F1 to LPF6 correspond to each switch for on / off
S1 to S6 are provided, and signals from LPF1 to LPF6 are given to four BBD1 to BBD4 via an amplifier A1. It seems economical to realize the low-pass filters LPF1 to LP6 with active filters. As shown in FIG. 14, the low-pass filters LPF1 to LPF6 are created according to the key shift amounts, and the outputs are switched according to the key shift amounts. In addition to this, it is possible to use a low pass filter, but there is still a problem in terms of price. However, in the future, it seems likely that Suitichidokiya Pashita will also be promising.

The sampling frequency conversion circuit requires at least two BBDs in order to input and output an input signal and an output signal at two sampling frequencies. further,
It is necessary to drive at least two BBDs with the same clock in order to perform the waveform connection processing required for waveform thinning and insertion. Further, one more line is required for phase matching. Therefore, the four BBD1 to BBD4 driven by the clock signals CK1 to CK4 are switches S7 to S10 for sequentially inputting input signal waveforms to the BBD1 to BBD4 as shown in the fourteenth diagram.
BB to output the contents of each BBD1 to BBD4 repeatedly.
Switch S19 for returning the output of D1 to BBD4 to the input
˜S22 are provided, and switches S11 to S18 for deriving the outputs of BBD1 to BBD4 to the digital attenuators ATT1, 2 are provided. The output from BBD1 is amplified by the amplifier A2 and is led to the digital attenuator ATT1 via the switch S11. Further, it is led to the digital attenuator ATT2 via the switch S15. The outputs from BBD2 to BBD4 are also amplified by the corresponding amplifiers A3 to A5, and the output of BBD2 is led to ATT1 via switch S12 and to ATT2 via switch S16. The output of BBD3 is led out to ATT1 via switch S13 and via switch S17.
Derived to ATT2. The output of BBD4 is A via switch S14.
It is led to TT1 and then to ATT2 via switch S18. With such a configuration, it is possible to drive at two frequencies. A CCD may be used instead of the BBD. BBD2
It has been confirmed by simulation that the CCD length is 1024 and good results can be obtained by performing phase matching.

The waveform connecting circuit requires a multiplier for multiplying the output waveform and the weighting function in order to connect the waveforms along the attenuation curves of the weighting functions W1 (n) and W2 (n) described above. First
In FIG. 14, digital attenuators ATT1 and ATT2 are used, and the weighting function is given from the microcomputer 3 as a digital signal. As an economical realization method of the multiplier, a photo coupler is used for the resistance of the operational amplifier A6, and the signal from the computer is passed through the D / A using the resistance ladder.
This is a method of inputting this signal to a photo coupler to control the gain of the operational amplifier. The outputs of the two ATTs 1 and 2 are added by the operational amplifier A6, and the waveforms are connected.

The phase detection circuit consists of diodes D1, D as shown in Fig. 14.
2, the capacitors C1, C2, C3, the resistors R1, R2, R3, R4, and the comparators K1, K2, the operation of which is as described in the item (f) Phase detection.

The sampling clock frequency dividing circuit N is a frequency dividing circuit for making a sampling clock of an input waveform and an output waveform from a basic clock and is composed of two counters 5 and 6 and one preset register 7. The sampling frequency O of the output waveform is fixed (24 kHz), but the sampling frequency I of the input waveform must be changed to the frequency shown in Table 1 according to the key shift amount. This change is performed by sending data from the microcomputer to the preset register 7.

The control unit is composed of a one-chip micro computer 3, takes in the key shift amount from the signals KS2, KS1, KS0, and switches analog switches, BBD or CCD clocks (sampling clocks) according to the procedure described later. ) Switching, phase matching, and weighting function output. The synchronization between the microcomputer 3 and other circuits is mainly O,
This is performed by a counter interrupt by the counters 8 and 9 driven by I and an interrupt by the phase detection signal, and the program corresponding to each interrupt is executed. The control signal from the microcomputer 3 is given to four sets of gate circuits G1 to G4. The gate circuit G1 has AND gates 10 to 13, the gate circuit G2 has AND gates 13 to 15, and the gate circuit G3 has an AND gate.
The gate circuit G4 has AND gates 19 to 21.
Have. The clock signal CK1 is derived through the gate circuit G1, the clock signal CK2 is derived through the gate circuit G2, the clock signal CK3 is derived through the gate circuit G3, and the clock signal CK4 is derived through the gate circuit G4. To be done. Data signals T1 to T8 and S'1 to S'22 are sent from the terminals Q1 to Q26 to the microcomputer.

