JP3707121B2 - Pitch detection device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To accurately find a pitch cycle fast with inexpensive constitution even when a speech waveform is a complicated waveform containing an overtone component. SOLUTION: A digital speech signal is oversampled by four-times oversampling 7 and the result is binarized by a binarization part 8; and inversion intervals of the signal obtained by the binarization are measured by a timer 9 to find successive zero-crossing intervals of the digital speech signal, and the intervals are stored in a RAM 10. A pitch arithmetic part 11 assumes that the pitch cycle is the sum of 2n pieces of zero-crossing interval data as to n=1-4, calculates a reproduction rate as the degree of matching in each pitch cycle of each zero-crossing interval data constituting one pitch cycle, and employs the assumption by which the highest reproduction rate is obtained to find the pitch cycle.

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a pitch detection device that detects a pitch period or a pitch frequency of a speech waveform.
[0002]
[Prior art]
One of the parameters characterizing the speech waveform is a pitch period (or pitch frequency), and a technique for detecting the pitch period of the speech waveform is generally used in speech analysis / synthesis systems, speech coding systems, and the like. Recently, some karaoke systems detect the pitch period of a singer's voice and are used for singing.
[0003]
Conventionally, there are the following methods for detecting the pitch period of speech.
(1) Zero cross method
Assuming that the speech waveform is very close to a sine wave, the speech waveform crosses the zero level line from the negative direction to the positive direction, then crosses from the positive direction to the negative direction, and then crosses from the negative direction to the positive direction again. Is repeated, a pitch period is given by a time interval crossing the zero level line in the same direction. In accordance with this idea, the zero cross method is a method in which two pitch intervals are simply measured to obtain a pitch period. Further, as a similar idea, there is a method of measuring a time interval at which an instantaneous value of a speech waveform becomes a maximum value or a minimum value to obtain a pitch period.
[0004]
(2) Autocorrelation method
In this autocorrelation method, the following autocorrelation function R (r) is calculated using time series samples x (1), x (2),... Obtained by sampling a speech waveform at a constant sampling period. To obtain the pitch period.
R (r) = 1 / N · Σ {x (n) · x (n + r)}
(In the above formula, Σ is an operator for calculating the sum in {} in the range of n = 1 to N · r.)
That is, r is changed variously, an autocorrelation function R (r) is obtained for each r, and the pitch period of the speech waveform is calculated from r when R (r) becomes maximum (that is, autocorrelation is maximum).
[0005]
[Problems to be solved by the invention]
By the way, the above-described zero cross method can detect the pitch period relatively inexpensively and at the same time, but human speech contains many overtone components and cannot detect an accurate pitch period. There was a problem. Moreover, although the above-described autocorrelation method can detect the pitch period to some degree of accuracy, the calculation amount is enormous and the detection time is long. In addition, the cost increases.
[0006]
The present invention overcomes the above problems, basically adopts a method for determining the pitch period by measuring the zero cross interval, and takes measures to prevent the harmful effects caused by adopting such a method, An object of the present invention is to provide a pitch detection device capable of detecting a pitch cycle accurately and at high speed with an inexpensive configuration.
[0007]
[Means for Solving the Problems]
  The present invention relates to an oversampling means for outputting a digital audio signal with a sampling frequency multiplied by a predetermined number, and a binary for comparing the digital audio signal output by the oversampling means with a predetermined level and converting it to a binary signal. And a continuous zero cross interval t of the digital audio signal based on the binary signal₁, T₂,..., And various changes in n (n is an integer of 1 or more), and for each n, the total of 2n zero cross intervals T = (t₁+ T₂＋ ・ ・ t_2n) Is assumed to be the pitch period,next toFor m periods (m is an integer of 2 or more) in contactAssumed abovePitch periodThe degree of coincidence between the corresponding zero-crossing intervals included in eachThe degree of coincidence of the waveform of the digital audio signalAsThe gist of the present invention is a pitch detection device comprising pitch calculation means that calculates and selects a pitch period by selecting n having the highest degree of waveform matching.
[0008]
DETAILED DESCRIPTION OF THE INVENTION
Embodiments will be described below for easier understanding of the present invention.
Such an embodiment shows one aspect of the present invention, and is not intended to limit the present invention, and can be arbitrarily changed within the scope of the present invention.
[0009]
A. Configuration of the embodiment
FIG. 1 is a block diagram showing a configuration of an embodiment in which the present invention is applied to a karaoke system. The present embodiment relates to a part for scoring the song of the singer among the constituent parts of the karaoke system. In FIG. 1, reference numeral 1 denotes a CD (compact disc) on which a digital music signal is recorded. The digital music signal recorded on the CD 1 is sequentially reproduced in synchronization with a clock having a sampling frequency fs = 44.1 kHz. A vocal extraction unit 2 extracts a signal corresponding to a vocal sound (hereinafter referred to as a digital model signal) from a digital music signal reproduced from the CD 1. As an example, a digital model signal can be obtained by a process of extracting a signal in a frequency band including a voice band of a digital music signal reproduced from the CD 1 using a band pass filter. In addition, when a medium on which only vocal sounds are recorded can be used, a digital music signal reproduced from the medium may be used as a digital model signal as it is. A microphone 3 collects the singing voice of a singer who sings along with the reproduction of the CD 1 and outputs it as an analog audio signal. Reference numeral 4 denotes an A / D converter, which samples an analog audio signal from the microphone 1 in synchronism with a clock having a sampling frequency fs = 44.1 kHz as in the case of CD1, and converts it into a digital audio signal.
