JPH0342412B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION This invention relates to signal analysis, and more particularly to pitch (fundamental frequency) analysis for musical performers.

By analyzing and displaying musical pitch information, it is possible to provide extremely useful feedback to music performers. To understand this invention well, it is necessary to clearly define what pitch is in music. All musical tones with perceivable pitch consist of periodic sound pressure waveforms. The simplest periodic waveform is a sine wave.
Harmonics (harmonics, i.e., sine waves with frequencies that are integral multiples of the fundamental frequency) are added to this basic sine wave.
By adding any number of , complex waveforms can be created on the time axis. Even if these harmonics exist, the pitch of the sound can be recognized as the fundamental frequency of the waveform. In fact, even if the fundamental frequency of a musical tone is weak or missing, the human pitch detection mechanism can estimate its fundamental pitch from the harmonics present. A simple pitch measuring device based on the frequency domain responds to all frequencies included in the waveform and rarely provides clear measurement results. Even if you use a method that displays the lowest frequency present, if the energy content at the fundamental frequency is smaller than the other harmonic components, the displayed frequency will not match the perceived pitch.

A better way to obtain the sensed pitch is to measure the time interval during which the waveform remains periodic. This method is a model closer to the human pitch detection mechanism. However, this method also has flaws. First, in naturally occurring sounds, the frequencies of overtones are often not exact integral multiples of the fundamental frequency, and therefore cannot be strictly called harmonics. This inaccuracy is due to the fact that the phase of the harmonics is constantly changing with respect to the phase of the fundamental frequency.
This creates a waveform structure that changes dynamically on the time axis. Therefore, the waveform changes completely over an interval of a few cycles, but the waveform over an interval of the next few cycles is similar. Furthermore, the overtone structure of musical tones, especially in the case of the human voice, changes significantly over a relatively short period of time. In this case as well, the waveform changes over several cycles.

Further complicating measurements is that the amplitude of naturally occurring acoustic waveforms is likely to be modulated by random fluctuations. There are also periodic amplitude fluctuations and periodic frequency fluctuations. That is, tremolo and vibrato.
The human singing voice typically contains some degree of all three of these effects.

None of the pitch measurement methods described above solve all of these problems. Many people are aware of the drawbacks of measuring on the frequency axis and are trying to measure the period of the waveform on the time axis. Many methods are based on detecting the amplitude peaks of periodic waveforms, such as Merritt, US Pat. No. 4,028,985 and Slepian and Weldon, US Pat. No. 4,217,808.
The peak detection method has several drawbacks.
First, an acoustic waveform containing many overtones has several peaks in one cycle, and the shape and amplitude of these peaks are constantly changing as described above. Therefore, a peak detected within one cycle does not necessarily correspond to a peak detected within an adjacent cycle, resulting in a large measurement error. Similarly, random or periodic fast amplitude fluctuations may cause peaks to be missed or other peaks to be detected instead of the main peak to be detected. Even if the peak is detected and intact, the waveform generally has a gentle slope near the peak, so small changes in amplitude will cause errors in time measurements.

Additionally, waveform amplitude methods often require automatic gain control (AGC) circuitry to adapt to changes in input signal level. The AGC circuit has a fast attack time (amplification response time to changes in input level) to avoid waveform distortion.
and designed to have a slow decay time. Therefore, the AGC circuit cannot follow the small but fast amplitude changes of naturally occurring acoustic waveforms. This is not a problem in normal audio applications since sound is only judged by the human ear, which is insensitive to gradual amplitude changes. However, small amplitude changes introduce significant errors to the peak detector. By reducing the decay time of AGC, it is possible to follow faster amplitude fluctuations, but level-dependent low frequency waveform distortion occurs. In order to minimize such problems, it is necessary to limit the measurable pitch range or provide a means to adjust the AGC time constant according to the input pitch.

It is known to perform frequency analysis by measuring the time when a waveform crosses a zero level. This zero crossing of the waveform is completely unaffected by the amplitude of the waveform. Although this method is suitable for relatively pure sounds, problems arise with waveforms that have several zero crossings during one cycle. Some filtering means can be used to remove the overtones so that only two zero crossings occur per cycle, but this requires operator intervention or introduces undesirable characteristics of the AGC circuit. All require automatic filter means.
Thus, while the zero-crossing method avoids the problems of the peak detection method, it does encounter other problems.

