JPH07287571A - Sound signal conversion device - Google Patents

Abstract

PURPOSE:To convert a sound signal into a correct key code irrespective of the kind of the inputted sound signal. CONSTITUTION:This sound signal conversion device is provided with a sound input means which inputs sound and generates the sound signal, a pitch extracting means which divides the sound signal generated by the sound input means into specific time sections and extracts pitches in the respective time sections, a mean pitch arithmetic means which calculates the mean pitch of the pitches extracted by the pitch extracting means, and a pitch information converting means which converts the mean pitch calculated by the mean pitch arithmetic means into pitch information. Further, the device has a sound input means which inputs a sound constituting music and generates a sound signal, a pitch extracting means which extracts a pitch from the sound signal generated by the sound input means, a pitch information converting means which converts the pitch extracted by the pitch extracting means into pitch information, a tone information supply mean which supplies tone information, and a pitch information correcting means which corrects the pitch information according to the tone information supplied from the tone information supply means.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice signal converter, and more particularly to a converter for converting an input voice signal into a key code.

[0002]

2. Description of the Related Art A voice signal converter is a device for converting a voice signal input from the outside into a key code (pitch information). The converted key code can be treated as performance data in MIDI format, for example. If the tone generator converts such performance data into a tone signal, it becomes a signal that can be sounded.

The voice signal converter performs a recognition process for detecting what key code the voice signals that are successively input are. The key code recognition rate, which indicates the rate at which a correct key code can be recognized from a voice signal, greatly differs depending on the type of voice signal input to the voice signal conversion device.

If the input voice signal is generated by playing a musical instrument, the key code recognition rate can be high to some extent. However, when the input voice signal is a human voice, the key code recognition rate becomes lower than that of the voice detection by playing a musical instrument.

The voice uttered from the human throat usually does not have the same key code continuously and becomes a fluttering sound. Even if it is a singing voice by a professional singer, since the fundamental frequency in the voice signal emitted fluctuates to some extent,
It is difficult to accurately recognize the key code for each note from the singing voice.

[0006]

An audio signal generated by synthesizing electric signals, like an electronic musical instrument, is a signal in which fluctuations in each frequency component are small and almost no noise is included. Therefore, it is relatively easy to recognize the corresponding key code from the audio signal from the electronic musical instrument.

However, it is difficult to recognize the key code in the voice signal generated by human voice, etc., because the frequency component has a large variation. An object of the present invention is to provide an audio signal conversion device capable of converting an audio signal into a correct key code regardless of the type of the input audio signal.

[0008]

SUMMARY OF THE INVENTION An audio signal converter according to the present invention has a predetermined audio input means for inputting an externally supplied audio to generate an audio signal and an audio signal generated by the audio input means. Pitch extracting means for dividing the time interval into a plurality of time intervals to extract a pitch in each time interval, an average pitch calculating means for averaging a plurality of pitches extracted by the pitch extracting means to calculate an average pitch, and an average Pitch information conversion means for converting the average pitch calculated by the pitch calculation means into pitch information.

Further, the audio signal conversion apparatus of the present invention extracts the pitch from the audio input means for externally inputting the audio forming the music to generate the audio signal and the audio signal generated by the audio input means. For extracting pitch information, pitch information converting means for converting the pitch extracted by the pitch extracting means into pitch information, key information supplying means for supplying key information, and key information supplying means. And pitch information modifying means for modifying the pitch information according to the key information.

[0010]

The pitch extracting means divides the voice signal input to the voice signal input means into predetermined time intervals and detects the pitch for each time interval. The detected pitch is averaged over a plurality of time intervals by the average pitch calculating means,
Converted to pitch information. As for the pitch information, it is possible to obtain correct pitch information by using the averaged pitch rather than directly using the detected pitch.

Further, after the pitch detected from the input voice signal forming the music is converted into pitch information, the pitch information is modified according to the music rule of the supplied key information.
Correct pitch information can be obtained by correcting the pitch information.

[0012]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows an example of the system configuration of an audio signal converter for implementing the present invention. The audio signal converter is
A voice signal input from the microphone 1 can be converted into a key code. The converted key code is converted into a tone signal of the corresponding key code in the sound source 4, and is sounded by the sound system 5.

A voice signal such as a singing voice is input to the microphone 1 from the outside. The input voice signal is, for example, an analog signal of a musical sound produced by a human singing voice or playing a musical instrument. Pressing the start switch starts the input of the audio signal, and pressing the stop switch ends the input of the audio signal. The voice analog signal input to the microphone 1 is converted into a digital signal in the D / A converter 2 and supplied to the bus 11.

The CPU 7 performs various control processes in order to convert the audio digital signal supplied to the bus 11 into a key code. C to convert to the correct key code
In addition to the normal key code conversion process, the PU 7 performs the average pitch detection and correction process to the in-key key code.

