JPH1049148A - Frequency detecting device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve counting precision while counting a period of a sound signal with a count value of a sampling clock by repeatedly executing period measurement in a prescribed period and finding weighted mean of plural obtained count values. SOLUTION: In a period detecting part 40, a noise property decision part 60 decides a singing sound signal having repeated irregular zero cross of too short period as a noise signal. An LPF 61 removes the noise signal and a harmonic component to supply the fundamental frequency of the singing sound signal to a count operation part 62. The count operation part 62 counts the zero cross interval of the inputted singing sound signal at the period of the sampling clock, and weighted mean of plural zero cross intervals is found. The number of average zero cross intervals is not fixed, and all timewise intervals of the zero cross occurring for a fixed time just before counting are averaged. When the frequency (period) of the sound signal is detected, as the period of the weighted mean, about 20ms is suitable.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a frequency detector for detecting a frequency (period) of an input audio signal.

[0002]

2. Description of the Related Art In a karaoke apparatus or the like that is currently in practical use, a harmony voice signal three or five times higher than a singer's singing voice signal is used to excite the singing or to make the singing sound better. Some have a function of generating and outputting. In the harmony voice signal addition function, a harmony voice signal is formed by frequency-shifting a singing voice signal input from a microphone in order to generate a harmony voice having a tone similar to the singing voice of the singer and having the same tempo. Things are common.

In order to realize the function of adding the harmony voice signal, it is necessary to detect the frequency (period) of the singing voice signal and shift this frequency to the frequency of the harmony melody. As a method for detecting the frequency of a singing voice signal, a method shown in FIG. 11 has been conventionally used. In this method, a timing at which an input signal passes through a 0 level line and is inverted from a negative value to a positive value is monitored as a zero cross timing, and an interval (a zero cross interval) from the immediately preceding zero cross timing to the current zero cross timing is set to one. This is a method of measuring as a cycle. Actually, since a signal discretized at the sampling period by the digitization process is input, the sampling value obtained at each sampling timing is monitored, and when this sampling value turns from a negative value to a positive value. The zero-cross timing is measured in sampling period units (integer values).

[0004]

However, this frequency detection method has a problem that the accuracy is limited by the sampling period. That is, in FIG. 11, the sampling frequency is set to Fs Hz and the sampling period T
s is the reciprocal of Ts = 1 / Fs sec, and if the true period of the input signal is Tin sec, the true zero-cross interval P
₀ samples is P ₀ = Tin / Ts. However, the value P actually measured is the count value (integer value) of the sampling frequency, and P ₁ = (value obtained by rounding down the decimal point of Tin / Ts) or P ₂ = (rounded up the decimal point of Tin / Ts) Value) = P
It becomes ₁ +1.

[0005] Each time the zero-cross is measured, but P ₁ or P ₂ is obtained, the variation of _{P 2 / P 1 = 1 +} 1 / P 1 occurs in the detected the frequency (period). For example, when Fs = 44.1 kHz and the signal frequency is 500 Hz, P ₀ = 88.2 P ₁ = 88 P ₂ = 89, and the fluctuation of the detected frequency is 89 /
When converted to 88 cents, this is a large value of 19.56 cents. When a harmony sound signal is generated based on this, the pitch of the harmony sound signal deviates by 19.56 cents, thereby causing a main melody and a beat, thereby deteriorating the sound quality. The higher the frequency of the audio signal input P ₁ is small, the pitch fluctuation 1 + 1 / P ₁ increases.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a frequency detecting device which has an improved accuracy while counting the period of an audio signal with a count value of a sampling clock.

[0007]

According to a first aspect of the present invention, there is provided a counter for inputting an audio signal digitized by a predetermined sampling clock and measuring a period of the audio signal by a count value of the sampling clock. In the frequency detecting device provided with the means, the counting means is repeatedly executed within a predetermined period, and the weighted averaging means for determining the frequency of the audio signal by weighted averaging a plurality of count values obtained thereby. It is characterized by the following.

