JPS60113295A - Musical sound generator by voice input - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] [44/Technical Field of the Invention] The present invention relates to a musical tone generation device powered by a single voice. [Prior art and its problems] Recently, the voice input from a microphone, which is the singing voice of a single person, has been developed. A musical sound generating device using voice input has been considered, which extracts a fundamental frequency (pitch) from a musical instrument and generates a musical instrument sound according to the extracted fundamental frequency. However,
If you want to extract the fundamental frequency, or pitch, of the audio input.

Conventionally, the pitch is generally determined by extracting input data through a low-pass filter, sampling it at a low sampling frequency, and then performing correlation calculations.

However, in the above conventional pitch extraction method for input speech, it is necessary to calculate correlations over a wide range of input data, and as a result, when attempting to extract pitch in real time, the extraction resolution becomes coarse. There is. For example, the pip frequency of a human singing voice is 80 to 8.
00) (z, and the sampling frequency is 32,768
In the case of HZ, the delay time τ for pitch search is 40 to 4
00 over a wide range. 7L/-mu 4 to calculate the correlation
If Q Q point = 12 m sec, it is impossible to calculate the correlation over the entire range within this time, and therefore the pitch extraction resolution becomes coarse. If the extraction resolution of the audio pitch is coarse, there will be many erroneous pitch extractions, and subtle musical tones such as vibrato and voltament cannot be generated.

[Purpose of the invention]

The present invention has been made in view of the above points.

When extracting pitches in real time, the extraction resolution can be increased to the maximum determined by the sampling frequency, and the erroneous extraction of pitches can be reduced, as well as a musical sound generation device using voice input that can generate subtle musical sounds. The purpose is to provide.

[Key points of the invention]

The present invention divides the pitch extraction analysis frame into a frame in which correlation is calculated over a wide range of delays with coarse resolution, and a frame in which correlation is calculated for a wide range of delays with coarse resolution, and a delay time in a narrow range with maximum resolution based on the delay time τ for the maximum correlation value extracted in that frame. [Embodiment of the Invention] An embodiment of the present invention will be described with reference to the drawings below. explain. First, the general structure of the entire musical tone generator will be explained with reference to FIG. In the figure, reference numeral 11 is a microphone, and audio input from this microphone 1 is sent to a preprocessing section 13 in a pitch extraction section 12. The preprocessing section 13 is composed of, for example, an amplifier equipped with an AGC circuit, a low-pass filter circuit, etc., and outputs the processed data to the A/D conversion section 14. This A/D converter 14 has a frequency of 1, for example, 32 KHz (
The input signal is sampled at a frequency of 32,768 Hz), converted into a 6-pit digital signal, and output to the ternary correlation section 15. This ternary correlation unit 15 compares the input data with a preset threshold level.

Convert to ternary data of r+xJ, rOJ, and rtJ. The data quantized by the ternary quantization unit 15 is sent to the voice/silence determination unit 16 and the autocorrelation calculation unit 17. The autocorrelation calculating section 17 calculates a correlation based on the input data, and outputs it to the maximum value determining section 18, although the details will be described later.

This maximum value determination unit 18 determines that the correlation value is the maximum T when τ is other than 0.
Then, Z・delay time τ=T is determined to be the pitch. and,
The determination result of this maximum value determination unit 18 is determined by the pitch extraction unit 12.
It is sent to the CPU 19 as an output. Also, the above sound/
The silence determination unit 16 counts the signal from the ternary quantization unit 15 to determine its power, determines voice/silence based on the power, and outputs it to the autocorrelation calculation unit 17 and CPU 19.

The CPUW 9 controls the musical piece generation unit 2 to generate musical instrument sounds with a period corresponding to the pitch according to the input data.
Control 0.

output from the speaker 21.

Next, the above autocorrelation “(For the details of calculation section 17, please refer to the second section.
The data sent from the ternary quantization unit 15, which will be explained with reference to the figure, is stored in the R, kM 31 in the autocorrelation calculation unit 17.
Both ζ input to the input signal ζ are input to the voice/silence determining section 16 . The sound/silence determining section 16 is composed of a counter 16a and a judgment circuit 16b. In step 16b, presence/absence of sound is determined from the power.