In the average pitch detection, the voice signal input to the microphone 1 is quantized and the pitch is detected for each quantized voice signal. Then, the average pitch of the plurality of pitches detected within the predetermined time is obtained, and the key code is calculated from the average pitch. By detecting the average pitch, it is possible to detect the correct pitch even in the case of a voice signal having a property in which the pitch easily changes with the passage of time such as a human singing voice.

In the correction processing for the in-key key code, first, the audio signals continuously input to the microphone 1 are regarded as music, and the key of the music is detected. When a tone of a musical tone is detected, it is possible to predict a key code that is easy to use for that tone. Using the knowledge of the music rules, the detected key code is corrected. Thereby, the key code detected by the erroneous recognition can be corrected to the correct key code.

A data memory (RAM) 10 stores data necessary for detecting a key when the input voice signal is regarded as a musical tone. For example, note name data that constitutes each key such as C major and C minor is stored, and a key is detected by checking which key the input voice signal corresponds to. .

After converting the input voice signal into a key code, the CPU 7 generates performance data based on the converted key code. The generated performance data is supplied to the sound source 4, and a musical tone signal is generated. Sound system 5
A musical tone is emitted according to the musical tone signal supplied from the sound source 4.

When the output switch of the output terminal 3 is pressed, the generated performance data is MIDI-generated as necessary.
The data is converted into format data or the like and output from the output terminal 3 to an external device.

A program memory (ROM) 8 stores a calculation program. The CPU 7 performs the above-described various arithmetic processes using a working memory such as a register and a buffer memory provided in the working memory (RAM) 9 according to the arithmetic program.

The timer 6 generates a timing signal and supplies an interrupt signal to the CPU 7 at predetermined time intervals. The timer 6 supplies an interrupt signal to the CPU 7 every 16 minutes (notes), for example, to execute interrupt processing. The interrupt processing can also generate a metronome sound that is necessary when a person inputs a singing voice into the microphone 1 from the outside.

The CPU 7 controls the D / A converter 2, the program memory 8, the working memory 9, the data memory 10, the sound source 4, and the output terminal 3 via the bus 11. FIG. 2 is a flowchart showing the main routine of the input conversion process performed by the CPU. First, the CPU executes step P1.
In step 1, initialization processing such as initialization of registers is performed.

In step P2, it is checked whether or not the start switch for instructing the start of inputting a voice signal from the microphone has been pressed. If the start switch has not been pressed, the process proceeds to step P9 without performing voice input processing.

If the start switch is pressed, the process proceeds to step P11 to initialize registers and flags necessary for voice input processing, and then proceeds to step P3.
In step P3, voice input processing and average pitch detection processing are performed. The voice input process is a process of taking a voice signal input from the microphone into the working memory. The average pitch detection is a process of quantizing the input voice signal and detecting the average value of the pitch of each quantized voice signal. Details of the processing will be described later with reference to the flowchart.

In step P4, it is checked whether or not the stop switch for instructing the end of the voice signal input from the microphone has been pressed. If the stop switch is not pushed, the above-mentioned voice input process and average pitch detection process are repeated for voice signals sequentially input to the microphone, and the process is repeated until the stop switch is pushed.

When the stop switch is pressed, the process proceeds to step P5, and the end code indicating the end of the input of the voice signal is stored in the memory MEM () in the working memory. The average pitch for each predetermined time detected in the above step is
It is already stored in the array memory MEM (). When the stop switch is pressed after all the average pitches of the input audio signal are stored, the memory MEM of the last array is stored.
The end code is stored in (I).

In step P6, the average pitch of the audio signals stored in the memory MEM () is converted into a key code. Since the average pitch is represented by a frequency, the value represented by the frequency is converted into a key code represented by C4 or the like. Details will be described later.

In step P7, a process for correcting a key code that is erroneously converted in the converted key code is performed. First, the plurality of key codes obtained in the previous step are regarded as music, and the key of the musical sound is detected. And for those that do not correspond to the detected tones, it is determined that they have been converted to the wrong key code,
Perform correction processing. Details of the processing will be described later with reference to the flowchart.

In step P8, after the key code is detected and corrected, performance data is generated based on the key code. In addition to the key code, the performance data includes data such as a key-on code indicating the start of sounding, a duration code indicating the duration of sounding, and a key-off code indicating the end of sounding. Details will be described later.

In step P9, it is checked whether or not the output switch has been pressed. If the output switch is pressed, the process proceeds to step P10 and the generated performance data is output from the output terminal. Then, the process returns to step P2 to check again whether the start switch for starting the voice input is pressed.

When it is determined in step P9 that the output switch is not pressed, the process directly returns to step P2 without outputting performance data from the output terminal.
FIG. 3 is a flowchart showing details of the voice input process and the average pitch detection process in step P3 of FIG.