According to a second aspect of the present invention, when there is a period during which no audio signal is input within the predetermined period, the weighted averaging means is invalidated, and the frequency is determined from the latest count value of the counting means. Means for performing the operation.

According to a third aspect of the present invention, the weighted averaging means performs weighted averaging only during a period during which the audio signal is input when there is a period during which the audio signal is not input within the predetermined period. It is characterized by having.

According to a fourth aspect of the present invention, the weighted averaging means reduces a rate of change of a weighting coefficient from an old count value to a new count value when a variation in the plurality of count values is small, and When the fluctuation of the count value is small, a means for increasing a change rate of the weighting coefficient is included.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A karaoke apparatus having a frequency detecting apparatus according to an embodiment of the present invention will be described with reference to the drawings. This karaoke apparatus has a harmony addition function of adding a harmony voice signal to a singing voice signal of a karaoke singer. The harmony adding function converts the singing voice signal of the karaoke singer input from the microphone 27 into a digital signal by the A / D converter 29, detects this frequency by the voice processing DSP 30, and outputs the main melody (original singing voice signal). ) Is converted to a harmony voice signal having a pitch of 3rd or 5th. Then, this function is output together with the original singing voice signal, thereby making it possible to perform karaoke singing with harmony.

FIG. 1 is a block diagram of the karaoke apparatus. The CPU 10 for controlling the operation of the entire apparatus includes a ROM 11, a RAM 12, a hard disk storage (HDD) 17, an ISDN controller 16, a remote control receiver 13, a display panel 14, a panel switch 15,
Sound source device 18, audio data processing unit 19, effect DSP2
0, a character display unit 23, an LD changer 24, a display control unit 25, and a voice processing DSP 30.

The ROM 11 stores a system program, an application program, a loader, and font data. The system program is a program that controls the basic operation of the device and data transmission / reception with peripheral devices. The application program is a peripheral device control program, a sequence program, or the like. During a karaoke performance, a sequence program is executed by the CPU 10 to generate musical tones and reproduce images based on music data. The loader is a program for downloading music data from the host station. The font data is for displaying lyrics, song titles, and the like, and stores fonts of a plurality of character types, such as Mincho and Gothic. In the RAM 12, a work area is set. A music data file is set in the HDD 17.

The ISDN controller 16 is a controller for communicating with a host station via an ISDN line. The ISDN controller 16 downloads music data and the like from the host station. Further, the ISDN controller 16 has a built-in DMA circuit, and writes the downloaded music data and application program directly to the HDD 17 without passing through the CPU 10.

The remote control receiver 13 receives the infrared signal sent from the remote control 31 and restores the data. The remote controller 31 includes a command switch such as a music selection switch, a numeric key switch, and the like. When the user operates these switches, an infrared signal modulated with a code corresponding to the operation is transmitted. The display panel 14 is provided on the front of the karaoke apparatus, and displays the currently playing music code and the number of reserved music. The panel switch 15 is provided on a front operation unit of the karaoke apparatus, and includes a music code input switch, a key change switch, and the like.

The sound source device 18 has a CPU for performing karaoke.
A tone signal is formed on the basis of the event data input from. The event data is stored in a music track of the music data. Since a plurality of tone tracks are set as shown in FIG. 7, a plurality of event data are also input in parallel. The sound source device 18 receives these data and simultaneously forms a plurality of tone signals. The audio data processing unit 19 forms an audio signal having a specified length and a specified pitch based on audio data that is ADPCM data included in the music data. The audio data is obtained by directly digitizing and storing a signal waveform, such as a back chorus or a model singing sound, which is hardly generated electronically by the sound source device 18.

On the other hand, a singing voice signal input from a singing microphone 27 is amplified by a preamplifier 28 and converted to a digital signal by an A / D converter 29, and then converted to a digital signal for effect.
It is input to the SP 20 and the DSP 30 for audio processing. The harmony information is also input from the CPU 10 to the voice processing DSP 30. The voice processing DSP 30 detects the frequency (period) of the singing voice signal, cuts out the waveform element data, and synthesizes the waveform element data with the frequency of the harmony information to form a harmony voice signal. This harmony audio signal is output to the effect DSP 20.