[0010]
Reference numeral 5 denotes a DC removal unit, which performs a DC removal process on sequentially supplied digital audio signals and digital example signals, and removes a low frequency band that can be regarded as DC, for example, a digital audio signal from which components in the 0 Hz to 50 Hz band are removed. And a digital model signal are output. Reference numeral 6 denotes an LPF (low-pass filter), for example 500 from each of the digital audio signal and the digital model signal output by the DC removing unit 5.HA component having a frequency equal to or higher than z is removed and output. These DC removal unit 5 and LPF 6 select and output only the components within the 50 to 500 Hz band for each of the digital audio signal and the digital model signal.
[0011]
7 is a 4-times oversampling unit, which performs an interpolation operation on the digital audio signal and digital sample signal (both sampling frequency fs = 44.1 kHz) that have passed through the LPF 6 and converts them to a signal having a 4-times sampling frequency. Output.
[0012]
FIG. 2 exemplifies a circuit configuration necessary for processing one of the digital audio signal and the digital model signal (hereinafter referred to as an input digital signal) in the 4 × oversampling unit 7. In this figure, a latch 71 receives and holds an input digital signal when a clock corresponding to the sampling frequency fs is given. The delay devices 72, 72,... Are cascade-connected to the subsequent stage of the latch 71 as shown. These delay units 72 are each provided with a clock having a frequency four times the sampling frequency fs, thereby sequentially shifting the input signal held in the latch 71 and delaying the input signal sequentially by one clock period. Each signal is output. .. Are multipliers, and 74, 74,... Are adders, which perform an interpolation operation that convolves a predetermined interpolation coefficient sequence with the output signals of the latch 71 and delay units 72, 72,. . With the above configuration, interpolation calculation is executed in synchronization with a clock having a frequency four times the sampling frequency fs, and the interpolated digital signal is sequentially output from the adder 74 at the final stage.
[0013]
The quadruple oversampling unit 7 is a means provided in order to increase the accuracy in obtaining the pitch period. That is, in the present embodiment, the pitch period of each digital signal is obtained by measuring the time interval between the zero cross points of the digital audio signal and the digital model signal. For this reason, in order to increase the measurement accuracy of the pitch period, it is necessary to increase the detection accuracy of the position of the zero cross point on the time axis. Therefore, by interposing the 4-times oversampling unit 7, the time density of each sample of the digital audio signal and the digital model signal is quadrupled, and the detection accuracy of the position of each zero cross point is increased. In this example, oversampling is performed by curve interpolation. However, in view of the problem of cost, linear interpolation that can obtain a certain degree of accuracy can also be used.
[0014]
A binarization unit 8 binarizes the levels of the digital audio signal and digital model signal output from the 4 × oversampling unit 7. This binarization basically determines whether the input digital signal is positive or negative with reference to the zero level, and outputs “1” if the input digital signal is positive and “0” if it is negative. is there. That is, the binarizing unit 8 is a means for outputting a binary signal in which “0” / “1” is inverted every time the input digital signal crosses the zero level. However, in this embodiment, when binarization is performed, the range of ± Δ centering on the zero level is used as a masking band, and even if the input digital signal has minute vibrations within the masking band of ± Δ The binary signal is not inverted by minute vibrations.
[0015]
FIG. 3 exemplifies a circuit configuration necessary for processing one of the digital audio signal and the digital model signal (hereinafter referred to as input digital signal) in the binarization unit 8. In this figure, reference numeral 81 denotes an absolute value detector for detecting the absolute value of the input digital signal. Reference numeral 82 denotes a comparison unit, which compares the absolute value of the input digital signal detected by the absolute value detection unit 81 with a predetermined value Δ. When the absolute value exceeds Δ, “1” is not exceeded. "0" is output to. Reference numeral 83 denotes a sample hold unit, which outputs an input digital signal as it is during a period in which “1” is output from the comparison unit 82 (sample state), and in a period in which “0” is output from the comparison unit 82 The input digital signal immediately before the output signal 82 changes from “1” to “0” is held and output (hold state). Reference numeral 84 denotes a comparison unit that determines whether the output signal of the sample hold unit 83 is positive or negative with reference to a zero level, and outputs a binary signal of “1” if positive and “0” if negative.
[0016]
According to the above configuration, when the input digital signal is outside the range of ± Δ, it is output as it is via the sample hold unit 83. When the input digital signal falls within the zero level ± Δ masking band, the value of the input digital signal immediately before is held by the sample hold unit 83, and the comparison is performed during this holding operation. The binary signal output from the unit 84 is not inverted. Therefore, when the input digital signal changes across the masking band of zero level ± Δ, the binary signal is inverted when the masking band has been crossed. On the other hand, if the input digital signal enters the masking band of zero level ± Δ but moves up and down in the masking band without crossing it, the sample hold unit even if the input digital signal crosses the zero level Since the output signal value 83 does not cross the zero level, inversion of the binary signal does not occur.
[0017]
3 may be replaced with the circuit shown in FIG. In FIG. 4, reference numerals 85 and 86 denote comparison units, which respectively compare the input digital signal with a reference level, and output “1” when the input digital signal is higher than the reference level, and output “0” when lower than the reference level. To do. The comparison unit 85 is given + Δ as the reference level, and the comparison unit 86 is given -Δ as the reference level. Reference numeral 87 denotes a latch that holds an input digital signal, and reference numeral 88 denotes a selector that selects and outputs the input digital signal or the output signal of the latch 87. A control unit 89 controls the latch 87 and the selector 88 based on the output signals of the comparison units 85 and 86. That is, it is as follows.