In the main routine of FIG. 2, before the voice input process is performed, the initial setting for the voice input process is already performed in step P11. In the initial setting, the key-on flag KON, the pitch register F, the flag Q, the counter I, and the counter N are initialized. The initial setting method of each flag is shown below.

First, the key-on flag KON is set to 0.
The key-on flag KON indicates 1 while any audio signal is input from the microphone, and indicates 0 when the input audio signal is almost 0.

Next, the pitch register F is cleared to 0. The pitch register F is a register for storing the pitch detected from the audio signal input from the microphone. Pitch is a frequency that represents a pitch.

Then, the flag Q is set to 0. Flag Q
Is set to 1 at a predetermined time interval (for example, the length of a 16th note) by an interrupt process triggered by an interrupt signal from the timer. Since the flag Q becomes 0 after the predetermined processing is performed, 0 is usually set,
1 is set momentarily at a predetermined time interval.

Further, the counter I and the counter N are set to 0.
Reset to. The counter I indicates the array number of the array memory MEM () that stores the calculated average pitch. The counter N counts the number of quantized audio signals input within a predetermined time.

After the above initialization, the processing for voice input and average pitch detection is performed in step Q2.
Start with. In step Q2, the key-on flag KO
Check whether N is 0 or not. If the key-on flag KON is 0, the process proceeds to step Q3. Initially, the key-on flag KON is initially set to 0, so the routine proceeds to step Q3.

In step Q3, it is checked whether or not the input level of the voice signal input from the microphone exceeds the threshold value # 1. The threshold # 1 indicates the noise level of the voice input level. If the threshold # 1 is not exceeded, it is determined that the input is in a silent state, and the processing of the input audio signal is not performed, and the process proceeds to step Q10.

If the voice input level exceeds the threshold value # 1, it is determined that there is a proper voice input, and the process proceeds to step Q4 to start the processing of the input voice signal. In step Q4, the key-on flag KON is set to 1 to indicate the start of proper voice input. Then, it progresses to step Q5.

In step Q5, pitch detection is performed based on the input voice signal. For an input voice signal represented by a digital signal sequence, a signal within a predetermined time is cut out by a Hamming window, a Hanning window, etc., and frequency analysis is performed. Pitch (pitch information) is detected by performing frequency analysis.

At step Q6, the detected pitch is added to the pitch register F. Since the pitch register F is initially set to 0, the value of the detected pitch itself is initially stored in the pitch register F.

At step Q7, the counter N is incremented. Then, it progresses to step Q10. In step Q10, it is checked whether or not the flag Q is 1. Since the flag Q is set to 0 in the initial setting,
The process once returns to the process of the main routine of FIG. If the stop switch for instructing the end of voice input is not pressed, the processing from step Q2 is restarted to repeat the above-mentioned voice input processing. By repeating the voice input process until the stop switch is pressed, the voice signals sequentially input from the microphone can be processed in real time over time.

In order to restart the voice input process, it is checked again in step Q2 whether the key-on flag KON is 0 or not. Since the key-on flag KON has already been set to 1 in the above-mentioned processing of the first step Q4,
Go to step Q8.

In step Q8, it is checked whether or not the voice input level input from the microphone this time is smaller than the threshold value # 2. The threshold # 2 indicates the noise level of the voice input level. If the threshold # 2 is exceeded, it indicates that the proper audio signal has been continuously input since the last time. On the other hand, if the threshold # 2 is not exceeded, it indicates that the input from the microphone is in the silent state.

In step Q8, if the voice input level exceeds the threshold value # 2, it means that the voice is continuously input from the microphone. Therefore, the process proceeds to step Q5, and the average pitch of the voice signals input this time. Perform detection processing.

In step Q5, the pitch of the input voice signal is detected, and the detected pitch is added to the pitch register F in step Q6. In step Q7, the counter N is incremented.

Until the stop switch is pressed,
The pitch of the voice signal sequentially input from the microphone to the pitch register F is accumulated. Since the counter N is incremented each time the pitch is accumulated, the counter N stores the number of accumulated pitches in the pitch register F.

Thereafter, in step Q10, the flag Q
Check whether is 1. The value of the flag Q is determined by the interrupt processing described below. FIG. 4 is a flowchart showing an interrupt process performed by the CPU. The CPU receives an interrupt signal from the timer at a predetermined time interval of, for example, 16 minutes (time length of 16th note). Upon receiving the interrupt signal, the CPU performs the following interrupt processing. here,
It is assumed that interrupt processing is performed at 16-minute intervals.

First, in step R1, the flag Q is set to 1
Set to. Then, in step R2, a click sound is generated. After that, the interrupt processing is ended and the processing of the original voice input is restarted.

As described above, in the interrupt process, the flag Q is set to 1 at 16-minute intervals and a click sound is generated. The click sound plays the role of a metronome. When a voice signal is input from the microphone, the input may be performed in accordance with the tempo of the click sound that sounds every 16 minutes.