The effect DSP 20 includes a tone signal formed by the tone generator 18, a sound signal formed by the sound data processing section 19, a singing sound signal converted by the A / D converter into a digital signal, and a harmony sound formed by the sound processing DSP 30. A signal is input. The effect DSP 20 gives effects such as reverb and echo to these input audio signals and tone signals. The type and degree of the effect provided by the effect DSP 20 are controlled based on the event data (DSP control data) of the effect track of the music data. The DSP control data is input to the effect DSP 20 at a predetermined timing by the CPU 10 based on the DSP control sequence program. The tone signal and audio signal to which the effect has been added are converted into analog signals by the D / A converter 21 and then output to the amplifier / speaker 22. The amplifier / speaker 22 amplifies this signal and emits sound.

The character display unit 23 generates character patterns such as song titles and lyrics based on the input character data.
The LD changer 24 reproduces the background video of the corresponding LD based on the input video selection data (chapter number). The video selection data is determined based on the genre data of the karaoke song or the like. The genre data is written in the header of the music data, and is read by the CPU 10 when the karaoke performance starts. CPU
10 determines which background video is to be reproduced based on the genre data, and outputs video selection data specifying the background video to the LD changer 24. The LD changer 24 contains about five (120 scenes) laser disks and can reproduce about 120 scenes of background video. One background video is selected from among them according to the video selection data, and is output as video data. The character pattern and the video data are input to the display control unit 25. The display control unit 25 superimposes these data in superimposition and displays them on the monitor 26.

Here, the configuration of music data used for karaoke performance in the karaoke apparatus will be described with reference to FIGS. FIG. 6 is a diagram showing a configuration of music data. FIGS. 7 and 8 are diagrams showing a detailed configuration of music data.

In FIG. 6, one piece of music data includes a header, a musical tone track, a main melody track, a harmony track, a lyrics track, an audio track, an effect track, and an audio data section.

The header is a portion in which various data relating to the music data are written, and data such as a music title, a genre, a release date, and a music playing time (length) are written therein. The CPU 10 determines a background video to be displayed on the monitor 26 based on the genre data when executing the main sequence program, and transmits a chapter number of the video to the LD changer 24. The method of determining the background image is to select an image of a snowy country in the case of enka on the theme of winter, and to select an image of a foreign country in the case of pops.

As shown in FIGS. 7 and 8, each track from the musical sound track to the effect track is composed of sequence data including a plurality of event data and duration data Δt indicating a time interval between the event data. The event data of each track is read out by the CPU 10 based on a sequence program during a karaoke performance.
The sequence program is executed at a predetermined tempo clock at Δt.
This is a program for reading out and outputting to a predetermined processing unit, assuming that when the time Δt is counted up, it is the readout timing of the event data subsequent thereto.

On the musical sound track, tracks of various parts including a melody track and a rhythm track are formed. The CPU 10 outputs to the tone generator 18 the event data read by the tone sequence program. The sound source device 18 selects a sounding channel based on the channel designation data included in the event data, and executes the event for the sounding channel.

In the main melody track, the main melody of the karaoke song, that is, the sequence data of the melody to be sung by the singer is written. The karaoke apparatus generates a guide melody based on the main melody data. The configuration of the harmony track is the same as that of the main melody track, and the sequence data of the harmony melody of this karaoke song is written. This data is also input from the CPU 10 to the voice processing DSP 30. The voice processing DSP 30 determines the frequency (pitch) of the harmony voice signal based on this data.
To determine.

The lyrics track is a track in which sequence data for displaying lyrics on the monitor 26 is stored. Although this sequence data is not tone data, this track is also described in the MIDI data format in order to unify the implementation and facilitate the work process. The data type is system exclusive
It is a message. In the data description of the lyrics track, usually, one line of lyrics is treated as one piece of lyrics display data. The lyrics display data is composed of character data (character codes and display coordinates of the characters) of one line of lyrics, display time of the lyrics (usually around 30 seconds), and wipe sequence data. The wipe sequence data is sequence data for changing the display color of the lyrics in accordance with the progress of the song. The timing of changing the display color (the time from when the lyrics are displayed) and the change position (coordinates) ) Is data sequentially recorded over the length of one line.