[0018]
a. When the output signals of the comparators 85 and 86 are both “1” or both are “0”
This is the case when the input digital signal is outside the zero level ± Δ masking band. In this case, the control unit 89 sets the latch 87 to the sample state and causes the selector 88 to output an input digital signal.
b. When the output signal of the comparison unit 85 is “0” and the output signal of the comparison unit 86 is “1”
This is the case when the input digital signal is inside the masking band of zero level ± Δ. In this case, the control unit 89 puts the latch 87 in the hold state when the input digital signal enters the masking band, and causes the selector 88 to output the output signal of the latch 87.
[0019]
In FIG. 1, reference numeral 9 denotes a time interval at which each output signal of the binarization unit 8 corresponding to a digital audio signal and a digital model signal occurs, that is, a time interval at which a zero cross point of each digital signal occurs. 10 is a RAM for storing the time measurement result of the timer 9.
[0020]
FIG. 5 is a block diagram showing the timer 9 and the RAM 10 together with their control systems. This figure shows only a portion necessary for processing corresponding to one of the digital audio signal and the digital model signal. In FIG. 5, 91 is a delay device, and 92 is an exclusive OR circuit. These constitute a differentiating circuit 90 for differentiating the binary signal output from the binarizing unit 8 and outputs a pulse every time the inversion of the binary signal occurs. The timer 9 is reset every time an output pulse from the differentiating circuit 90 is given, and after this reset, the clock with a constant frequency of 4 fs is counted until the next reset.
[0021]
The count value of the timer 9 is given as input data to the latch 93. The latch 93 receives and holds the count value of the timer 9 immediately before the reset, when the output pulse from the differentiation circuit 90 is given. The count value held in the latch 93 is the number of clocks having a frequency of 4 fs output from when the previous inversion of the binary signal is detected until the current inversion is detected. It can be said that this represents a time interval in which the occurrence of. Therefore, hereinafter, the data held in the latch 93 is referred to as zero cross interval data.
[0022]
The write controller 94 sequentially reads the zero cross interval data in the latch 93 each time an output pulse from the differentiation circuit 90 is given, and the zero cross interval data within a certain range is equal to or greater than a predetermined value (the count value of the timer 9). Is set and written to the RAM 10, and when it is less than a predetermined value (the count value of the timer 9 is small), a limit is set and the RAM 10 is not written and discarded. The reason why only the zero cross interval data within a certain range is written in the RAM 10 is that the zero cross interval data that is not valid as the time interval of the zero cross point of the audio signal is used for the calculation, and an incorrect pitch period is set. This is to prevent the calculation.
[0023]
The pitch calculation unit 11 in FIG. 1 calculates the pitch period of each of the digital audio signal and the digital model signal by referring to the zero cross interval data stored in the RAM 10.
[0024]
Here, if the digital audio signal or the like is a sine wave, it crosses the zero level line at the start point and end point of one cycle of the sine wave, and crosses the zero level line only once in the middle of these zero cross points. To do. Therefore, the pitch period can be obtained by adding two consecutive zero cross interval data.
[0025]
However, since a digital audio signal or the like representing a human audio waveform contains many overtone components, the waveform for one pitch period has three or more zero cross points between the start and end points of the pitch period. In such a case, even if two consecutive zero cross interval data are added, a correct pitch period cannot be obtained.
[0026]
Therefore, in the present embodiment, it is assumed that for each of a plurality of types of integers n, one pitch period has a length corresponding to the sum of 2n zero cross interval data. Then, the pitch period is obtained under each assumption, and how much the occurrence timing of each zero cross point within one pitch period is matched between the pitch periods. Details of detection of the degree of coincidence of the occurrence timing of the zero cross point will be described later. Then, the pitch period having the highest degree of coincidence is selected as the true pitch period. This presupposes the property of the audio signal that a large waveform change does not occur within a short time.
[0027]
Next, in FIG. 1, reference numeral 12 denotes a level detection unit that detects the levels of the digital audio signal output by the A / D converter 4 and the digital model signal output by the vocal extraction unit 2. A signal representing is output.
[0028]
Reference numeral 13 denotes a scoring unit, which comprehensively evaluates the pitch period deviations of the digital audio signal and the digital model signal obtained by the pitch calculation unit 11 and the deviations of both signal levels obtained by the level detection unit 12 to sing A person's song. This scoring result is displayed on the display unit 14.
[0029]
B. Operation of the embodiment
The operation of this embodiment will be described below. When a song is selected by a singer, digital music signals are sequentially reproduced from the CD 1 corresponding to the song. Then, a digital model signal is extracted from the digital music signal by the vocal extraction unit 2 and is output to the DC removal unit 5 and the level detection unit 12. On the other hand, the singer starts singing by playing CD1, and the singing voice is collected by the microphone 3 and output as an analog audio signal. This analog audio signal is converted into a digital audio signal via the A / D converter 4 and output to the DC removing unit 5 and the level detecting unit 12.
[0030]
The digital audio signal and the digital model signal are sequentially passed through the DC removal unit 5 and the LPF 6 so that unnecessary frequency band signals are removed, and a digital signal representing a waveform composed only of components in the human voice frequency band and Are output to the 4-times oversampling unit 7 respectively.