In step Q10 of FIG. 3, it is determined that the flag Q is 1 when 16 minutes have passed, and step Q10
Proceed to 11. In step Q11, it is checked whether the counter N is larger than 0. The counter N stores the number of pitches detected by quantization during 16 minutes. If it is determined that the input audio signal from the microphone is silent for 16 minutes, the counter N shows 0.

As described above, when a proper voice is input to the microphone, since the counter N stores a number of 1 or more, the process proceeds to step Q12. In step Q12, the average pitch is calculated by the following equation and stored in the average pitch register AF.

AF = F / N The pitch register F stores the accumulated value of the pitch detected during the 16-minute length. By dividing the cumulative value F of pitches by the quantization number N, the average value AF of pitches can be obtained. Then, it progresses to step Q14.

On the other hand, in step Q11, when the silent voice input state continues for 16 minutes, the counter N is 0, so the routine proceeds to step Q13. Step Q13
Then, all 1s are stored in the average pitch register AF.
Then, it progresses to step Q14. All 1 indicates a value that makes all the bits configuring the register 1s. If all 1s are stored in the average pitch register AF,
The 16-minute rest, which is a 16-minute silent state, is input from the microphone.

At step Q14, the calculated average pitch AF is stored in the array memory MEM (I). Since the counter I is initially set to 0, the average pitch AF is stored in the memory MEM (0) of the 0th array element number. The counter I indicates the obtained average pitch number, and sequentially increases from 0 with the passage of time.

At step Q15, the counter I is incremented to set the array number of the next average pitch. In step Q16, 0 is set in each of the pitch register F, the counter N, and the flag Q, and initial setting for obtaining the next average pitch is performed. After that, the process returns to the process of the main routine of FIG. 2 and the process of obtaining the average pitch is repeated until the stop switch is pressed.

Next, the processing when the voice input from the microphone becomes silent will be described. In step Q2,
If there was a proper voice input at the previous time, the key-on flag KO
Since N is 1, the process proceeds to step Q8. If the voice input level is smaller than the threshold value # 2, it is determined in step Q8 that there is no sound, and the process proceeds to step Q9. In step Q9, the key-on flag KO
If N is set to 0 and pitch detection is not performed, step Q10
Go to. In step Q10, the process of obtaining the average pitch for every 16 minutes is performed as described above.

As described above, when there is a proper voice input from the microphone, the input voice signal is quantized within a 16-minute length range, and the pitch of each quantized signal is detected. Then, the accumulated value F of the detected pitch is converted into the quantization number N
The average pitch AF is obtained by dividing by.

Since the voice signal has a large fluctuation in human singing voice and the like, when the local pitch of the voice signal is detected, the pitch is frequently changed, and erroneous recognition of the pitch is likely to occur. Therefore, the correct pitch can be detected by quantizing the input voice signal, detecting the pitch for each quantization, and then obtaining the average pitch. The calculated average pitch is stored in the array memory MEM ().

FIG. 5 shows a configuration example of the memory MEM (). The memory MEM () sequentially stores the average pitch obtained over time. The memory area 15 stores a frequency indicating the obtained average pitch. The average pitch is an average value of pitches obtained when a proper voice input is made within 16 minutes.

The memory area 16 stores all 1's data. The all-1 data is data stored when silent voice input continues for 16 minutes, and indicates a 16-minute rest.

The frequency stored in the memory MEM () is a frequency indicating the pitch of a sixteenth note. However,
When all 1's are stored, it means a 16th rest instead of indicating the pitch of a 16th note.

The end code 17 is stored in the last area of the memory MEM () in step P5 of FIG. 2 after the stop switch has been pressed and all voice input has been completed. The end code 17 indicates the end of the average pitch data continuously stored in the memory MEM ().

FIG. 6 is a flow chart showing details of the key code conversion process performed in step P6 in the main routine of FIG. Key code conversion is performed in the memory MEM ()
This is a process for converting the frequency of the average pitch stored in the key code into a key code.

In step S1, the average pitch showing the lowest frequency in the memory MEM () in which the average pitch frequency is stored is extracted, and the lowest frequency register Fm is extracted.
The lowest frequency is stored in in.

In step S2, the frequency stored in the memory MEM () is read and stored in the register F. Then, the relative key code KC is obtained by the calculation of the following equation. KC = 12log ₂ (F / Fmin) Relative key code KC indicates a value of 1/100 of the normal cent, and with the lowest frequency Fmin as a reference,
It shows how high or low the frequency F is. If the frequency F has the same value as the minimum frequency Fmin, the relative key code KC becomes 0, and if the frequency F is semitone higher than the minimum frequency Fmin, the relative key code KC becomes 1.