The present invention provides an apparatus and method for measuring the pitch of a musical tone with sufficient accuracy and speed, and displaying the pitch and pitch error completely and clearly intuitively, so that instrumentalists and singers can easily measure the pitch. identify,
The purpose is to provide immediate feedback as pitch information to help correct errors.

The device according to the invention has two distinct components. namely, an analog signal processing circuit and a digital calculation and display circuit. Analog signal processing circuitry receives a signal from a suitable signal source, amplifies the signal if necessary, removes frequency bands outside the region of interest, and generates a digital reference signal representative of zero crossings of the analog signal. The digital calculation circuit uses the zero crossing time data to perform an analysis to determine the fundamental pitch of the input signal. In practice, this is accomplished by sequentially delaying the digital reference signal by an amount corresponding to the spacing between zero crossings and correlating the delayed signal with the digital reference signal.
High correlation corresponds to a delay approximately equal to an integer number of fundamental periods. Furthermore, the digital calculation circuit converts the pitch information into an appropriate display drive signal and provides it to the display device via a buffer, if necessary.

The apparatus of this invention measures the pitch of an audio frequency signal and displays the pitch accurately and immediately on a display so that it can be easily read by an untrained operator. Pitch measurement and display are all done automatically, so the operator does not have to make any adjustments or have any trouble with operations.
The pitch measurement range is at least seven octaves, and its display accuracy is constant for an average tuned scale. For comparison, in the method using beat tones, a given pitch error produces a beat frequency that is proportional to the pitch of the tones. The pitch is preferably indicated by light on the musical staff (staff), and an indicator is also provided to indicate the octave difference and error. This pitch display is easy to read and has an intuitive naturalness for the musician. The display is updated frequently enough to give the impression that it is responding quickly to pitch changes. This device also recognizes musical tones that have a pitch that can be felt in the mind (hereinafter referred to as estimated pitch).
No display for all other inputs. Therefore, transient noise or other erroneous data will only temporarily erase the display and will not adversely affect subsequent measurements. This device is capable of accurately measuring the estimated pitch of a wide variety of acoustically generated sounds. A sound can contain any number of overtones. The frequencies of the overtones may deviate from exact integer multiples of the fundamental frequency, and the fundamental frequency may be weak or absent. Furthermore, the amplitude of the sound may be spread over a wide range, and even if it changes rapidly randomly or periodically, the accuracy of the pitch measurement will not be affected.

FIG. 1 is a perspective view showing the external appearance of the pitch analyzer 1. As shown in FIG. Broadly speaking, the analyzer 1 includes an electric circuit within a cabinet 2, which will be described below, and analyzes and displays pitch information of musical tones. Musical sounds are detected by a built-in microphone 4 and converted into electrical signals. Alternatively, instead of built-in microphone 4, an external microphone or electronic device may be connected to input jack 5. The display section includes a first LED display row 6, a second LED display row 7, and a third LED display row 8.
have. The pitches may be averaged digitally by operating the selection switch 9 (see FIG. 7).

The LED display 6 corresponds to the notes (notes) in the musical scale. These LED indicators 6 are arranged so that the horizontal direction corresponds to the piano keyboard, and the vertical direction corresponds to the musical staff. LED display 7 is in the center C
Corresponds to the octave difference from the octave starting from the C note. In a preferred embodiment, tones up to three octaves below the middle C note and up to three octaves above the B note above the middle C note can be displayed.
The LED display 8 displays how much the input note deviates from the nearest standard average rate tuned note.

FIG. 2 shows the main blocks of the electrical circuit of the analyzer 1. An input signal from a microphone, pickup device, or other signal source is applied via a preamplifier 10 and a comparator 11 to a pulse circuit 12 which generates a pulse each time the input signal crosses the zero level. Time measuring device 13 measures each zero crossing time of the input signal represented by a pulse from pulse circuit 12 and stores these values in memory 14 . The actual pitch determination is made by microprocessor 15 which analyzes the time values stored in memory 14. The microprocessor 15 executes this analysis using a stored program 16.
This is done by executing. The result of the analysis is stored in output latch 17 as a number representing the pitch value. Decoder/driver 18 and display device 19
(including LED indicators 6, 7, and 8) convert pitch data into information that can be understood by the operator.

The first block, preamplifier 10, is provided with a signal at its input, for example, which is periodic and has a pitch within the range of interest. The signal source may be a microphone, a transducer, or any other suitable generator for generating electrical signals. Preamplifier 10 amplifies the signal and removes frequencies outside the pitch range of interest.