[0031]
Then, the digital audio signal and the digital model signal are each interpolated on the time axis by the 4 × oversampling unit 7, converted into a signal having a 4 × sampling frequency, and output as a binary signal by the binarizing unit 8. Is converted to
[0032]
FIG. 6 illustrates the operation of the 4 × oversampling unit 7. In FIG. 6A, the horizontal straight line is a zero level line. In addition, a ◯ -plot is shown along the sinusoidal signal waveform, but the latter plot represents the individual original samples that make up the digital audio signal (digital model signal), and the former represents these original samples. It represents the original signal waveform that is the matrix of the sample. In addition, three x-marked plots are inserted between the ◯ -marked plots representing the original samples, and these represent the interpolated samples obtained by the 4-times oversampling unit 7. .
[0033]
FIG. 6B shows a binary signal obtained when the oversampling is not performed four times and only the original sample (circle mark) is given to the binarization unit 8. FIG. A binary signal obtained when double oversampling is performed and the original sample (◯ mark) and the interpolation sample (x mark) are given to the binarization unit 8 is shown. For convenience of explanation, these drawings show an example in which the digital audio signal (digital example signal) does not include a vibration having a level smaller than the masking band of the binarization unit 8.
[0034]
Here, the digital audio signal or the like is sampled at a constant sampling period regardless of the signal waveform. Therefore, when the digital audio signal or the like repeats the same waveform, as shown in FIG. 6A, the timing at which the instantaneous value is sampled varies depending on each waveform. For this reason, when the sampling period is rough, as shown in FIG. 6B, when the pitch period is switched, binary signals having different waveforms may be obtained even though the waveform is the same. However, when binarization is performed after 4 times oversampling of a digital audio signal or the like as in the present embodiment, inversion is performed at a timing close to the original zero cross point as shown in FIG. The binary signal to be obtained is obtained, and the problem as shown in FIG. 6B is prevented.
[0035]
7A to 7D illustrate the operation of the binarization unit 8. First, in FIG. 7A, a sinusoidal signal waveform represents a digital audio signal (digital example signal) output from the quadruple oversampling unit 7, and a horizontal line represents a zero level line. FIG. 7B shows the operation of the sample hold unit 83 in FIG. As shown in this figure, the sample hold unit 83 is in the sample state when the digital audio signal (digital model signal) as an input signal is outside the zero level ± Δ masking band (“S” in the figure). If it is inside the masking band of zero level ± Δ, it is in a hold state (indicated as “H” in the figure). As a result of such control of the sample and hold unit 83, the signal waveform input to the comparison unit 84 is as illustrated in FIG. 7C, and the binary signal obtained from the comparison unit 84 is illustrated in FIG. It will be illustrated as follows. As described above, when the digital audio signal (digital model signal) changes across the masking band of zero level ± Δ, the binary signal is inverted when the masking band is crossed. Even if the digital audio signal (digital example signal) includes a minute vibration part with an amplitude of ± Δ or less, the digital audio signal (digital example signal) is within the masking band of zero level ± Δ. Since the sample hold unit 83 performs the previous value holding operation, the binary signal is not inverted in the vibration portion.
[0036]
In this embodiment, since the pitch period is calculated using zero cross intervals, if too many zero cross intervals are detected for the input digital signal waveform corresponding to one pitch period, the burden of calculating the pitch period Will become bigger. However, in the present embodiment, since the binary signal is generated by the binarization unit 8 having the masking band as described above, in the input digital signal, fine movement near the zero level that is not important for the calculation of the pitch period is generated. A binary signal that is ignored and does not contain “0” / “1” inversions more than necessary is obtained, and it is possible to detect a number of zero cross intervals that are appropriate for the calculation of the pitch period.
[0037]
As described above, a binary signal is generated based on each of the digital audio signal and the digital model signal. Then, for each binary signal, the time intervals at which the “1” / “0” inversion occurs are sequentially counted by the timer 9, and the zero cross interval data as the timing results are sequentially held in the latch 93 shown in FIG. In this way, the zero cross interval data sequentially held in the latch 93 is sequentially written into the RAM 10 under the control of the write control unit 94. That is, the write control unit 94 executes a write control routine whose flow is shown in FIG. 8 in response to the pulse output from the differentiation circuit 90 by the inversion of the binary signal. First, the write control unit 94 takes in the zero cross interval data t from the latch 93 (step S1), and determines whether the zero cross interval data t is equal to or greater than the lower limit “8”. If this determination is “NO”, the routine is terminated without writing the zero-crossing interval data t. If the determination result in step S2 is “YES”, the process proceeds to step S3, and it is determined whether or not the zero crossing interval data t is larger than the upper limit value “8192”. If the determination result is “NO”, the zero cross interval data t is written to the RAM 10 (step S4), and the routine is terminated. On the other hand, if the determination result in step S3 is “YES”, “8192” is written in the RAM 10 instead of the fetched zero cross interval data t (step S5), and the routine is terminated. With the above control, only zero cross interval data within the range of “8” to “8192” is written to the RAM 10, and therefore zero cross interval data that is not valid as the time interval of the zero cross point of the audio signal is used for the calculation. It is possible to prevent an erroneous pitch period from being calculated.
[0038]
In this way, the zero cross interval data stored in the RAM 10 is referred to by the pitch calculation unit 11, and the pitch periods of the digital audio signal and the digital model signal are obtained. Here, with reference to FIG. 9, the outline of the calculation process of the pitch period of the digital audio signal will be described as an example. If a digital audio signal as illustrated in FIG. 9A is given to the binarization unit 8, zero-crossing interval data t generated up to the present time₁, T₂,... Are stored in the RAM 10. The pitch calculation unit 11 calculates the zero cross interval data t₁, T₂,... And the pitch period of the digital audio signal, the following four assumptions are made, and the pitch period is obtained according to a procedure for examining the validity of each.
