JPS60158490A - Musical tone generator with 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] [Technical Field of the Invention] The present invention relates to an electronic musical instrument that generates musical tones by extracting voice pitch data, and particularly relates to an electronic musical instrument that generates musical tones by extracting voice pitch data, and particularly relates to an electronic musical instrument that generates musical tones by extracting voice pitch data, and particularly relates to an electronic musical instrument that generates musical tones by extracting voice pitch data, and particularly relates to an electronic musical instrument that generates musical tones by extracting voice pitch data, and particularly relates to an electronic musical instrument that generates musical tones by extracting voice pitch data. The present invention relates to a musical tone generator.

[Prior art]

Advances in electronic technology have made it possible to generate musical sound waveforms electronically and to produce musical tones through speakers. This aforementioned device is generally called an electronic musical instrument. Generally, electronic musical instruments use keys to select the musical tones to be generated, and furthermore, lever switches or the like can be used to specify and generate piano tones or the tones of other musical instruments.

[Problems with conventional technology]

As mentioned above, this electronic musical instrument uses keys to select musical tones, so in order to play it, it is necessary to practice piano skills and how to operate the keys. In other words, it was not possible to simply accompany someone's song, and only those who had the skills to operate electronic musical instruments could do such a thing.

That is, conventional electronic musical instruments operated by keys have a problem in that they cannot be easily played by anyone.

[Object 1 of the Invention] The present invention solves the above-mentioned problems, and provides an electronic musical instrument that can be easily created, played, and accompanied by one without having to have masterful electronic musical instrument technology. In particular, an object of the present invention is to provide a device for generating musical tones by manual voice input for generating appropriate musical tones from voice input data.

[Structure of the invention]

According to the present invention, the above objects include detecting means for detecting the temporal position of at least the start and end portions of a voice; an extraction stage for outputting pitch data of the voice;
a plurality of machining means for applying different pressures to the pitch data; and means for sequentially selecting L pitch data of the plurality of machining means corresponding to each of the detected temporal positions;
The present invention is characterized in that it has a musical tone generation means for generating musical tones based on the output from the selection means and the musical tone output from the selection means.

[Embodiments of the invention]

In the present invention, instead of selecting the scale and timing of a musical note using a key as in the past, the scale and timing of a musical note can be adjusted by one person or voice.

FIG. 1 is a circuit diagram of an embodiment of the present invention.

This figure shows the configuration of an electronic musical instrument that generates musical tones in response to audio input. Microphone 1 that converts audio into electrical signals Bon 1
The output is applied to the preprocessing section 2. The output of the preprocessing section 2 is connected to the pinch extraction section 3. The output of the pitch extractor 3 is applied to the p'1senosa((/])U)J via the latch 4.

Output υ of Brosesori 5, 1 pitch extraction 111 part 3. Storage section 6. Error removal section 7. Moving average calculation unit 8. It is connected to the flag creation section 9.

The output of the storage section 6 is applied to the error removal section 7 and the pitch data control section IO, and the output of the chorus removal section 7 is applied to the moving average calculation section 8 and the pitch data control section 10, respectively. The output of the moving average calculation section 8 is connected to a binochitator control section 10 and a co-generator L--ta 11.

The output of the pitch data control g+++o is also connected to the low l shifter 11. The output of the yaw 1 generator 11 is connected to a musical tone generating section 12, and the output of the musical tone generating section 12 is connected to a speaker 13 which converts the electric signal to 75 degrees centigrade.

The audio signal converted into electric signal by the microbone 1 is applied to a preprocessing section 2, where it is preprocessed. In this preprocessing, first, signals outside the band are removed using, for example, a 1 corpus filter (1° PF).

This out-of-band removal is done to prevent malfunctions in the pinot extraction performed by the next stage. ].

Furthermore, the second pre-processing is to amplify the input audio signal, from which the outside of the 11 range has been removed, to a specific amplitude value by means of an automatic knock-in con-recall circuit.

Third, the input audio signal having the above-mentioned specific amplitude value is converted into digital data by an analog/digital conversion circuit. The above-mentioned amplification to achieve the specific amplitude value is performed in order to make the output high 1-loss of this analog 1 cog/digital conversion circuit effective.