FIG. 4 shows a preferred example of the preamplifier 10.
The input signal from the microphone 30 is amplified by a microphone amplifier 31 and applied to a high-pass filter 32 and a low-pass filter 33, one of which is selected by a switch 34 under the control of the microprocessor 15. be done. Which filter is selected is determined by which filter provides the best pitch data. The low pass filter 33 is used to remove high frequencies from the predominantly low temperature components. The high pass filter 32 is used to remove low frequency interference from the power line and random low frequency fluctuations superimposed on the mid and high frequencies. The buffer 35 further amplifies the signal and outputs it as the output of the preamplifier 10.

The output of the preamplifier 10 is applied to a comparator 11, which generates a binary reference signal that takes a high level if the input signal is higher than the zero level and a low level if it is lower. In a preferred embodiment, comparator 11
has very little hysteresis to obtain sharp and non-oscillating transitions (logic inversion between low and high levels).

The output of comparator 11 is provided to pulse circuit 12 which generates a pulse at each zero crossing point. The characteristics of this pulse are determined by how the time measuring device 13 for measuring the zero crossing time is configured. For the time measuring device 13 constructed in hardware, the pulse circuit used is, for example, a monostable multivibrator that is triggered by positive and negative transitions of the signal.
In FIG. 5, the time measuring device 13 is a microprocessor 15.
When configured as part of a program 16 for (i.e. configured as software)
This is a preferred embodiment of the pulse circuit 12 of FIG. The output from the comparator 11 is applied to one input of an exclusive logic gate (XOR gate) 40, the output of which is applied to the interrupt input of the microprocessor. When given an interrupt, the microprocessor 15 will
inverts the logic state of the signal applied to the other input of XOR gate 40, thereby causing XOR gate 40 to
terminates the interrupt, which is the output pulse of , and the pulse circuit 12 is ready to respond to the next zero crossing.

Time measuring device 13 for measuring each zero crossing time of the signal
is configured as a counter, and its counting output is stored in the memory 1 every time a pulse is given from the pulse circuit 12.
4 is stored. Means are provided to prevent data acquisition once a predetermined number of zero crossing times have been stored in the memory 13. The number of zero-crossing times must be large enough to represent at least two cycles of the input waveform and large enough that the waveform at the beginning of the sample period is substantially different from the waveform at the end. It shouldn't be.

In the preferred embodiment, when the time measuring device 13 is constituted by a microprocessor 15 executing part of the program 16, the numbers stored in the memory 14 are clocked at an accurate rate proportional to the clock frequency of the microprocessor 15. Incremented. When an interrupt occurs due to a zero crossing, the incremented number representing the time of the zero crossing is stored in another area within memory 14. The counter stops each time an interrupt occurs. The number of counts lost while the interrupt is valid must be compensated for. Thus, a total number of lost counts is kept and added to each time value before being stored in memory.

Once the zero crossing time data is stored in memory 14, microprocessor 15 uses program 16 to perform an analysis of the data to determine the pitch value of the input signal. The pitch information output is latched into the output latch and new samples of the waveform are processed. The following types of pitch information can be output. For example, data sent to another device via a communication link, a graphical display shown as notes on a staff (staff) or notes on a keyboard, a line display such as a thermometer, or conveying pitch information to the user. means etc. As mentioned above, in the preferred embodiment the pitch information includes 7 values for pitch octave, 12 values for note name values, and 8 values for pitch error in cents (percent of semitone distance). It is expressed as the value of

In a preferred embodiment the decoder/driver 1
8 converts the binary number latched in the output latch 17 into a signal for driving the display device 18. In the preferred embodiment, display 18 uses LEDs, but other display elements may be used.

Although this invention is shown in a block diagram as shown in FIG. 2, this does not mean that each component block of the device can actually be physically separated, but merely that it can be carried out to achieve the object of this invention. It merely illustrates the functions to be performed. As one specific embodiment of the present invention, an example in which an individual transistor amplifier is used for the preamplifier 10 and a monolithic IC is used for the comparator 11 and the pulse circuit 12 is shown in FIGS. 6 and 7. Microprocessor 15 is 6502 type
It is a MOS silicon IC. The storage program 16 is
It is a 2716 type PROM. Number 50 is a clock generator. The time measuring device 13 is implemented as part of a storage program 16 executed by a microprocessor 15. Memory 14 and output latch 17 are implemented in a single 6532 type IC designed to be used with microprocessor 15. The decoder/driver 18 is a TTL IC and drives the LED display device 19. Embodiments suitable for mass production can be designed to use one monolithic IC for the necessary analog operations and one monolithic IC for the digital operations.