[0039]
(1) Assumption 1
The pitch period of the digital audio signal is two zero cross interval data t₁, T₂Length T equivalent to the sum of₁Have That is, the time T shown in FIG.₁₁, T₁₂,... Are the pitch period of the digital audio signal.
(2) Assumption 2
The pitch period of the digital audio signal is four zero cross interval data t.₁~ T_FourLength T equivalent to the sum of₂Have That is, the time T shown in FIG._{twenty one}, T_{twenty two},... Are the pitch period of the digital audio signal.
(3) Assumption 3
The pitch period of the digital audio signal is six zero cross interval data t₁~ T₆Length T equivalent to the sum of_ThreeHave That is, the time T shown in FIG.₃₁, T₃₂,... Are the pitch period of the digital audio signal.
(4) Assumption 4
The pitch period of the digital audio signal is 8 zero cross interval data t₁~ T₈Length T equivalent to the sum of_FourHave That is, the time T shown in FIG.₄₁, T₄₂,... Are the pitch period of the digital audio signal.
[0040]
The examination of the validity of each of the above assumptions and the calculation of the pitch period based on the examination results are executed according to the flow shown in FIG. First, the pitch calculation unit 11 calculates the reproducibility CR1 of the waveform of the digital audio signal when the above assumption 1 is assumed (step S101). This reproduction rate is a numerical value indicating how much the digital audio signal waveforms corresponding to each pitch period match in accordance with the above assumptions. In this embodiment, the zero cross interval data t₁, T₂Calculate based on.
[0041]
Here, with reference to the flowchart of FIG. 11, the procedure of the calculation which calculates | requires the reproduction rate CR1 performed in step S101 is demonstrated. First, in step S201, "0" and "1" are set as initial values for the counter CNT and the control variable i, respectively.
[0042]
In step S202, the control variable i is increased by “2”, and i = “3”. Next, proceeding to step S203, 0.9t₁-T_i<0 is satisfied, that is, zero cross interval data t_ThreeIs zero cross interval data t₁It is judged whether it is larger than 90%. If the determination result is “YES”, the counter CNT is incremented by “1” (step S204), and the process proceeds to step S205. If “NO”, the process proceeds to step S205 without going through step S204. Next, in step S205, -1.1t₁+ T_i<0 is satisfied, that is, zero cross interval data t_ThreeIs zero cross interval data t₁It is judged whether it is smaller than 110%. If the determination result is “YES”, the counter CNT is incremented by “1” (step S206), and the process proceeds to step S207. If “NO”, the process proceeds to step S207 without going through step S206.
[0043]
Next, in step S207, it is determined whether or not the control variable i is “7”. If the determination result is “NO”, the process returns to step S202. Thereafter, steps S202 to S207 are executed twice, and zero cross interval data t_FiveAnd t₇Are determined in steps S203 and S205, and each zero cross interval data is converted into zero cross interval data t.₁The counter CNT is incremented when it is larger than 90% or smaller than 110% (steps S204 and S206).
[0044]
When i = “7”, the determination result in step S207 is “YES”, the process proceeds to step S208, and “2” is set to the control variable i.
[0045]
Next, in step S209, the control variable i is increased by “2”, and i = “4”. Next, proceeding to step S210, 0.9t₂-T_i<0 is satisfied, that is, zero cross interval data t_FourIs zero cross interval data t₂It is judged whether it is larger than 90%. If the determination result is “YES”, the counter CNT is incremented by “1” (step S211), and the process proceeds to step S212. If “NO”, the process proceeds to step S212 without passing through step S211. Next, in step S212, -1.1t₂+ T_i<0 is satisfied, that is, zero cross interval data t_FourIs zero cross interval data t₂It is judged whether it is smaller than 110%. If the determination result is “YES”, the counter CNT is incremented by “1” (step S213), and the process proceeds to step S214. If “NO”, the process proceeds to step S214 without going through step S213.
[0046]
Next, in step S214, it is determined whether or not the control variable i is “8”. If the determination result is “NO”, the process returns to step S209. Thereafter, steps S209 to S214 are executed twice, and zero cross interval data t₆And t₈Are determined in steps S210 and S212, and each zero cross interval data is converted into zero cross interval data t.₂The counter CNT is incremented when it is larger than 90% or smaller than 110% (steps S211, S213).
[0047]
When i = “8”, the determination result in step S214 is “YES”, the process proceeds to step S215, and the value of the counter CNT is normalized by the number of determinations regarding the zero cross interval data, and the result is reproduced. Let CR1. In this flow, since the determination is performed 12 times, CNT / 12 is set as the recall rate CR1.
[0048]
Here, the length of the pitch period is the sum T of the two zero cross interval data.₁In an ideal state where the assumption is correct and the waveform of the digital audio signal does not change even when the pitch period is switched four times, t₁= T_Three= T_Five= T₇And t₂= T_Four= T₆= T₈It becomes. Accordingly, in this case, the reproduction rate CR1 obtained by the above processing is 100%. Even if there is some error in each zero cross interval data, t_Three, T_FiveAnd t₇Is t₁Within ± 10% and t_Four, T₆And t₈Is t₂When it is within the range of ± 10%, the recall ratio CR1 is 100%. On the other hand, if the above assumption is incorrect, a large difference occurs between the zero-crossing interval data corresponding to each other by switching the pitch period. For this reason, it becomes easy to make a negative determination in the above-described step S203 and the like, and the reproduction rate CR1 decreases as the number of such negative determinations increases.