The sound/non-sound determination signal outputted from the determination circuit 16b is sent to the CPUz9 as well as to the timing control section 32 in the autocorrelation calculation section 17. The timing control unit 32 includes a clock pulse φ8, a frame signal φf, a load signal φL, timing signals φB, φG, read/write commands R/W, etc. for performing gate control, etc.
Generate R/V' etc. The frame signal φf outputted from the timing control section 32 is, for example, 4
It is sent to the frame counter 33 in base 00. This frame counter 33 is.

A count from 0 to 399 is performed using the frame signal φf, and the count contents are outputted to the address combination section 34 and also to the RAMJJ via the gate circuit 01 whose gate is controlled by the timing signal φB. Furthermore, the clock pulse φ1 outputted from the timing control section 32 is inputted as a count pulse to, for example, an octal counter 35. Also, this counter 35
is given the above-mentioned timing signal φ○ as a clear signal, and also given data of "τ-m" from the initial value setting circuit 36. In this case, τ is the maximum pitch delay value extracted in the phase frame. Yes 1m is a constant, for example "3
9”. Then, the initial value setting 1 circuit 36
“τ−m” outputted from the timing control unit 32
Low toy No. 3 φL (2 frame counter 3
When the above-mentioned "τ-m" is loaded, this frame counter 35, which is loaded into the frame counter 5, counts "0 to 79", that is, counts 1τ-n1 to τ-m+79J, and outputs the count output. At the same time as being input to the ROMJ7, the address calculation unit 34 and RO are input via the gate circuit G2.
Input to MJ8. The above ROM37 contains I'l
Delay time τ0 to τ1 for l frame. The stored contents are read out according to the count value of the counter 35 and sent to the address calculation unit 34 and ROM 38 via the gate circuit G3. This ROM38+- has the above-mentioned "τ0 ~ τ, J, r τ -m ~ τ -m-1
The window function for −79Jl2 is stored,
The stored data is transferred to the addition/subtraction circuit 4 via the logic gate 39.
It is input to input terminal a of 0. The gate circuit G is gate-controlled by the timing signal φA, and the gate circuit G is gate-controlled by the timing signal φ applied via the input terminal 41. , the count value j of the counter 33 and the data τ1 sent via the gate circuit 02 or 0. are added, and the addition result is used as the timing signal φ.
RA via gate circuit O3 gate-controlled by ,
Output to MJz.

Further, this RAM 31 is given read/write 4M numbers R, 'W, and 1 from the timing control section 32.

The above RAMJJ contains −, ternary quantized data according to the above address data, for example I) j +o to Dj-1-sa
y is written. The data read from RAMJJ is sent directly and via the delay circuit 42 to the multiplication circuit 43. This multiplication circuit 43 multiplies the input data and sends the multiplication result data to the logic gate 39. A sign pit indicating positive or negative is input to the subtraction %1 control terminal sub of the addition/subtraction circuit 40. The output of the input circuit 40 is sent to the RAM 45. Address data is input from the address counter 46 to this RAM 45, and a sleep/out/write command R/W 2 is input from the evening/f mink control section 32. The address counter 46 is cleared by the timing signal φC, and performs a count-up operation by the clock pulse φ1. ”1 memory 45 stores the 1st correlation value R6(τ
)~■<77(τ).

The stored data is input to the input sentence 1.1 child of the addition circuit 40, and is also sent to the maximum value detection '&1 BIS 47. The maximum value detection unit 47 detects the delay time τ at the maximum value from the input autocorrelation (r+a R6(τ) to R,,(τ)), and outputs it to the initial value setting circuit 36 and to the CPUJ9. Output.

Next, 11 works of the above embodiment will be explained. Microphone 1
1, the preprocessing section 13 removes noise and the like outside the audio signal band, and amplifies the audio data to a specific amplitude value.

This preprocessed input data is then converted into digital 4J data by the A/D converter 14 and sent to the ternary quantizer 15 . This ternary quantization unit 15 compares the input data with a preset threshold level of 2, and calculates l+lJ.

rOJ I Convert to r-IJ ternary data. As shown in the primary table, this ternary data consists of a total of 2 bits, the 'F3 bit and the data bit, and is expressed as [+lJ or (1)].
When , the code pit is "0".

r-1'', the shrine number bit becomes Ill. Also, the data bit indicates its absolute value, and the 3(++'j data is "
The height of ±l” becomes “0” when roJ.