The digital audio signal processed by the preprocessing section 2 is then applied to the pinch extraction section 3. The pitch extractor 3 extracts the pitch of the fundamental frequency of the human voice signal under the control of a processor (CPIJ) 5. The pitch extracting section 3 uses 2, for example, Japanese Patent Application No. 58-31284, Japanese Patent Application No. 58-31285.
The details are described in the respective publications of No. 58-31286 and Japanese Patent Application No. 58-31286.
It has a value correlation processing circuit, and performs ), -ζ pitch extraction on these circuits. The data quantization circuit υJ sequentially obtains the maximum and minimum values of the audio digital data at -2 over a specific period of time, and uses the maximum and minimum values to perform ternarization (for example). ＼Determine. Then, the input signal is ternarized using the threshold value. The scale extraction circuit is a circuit that delays audio ternary data that has been ternarized using the audio digital take by -U corresponding to the scale to be extracted, and multiplies the delayed data by the input signal. This circuit provides time correlation. This data simply represents the correlation in terms of time between the delayed data and the input data, and is further processed to perform extractions. The above-mentioned ternary correlation processing circuit performs the extraction. The ternary correlation processing circuit is a circuit that performs window processing etc. for pitch extraction. That is, the above-mentioned scale extraction circuit outputs a plurality of data corresponding to the scale to be extracted.-1 This ternary correlation processing circuit performs window processing on the plurality of data.
In other words, it is related to the musical scale, and it is 2 and then the weight is +3.
Furthermore, 9.1 is accumulated for a fixed time (one frame).

Then, find the maximum value among the accumulated values (accumulated values corresponding to the delay time).

The pitch is determined based on the delay time corresponding to the maximum value and output. This output is data related to the pitch of the fundamental wave of the input voice (including harmonics).

The pitch data output from the pitch extraction section 3 described above is sent to the latch 4. The pitch data is stored in the storage unit 6 via the pitch control unit 11 through the pitch control unit 5.
While being transferred to 0, the error goods are transferred to the central part 7-. The error removing unit 7 uses the data stored in the storage unit 6 to detect a change in specific pitch data, especially a change of one octave or more in one frame Q, and when that change is detected. There is a circuit 0 where the data is the same as the data before it is changed. It should be noted that there may be sudden changes in the pitch, but in error removal n137, if the pitch take changes suddenly for two consecutive frames or more.

'', operation 4.1 is not performed, and at this time, the data changes with a delay of one frame. The error removing unit 7 removes frame j.

The data from which errors have been removed are transferred to the pinpoint index control section 10 and averaged at the moving average calculation section 8. Then, the output obtained by performing a plurality of averaging processes is output. That is, each pin inputted from the error removal unit 7 as the first 111 force.

An average is performed on the current data over the past four frames L--A including the current one, and a result is output. In addition, as a second output, the past 6 solenoids 11 including the current one for each pitch take, and the previous summation average output are appropriately weighted and averaged. Output. For example, the past G frame and the previous addition average output value are 2
The multiplied values are accumulated and averaged, resulting in an 8-frame J/moving average output with hysteresis.

These processes are combined with shift 1L and sister, which sequentially stores the pitch take input to frame iυ! :3. u
It is carried out by Nato.

In this way, the i17 equalized data G; I shift J is output from the average calculating section 8, and the pitch take control t:++
1. .

Is there any processing required to generate musical tones? ).

In other words, it is a circuit that detects and removes malfunctions of nIJ and error removal 1) 1 (7 is 1 so, 1, y of pitch extraction by external wholesaler 8 sounds, etc.), and eliminates it in a pinch. Some remove large step changes in data as a result of extraction malfunctions. Then, the moving average calculation unit 8 appropriately stabilizes changes in the pitch taker due to a slight change in pitch erroneous extraction due to strange pitch changes in the input audio, etc. (by performing multiple averaging processes). In particular, it is a circuit that allows the above hysteresis (J8 Frame J, the fact that the extracted data for each frame fluctuates finely with a width of about one scale around a specific scale that is electrostatically extracted by the moving average. Preventing.

The human voice is not limited to singing; it also makes consonants and vowels.

There are many fundamental wave pinches in vowels, but there are fewer pinches in consonants. For this reason.

While producing a consonant (for example, while producing °'S' when producing "SA"), a pinch of the fundamental wave may be erroneously extracted. Pinch data eliminates these malfunctions of musical tones caused by erroneous extraction and makes them match the scale of the input voice.