The audio track is a sequence track for specifying the generation timing of the audio data n (n = 1, 2, 3,...) Stored in the audio data section. The voice data section stores human voices such as back chorus and harmony singing that are difficult to synthesize by the sound source device 18.
In the audio track, the audio designation data and the reading interval of the audio designation data, that is, the duration data Δt that designates the timing of outputting the audio data to the audio data processing unit 19 and forming the audio signal, are written. The voice designation data includes a voice data number, pitch data, and volume data. The audio data number is an identification number n of each audio data recorded in the audio data section. The pitch data and the volume data are data indicating the pitch and volume of the audio data to be formed. In other words, a back chorus without words, such as "Ah" or "Wawa Wawa", can be used many times by changing the pitch or volume. The pitch and volume are shifted based on and used repeatedly. The audio data processing unit 19 sets the output level based on the volume data, and sets the interval of the audio signal by changing the reading interval of the audio data based on the interval data.

In the effect track, DSP control data for controlling the effect DSP 20 is written. The effect DSP 20 applies reverberation or other reverberation effects to signals input from the sound source device 18, the audio data processing unit 19, and the audio processing DSP 30. DSP
The control data includes data designating the kind of the effect and data of a change amount thereof.

FIG. 2 is a diagram for explaining the function of the voice processing DSP 30. The voice processing DSP 30 forms a harmony voice signal for the input singing voice signal based on a built-in microprogram, and this microprogram can be represented as shown in FIG.

The singing voice signal input from the microphone 27, amplified by the amplifier 28, and converted into a digital signal by the A / D converter 29 is converted into a period detecting section 40, a peak detecting section 41, and a phoneme detecting section 42 of the voice processing DSP 30. , Are input to the average volume detector 43 and the multiplier 45.

The cycle detecting section 40 detects the cycle T based on the waveform of the input singing voice signal (see FIG. 9A). The cycle detector 40 outputs the detected cycle information to the peak detector 41 and the window function generator 44. The details of the function of the cycle detection unit 40 will be described later with reference to FIGS.

The peak detector 41 detects a local peak in one cycle of the input singing voice signal (see FIG. 9A). The interval of one cycle is determined by the cycle information input from the cycle detector 40. The peak detector 41 outputs the detected peak timing information to the window function generator 44.

The phoneme detector 42 detects a phoneme break based on a level break or a change in frequency component of the input singing voice signal. Here, a phoneme refers to a section in which pronunciation is divided into individual consonants and vowels. In FIG. 9 (B), the lyrics “Akashiya no” are “A” and “K”, respectively.
It consists of five syllables, "shi", "ya", and "no", and these syllables are "a", "k", "a", "sh", "i", and "y".
It can be divided into nine phonemes "a", "n" and "o". Between each syllable, there is a break in which the level decreases, and phonemes are divided based on the fact that the consonant has a non-periodic waveform like a white noise, while the vowel has a periodic waveform. When detecting a break between phonemes, the phoneme detecting unit 42 outputs information indicating that the phoneme is a break to the window function generating unit 44.

The average volume detector 43 detects the average volume by smoothing the amplitude level of the input singing voice signal. The average volume detection unit 43 uses the detected average volume information as the volume control unit 5
Output to 0.