Table 1 However, the two sets of quantization by the ternary quantization unit 15 are n>'L, 30.5 tt sec I 32 KH
The signal is sent to the autocorrelation calculation unit 17 and the voice/silence determination unit 16 at a period of 8.z). The voice/silence determination unit 16 calculates the power of the input data to determine voice/silence, and outputs the determination result to the timing control unit 32 in the CP (JI 9 and autocorrelation 81 calculation unit 17). Autocorrelation 81 Calculation I7
is 781i for each data when the data is input.
Calculate the correlation for the delay τ of d(, and calculate the autocorrelation for 398 data as 1((Ti)=”3L)fJ)・1Jfj+4i)
・w(fi) −=−(11j=0 R(Ti) is an autocorrelation value, D(jl ternary quantized data, W(τl) is a window function (weighting function).

The operation of the autocorrelation calculation section carried out in the autocorrelation calculation section 17, the details of which are shown in FIG. 2, will be explained below. First, the gate of the gate circuit G is opened by the timing signal φB, the address of RAMJJ is specified by the output of the frame counter 33, and the ball value graduation data D (jl) is hidden in the specified address.
Quantized data D (jl) is read out from M3) and latched in the delay circuit 42. Then, the timing signal φ0 is output, the gate of the gate circuit 04 is opened, and the address of the RAM 31 is determined by the output of the address calculation unit 34. Specified.At this time.

The address calculation unit 34 uses the output j of the frame counter 33
and the output τ1 of the ROM 5y are added and output. Therefore, the address of RAMJJ is specified by the addition output (j+τi), and the next quantized data D(j+τi) is written.

Then, data D(i+τ
t) is read out to the multiplication circuit 43.

Multiplication is performed with the data D(j) held in the delay circuit 42, that is, multiplication by D(jl·D(j+τi)), and the data output from the multiplication circuit 42 is sent to the logic gate 39. Multiplying is performed between the window value sent to 438 and the window value read from 438. In this case, the output of the multiplier circuit 43 is 3i1〆i (l l
, 00, 01), so the data part is "engineering".
When , input from ROM38 (direction 1) p
Output from gate 39.

When it is "0", the output of the logic gate 39 is inhibited.

The output of the logic gate 39 is then sent to an addition/subtraction circuit 40.
Human power ☆; 1 (input to 1 child 21. This addition/subtraction circuit 40 adds when the signal pick 1 sent from the 1 multiplication circuit 43 is "()", and when it is "1", lA'!, ′#
! 111 works are now available. And the above addition/subtraction 1111
1! '1 < 40 through the output delay circuit 44
At the same time as inputting N into A k*145, the stored data is read out to the input terminal of the addition/subtraction circuit 40, and addition/subtraction is performed between the next data and the accumulated value. Hereinafter, the above calculation will be performed using the count value j-(]~39 of the frame counter 33.
7 is performed on the data of [398j, and the correlation calculation for l frames is completed. Then, at the timing of %J2398, the maximum value detection unit 47 detects the maximum autocorrelation value R (τ1) from the data stored in the RAM 45, and the delay τ1 is detected by the CPU 79 and the initial value setting circuit 3.
Output to 7.

Therefore, as shown in Figure 4, the R1 calculation of the autocorrelation value is based on the coarse frame in which a correlation is expressed over a wide range of delays with coarse resolution, and the delay relative to the maximum correlation value extracted in the Qth 11th frame. A narrow range delay with maximum resolution based on the time τ is carried out in a two-stage configuration of Wi frames with 51 correlations between them.

The only difference is the address H1'# and the value of Ti sent to section 34. The above delay! 1l-1Jτi is output according to the counting operation of the counter 35. The counter 35 operates according to the clock pulse φ,
As shown in FIG. 3, 80 counts are made for each frame data. The counter 35 is cleared by the timing signal φC at the final timing of each frame, and starts a new counting operation in the first frame. In this case, if it is a coarse frame, -counter 35 is.