With such a configuration, the detection of the pitch phase is performed by a simple peak detection circuit including a diode, a resistor, a capacitor, and a comparator, which can be realized at a very low cost. Also, the waveform connection for pitch phase synchronization is 4
Since it is performed by switching the clock frequency and input / output of the analog delay line of BBD or CCD of the book,
Since the signal value of the waveform is processed without being quantized, the distortion due to the quantization error is small. Further, since the clock frequency of the analog delay line and the switching of the input and output can be controlled by the microcomputer, there is great flexibility.

(2) Circuit Operation FIGS. 16 and 17 are diagrams for explaining the operation of the hardware shown in FIGS. 14 and 15. Figure 16 shows I
> O, that is, the case of pitch conversion to the lower one,
The figure shows the pitch conversion to I <O, that is, the higher one. First
In Figure 16 In Figure 17 , And M = 1024 and N = 512. 16th
In Figure (1), BR1 is set as input and BR4 is set as output as initial conditions. BR1 is full in Figure 16 (2) and the input register is switched to BR2, and output is shown in Figure 16 (3). Since BR4 is empty, it is switched to BR1, and BR2 is full in FIG. 16 (4), so the input register is switched to BR3. In Fig. 16 (5), the waveform of BR1 is exhausted, and at the same time BR3 becomes full, so the input is moved to BR4 and the condition is satisfied, so BR2 and BR3
To start the thinning process. Shift BR2 and BR3 at the same time. When the peak of the waveform stored in BR3 in Fig. 16 (6) appears as LC / 2, the clock of BR3 is stopped, and BR2 of the same polarity as the peak of BR3 in Fig. 16 (7).
The peak of appears at LC / 2, and phase matching is completed, so connection processing is started. BR2, BR in Fig. 16 (8)
Since the connection process of 3 is completed, the output register is moved to BR3, and BR4 is full in Figure 16 (9), so the input register is moved to BR1 and BR3 is empty in Figure 16 (10). Since it is Natsuta, the output register is moved to BR4. Next, in Figure 16 (11), BR1 is full, so the input register is moved to BR2. In Figure 16 (12), BR2 is full, so the input register is moved to BR3. At (13), when BR4 becomes empty and the output register is moved to BR1, the condition is satisfied, so the thinning process is started. In Fig. 16 (14), the peak of BR2 has come to the position of LC / 2.
Stop the shift. In FIG. 16 (15), the peak of the same polarity of BR1 comes to the position of LC / 2 and the phases match, so connection processing is started. BR in the middle of the connection process in Fig. 16 (16).
Since 3 is full, the input register is moved to BR4 and the 16th
Since the connection processing is completed in FIG. 17 and the output register is moved to BR2, BR2 is empty in FIG. 16 and the output register is moved to BR3.

In FIG. 17, first, BR1 is set as an input buffer and BR2 is set as an output buffer register in FIG. 17 (1).
In FIG. 17 (2), BR2 is empty, so the output register is moved to BR3. In FIG. 17 (3), BR1 is full, so the input register is moved to BR2. Next, in FIG. 17 (4), BR3 is empty, so output register B
Move to R4, BR2 is full in Fig. 17 (5), so input register is moved to BR3 and BR4 is empty at the same time, output register is moved to BR1 and BR1 is shown in Fig. 17 (6). Is empty, so move the output register to BR2. Next, in FIG. 17 (7), BR3 is full, so the input register is moved to BR4, and in FIG. 17 (8), BR2 is empty, so the output register is moved to BR3. Fig. 17 (9)
In BR, the input register was moved to BR1 and the condition was satisfied when BR3 became empty at the same time, so the repetitive processing was started. In Fig. 17 (10), the peak of the waveform of BR3 is LC / Since I came to position 2, stop the shift and
In Fig. 17 (11), the peak of the same polarity in the BR4 waveform is LC / 2.
Since the phases are aligned at the position of, connection processing is performed. Next, in FIG. 17 (12), the connection processing is completed, so the output register is moved to BR3, and in FIG. 17 (13), BR3
Is empty, move the output register to BR4. Furthermore the first
In Fig. 7 (14), BR1 is full and the input register is moved to BR2. In Fig. 17 (15), BR4 is empty and the output register is moved to BR1.