[0049]
When the calculation of the reproduction rate CR1 is completed in this way, the process returns to the flow of FIG. 10 and proceeds to step S102, where the reproduction rate CR2 of the digital audio signal waveform on the assumption of the above assumption 2 is calculated. That is, a length T corresponding to the sum of four zero-crossing interval pitch pitch data₂Is assumed to have The zero cross interval data t corresponding to the first pitch period₁~ T_FourAnd zero cross interval data t corresponding to the second, third and fourth pitch periods, respectively._Five~ T₈, T₉~ T₁₂And t₁₃~ T₁₅It is determined whether or not each of them matches the reference within a predetermined error range. Then, the number of times that a positive determination result is obtained is counted, normalized by the total number of determinations, and the recall ratio CR2 is obtained.
[0050]
In an ideal state where the assumption that the length of the pitch period is the sum of the four zero-cross interval data is correct and the waveform of the digital audio signal does not change even when the pitch period is switched four times,
t₁= T_Five= T₉= T₁₃
t₂= T₆= T_Ten= T₁₄
t_Three= T₇= T₁₁= T₁₅
t_Four= T₈= T₁₂= T₁₆
All the conditions are satisfied, and the reproducibility CR2 is 100%. In addition, even if there is some error in each zero cross interval data, the reproduction rate CR2 is 100% if it is within ± 10%. When the zero pitch interval data greatly deviates from the reference (that is, zero cross interval data corresponding to the first pitch cycle) is generated by switching the pitch cycle, the recall ratio CR2 is lowered according to the number of the zero cycle interval data. It will be.
[0051]
In step S103, the digital audio signal waveform reproduction rate CR3 when the above assumption 3 is assumed is calculated. That is, a length T corresponding to the sum of six zero-cross interval data whose pitch period is six._ThreeIs assumed to have The zero cross interval data t corresponding to the first pitch period₁~ T₆And zero cross interval data t corresponding to the second, third and fourth pitch periods, respectively.₇~ T₁₂, T₁₃~ T₁₈And t₁₉~ T_{twenty four}It is determined whether or not each of them matches the reference within a predetermined error range. Then, the number of times that a positive determination result is obtained is counted, normalized by the total number of determinations, and the recall ratio CR3 is obtained.
[0052]
This recall CR3 is
t₁= T₇= T₁₃= T₁₉
t₂= T₈= T₁₄= T₂₀
t_Three= T₉= T₁₅= T_{twenty one}
t_Four= T_Ten= T₁₆= T_{twenty two}
t_Five= T₁₁= T₁₇= T_{twenty three}
t₆= T₁₂= T₁₈= T_{twenty four}
If all the above conditions are satisfied, or if there is some error in each zero-crossing interval data, but the error is within a range of ± 10%, the recall ratio CR3 is 100%. In addition, when zero-cross interval data having a large error occurs, the recall CR3 is lowered according to the number of the data.
[0053]
In step S104, the digital audio signal waveform reproduction rate CR3 when the above assumption 4 is assumed is calculated. That is, the length T corresponding to the sum of the zero cross interval data whose pitch period is eight._FourIs assumed to have The zero cross interval data t corresponding to the first pitch period₁~ T₈And zero cross interval data t corresponding to the second and third pitch periods, respectively.₉~ T₁₆And t₁₇~ T_{twenty four}It is determined whether or not each of them matches the reference within a predetermined error range. Then, the number of times that a positive determination result is obtained is counted, normalized by the total number of determinations, and the recall ratio CR4 is obtained.
[0054]
In each of the processes in steps S101 to S103, four pitch periods are targeted for processing, but in this step S104, three pitch periods (T in FIG. 9 (b4)).₄₁~ T₄₃). This is due to the following reason. That is, in step S104, a long time corresponding to eight zero-cross interval data is assumed as the pitch period. Accordingly, assuming that four pitch periods are to be processed in step S104, even if assumption 4 is correct, the digital audio signal waveform is stable over an extremely long period of four pitch periods. If not, the reproducibility CR4 is lowered. However, the waveform of the digital audio signal can maintain the same waveform for a certain short period of time, but the waveform changes after a certain period of time. For this reason, when four pitch periods are processed, even if Assumption 4 is correct, an unreasonably low recall ratio CR4 is calculated due to the influence of temporal changes in the waveform of the digital audio signal. Probability is high. Therefore, in step S104, as described above, three pitch periods are targeted for processing.
[0055]
In step S104, the recall CR4 is
t₁= T₉= T₁₇
t₂= T_Ten= T₁₈
t_Three= T₁₁= T₁₉
t_Four= T₁₂= T₂₀
t_Five= T₁₃= T_{twenty one}
t₆= T₁₄= T_{twenty two}
t₇= T₁₅= T_{twenty three}
t₈= T₁₆= T_{twenty four}
If all the above conditions are satisfied, or if there is some error in each zero-crossing interval data, but the error is within the range of ± 10%, the recall ratio CR4 is 100%. Further, when zero-cross interval data having a large error is generated, the recall ratio CR4 is lowered according to the number of the data.