Control 10. This control unit uses data such as audio power and audio input detection to perform the following 1. The data stored in section a 6, the output take of error removal section 7, or the output data of moving average calculation section 8 are selected and output.

The yaw 1 generator 11 converts the output of the pinch control section 10 into a musical tone to be generated in the musical tone generating section 12.
This circuit converts the data into musical sound data.

The above-mentioned binoji extraction section 3. Storage section 6. Error removal section 7.
The moving average calculation unit 8 is controlled by the processor 5. Further, the pitch data control unit 10° code generator 11 is controlled by the processor 5 via the flag generation unit 9. In other words, a flag or the like is set in the flag creation section 9 by the processor 5, and the aforementioned pinoko data control section 10. The motor 1 and generator 11 operate.

The output of the code generator 11, ie, the converted data, is applied to the musical tone generator 12, and specifies the musical tone to be output by the speaker 13. In other words, the musical tone generator 12
Then, the second generator 11 generates a musical tone electrical signal (analog 1 cog) specified by the conversion data of the generator 11 and applies it to the speaker 13. As a result, the speaker 13 generates a musical tone corresponding to the human voice.

The circuit shown in FIG. 1 represents the configuration of an electronic musical instrument that generates musical tones in response to audio input.

This electronic musical instrument is made possible by the circuitry described in more detail below.

The invention particularly relates to the pitch data control section 10 of the embodiment of the invention shown in FIG.

Below is the pitch data system fffll! Jl≦IO will be explained in detail.

At the beginning of a human voice's utterance, even if an attempt is made to pronounce it at a specific pitch, the pitch is often unstable. Moreover, as mentioned above, in consonant parts, the pinch is unclear, so erroneous extraction is common.

On the other hand, the hysteresis moving average output in the moving average calculating section 8 is effective for stabilizing pisochitake for a steady scale, but is not effective for a portion with large changes at the beginning.

Furthermore, the output of the error removing section 7 can appropriately remove errors from one frame of pitch data and can follow sudden changes in musical scale, but cannot follow irregular changes at the beginning of the audio. . Furthermore, the 4-frame moving average output from the moving average calculating section 8 has a property intermediate between the above two outputs.

Therefore, for example, in the first 1 to 3 frames after the start of nine utterances, the unprocessed pitch data from the storage section 6 is output as is to the court generator 11.

For the next 4 to 7 frames, pitch data is outputted through the error removing section 7.

Furthermore, the moving average calculation unit 8
Outputs the 4-frame moving average output from .

Then, for the 12th frame and the first step, the 8-frame moving average output with hysteresis of the calculation unit 8 is outputted. As the vocalization progresses from the beginning to the steady state, suitably processed pitch data is selectively output to the "~I' generator 11.

In addition, when one person pronounces, for example, "patoso", when moving from the steady part of "t'" to "'so", there is a consonant at the beginning of "so", so the pitch data is interrupted at that part. It may be stored away. However, when actually making a musical note sound like padoso゛゛, 1"
and ``so'' without interruption.

It is desirable that it connects smoothly. When a consonant part is detected at that point, the pitch data immediately before the consonant part is output as a continuation of the vowel part of the 1. to 3rd frames J. following that part point. After that, in the transition from the 4th to the 7th frame, the pitch may be interrupted in the consonant part and the change in pinch may be large, so similar to the process of adding 1 to the br; The data is selectively output from the code generator 11-.I! Ru.

In order to realize the operations described in - by the pitch data control section 10, various flags are created in the flag creation section 9 (FIG. 1). The various flags created are 1 power flag PF, correlation value flag CF, key-on flag K OF,
Key attack flag I (ΔTF.

It consists of a key-off flag K OF F. First, the power flag PF will be explained.

From the pitch extraction section 3, the output signal is output via a latch 4.
f! In addition to the binozi data, the power of the ternary quantized data for each arene (accumulated value that falls within the square value of the ternary amount quantized data) is output in 7th and 5th. be done. When this power value exceeds a certain darkness value.

Power flag PF is high rail (-hereinafter referred to as I+)
becomes. This power flag PF serves as an indicator of the presence of speech regardless of whether it is a consonant or a vowel.

Next, from the pitch extracting section 3, the maximum correlation value in the delay time corresponding to the pitch is sent via the latch 4 to Pl'J (
! It is output to the border 5. When this maximum correlation value exceeds a certain dark value, the correlation value flag CF becomes 11. This correlation value flag CF indicates the degree of presence of pitch; in the case of a vowel, etc., the CF is H, but in the case of a consonant or silence, the CF is 1''-leher (hereinafter referred to as 1.).