The obtained relative key code KC is stored again in the memory MEM () where the frequency was originally stored. Similarly, all the frequencies stored in the memory MEM () are converted into the relative key code KC. Memory M
In EM (), the lowest frequency is 1 / based on Fmin.
100 cent representations are stored.

In step S3, the lowest frequency Fmin is converted into a key code closest to the frequency, and the lowest frequency key code is stored in the register KCmin. For example, if the frequency is 440 [Hz], it is converted into an A4 key code and stored in the register KCmin.

In step S4, the relative key code stored in the memory MEM () is read out and stored in the register KC.
To store. Then, the key code K is obtained by the calculation of the following equation.

K = KC + KCmin The key code K is the lowest reference key code KCmin.
It is an absolute key code obtained by adding the relative key code KC to.

The obtained key code K is stored again in the memory MEM () where the relative key code was originally stored. Similarly, all the relative key codes stored in the memory MEM () are converted into absolute key codes K.
The memory MEM () stores a key code for every 16 minutes. Then, the process returns to the main routine of FIG.

FIG. 7 is a flow chart showing the details of the correction processing for the in-key key code in step P7 in the main routine of FIG. In step T1, a set of audio signals continuously input from the microphone is regarded as a musical piece, and the key of the musical piece is detected. In the data memory, note name data indicating the constituent notes of the key are stored.

FIG. 8 shows the note name data of the constituent tones stored in the data memory. Key is 12 note names (C, C
There are major and minor keys respectively when #, ..., B) are tones, and data for each key is stored in the data memory. In addition, you may divide and store in a finer tone.

For example, in the C major memory area 21, C
The chords C, D, E, F, G, A, and B, which are major constituent tones, are stored. Similarly, memory area 2 in C # major
A code of a constituent sound of C # major is stored in 2, and a code of a constituent sound of C minor is stored in a memory area 23 in C minor.

In the key detection, it is checked which key among the keys stored in the data memory the key code stored in the memory MEM () best matches. Memory ME
The key code of M () is compared with the constituent tones of the key in the data memory, and the key with the highest degree of coincidence is used as the key of the input voice signal.

The tonic register TN stores tones (C, C #, D, etc.) of the detected key. Mode register MD
Stores the detected key mode (major or minor). After the key detection is performed, the process proceeds to step T2.

The key detection method is not limited to the above-mentioned method, and the key detection may be performed by another method. Step T
In step 2, the detected key is used to round the key code into the constituent sounds. When the key code corresponding to the input voice signal is not the constituent sound of the detected key, the key code is corrected to the key code of the constituent sound. A detailed description of the process is given below.

FIG. 9 is a flow chart showing details of the in-tone tone rounding processing in step T2 of FIG. First, the last sound of the reference music is rounded, and then the remaining sounds are rounded. The processing procedure will be described below.

At step U1, m of the memory MEM () is stored.
The second (the key code number stored in the memory MEM () and the last stored key code number) key code is read.

In step U2, it is checked whether or not the read key code is an in-tone (constant tone of the detected tone). If it is an in-tone, it is determined that the key code has been correctly detected, and the process proceeds to step U6 to store the read key code itself in the register KC. Then, it progresses to step U7.

On the other hand, if the read key code is not an in-tone, it is determined that it is a key code detected by erroneous recognition, and the process proceeds to step U3. In step U3, the read m-th key code is rounded to the in-tone. First, the read key code is converted into an in-tone tone that is closest to the key code. If there are a plurality of intones that are close to each other, they are converted into intones according to the next priority.

1 degree> 5 degrees> 3 degrees> 2 degrees> 6 degrees> 7 degrees> 4 degrees For example, if the detected key is C major, one tone is C and five tones are It is G, and the third tone is E. 1st sound is the highest priority, followed by 5th, 3rd, ...
・ It is converted into the internal tone in the order of.

There is a rule that the last note of a musical composition is likely to end with one note of that key. Therefore, the 1st sound is given the highest priority, and then the 5th, 3
Conversion is performed by prioritizing the sounds in the order of degrees, ... And
In step U4, the key code converted into the in-tone is stored in the register KC.

At step U5, the key code stored in the register KC is stored again in the memory MEM () of the array number from which the original key code is read. In the register KC, the key code corrected to the in-tone is stored.
However, when the detected key code is an in-tone tone, it is determined that the correct recognition is performed and the detected key code itself is stored in the register KC without correction.

At step U7, the number register m is decremented. By reducing the value of number m by 1,
It can be changed to the number m representing the immediately preceding sound. The process for the previous sound indicated by the number m is performed next.

At step U8, the m-th key code stored in the memory MEM () is read. The read key code is the penultimate sound. In step U9, it is checked whether the read key code is an in-tone tone. If it is an in-tone, it is determined that the key code has been correctly detected, the process proceeds to step U14 without correcting the m-th key code of the memory MEM (), and the register K
After storing the key code in C, the process proceeds to step U13.