[0056]
Next, the process proceeds to step S105, and it is determined which of assumptions 1 to 4 is appropriate based on the recall rates CR1 to CR4 obtained as described above. A detailed flow of this determination is shown in FIG. First, in step S301, it is determined which of the recall rates CR1 to CR4 is the maximum. If the recall ratio CR1 is the maximum, it is determined whether or not this CR1 is greater than a predetermined reference value ref (step S302). If this determination result is “YES”, the assumption 1 is followed. That is, the length T of the zero cross interval data for two pieces₁Thus, the pitch period is obtained. The same applies to the case where the other reproduction ratios CR2 to CR4 are maximum, and it is determined whether or not CR2 or the like is larger than a predetermined reference value ref (steps S303 to S305), and the determination result is “YES” Includes the length T of the four zero-crossing interval data in accordance with the assumptions used to calculate each recall.₂, The length T of the zero cross interval data for 6 pieces_ThreeAlternatively, the length T of the zero cross interval data for 8 pieces_FourThus, the pitch period is obtained. If the recall is the same, the priority order is CR1> CR2> CR3> CR4 (CR1 is the highest priority).
[0057]
On the other hand, when the maximum one of the recall ratios CR1 to CR4 is equal to or less than the reference value ref, the determination result is “NO” regardless of which of the steps S302 to S305 is performed. In this case, a conclusion cannot be drawn as to which of assumptions 1 to 4 is valid, and the determination result is “not applicable”.
[0058]
When the above determination is completed, the process returns to the flow shown in FIG. 10 and proceeds to the step corresponding to the determination result. That is, the length T of the zero cross interval data for two pieces₁If it is determined that the pitch period is to be obtained, the process proceeds to step S106, where four pitch periods each consisting of two pieces of zero cross interval data are obtained (T in FIG. 9B1).₁₁~ T₁₄These average values are used as the pitch period of the digital audio signal. Also, the length T of the zero cross interval data for four pieces₂If it is determined that the pitch period is to be obtained, the process proceeds to step S107, and four pitch periods are obtained according to this determination result (T in FIG. 9B2)._{twenty one}~ T_{twenty four}These average values are used as the pitch period of the digital audio signal. In addition, the length T of the zero cross interval data for six pieces_ThreeIf it is determined that the pitch period is to be obtained, the process proceeds to step S108, and four pitch periods are obtained according to the determination result (T in FIG. 9B3).₃₁~ T₃₄These average values are used as the pitch period of the digital audio signal. And the length T of the zero cross interval data for 8 pieces_FourIf it is determined that the pitch period is to be obtained, the process proceeds to step S109, and three pitch periods are obtained according to the determination result (T in FIG. 9 (b4)).₄₁~ T₄₃These average values are used as the pitch period of the digital audio signal.
[0059]
When the above process ends, the process returns to step S101, and the same process is repeated. In this way, the pitch period of the digital audio signal is output continuously. On the other hand, in the determination of FIG. 12, if the conclusion “not applicable” is obtained, the pitch period is not calculated, a signal indicating that the pitch period is not calculated is output, and the process returns to step S101. . In the above description, the pitch period calculation process has been described by taking the case of a digital audio signal as an example. However, the pitch period is also calculated for a digital model signal by exactly the same process.
[0060]
As described above, in the present embodiment, the recall is obtained for all assumptions 1 to 4, the assumption with the highest recall is selected, and the pitch calculation based on the selected assumption is within the allowable range. This is a cautious procedure that is implemented only when it is within the range and not performed when it is out of the allowable range. The reason for taking such a careful procedure is as follows.
[0061]
a. As a procedure other than the above procedure, for example, the respective recall rates corresponding to assumptions 1 to 4 are sequentially calculated, and when the recall rate within the allowable range is obtained, the calculation is terminated. An alternative is possible in which the pitch period is determined by selecting. However, depending on the speech waveform, for example, a situation may occur in which the recall rate corresponding to Assumptions 1 and 3 is within an allowable range, and the recall rate corresponding to Assumption 3 is higher than that of Assumption 1. If this alternative is to be followed in such a case, assumption 1 is selected and an incorrect pitch period is obtained. Although it is conceivable to set the allowable range narrow so that the selection of the assumption is made correctly, there may be cases where “not applicable” is determined in that case.
[0062]
b. An alternative is also conceivable in which all recall rates corresponding to assumptions 1 to 4 are calculated, the assumption with which the maximum recall rate is obtained is unconditionally adopted, and the pitch period is obtained. However, the recall rate corresponding to any assumption is uniformly low, and there may be cases where the recall rate corresponding to a specific assumption is slightly better than others. Even if the pitch period is forcibly obtained by adopting, there is no guarantee that an accurate pitch period will be obtained. For example, when the waveform of the digital audio signal changes abruptly in the pitch period, the reproducibility is likely to be low in any of the above assumptions.
[0063]
c. Therefore, in the present embodiment, the pitch period is calculated according to the above-described procedure, thereby preventing an inappropriate pitch period from being output.
[0064]
The pitch periods of the digital audio signal and the digital model signal obtained as described above are sequentially reported to the scoring unit 13, and the deviation of the pitch period of both signals and the deviation of both signal levels obtained by the level detection unit 12 are reported. According to the overall evaluation, the song of the singer is scored, and the scoring result is displayed on the display unit 14.