Although the phase detection circuit is not shown in FIG. 16 and FIG. 17, when the output of the phase detection circuit changes from 0 to 1 when the input waveform is input to the BBD or CCD of 1024 cells,
Interrupt the microcomputer and store the peak location. The location of the peak appears between the 256th and 768th of 1024, and the location of the peak may be stored for each of PP and NP. 256th to 768th
Since the number of peaks appearing between 512 sample points is 12th even if the pitch is 4 kHz, a table of 12 × 2 × 3 × 2 = 144 bytes is sufficient for storing this. These pieces of information are retrieved and used by a table search at the time of phase matching.

The control of the circuit system switches the switches for each event that the register becomes full or empty. Each event is fully predictable, so the next time
If two or three events, such as the state of the switch to be switched, are calculated in advance, it seems that even a relatively slow microcomputer can control.

(I) Conditions for Phase-Synchronous Connection of Waveforms FIG. 18 is an output waveform diagram of the output buffer register BR. The condition for thinning processing (ρ> 1) is I = O /
ρ and I> O.

(1) Thinning process start condition When the contents of the output buffer register BR1 are exhausted and the output is switched to the next buffer register as shown in the eighteenth diagram, two registers BR2 including that register are provided. If a waveform is input to BR3, the thinning process is started.

(2) Restriction conditions The following conditions must be satisfied for the peak positions (P2, P3) for phase matching.

(A) BR2 must be emptied before the input waveform can fit into BR4 and BR1. Therefore, the tenth formula is established.

Here, since I = O / ρ, the eleventh formula is established.

Putting, the range of W1 is shown by the 12th formula.

Here, since W1> LC / 2 must be satisfied, equations 13 and 14
The formula holds.

(B) BR2, B before the input waveform can fit into BR4, BR1, BR2
R3 must be empty. Therefore, the 15th formula is materialized.

Here, since I = O / ρ and I> O, the 16th
The formula holds.

When ρ (X + 2LR) −LR> R2−P3 ≧ (16) W2 = ρ (X + 2LR) −LR, the range of W2 is expressed by Expression 17.

LR-LC> W2 ≧ (2ρ−1) LR Since W2> 0 must be satisfied here, the 18th and 19th equations are established.

(2ρ-1) LR> 0 (18) Where LR is the length of the analog register (BBD) and LC
Is the length of the crossover region, Pi is the position of the peak of the waveform stored in the i-th analog register, and Xi is i
This is the length of the free area of the th analog register.

Before starting the thinning process, it is necessary to find the peak positions that satisfy the expressions 11 and 16. If no peak position that satisfies these conditions is found, set both peak positions to LC / 2 and perform thinning processing.

The probability that a peak position satisfying the conditions of Expressions 11 and 16 can be detected increases as W1 and W2 increase, and this effect is particularly large at low pitch frequencies. To detect the peak position with high probability, W1 and W2 need about 1.5 times the pitch period. The values of W1 and W2 are shown in Table 2.

On the other hand, in the 4-buffer register configuration, There is a restriction of, and it is necessary to satisfy the condition of Formula 14.

(J) Condition in Iterative Process (ρ> 1) The condition in the iterative process is expressed by Equation 20a and Equation 20b.

I = O / ρ (20a) I <O (20b) (1) Iterative processing start condition As shown in the 19th drawing, the contents of the output buffer register BR2 are completely output, and the next buffer register BR3 is output. When the output is switched, the iterative process is started when the next buffer register BR4 is being input and the empty area X satisfies the expression (21).

(2) Restriction conditions The following conditions must be satisfied for the peak values (P2, P3) for phase matching.

(A) Through repeated processing, BR4 must satisfy the waveform before the waveforms of BR2 and BR3 are exhausted. Therefore, formula 22 is established.

Equation 23 is established from equations 20a and 22.

P3−P2> ρX−2LR (23) If ρ <2, the equation 23 is unconditionally satisfied.

(B) BR2 must be emptied before the input waveform can fit into BR4 and BR1. Therefore, formula 24 is established.

From Equation 20a, Equation 20b, and Equation 24, Equation 25 is established.

When ρ (LR + X) −LR> P3−P2 ≧ 0 (25) W3 = ρ (LR + X) −LR, the range of W3 is represented by the formula 26.

(2ρ-1) LR ≧ W3> ρLR (26) Here, since ρLR> LR-LC, Equation 25 is not a limitation.