On the other hand, if the read key code is not an in-tone, it is determined that it is a key code detected by erroneous recognition, and the process proceeds to step U10. In step U10,
A process of rounding the read m-th key code to an in-tone is performed In step U10, the read key code is converted into an in-tone that is closest to the key code. If there are a plurality of intones that are close to each other, they are converted into intones that are closest to the key code stored in the register KC.

The register KC stores the key code stored in either step U4 or step U6. That is, the corrected key code of the last sound of the music is stored. The key code closest to the key code of the last sound is converted preferentially.

In step U11, the key code corrected and converted into the in-tone is stored in the register KC, and step U1
2 stores the key code stored in the register KC in the memory ME
Restore in the original memory location of M ().

In step U13, it is checked whether or not the number m is 1. The fact that the number m is 1 indicates that the correction process of the first sound of the music (the sound of the audio signal first input from the microphone) to the in-tone is completed. Since the correction process is performed in order from the last sound to the first sound of the music, when the processing for the first sound is completed, the processing for all the sounds is completed.

If the number m is not 1, the process returns to step U7, the number m is decremented, and the correction process for the key code of the preceding sound is repeated. Memory MEM () m
The second key code is read out, and if the read key code is within the tone, the key code is not corrected. If the read key code is not the in-tone, the key code of the closest in-tone is converted. When there are a plurality of similar in-tone tones, the key code close to the corrected key code of the preceding sound is preferentially converted.

As described above, the priority of the last note of the music is determined on the basis of the 1st and 5th notes of the key, but the priority of the notes other than the last note is the previous one. The key code is decided by giving priority to the sound close to the key code of the sound. This is because only the last sound is likely to be a key-related sound.

When the erroneously recognized key code is corrected by the above correction processing, it is stored in the memory MEM ().
When the correction process is completed, the process returns to the original process of FIG. 7. In addition,
In the above key code correction processing, the case has been described in which all out-of-tune tones are corrected into in-tone sounds, but it is not necessary to correct all. For example, there is a case where it is desired to partially use the extratone by using accidentals such as # and ♭ which are effective only within one measure. In such a case, in the detected key, only the prohibited sound (note name) may be corrected to the prohibited sound.

In addition to correcting the key code by using the key, a scale is detected to detect a plurality of pitch ranges to be used, and a pitch outside the detected pitch range has a predetermined pitch. You may perform the process which corrects to the sound within width. Since the scale is determined by the tonality and chord, and the pitch name used is determined by the scale, the pitch range used can be determined by giving each pitch corresponding to this pitch a width.

FIG. 10 is a flow chart showing details of the performance data conversion process of step P8 in the main routine of FIG. The performance data conversion is performed based on the key code detected from the input voice signal and converted into a performance data format capable of producing sound.

At step V1, the array element number register P,
The duration register D and the key-on flag KON are set to 0, respectively. The array number register P indicates the array number of the memory MEM () in which the key code is stored. The duration register D stores a duration time during which the same key code is continuously sounded or a rest time time during which no sound is generated. The key-on flag KON becomes 1 when any key code is being sounded and becomes 0 when no key code is sounded.

In step V2, a write pointer indicating the writing position in the performance data memory is set at the beginning of the performance data memory to prepare for writing the first performance data.

In step V3, it is checked whether the end code is stored in the memory MEM (P). The array element number P is set to 0 by default. Memory MEM
In (), a key code which is corrected as needed after being detected from the input voice signal is stored. if,
If the end code is stored in the memory MEM (),
Indicates that voice input is completed at the position of that array number.

If the end code is not stored in the memory MEM (P), the process proceeds to step V4 and the memory MEM (P) is reached.
Based on the key code stored in (P), it is converted into performance data. The performance data conversion method for the array element number P will be described later.

After the performance data is generated for the key code of the memory MEM (P), the process proceeds to step V5,
The array number register P is incremented to prepare for the generation of performance data for the next array number. Also, the duration register D is incremented to count the duration of sound generation. Increasing the duration register D by 1 increases the duration by a length of 16 minutes. Memory M
This is because the key code is stored in EM () every 16 minutes.

Then, the process returns to step V3, and it is checked whether the key code (MEM (P)) stored in the next array element number P is the end code. If it is not an end code, the process proceeds to step V4, and performance data is generated based on the key code stored in the memory MEM (P). Then, in step V5, the array number register P and the duration register D are each incremented.

The duration register D stores the number of consecutive identical key codes stored in the memory MEM (). For example, when only one key code of A4 appears and the key code changes next time, 1 is stored in the duration register D to indicate the 16th note of A4. If two A4 key codes continue in succession, 2 is stored in the duration register D and A
The eighth note of 4 is shown.