(b) Summary of calculation method The basic idea used in the storage program is as follows.
When a segment of a periodic waveform containing several cycles is delayed by exactly one cycle time and compared to the original waveform, the original and delayed waveforms are very different at every point along the segment. It is to show good match or correlation. Of course, the same can be said when the amount of delay is an exact integral multiple of one cycle. Other delay times show a small correlation.

Another important assumption is that the zero crossings of a naturally occurring acoustic waveform are such that the true period of that waveform can be obtained from a binary waveform with zero crossings at the same point as the input waveform, using the correlation technique described above. This means that it contains sufficient information about the waveform. This assumption turned out to be correct.

FIG. 3 diagrammatically illustrates the method used to calculate correlation and determine pitch. The input waveform is shown as arbitrary time memories, each memory representing a unit time. Zero crossings in the input waveform are indicated by logic state changes (transitions) in the reference waveform. The reference waveform is that obtained at the output of comparator 11 in FIG.

The true period of the input waveform can be found by delaying the reference waveform by various values and calculating the correlation corresponding to each delay. Delays exhibiting high correlation can be considered approximately integer multiples of one cycle.

Correlation calculation is performed for the first delayed waveform when the reference waveform is delayed so that the first positive zero crossing of the delayed waveform corresponds to the second positive zero crossing position of the reference waveform at T3. The correlation between this first delayed waveform and the original reference waveform is such that at every point between the start point T3 of the first delayed waveform and the end point T14 of the reference waveform (28 units of time), the two waveforms Calculated by comparing . Within this 28 time unit range, the waveform has the same logic state for 16 time units, as shown by the plus sign in the first delayed waveform, and the opposite logic state for 12 time units, as shown by the minus sign. have. Therefore, the value of correlation is (16-12)/28
= 4/28 or 14%. This is considered to be a poor correlation.

The second delayed waveform delays the reference waveform until the next positive zero crossing, starting at T5. This second delayed waveform and the reference waveform have the same logic state over a total of 25 time units from T5 to T14. Therefore, the correlation value is 25/25, or 100%, indicating perfect correlation. This delay time corresponds to just one cycle after the original waveform.

In the third delayed waveform, from T7 to T14, there are the same logic state for 10 unit times and the opposite logic state for 8 unit times. Therefore, the correlation is (10-2)/18=2/
18, or 11%, and the correlation is getting worse again.

The fourth delayed waveform is delayed by just two cycles and has perfect correlation in the time range from T9 to T14.

Naturally occurring acoustic waveforms rarely have a perfect period, so perfect correlations are rarely obtained, but there is generally a clear difference between correlations caused by a one-cycle delay and correlations caused by less than one-cycle delay. It will be done. To explain this, let us assume that there is an imperfection in the waveform that fails to cross zero from T7 to T8. When calculating the correlation for the second delay, both waveforms are in the same logic state for 23 units of time, and 2 units of time are in the opposite logic state. Therefore, the correlation is (23−
2)/25 or 84%, which is clearly perfect 1
This is a much higher value than the correlation without cycle delay.

After calculating this first group of correlations, further analysis can be performed by calculating a second group of correlations for the same reference waveform with the starting point at T2 instead of T1. Starting at T3 for the third group, the analysis can be repeated in a similar manner until sufficient data is obtained for valid calculations.

While performing the correlation calculation, only the delay time values that give good correlation results are kept in the list. This list ranks by correlation and holds the delay times that give the highest correlation. Alternatively, the list of valid delay times may be compiled by storing only delay times corresponding to correlations exceeding a predetermined threshold. In order to save the amount of memory allocated to the list, the delay time corresponding to one correlation threshold is accumulated from one end of the list, and the delay time corresponding to a higher threshold is accumulated from the other end of the list. Good too. Once the list is filled, the delay times corresponding to the higher thresholds are written over the delay times corresponding to the lower thresholds in the list, thereby continually improving the quality of the data in the list.

Once the list is complete, compare the lowest delay time with good correlation to all others. If this minimum delay time is extremely smaller than the maximum one, it is considered that the data is not due to a complete one-cycle delay but is a result of a very small portion within one cycle. Such a delay is determined to be invalid and the remaining data is examined to find a delay time that would correspond to one complete cycle of the waveform. Delay times that are close to an integer multiple of the minimum effective delay time are normalized by dividing by an exact integer value. Delay times that are not close to integral multiples are discarded. The pitch of the sound is then calculated by taking the average of all these normalized times.