After being cleared by the timing signal φC, the initial value "0"
'' to ``79''. At this time, depending on the count value of the counter 35, ter< o Ms 7
address is specified. In this 1<0M5v, the delay time τi (τ.~r77) for the coarse frame is stored, and the stored contents are sent to the address calculation section 34 via the gate circuit 03 according to the count-up operation of the counter 35. Read out. This address calculation unit 34 is RO
The delay time τl read from M37 and the output of the counter 33 are added, and the address of the RAM 31 is designated by the addition result rj+Tij. In addition, the above ROMa
The address of the find function 38 is specified by the output τi of v. And the delay time τ. When the process from ~τ77 is completed and the count value of the counter 35 reaches "78" as shown in FIG.
W is given, and the contents of RAM3 are read out.

Next, when the value of the counter 35 counts up to 179'', the timing control unit 32 outputs a timing signal bar at the final timing, and the counter 35 is cleared. ” In the rough frame as described in 1.

A calculation is performed to calculate the above autocorrelation value according to the delay time τ1 stored in 'i<0M3v.

When the processing for the coarse frame is completed.

Dense frames are selectively displayed according to the state of silent/absent frames at that time. That is, as shown in FIG. 4, the first sound frame that changes from a silent frame to a sound frame is a coarse frame, and then becomes a dense frame. After that, as long as the sound frame continues, coarse frames and fine frames are alternately repeated Z. And in silent frames.

However, in the case of a dense frame, the load signal φL is output from the timing control section 32 and the initial value setting circuit 36 outputs [τ-m j
The value of is loaded into the counter 35. The initial value setting circuit 36 reduces the delay time τ at the maximum +14 function value obtained in the previous sum frame by a constant value m, and the subtraction result is “
Z・ outputs τ-1nJ as the initial value. When the counter 35 is loaded with the value of [°τ-ml], the counter 35 uses this value as an initial value and then counts r+80J using the clock pulse φ. Now, for example, the delay time r+=rs6 extracted from the set frame, and "m J
If the value of is set to "39", then "r-m=
186-39=I47J, and the value of 147J is loaded into the counter 35. Therefore.

The counter 35 is "r226j with j47J as the initial value.
It counts up to r147J ~ "224"
o) τi of 78 buildings is sent to the address calculation unit 34 and ROM 5s via the gate circuit 02.・This allows the pitch extracted in the previous coarse frame, that is, the delay time τ
Correlations for 78 types of delays are calculated with i as the center.

Then, the delay time τ with respect to the maximum correlation value at that time is CP
Sent to U19.

The C'PU 19 calculates the delay time τ with respect to the maximum correlation value sent from the maximum value determination unit 18 and the sound/silence determination unit 16.
Control the musical tone generator 20 according to the sound/silence signal from 1. A musical instrument sound having a period corresponding to the detected pitch is output from a speaker 2.

〔Effect of the invention〕

As described above, according to the present invention, correlations are calculated over a wide range of delays with coarse resolution, and the correlation is calculated with maximum resolution based on the delay time τ for the maximum correlation value extracted in that frame. Since it has a two-stage configuration of frames that measure the correlation over a narrow range of delay times, when extracting pitches in real time, the extraction resolution can be increased to the maximum determined by the sampling frequency. In addition to preventing erroneous extraction of pitches, it is also possible to generate delicate musical tones such as vibrato and voltamento, which would not be possible if pitch extraction were based only on scale correspondence.

[Brief explanation of the drawing]

The drawings show one embodiment of the present invention, and FIG.
Figure 2 is a block diagram showing the details of the autocorrelation calculation section in Figure 1, Figure 3 is a diagram showing the overall schematic configuration, Figure 3 is a block diagram showing the details of the autocorrelation calculation section in Figure 1,
The figure is a timing chart for explaining the operation. 11°°゛microphone, 12...pitch extraction section. 16...Sound/silent piece determination section 12...Autocorrelation calculation section, 31...RAJ33 for storing ternary lattice data
...Soleno counter, 34...Address calculation section,
35...Counter, 36...Initial value setting circuit. 37...ROM for storing set frame delay time, s8...
- ROM for storing the number of window gates, 39...logic gate, 4o...addition circuit, 4.2.44...delay circuit, 45...correlation value storage hI RAM.

Claims (1)

[Claims] Voice pitch information is output from voice input information. In a musical tone generating device that generates a musical tone with a local period corresponding to this voice pitch information, when extracting the voice pitch information, a means for performing correlation H1 on scales in the σ range and extracting pip information for the maximum correlation value. and a means for extracting pitch information by performing a correlation meter p with high resolution on a spherical range around the pitch information extracted by this means. .