Next, the key-on flag I<OF is a flag indicating that the instrument is in the sounding state, and when the power flag PF and the correlation value flag CF become j(to 11), the key-on flag I
i, and when J( becomes L, it becomes 1.

No change at other times. Key attack flag KA'
T'F is the key-on flag KOF) -" is 11 auspicious beginnings, and the month is -1. True: Key" flag KOF F is the key-on flag + < o Ir is 1.
The first number is 1:1, which is b month■.

These two flags serve as indicators indicating the beginning and end of pronunciation. Note that since two or more flag creation operations are performed for each frame, they are all synchronized by an array signal having a time width of a frame interval.

The start frame of pronunciation is determined by the above five flags.

Consonant frames and pronunciation end frames are determined.

In other words, the start frame of the sound is set to the key attack flag.
This is the frame in which the ATF becomes 11, and the end frame of one sound generation is the frame immediately before the key-off flag KOFF becomes 11. Further, the consonant frame is a frame in which the power flag PF and key-on flag KOF are J-1, and the correlation value flag CF is L.

In the pitch control section 10 to realize the pitch data control operation based on the flag.

Two counters are used: a counter ΔNC (14,) that counts from the pronunciation start frame to the 11th frame, and a counter SNC (15) that counts from the frame following the consonant frame to the 11th frame.

Thus, the unprocessed pitch data N1 from the storage section 6 (FIG. 1) is transferred to the latch 2 through the latch 18 and gate 26.
Connected to 2. Pinoch take N2 from the error remover 7 (FIG. 1) is connected to the latch 22 via the latch 19 and gate 27.

The 4-frame moving average output pisochi data N4 from the moving average calculating section 8 is connected to the latch 22 via the latch 20 and the gate 28. Similarly, the hysteresis moving average output pinge data Nθ from the moving average calculation FIG. 8 is connected to the latch 22 via the latch 21 and gate 29.

The output of latch 22 is output to code generator 11 (FIG. 1), and latch 23. It is fed back to its own input via game 1-30. Flag creation section 9
The key-on flag KOF from (Figure 1) is AND circuit 3
1 and 32 to the count enable human power EN of counter 14 and counter 15, respectively. Furthermore, this signal KOF is input to the OR circuit 44 via an inverter 46. Both the power flag PF and the correlation value flag CF from the flag generation unit 9 are input to the NAND circuit 45, and the output thereof is input to the count reset manual R of the counters 14 and 15 via OR circuits 36 and 39, respectively. It is also input to the OR circuit 44. Further, the key attack flag KATF from the flag creation section 9 and the output from each bin 1 of the counter 14 via the OR circuit 37 are output from the OR circuit 35. and is inputted to the count enable manual power EN of the counter 14 via the AND circuit 31. On the other hand, the correlation value flag CF from the flag generating section 9 is connected to a latch 24 operated by the frame clock 1:I block φ, and its output is connected to a latch 25. The output of the latch 24 and the output of the lunch 25 via the input terminal 4B are input to the AND circuit 34.

The output of the tent circuit 34 and each bit output of the counter 15 via the OR circuit 40 are the OR circuit 38 and the AND circuit 3.
2, the counter 15 counts enable human power E
Enter N. Counter 14 and counter 15 count based on frame clock ψ, and their respective hint outputs are input to logic circuits 16 and 17, respectively. 1
Outputs 49 and 54, which become H when each count value is 0 in 6 and 17, both control the gate 29 via the OR circuit 41 and the ant circuit 33. At this time, the output of the OR circuit 44 is input to the other input of the AND circuit 33 via the inverter 47.

When the count value is 1 to 3 in the logic circuit 16, the output 50 which becomes ■] controls the case 1-26, and also controls the logic circuit 1.
7, the output 55 becomes H when the count value is 1 to 3.
input to the OR circuit 44. In the logic circuits 16 and 17, the outputs 51 and 56, which become H when the value is 4 to 7, control the gate 27 via the OR circuit 42. Further, in the logic circuits 16 and 17, the count value is 8 to
Outputs 52 and 57 which become 11 when II is OR circuit 43
The gate 28 is controlled via the gate 28. Further, outputs 53 and 58 which become H when the count value reaches 12 in the logic circuits 16 and 17 are input to the inputs R of the counters 14 and 15 via OR circuits 36 and 39, respectively. On the other hand, the output of the OR circuit 44 controls the gate 30.