In step V3, the memory MEM
If it is determined that the end code is stored in (P), this indicates the end of the key code, and the process proceeds to step V6. In step V6, it is checked whether or not the key-on flag KON is 1. If the key-on flag KON is 1, the performance data indicates that one of the key codes is currently being sounded, so the process proceeds to step V7.

In step V7, the value of the duration register D indicating the sounding duration of the key code currently being sounded is written in the performance data memory together with the duration code. The duration code is a code indicating that the data of the duration register D indicating the sounding duration continues in the performance data memory, and the duration code and the data of the duration register D are always written in the performance data memory as a set. After the writing is completed, the write pointer to the performance data memory is advanced to prepare for the next writing.

At step V8, a key-off code indicating the end of tone generation and an end code indicating the end of performance data are written in the performance data memory. After the writing is finished, the performance data generation is finished and the process returns to the main routine of FIG.

At step V6, the key-on flag K
When it is judged that ON is not 1, it means that the note is not being sounded at present, so the routine proceeds to step V9, and only the end code is written in the performance data memory. After the writing is completed, the process returns to the process of the main routine of FIG.

FIG. 11 is a flow chart showing details of the performance data conversion process in step V4 of FIG. When the key code is stored in the memory MEM (P) instead of the end code, the procedure for converting into the performance data based on the key code in the memory MEM (P) is shown.

In step W1, it is checked whether or not all 1's are stored in the memory MEM (P). If all 1's are stored, it indicates that the input is silent.
If all 1's are not stored in the memory MEM (P), some key code is stored, so the process proceeds to step W2 to generate performance data based on the key code.

At step W2, the key-on flag KON
Check whether is 1. Key-on flag KON is 1
If not, since the key is off and no sound is being generated, the process proceeds to step W6, the key-on flag KON is set to 1, and the process proceeds to step W7.

In step W7, the duration code indicating the rest length indicating the previous key-off and the value of the duration register D are written. After finishing writing
The write pointer of the performance data memory is advanced, and the process proceeds to step W8.

At step W8, a key-on code indicating the current key-on and the memory MEM are stored in the performance data memory.
Write the key code stored in (P). After the writing is completed, the write pointer of the performance data memory is advanced, and the process proceeds to step W9.

At step W9, the key code of the memory MEM (P) written in the performance data memory is stored in the register K.
Store in C. The key code stored in the register KC is used to determine whether or not the same key code is consecutive in the processing of the next array element number P + 1.

At step W10, the duration register D is reset to 0 in order to re-count the duration of the key code written together with the key-on code in the performance data memory this time. After that, the process returns to the process of FIG. 10 to perform the performance data generation process for the next array element number P + 1.

Next, the case where the same key code as the key code written in the performance data memory last time is stored in the memory MEM (P) will be described. That is, the memory M
This is the case where the key code of EM (P) and the key code of MEM (P-1) are the same.

In this case, it is determined in step W1 that all 1's are not stored in the memory MEM (P), and further in step W2, it is determined that the key is still on from the previous time, and the process proceeds to step W3.

In step W3, it is checked whether or not the key code stored in the memory MEM (P) is the same as the key code stored in the register KC. The register KC stores a key code that has been key-on since the last time.

If the key code of the memory MEM (P) and the key code of the register KC are the same, the current array element number P
Also indicates that the same key code as the previous time is to be sounded, so nothing is written to the performance data memory, the process returns to the process of FIG. 10, only the value of the duration register D is incremented, and the key code continues to sound. Count the time.

Next, a case where a key code different from the key code written in the performance data memory last time is stored in the memory MEM (P) will be described. That is, this is a case where the key code to be sounded changes during key-on.

In this case, it is determined in step W1 that all 1's are not stored in the memory MEM (P), and further in step W2, it is determined that the key is still on from the previous time, and the process proceeds to step W3.

At step W3, it is checked whether or not the key code stored in the memory MEM (P) is the same as the key code stored in the register KC. Memory M
If the key code of EM (P) and the key code of the register KC are not the same, the sounding instruction for the array element number P this time means sounding of a key code different from the previous one, so the process proceeds to step W4.

At step W4, the value of the duration register D indicating the duration of the key code sounding up to the last time is written in the performance data memory together with the duration code. After the writing is completed, the write pointer of the performance data memory is advanced to prepare for the next writing.

In step W5, a key-off code indicating the end of sounding of the previous key code is written in the performance data memory, and the write pointer of the performance data memory is advanced.

At step W8, the key-on code indicating the start of sounding for the current key code and the memory MEM are displayed.
The key code stored in (P) is written in the performance data memory, and the write pointer of the performance data memory is advanced.

Then, in step W9, the memory ME
The key code stored in M (P) is stored in the register KC, the duration register D is reset to 0 in step W10, and the next sounding duration time is counted again. Then, returning to the processing of FIG. 10, the next array element number P +
The process for 1 is performed.