The window function generator 44 outputs a window function as shown in FIG. This window function is output to the multiplier 45. Since the singing voice signal is input to the multiplier 45 as described above, only the window function portion of the singing voice signal is cut out (see FIG. 9C). It is desirable to use a function that is differentially continuous from the start to the end as the window function. If a differentially continuous function is used, noise is not generated at the boundary of the cut even if only a part (one cycle) of the singing voice signal is cut. Therefore, in this DSP 30, sin ² (ωt / 2)
(T = 0 to T: T is one cycle of the singing voice signal). As is apparent from this equation, the length of the window function is one cycle of the singing voice signal. The length of one cycle is given by the cycle information input from the cycle detector 40. Further, the window function generating section 44 repeatedly generates a window function at an appropriate interval of several tens ms to 100 ms. The reason why the window function is generated with a certain time interval is that if the same waveform element data is not continued to some extent, the timbre of the waveform element data will not be recognized by the listener. On the other hand, whenever information indicating a break between phonemes is input from the phoneme detection unit 42, a window function is generated to cut out waveform element data of a new phoneme. This is because the timbre changes completely when the phoneme is switched, so that it follows the change. Also,
The start timing of the window function is controlled so that the peak input from the peak detection unit 41 is located at the center of the window function, that is, the midpoint between the peak and the next peak, that is, the lowest level point. The waveform element data cut out by the window function as described above is obtained by storing the timbre of the singing voice signal, that is, the formant (overtone component) almost as it is.

The window function generating section 44 outputs the information to the effect that the window function is to be generated and its length to the writing control section 47 at the same time as generating the window function. The write control unit 47 inputs a write address that advances in synchronization with the sampling clock (44.1 kHz) from the start to the end of the window function in accordance with this information. By inputting the write address, the waveform element data cut out by the multiplier 45 is stored in the memory 46.

With the above configuration, the memory 46 stores the waveform element data for one cycle of the singing voice signal at that time. By repeatedly reading out the waveform element data at an arbitrary cycle, it is possible to synthesize an audio signal having the fundamental frequency of the arbitrary cycle and having the timbre (overtone configuration) of the singing voice signal. Therefore, by repeatedly reading out the waveform element data from the singing voice signal at a harmony frequency cycle having a consonant frequency relationship such as 3rd or 5th, a harmony voice signal having the same frequency and the same timbre as the singing voice signal is formed. be able to.

The read control of the memory 46 is performed by the read control unit 4.
8 does. Harmony data is input from the CPU 10 to the read control unit 48. This harmony data
This is event data read from the harmony track of the musical sound data. The read control unit 48 repeatedly accesses the memory 46 at the frequency of the harmony data. That is, the waveform element data is repeatedly read out only for the frequency of the harmony data in one second. When the melody of the harmony has a lower frequency than the main melody, the harmony voice signal has a waveform in which the waveform element data is arranged at an interval of T1 longer than the data length T as shown in FIG. Become. When the harmony melody has a higher frequency than the main melody, the harmony sound signal has a waveform element data T2 shorter than the data length T as shown in FIG.
The waveforms are arranged so as to overlap each other at intervals of. Thereby, the fundamental frequency of the harmony audio signal is 1 / T1
And 1 / T2, but since the harmonic components in each waveform element data are stored as they are, a formant similar to the singing voice signal is formed. Further, since the window function is differentially continuous, no noise is generated.

The harmony sound signal formed by repeatedly reading the waveform element data from the memory 46 as described above is input to the multiplier 51 via the switch 49. The multiplier 51 receives volume control data from the volume controller 50. The volume control unit 50 is provided with the average volume detection unit 43.
And average volume information of the singing voice signal, and generates volume control data based on the average volume information. The volume control data is set to, for example, a value of 80% of the average volume information. The harmony sound signal whose volume has been controlled by the multiplier 51 is output to the effect DSP 20. The switch 49 is used when the output is forcibly set to 0 at a break between phrases.