The operation for performing the pitch data control by the pinch control section 10 having the above configuration will be explained with reference to FIG. 3.

FIG. 3 shows each flag, two counters 14 and 15, and a code generator II from the pitch data control section 10.
3 is a timing chart showing the relationship between the pitch data and the pitch data output to. PF by inputting voice more forcefully
becomes ■(with a delay of 2 frames, CF Rikitsu) (this delay depends on the circuit configuration). Then, with a delay of 2 frames, KOF becomes I], and at the same time, KATF becomes 1
Only the frame becomes I-1. Then KATF becomes 11 and KOF.

Counter 1 on the condition that both PF and CF are ■]
4 starts counting. This condition is determined by the count enable manual input via the main circuit 35 and the un1 circuit 31.
This holds true when becomes H.

Also, K A i' F, which starts counting, becomes L.

At least 1 bit out of 4 bits of counter output
becomes [I, so the output of the OR circuit 37 becomes 11.

The count enable input EN becomes H via the OR circuit 35 and the AND circuit 31 and continues counting. Extracted data is continuously selected and output by this counting operation.

Next, when the count value of this counter 14 is 1 to 3, the gate 26 is opened and input is made to the notch 18.
The unprocessed pitch data NI from the section 6 (FIG. 1) is selectively outputted to the court generator 11 via the latch 22. This operation is realized by the output 50 of the logic circuit 16 becoming H. Next, when the count value is 4 to 7, the output 51 of the logic circuit 16 becomes 11, and the off circuit 4
2 to the gate 270 control terminal, so gate 2
7 opens, and the pitch data N2 from the error removing section 7 input to the latch 19 is selectively output to the latch 22. Furthermore, when the count value is 8 to 11, the gate 28 is opened and the 4
The frame moving average output histitator Nd is selectively outputted to the back 22.

This operation is performed when the output 52 of the logic circuit 16 becomes tl and 14 is applied to the control terminal of the gate 28 via the OR circuit 43.

Then, when the count value of the counter 14 reaches 12, the output 53 becomes H and is output to the counter 14 via the OR circuit 36.
is cleared and the count ends.

After that, the gate 29 is opened on the condition that PF, CF, and KOF are all 11, and the 8-frame moving average output pinoch data N8 is selectively outputted from the moving average calculation unit 8 input to the latch 21. It will be output on Lunch 22. This operation is performed by outputting an output 49 via an OR circuit 41 and a NAND circuit 45 . .

OR circuit 44. The output through the inverter 47 becomes H, which causes the circuit 33 to become I].

This is achieved by applying it to the control terminal of gate 29.

Next, when a consonant frame is detected in the key-on state (KOF is old), that is, PF is H.

When the CF Rikigetsu is -, the game 130 opens and the pitch take BN of the frame immediately before that (Nθ in the case of Fig. 2) is displayed.
is selectively output to the latch 22. This operation is realized by connecting the NAND circuit 45 and the OR circuit 44 to the control terminal of the gate 30. Then, when the consonant frame ends and the ninth pronunciation frame arrives, the counter 15 starts counting. That is, when CF rises to H while KOF is at H, the counter 15 starts counting.

This operation is performed by the latch 24 and the latch 25 via the inverter 48 to generate a pulse corresponding to the AND circuit 34. This is realized by inputting it to the count enable manual power EN of the counter 15 via the OR circuit 38.

When counting starts, the count enable manual input signal is input via the OR circuit 40.48 and the AND circuit 32.
becomes II and continues counting.

Next, when the count value is 1 to 3, the pitch data BN for the consonant frame is selectively output to the latch 22. This operation is realized when the output 55 of the logic circuit 15 becomes 4 and the gate 30 is opened via the OR circuit 44.

Next, when the count value is 4 to 7, the gate 27 is opened in the same way as the counter 14, and the pitch data N2 is selectively output to the latch 22. This is realized by the output 56 becoming H and the gate 27 opening.

Further, when the count value is between 8 and 11, the gate 28 similarly opens, and the pitch gutter N4 is selectively set to the latch 22 to 11.
1 power Juru. The dirt becomes output 57 or II and is realized every time games 1 to 28 are opened. Then, when the number 1 exceeds 11, the output iJ of the counter 15 is similarly cleared by the output 58.