The present invention can be used not only in a karaoke device or the like, but also in a tape recorder for compressing or expanding the time axis of conversation. That is, in order to compress the time axis, the tape feeding speed is reproduced faster than the normal speed, and the pitch frequency raised thereby can be lowered to the original frequency by the pitch conversion method of the present invention. Good.
In order to extend the time axis, the tape feeding speed may be reproduced slower than the normal speed, and the pitch frequency lowered thereby may be raised to the original pitch frequency by the pitch conversion method.

Also, the pitch conversion method according to the present invention can be used as a harmonizer to enjoy the chorus by one person.

Effects As described above, according to the present invention, the waveform connection of the pitch phase synchronization can be smoothly performed. Therefore, for example, in karaoke accompaniment, when converting the pitch to a pitch that is easy for a singer to sing, a tremolo-like amplitude is obtained. Noise due to fluctuations can be prevented. In addition, the harmony can be generated by performing the pitch conversion on the singing voice of the singer and mixing the pitch-converted singing voice and the original singing voice at a certain ratio. Particularly in the present invention, when connecting the preceding signal waveform multiplied by the monotonically decreasing weight function and the subsequent signal waveform multiplied by the monotonically increasing weight function, the preceding signal based on the phase signal detected in real time by the peak tracking method. Since the phase of the pitch waveform of and the phase of the pitch waveform of the subsequent signal are matched, noise does not occur because the pitch cycle does not fluctuate rapidly and the envelope of the composite waveform does not fluctuate in amplitude. There is a remarkable effect that it does not occur.

[Brief description of drawings]

FIG. 1 is a diagram showing the relationship between the sampling of the input waveform and the output waveform at the pitch conversion rate ρ = 1.25, FIG. 2 (1) is a diagram showing the spectrum of the input waveform, and FIG. 2 (2) is the spectrum of the output waveform. FIG. 3, FIG. 3 is a diagram for explaining repetitive insertion processing, FIG. 4 is a diagram for showing a connecting method of repetitive insertion parts, FIG. 5 is a diagram for explaining decimation processing, and FIG. 6 is decimation. FIG. 7 is a diagram showing a connection method of parts, FIG. 7 is a diagram showing a Hanning function, FIG. 8 is a diagram showing a state in which waveforms are connected without phase matching, and FIG. 9 is a state in which waveforms are connected with phase matching. Figure, Figure 10 is a connection waveform diagram without phase matching,
FIG. 11 is a connection waveform diagram with phase matching, FIG. 12 is a graph for explaining the peak tracking method, FIG. 13 is a circuit diagram of phase signal generation by the peak tracking method, and FIG. 14 is according to the present invention. FIG. 15 is a circuit diagram showing the configuration of a pitch conversion type hardware data system, FIG. 15 is a circuit diagram showing the configuration of a pitch conversion type hardware control system according to the present invention, and FIGS. 16 and 17 are FIG. FIG. 18 is a diagram for explaining the operation of the hardware shown in FIG. 15, FIG. 18 is an output waveform diagram of the output buffer register BR in the decimation process, and FIG. 19 is an output of the output buffer register BR in the repeated insertion process. It is a waveform diagram. ρ: pitch conversion rate, PI: original pitch frequency, PO: converted frequency, I: input signal sampling frequency, O: output signal sampling frequency, Ri: repeated insertion part, Ci ... … Decimated part, W1 (n), W2 (n)
...... Weight function

Claims (1)

[Claims] 1. A pitch frequency is converted by making a ratio of a sampling frequency of an input signal and a sampling frequency of an output signal equal to a ratio of a pitch to be converted, and expansion and contraction of a time axis caused by a difference in sampling frequency of input and output,
In the method of converting the pitch frequency of a signal by correcting it by periodically thinning out or repeating a part of the signal and returning to the original time axis, connection processing of signal waveforms involved in thinning / repeating processing of partial signal waveforms As a weighting function that monotonically decreases from one logical value to the other
W1 (t) (0 ≦ t ≦ T) and a monotonically increasing weight function W2 (t) = W1 (T−t) (0 ≦
t ≦ T), multiply the preceding signal waveform by the former weighting function W1 (t), multiply the latter by a weighting function W2 (t) on the subsequent signal waveform, and add them by the peak tracking method. A pitch conversion system characterized by smoothly connecting signal waveforms by matching the phase of the pitch waveform of the preceding signal and the phase of the pitch waveform of the subsequent signal based on the phase signal detected in time.