FIG. 3 is a diagram showing the configuration of the cycle detecting section 40. Here, the period detecting unit 40 detects the period of the input singing voice signal. Since the period and the frequency have the same value in a reciprocal relationship, the period detecting unit is equivalent to the frequency detecting unit. is there. That is, the frequency can be easily detected by reversing the detected period. The cycle detection unit 40 includes a noise determination unit 60, a low-pass filter (LPF) 61, and a count calculation unit 62. The noise determination unit 60 determines the periodicity of the zero cross of the singing voice signal, and determines that the signal is a noise signal if the zero cross is repeated in a very short time and irregularly. L
The PF 61 removes a noise signal and a harmonic component and supplies the fundamental frequency of the singing voice signal to the count calculation unit 62. The count calculation unit 62 is a functional unit that counts the zero-cross intervals of the input singing voice signal at the cycle of the sampling clock and weights and averages a plurality of zero-cross intervals. The average number of zero-cross intervals is not constant,
All the time intervals of the zero cross generated in the immediately preceding fixed time are averaged. Thus, in the case of a high frequency, one zero-cross interval is short, so that many zero-cross intervals are averaged, and in the case of a low frequency, one zero-cross interval is long, so that a small zero-cross interval is averaged. This is because, in the case of high frequencies, the discretization error of the sampling clock occupying the zero-cross interval is large, so a large number of averages are required.In the case of low frequencies, the discretization error of the sampling clock occupying the zero-cross interval is small. Consistent with not having to take a large number of averages. When detecting the frequency (period) of the audio signal, about 20 ms is appropriate as the period of the weighted average.

The count operation section 62 has a register as shown in FIG. vflag is a flag indicating that the period (frequency) is detected. Zcol is a register that stores the previous zero-cross timing (free-run count value of the sampling clock). P (0) to P (N-1) are registers capable of storing up to N count values of zero-cross intervals. N is a value larger than the number of data that can be required for the weighted average.

FIG. 5 is a flowchart showing the operation of the cycle detector 40. First, it is determined whether the input signal is a periodic signal having a pitch or a noise signal (s
1). In the case of a noisy signal, since there are many zero crossings in a short time, a noisy signal can be determined using this. For example, if there are 100 or more zero crossings within 10 ms, the signal is determined to be a noise signal. If the input signal is a noise signal, vflag indicating that a periodic signal is input is cleared, 0 is written in the register P (n) of the address n, and the process returns (s12). The above processing is performed by the noise characteristic determination unit 60.

If the signal is not a noise signal, the following processing is performed on an input signal which has passed through a low-pass filter (LPF) 61 and from which noise components and harmonic components have been removed. The following processing is executed by the count calculation unit 62. First, it is determined whether there is a stationary signal that has passed through the LPF 61 (s2). This can be determined based on whether or not the amplitude level of the input signal is larger than a certain value. If the amplitude level is larger than the threshold value, it is determined that there is a signal, and if it is smaller, it is determined that there is no signal. If there is no input signal,
As in the case where the noise signal is input, vflag is cleared, 0 is written to P (n), and the process returns (s1).
2).

On the other hand, if there is a signal, it is determined whether or not a new zero cross has occurred (s3). This zero cross is a negative → positive zero cross, which can be determined by the shift of the sampling value from a negative value to a positive value (FIG. 10).
reference). If there is no zero cross, return without doing anything. If a zero cross occurs, the current vflag
Is checked (s4). If vflag is 0 (reset), it is the first zero cross after it is determined that there is an input signal, and the zero cross interval cannot be measured because there is no previous zero cross. In this case, vflag is set and the current time is written to zcold. As the current time, the count value of the sampling clock counted in free run is used. Then, an invalid value 0 is returned as the fixed frequency information Pave (s5). This Pav
Although e is output to the window function generator 44 as the period information T, when the value is 0, the window function generator 44 determines that the period has not been detected and does not generate the window function. That is, no harmony audio signal is formed.

If vflag has already been set in s4, the address n of the memory storing the count value of the zero-cross interval is incremented by one. When the address becomes N as a result of the count-up, the address is set to 0. The difference between the current time and the previous zero-cross time zcol is calculated, and the calculated difference is used as the zero-cross interval in the register P (n)
Write to. Thereafter, the current time is written to zold (s6).