When using the device for human voices or other complex sounds, additional calculations are performed on the pitch data to make the pitch analysis more effective. Display of pitch may be prohibited until at least two similar pitch values are obtained in succession and their average is taken. The next similar pitch data may be averaged such that the sum of the current pitch value and the difference between the newly obtained pitch and the current pitch divided by a certain value N is displayed as the pitch. The larger the value of N, the smaller the influence of the new pitch on the display pitch. In the preferred embodiment, N=4. There are several advantages to using this averaging method. First, it is possible to suppress meaningless pitch readings caused by sound transient onset phenomena. Second, the pitch can be stably displayed even when noise is superimposed or the pitch is originally unclear. In addition, wide pitch fluctuations in sounds with substantial vibrato are averaged out,
Provide an easily readable display.

In order to adapt to various pitches and tuning standards,
The clock frequency used to measure zero crossing time may be changed without changing the constants used in the program. Conversely, various reference pitches or tuning systems may be selected by changing the constants used in the calculation program.

(b) Overview of the calculation program The program has four main parts that perform four distinct tasks. The first main section records the time at which the zero crossing of the input waveform occurs. The second main part creates a list of delay times for which high correlations have been calculated. The third main part calculates the pitch from these delay times. The fourth main part is optional and performs the averaging of the series of calculated pitch values. These operations will be described in detail below.

A Recording the zero-crossing time of the input waveform A-1 Set a counter with a clock frequency in hardware or software so that overflow does not occur for the maximum expected time interval.

A-2 Using the counter, record the time at which each zero crossing of the waveform occurs in a transition time table set in memory 14 accessible by the microprocessor. The microprocessor performs correlation analysis.

A-3 Stop recording zero-crossing times when a predetermined number of zero-crossing times are recorded in the transition time table or a predetermined time has elapsed.

A-4 To reduce round-off errors in the following correlation calculations, each term in the transition time table is repeatedly shifted to the left (doubled) until the maximum number overflows, and the number of shifts is recorded in a register called an octave. .

B. Creation of a list of delay times corresponding to high correlations B-1. Set pointers named reference pointer and delay pointer to the first and third points in the table corresponding to the first and second zero crossings of the same logic state, respectively. Set to section.

B-2 Start calculating the correlation between two waveforms represented by data from the time value of the delay pointer in the transition time table.

B-2-a A variable called delay is set equal to the difference in time specified by the two pointers.

B-2-b Check whether the time difference between the current zero crossing and the next zero crossing is smaller than the data specified by the reference pointer or delay pointer. (Always subtract the delay time associated with the delay pointer so that the times from the transition time table can be easily compared.) B-2-c Two waveforms are known to start in the same logic state. Therefore, the time difference to this nearest zero crossing is added to a variable called the total correlation.

B-2-d Since both waveforms are in opposite logic states, step forward in time until the next closest transition to either waveform is found and subtract the time difference between this transition and the previous transition from the total correlation.

B-2-e Continuing to advance the time, subtract the transition time difference from the total correlation each time one pointer specifies an odd number in the table and the other pointer specifies a coefficient. If both pointers point to odd or even numbers, the time difference is added to the total correlation.

B-2-f When the end of the transition time table is reached, the total correlation is divided by the total time from the start time of the delay pointer to the end of the transition time table.

B-3 If the correlation is greater than a predetermined threshold, accumulate the delay time in the delay time list, otherwise discard it.

B-4 Set the reference pointer and delay pointer to the first and fifth terms of the transition time table and execute steps B-2 to B-3.

B-5 The delay pointer is sequentially set to the odd-numbered entry in the transition time table until a predetermined number of entries are entered in the delay time list or there is no valid data in the transition time table, and steps B-2 to B- Execute 3.

B-6 Set the reference pointer and delay pointer to the second and fourth positions of the transition time table, respectively.
, and perform operations similar to steps B-2 to B-5 using only even-numbered transitions.

B-7 Step B-7: Set the reference pointer for each term in the transition time table sequentially until a sufficient number of calculation results have been entered in the reference pointer delay time list or there is no valid data in the transition time table. 2~B-
Perform calculations similar to 6 for several groups.

C Determination of pitch from correlation data C-1 Find the longest delay and shortest delay from the delay time list.

C-2 If the ratio of these numbers is not too large (less than 8 in this example), the smaller one is considered to represent one cycle of the waveform.