[0065]
C. Evaluation result of the apparatus according to this embodiment
Various operation conditions were set for each part of the pitch cycle detection device described above, and the pitch cycle detection time and detection error were evaluated. FIGS. 13 to 16 show the results. First, FIG. 13 shows a result of measuring a pitch period detection error in a practical range by using a circuit that performs linear interpolation as the 4-times oversampling unit 7 and changing the oversampling frequency of the circuit variously. From this result, it can be seen that the detection error in the practical range can be sufficiently reduced by performing interpolation of about 4 times oversampling. Next, FIG. 14 shows how the pitch period is detected for each of the case where the pitch period is obtained from the correlation between the three periods (m = 3) and the case where the pitch period is obtained from the correlation between the four periods (m = 4). The result of measuring the delay time for each input frequency is shown. As shown in this experimental result, if m = 3 or 4, it is possible to keep the detection delay within a range where there is no problem. FIG. 15 shows the relationship between the number of times of averaging and the pitch cycle extraction error. FIG. 16 shows the result of an experiment of how many periods (pitch periods) in the past can be accurately extracted by comparing with the waveform for the past. This experimental result shows that there are many errors when comparing the past two cycles, and when comparing input waveforms over the past five cycles, the waveform is too old and the pitch cycle is wrong. Comparing input waveforms over four periods shows that it is optimal for accurate pitch extraction.
[0066]
D. Modified example
(1) In the above embodiment, it is assumed that the pitch period is the sum of 2n pieces of zero cross interval data depending on how much the zero cross interval data constituting one pitch period matches between the pitch periods. Judgment whether or not is appropriate. Instead of this method, for each n, a predetermined number of pitch periods are obtained by calculating the sum of 2n zero cross interval data, and the n with the least variation in these pitch periods is selected, and the pitch period is determined. You may make it select. That is, in FIGS. 9B1 to 9B4, T₁₁~ T₁₄T is the smallest variation of T₁₁~ T₁₄Is the pitch period, and T_{twenty one}~ T_{twenty four}T is the smallest variation of T_{twenty one}~ T_{twenty four}Is the pitch period, and so on. In addition, using the determination method based on the zero cross interval data disclosed in the above embodiment and the determination method obtained for the variation of the pitch period, the zero cycle interval data and the pitch period variation of the pitch period length are comprehensively evaluated, You may make it select a pitch period.
[0067]
(2) In the above embodiment, the width Δ of the masking band of the binarization unit 8 is fixed. However, since the amplitude of the minute vertical movement of the speech waveform generated near the zero level depends on the amplitude of the entire speech waveform, it may be difficult to determine an appropriate Δ. Therefore, the amplitude of the digital audio signal or the digital model signal is detected, the amplitude value is multiplied by a predetermined coefficient, and the result is set to Δ, and the control of the masking band width Δ of the binarization unit 8 is performed. It is preferred to do so.
[0068]
【The invention's effect】
  As described above, according to the present invention, a digital audio signal is oversampled, the oversampled digital audio signal is converted into a binary signal, and a continuous zero of the audio waveform is generated based on the binary signal. Measure the cross interval, and for each n, assume that the pitch period is the sum of 2n zero cross interval data.Find the degree of coincidence of the waveform between each pitch period and adopt the best assumption of coincidenceSince the pitch period is calculated, the zero-crossing interval can be measured accurately, and even when the speech waveform is a complex waveform containing harmonic components, the pitch period can be accurately and quickly set with an inexpensive configuration. There is an effect that it can be obtained.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of an embodiment of the present invention.
FIG. 2 is a block diagram showing a configuration example of a 4-times oversampling unit in the same embodiment.
FIG. 3 is a block diagram illustrating a configuration of a binarization unit in the embodiment.
FIG. 4 is a block diagram illustrating a configuration of a binarization unit in the embodiment.
FIG. 5 is a block diagram showing a timer, a RAM, and a control system thereof in the same embodiment.
FIG. 6 is a diagram showing an operation of a 4-times oversampling unit in the same embodiment.
FIG. 7 is a diagram showing an operation of a binarization unit in the same embodiment.
FIG. 8 is a diagram showing an operation of a write control unit in the same embodiment.
FIG. 9 is a diagram illustrating an outline of a pitch cycle calculation process in the embodiment.
FIG. 10 is a flowchart showing a pitch period calculation process in the embodiment.
FIG. 11 is a flowchart showing a pitch period calculation process in the embodiment.
FIG. 12 is a flowchart showing a pitch period calculation process in the embodiment.
FIG. 13 is a diagram showing a performance evaluation result of the embodiment.
FIG. 14 is a diagram showing a performance evaluation result of the embodiment.
FIG. 15 is a diagram showing a performance evaluation result of the embodiment.
FIG. 16 is a diagram showing a performance evaluation result of the embodiment.
[Explanation of symbols]
1 ... CD, 2 ... Vocal extractor, 3 ... Microphone,
4 ... A / D converter, 5 ... DC removal unit, 6 ... LPF,
7 ... 4 times oversampling unit, 8 ... binarization unit,
9 …… Timer (zero cross interval measuring means),
10: RAM (zero cross interval measuring means),
11: Pitch calculation unit (pitch calculation means),
12... Level detection section, 13... Scoring section, 14.

Claims (1)

Oversampling means for outputting a digital audio signal with a sampling frequency multiplied by a predetermined number;
Binarizing means for comparing the digital audio signal output by the oversampling means with a predetermined level and converting it to a binary signal;
Based on the binary signal, zero cross interval measuring means for measuring continuous zero cross intervals t ₁ , t ₂ ,... Of the digital audio signal;
n (n is an integer of 1 or more) is varied is the for each n, assuming that the total sum _{T = (t 1 + t 2} + ·· t 2n) the pitch period of the 2n zero cross interval, adjacent contact m cycles The degree of coincidence between the corresponding zero-crossing intervals included in each of the assumed pitch periods (m is an integer of 2 or more) is calculated as the degree of coincidence of the waveform of the digital audio signal. And a pitch calculation means for obtaining a pitch period by selecting n having the highest degree of coincidence.

Images