Next, weighted averaging is performed. First, Psum = 0, Asum = 0, c = 0, and i = n are set as initial values (s7). Here, Psum is the sum of the zero-cross intervals, Asum is the sum of the weighting coefficients for weighted averaging, c is the number of added zero-cross intervals, and i is the memory address. First, it is checked whether P (i) = 0. 0
Indicates that there is a period during which no signal is present or a period during which a noise signal is input within the weighted average calculation period. In such a case, it is determined that the period information Pave with high accuracy cannot be obtained even by taking a weighted average, and the latest count value P (n) of the zero-cross interval is output as the determined frequency information Pave (period T) (s1).
3) Return.

If P (i) is not 0, the following calculation is performed.

Psum = Psum + a (c) × P (i) Asum = Asum + a (c) c = c + 1 i = i−1 However, if i <0, the operation above i = N−1 is the period when Psum is the weighted average. The processing is repeatedly executed until L (sampling clock count value of 20 ms) is exceeded (s10). At least once during this period, P (i)
If = 0, the process proceeds to s13 in the judgment of s8. P (i) = 0
Is not found once, the sum Psum of the weighted zero-cross intervals and the sum Asum of the weighting coefficients during the weighted average period are calculated. That is, Psum = a (0) × P (n) + a (1) × P (n−1) + a (2) ×
P (n−2) +... Note that if the period of the weighted average is TL (= 20 ms), then L = TL × TS. Here, TS is a sampling clock frequency. Finally, by performing Pave = Psum / Asum, it is possible to calculate the period information Pave which is a weighted average value of the zero-cross interval (s11). The determined frequency information Pave is output to the window function generator 44 as the period information T and used for pitch conversion of the singing voice signal.

When there is a period in which there is no signal within the weighted average period, such as when the signal rises, in this example, the weighted average is stopped and the latest zero-cross interval count value P
Although (n) is output, weighted averaging may be performed by extracting only a certain period of the periodic signal within the weighted average period.

The weighting coefficient a (c) is defined as follows: a (0)> a (1)> a (2)>... What is necessary is just to make the frequency of the input signal follow up the change. Alternatively, the characteristics of the input signal may be used to change dynamically. That is, when a (k) = α ** k, when the frequency of the input signal is stable, α is set to a large value (a value close to 1) to approach the simple average, and the frequency of the input signal varies. In this case, α may be automatically changed to a small value according to the change.

More specifically, before performing the weighted averaging, the zero-crossing interval P measured during the same period as the weighted averaging is performed.
The ratio between the maximum value and the minimum value of (n) is calculated. When the ratio is small, the frequency of the input signal is stable. When the ratio is large, the frequency of the input signal is changing or stable. It is determined that the value has not been set, and the value of α may be determined accordingly.

For example, if R (= Pmax / Pmin), if R ≦ 1.03, α = 1.0 1.03 <R ≦ 1.06, and if α ≦ 0.95 1.06 <R ≦ 1, If 0.09, α = 0.90 1.09 <R ≦ 1.12, if α = 0.85 1.12 <R, then α = 0.80. By determining the weighting coefficient according to R in this manner, when the frequency of the input signal is stable, it becomes close to a simple average, and the calculation accuracy of the frequency (period) can be increased. Further, when the frequency of the input signal is fluctuating, the frequency tracking performance is required rather than the detection accuracy. Therefore, the weight of data close to the present is increased, and the data can be tracked well. Note that when α = 1.0, all weighting factors are equal, which is the same as the simple average.

In the above embodiment, the detection of the frequency (period) of the singing voice signal has been described. However, this may be used for detecting the frequency of other voice frequency signals such as musical instruments.

[0054]

According to the first aspect of the present invention, by averaging a plurality of count values, the count value and the fraction obtained by truncating the fraction from the true frequency (period) of the input audio signal are obtained. Since the count value obtained by rounding up is cancelled, the error is reduced and a highly accurate frequency (cycle) can be detected. In addition, when a plurality of count values are weighted and averaged, by increasing the weight of the new count value with respect to the old count value, it is possible to improve the followability when the frequency of the input audio signal fluctuates. .

According to the second aspect of the present invention, when a signal is not input at the start of sound generation or due to interruption of sound generation, the frequency is unstable in such a situation.
By using one count value instead of the average value, useless calculation can be eliminated.