C-3 If the ratio is too large, the smaller delay time is determined to be invalid and removed from the delay time list.

C-4 Repeat steps C-2 and C-3 until a valid minimum delay is found.

C-5 Examine each delay time and mark as invalid those that are not close to an integer multiple of the shortest valid delay time found in the previous step.

C-6 If a sufficient number of entries are not left in the delay time list, it is determined that there is no valid pitch in the current sound sample, the pitch calculation is stopped, and the process returns to the beginning of step A.

C-7 If there are sufficient entries in the delay time list, normalize all delay times by dividing each valid delay time by the nearest integer to represent one cycle of the waveform.

C-8 Calculate the pitch of the sound by taking the average of all valid normalized delay times.

D Averaging a series of pitch values (optional) D-1 If the new pitch value is not close to the previous value, erase the display and obtain new zero-crossing data.

D-2 If the new pitch value is close to the previous pitch value, take the average with that value and display the calculated pitch value.

D-3 If the new pitch value is the third or later value in the series of pitch values, calculate the new average pitch value using the following formula:

Average pitch = previous pitch + (new pitch - previous pitch)/N where N may be a fixed value or may be dynamically varied based on the number of consecutive pitches obtained.

As can be seen from the foregoing description, the present invention provides a pitch measuring instrument that extracts relevant features from an input waveform and displays these features as meaningful to the musician. Although the foregoing provides a disclosure of preferred embodiments of the invention, it will be understood that various modifications may be made without departing from the spirit of the invention. For example, although the invention has been described as a device for musical performers, the basic device of the invention could also be used for things such as speech therapy.

[Brief explanation of drawings]

FIG. 1 is an external view of a pitch measuring device according to a preferred embodiment of the present invention, FIG. 2 is a block diagram explaining the functional elements of the present invention necessary for performing pitch analysis, and FIG. 3 is a diagram showing pitch information from an input waveform. Figure 4 shows the preferred embodiment of the preamplifier, Figure 5 shows the preferred embodiment of the pulse circuit, and Figure 6 shows the analog circuitry of the device. FIG. 7 is a diagram showing the digital circuit section of this device. 6: 1st LED indicator, 7: 2nd LED indicator,
8: 3rd LED indicator, 10: Preamplifier, 11:
Comparator, 12: Pulse circuit, 13: Time measuring device,
14: Memory, 15: Microprocessor, 16...
...accumulation program, 17: output latch, 18: decoder/driver, 19: display device.

Claims (1)

[Claims] 1. Generating two types of reference waveforms in response to an audio input signal and having logical inversions in response to rising and falling edges at the zero level crossing point of the input signal at times corresponding to the zero level crossing point. means for measuring a respective time interval from a first logic inversion of said reference waveform to a plurality of subsequent logic inversions in the same direction as said first logic inversion; memory means for storing measured values; and a memory means for storing measured values; and a memory means for storing measured values; means for determining respective correlation values between a waveform and the plurality of delayed waveforms; selecting a set of time intervals from among the plurality of time intervals that provides a delayed waveform having a correlation value exceeding a predetermined threshold; and means for determining a characteristic period of the reference waveform based on the set of time intervals. 2 converting into a binary reference waveform having logical inversions corresponding to rising and falling edges at the zero level crossing point of the audio input signal at points corresponding to the zero level crossing point; a logic inversion followed by a plurality of logic inversions in the same direction as the first logic inversion; measuring a time interval from the first logic inversion of the reference waveform to the plurality of logic inversions in the same direction; Determining a correlation value between the reference waveform and a delayed waveform obtained by delaying the reference waveform by the time interval; and providing a delayed waveform having a correlation value exceeding a predetermined threshold. A method for measuring pitch of an audio input signal, comprising: selecting a set of time intervals from the plurality of time intervals; and determining a characteristic period of the reference waveform based on the set of time intervals. 3. The step of determining the characteristic period is to determine the minimum time among the time intervals for which the value obtained by dividing the maximum value in the set of time intervals by the time interval of the set is smaller than a predetermined constant value. selecting intervals; in said set of time intervals, time intervals whose values are separated by a predetermined value or more from the respective nearest integer multiple of said selected minimum time interval; normalizing each time interval to one cycle time by dividing each remaining time interval of said set of time intervals by an integer giving said nearest integer multiple; and normalizing 1. 3. The pitch measuring method according to claim 2, comprising the step of obtaining pitch by averaging cycle times.