Next, the performance data generation procedure when the memory MEM (P) stores not all the key codes but all 1s indicating a silent state (rest) will be described. In that case, it is determined in step W1 that all 1s are stored in the memory MEM (P), and the process proceeds to step W11.

At step W11, the key-on flag KO
Check whether N is 1. If the key-on flag KON is 1, step W12 is performed to instruct the key-off.
Then, the key-on flag KON is set to 0.

At step W13, the duration code indicating the duration of the key code sounding up to the last time and the value of the duration register D are written in the performance data memory. After the writing is completed, the write pointer of the performance data memory is advanced to prepare for the next writing.

In step W14, a key-off code indicating the end of sounding of the previous key code is written in the performance data memory, and the write pointer of the performance data memory is advanced. Then, the process proceeds to step W10, the duration register D is reset to 0, and the duration of the rest after the key-off is counted again. Then, the process returns to the process of FIG. 10 and the process for the next array element number P + 1 is performed.

After it is determined that all 1s are stored in the memory MEM (P), if it is determined in step W11 that the key-on flag KON is not 1, it indicates that the key-off state has continued from the previous time. ,
The process returns to the processing of FIG. 10 without writing anything in the performance data memory, increments the duration register D indicating the duration of the rest in the key-off, and the next array element number P +
The process for 1 is performed.

The procedure for generating performance data has been described above.
The format of the performance data is not limited to the above-mentioned format, and performance data in the MIDI format or the like may be generated. In the present embodiment, by detecting the average pitch after performing the quantization process on the voice signal input from the microphone, it is possible to detect the correct pitch without being adversely affected by the fluctuation of the pitch. it can.

Further, for the detected pitch, the pitch detected by erroneous recognition is corrected using the key obtained by the key detection. Pitch detection with a high recognition rate can be realized by performing pitch correction processing using a music rule determined by the key. The detected and corrected pitch is converted into playable performance data by performing performance data conversion processing.

In this embodiment, the key is detected from the input voice, but the player may input the key using a switch or the like. Although the present invention has been described above with reference to the embodiments, the present invention is not limited thereto.
For example, it will be apparent to those skilled in the art that various changes, improvements, combinations, and the like can be made.

[0133]

EFFECTS OF THE INVENTION By obtaining the average pitch of the pitches for each of a plurality of time intervals detected from the input voice signal, it is possible to reduce the adverse effect due to the fine fluctuation of the pitch. Correct pitch information can be obtained by generating pitch information based on the averaged pitch.

Further, after detecting the key based on the pitch information detected from the input voice signal constituting the music, the pitch information is corrected according to the music rule of the detected key.
Correct pitch information can be obtained by correcting the pitch information.

[Brief description of drawings]

FIG. 1 is a block diagram showing a system configuration example of an audio signal conversion device for carrying out the present invention.

FIG. 2 is a flowchart showing a main routine of input conversion processing.

FIG. 3 is a flowchart showing details of voice input processing and average pitch detection processing in step P3 of FIG.

FIG. 4 is a flowchart showing interrupt processing performed by a CPU.

FIG. 5 is a conceptual diagram showing a configuration example of a memory MEM ().

FIG. 6 is a flowchart showing details of a key code conversion process performed in step P6 in the main routine of FIG.

FIG. 7 is a flowchart showing details of correction processing for a key key code in step P7 in the main routine of FIG.

FIG. 8 is a conceptual diagram showing tone name data of a constituent tone of a key stored in a data memory.

FIG. 9 is a flowchart showing details of in-tone tone rounding processing in step T2 of FIG. 7.

10 is a flowchart showing details of the performance data conversion process of step P8 in the main routine of FIG.

FIG. 11 is a flowchart showing details of performance data conversion processing in step V4 of FIG.

[Explanation of symbols]

1 microphone, 2 D / A converter, 3 output terminals, 4 sound sources, 5 sound system, 6 timer, 7 CPU, 8 program memory,
9 working memory, 10 data memory,
11 bus

Claims (2)

[Claims] 1. A voice input means for inputting a voice supplied from the outside to generate a voice signal, and a voice signal generated by the voice input means is divided into predetermined time intervals to divide the time intervals. Pitch extracting means for extracting a pitch, average pitch calculating means for calculating an average pitch by averaging a plurality of pitches extracted by the pitch extracting means, and average pitch calculated by the average pitch calculating means Signal converting device having pitch information converting means for converting the sound into pitch information. 2. A voice input unit for externally inputting a voice constituting a song to generate a voice signal, a pitch extracting unit for extracting a pitch from a voice signal generated by the voice input unit, Pitch information conversion means for converting the pitch extracted by the pitch extraction means into pitch information, key information supply means for supplying key information, and key information supplied from the key information means A voice signal converter having pitch information correcting means for correcting the pitch information.