According to the third aspect of the present invention, even if a signal is not input for some reason, the output of the definite frequency information can be continued by averaging only the count value of the input signal. .

According to the fourth aspect of the present invention, when the variation of the plurality of count values is small, that is, when the frequency variation of the input signal is small, the rate of change of the weighting coefficient is reduced to approach the simple average, thereby obtaining the average value. Increase accuracy. When the frequency fluctuation of the input signal is large, the rate of change of the weighting coefficient is increased to increase the weight of the new count value. As a result, the followability of the determined frequency value to the input signal is improved.

[Brief description of the drawings]

FIG. 1 is a block diagram of a karaoke apparatus according to an embodiment of the present invention.

FIG. 2 is a diagram showing a configuration of a voice processing DSP of the karaoke apparatus.

FIG. 3 is a diagram showing a configuration of a period detecting unit which is a part of the DSP for audio processing;

FIG. 4 is a diagram showing a configuration of a register provided in the same period detection unit.

FIG. 5 is a flowchart showing the operation of the same period detection unit.

FIG. 6 is a diagram showing a configuration of music data used in the karaoke apparatus.

FIG. 7 is a diagram showing a configuration of music data used in the karaoke apparatus.

FIG. 8 is a diagram showing a configuration of music data used in the karaoke apparatus.

FIG. 9 is a diagram for explaining a method of extracting waveform element data from a singing voice signal.

FIG. 10 is a diagram illustrating a method of forming a harmony audio signal.

FIG. 11 is a diagram showing the principle of period detection.

[Explanation of symbols]

Reference Signs List 30 DSP for voice processing, 40 Period detector, 44 Window function generator, 60 Noise determiner, 61 Low-pass filter, 62 Count calculator

────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] October 17, 1996

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] Claim 4

[Correction method] Change

[Correction contents]

[Procedure amendment 2]

[Document name to be amended] Statement

[Correction target item name] 0010

[Correction method] Change

[Correction contents]

According to a fourth aspect of the present invention, the weighted averaging means reduces a rate of change of a weighting coefficient from an old count value to a new count value when a variation in the plurality of count values is small, and fluctuations in the count value is large
Characterized in that it comprises a means for increasing the rate of change of the weighting coefficients advance are.

[Procedure amendment 3]

[Document name to be amended] Statement

[Correction target item name] 0045

[Correction method] Change

[Correction contents]

If vflag has already been set in s4, the address n of the memory storing the count value of the zero-cross interval is incremented by one. When the address becomes N as a result of the count-up, the address is set to 0. The difference between the current time and the previous zero-cross time zcol is calculated, and the calculated difference is used as the zero-cross interval in the register P (n).
Write to. Thereafter, write the current time to the z c old (s6).

[Procedure amendment 4]

[Document name to be amended] Drawing

[Correction target item name] Fig. 5

[Correction method] Change

[Correction contents]

FIG. 5

Claims (4)

[Claims] 1. A frequency detecting device comprising a counting means for inputting an audio signal digitized by a predetermined sampling clock and measuring a period of the audio signal by a count value of the sampling clock. A frequency detecting apparatus comprising: weighted averaging means for repeatedly executing a counting means and averaging a plurality of count values obtained thereby to determine the frequency of the audio signal. 2. The apparatus according to claim 1, further comprising: means for invalidating the weighted averaging means and determining a frequency from the latest count value of the counting means when there is a period during which no audio signal is input within the predetermined period. 3. The frequency detection device according to 1. 3. The weighted averaging means is means for performing weighted averaging only during a period during which an audio signal is input, when there is a period during which no audio signal is input within the predetermined period.
3. The frequency detection device according to 1. 4. The weighted averaging means reduces a change rate of a weighting coefficient from an old count value to a new count value when the variation of the plurality of count values is small, and sets the weighted average value when the variation of the plurality of count values is small. 2. The frequency detection device according to claim 1, further comprising means for increasing a change rate of the weighting coefficient.