Power/Non 1 is completed.Then, the first key 129 opens and selectively outputs the pitch take N8 to the second gear 22.The operation is as follows every time the output 54 becomes T-1 and the first gear 1-29 opens. Realized.

Finally, in the last part of the pronunciation, PF is followed by I.
, and 1 until KOF reaches L with a delay of 2 frames.
1 gate 3 as the pronunciation may continue to be unstable.
0 opens, and pitch data B of Arene\ just before that
N (N e in the case of FIG. 2) is selectively output to the latch 22.

This operation is realized by opening -ζ-1-30 through the circuit 45 and the OR circuit 44.

FIG. 4 collectively shows the relationship between the counter values of the counters 14 and 15 and the type of pitch take to be selected. The count value is 1 at each force Y'
When it is 3, if the counter 14 is on, it means the beginning of sound generation, so the unprocessed pinot-y-take N1 is selected. Furthermore, since the frame follows one consonant, the counter 15 selects the pitch take BN of Fushi - -no immediately before the consonant.

Furthermore, the force at the beginning of each pronunciation: nont value 4 to 7.
8 to 11, unprocessed pitch take 1 frame J
・Pitch take with error removed Pitch take with moving average of 4 frames, pitch data with hysteresis (18 frames shifted) are selectively adopted in this order, so irregular changes and for both consonant parts.

Yaw 1' generator 1 for pitch take with less erroneous extraction
1). Also, in the consonant part and the final part of the pronunciation, the pitch take is seamless and connected smoothly.

The changes in the scale created by the chord generator 11'ζ become natural. In the embodiment of the present invention, the correlation value flag CF off is set for consonant parts, but this applies not only to consonant parts, but also to changes in scales generated by musical instruments, for example.

〔Effect of the invention〕

According to the present invention, the pitch is extracted from the voice input using a digital pinch extractor, and the beginning and end of pronunciation, the consonant part, etc. can be appropriately determined from each extracted information, and musical sounds are generated accordingly. It is possible to appropriately select pitch data for

According to the present invention, musical tones can be input by voice.

Musical sounds can be controlled without the need for finger training.

Furthermore, since it is a digital type, it can be connected to a digital sound source chip such as a VCO (voltage controlled oscillator).

[Brief explanation of drawings]

FIG. 1 is an overall circuit configuration diagram of an embodiment according to the present invention, and FIG. 2 is a detailed circuit diagram of the pitch data control section of FIG. 1. i'/
) Figure 3 is an operation timing chart of the present invention.
FIG. 4 shows the counter ΔNC (14). 5NC (15) count value and selected pitch data f!
It is a diagram of the relationship with the class. 10=-pinch take system falli7B, 14,
15...Counter, 16.17...Logic circuit,
], 8.19.20.21, 22, 23.24, . 25
・・Rudge, 2G, 27゜28.2'9.30・・
・Gate patent applicant Yoshiyuki Osuga, patent attorney for Casio Computer Co., Ltd. Figure 3 Figure 4

Claims (5)

[Claims] (1) A detection means for detecting the temporal position of at least the start and end portions of audio, an extraction means for extracting pitch data of the audio, and a plurality of processing means for performing different processing on the pitch data. and. means for sequentially selecting machining pitch data of the plurality of machining means in correspondence with each of the detected temporal positions; and musical sound generating means for generating musical tones based on the machining pitch data output from each selection means. 1. A musical sound generation device using voice input, characterized in that it has the following. (2) The detecting means includes extraction change detecting means for detecting that the amount of change in the output of the extracting means is a specific value (a or more) and determining the temporal position of the point of change. A musical tone generating device using voice input according to claim 1. (3) The selection means includes a counter that starts counting from a temporal position immediately after at least one of the temporal position of the start portion of the audio and the temporal position obtained by the extracted change detection means; 3. The musical tone generating device according to claim 2, further comprising means for selecting the processed pitch data according to a count output result. (4) The selection means has means for continuously outputting the processed pitch data of the end part of the audio or the part immediately before the temporal position immediately after at least one of the temporal positions obtained by the extracted change detection means. A musical tone generating device according to claim 2, characterized in that the musical tone generating device uses voice input. (5) The musical sound generation device according to claim 1, wherein the plurality of processing means are extracted data error removal means and extracted data averaging means.