JP3116381B2 - Musical sound compression method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a tone generator for generating tone signals by executing a sound source processing program based on an adaptive difference pulse code modulation system to generate difference data of adaptively quantized tone signals. The present invention relates to a method of compressing a musical sound waveform for performing a sound compression.

[0002]

2. Description of the Related Art Various electronic musical instruments having high performance have been realized by the development of digital signal processing technology and LSI processing technology. In particular, with the cost reduction of semiconductor memory, musical tone waveforms of natural musical instruments and the like are stored in the memory as digital signals,
An electronic musical instrument capable of producing realistic musical sound waveforms by reading this at the pitch etc.
It contributes to a huge increase in the music population regardless of amateur.

[0003] Such a sound source system includes a PCM (pulse code modulation) system, a DPCM (differential pulse code modulation).
And ADPCM (Adaptive Differential Pulse Code Modulation). In particular, the ADPCM method is an excellent sound source method because large data compression is possible.

[0004] Since the musical tone waveform generator employing the above-mentioned sound source system requires a large amount of high-speed digital operation, an architecture equivalent to a tone generation algorithm based on the required sound source system has conventionally been implemented by hardware. It consisted of a dedicated tone generator circuit. But,
The problem is that the hardware scale becomes large, the cost increases in the manufacturing stage, and the size of the musical tone waveform generator increases in terms of the yield at the time of manufacturing the LSI chip when it is realized by the LSI. Had.

On the other hand, in recent years, many high-performance microprocessors for performing general-purpose data processing have been realized, and sound source processing can be performed by software using such a microprocessor. For example, by performing parallel processing with a plurality of processor configurations, processing of an advanced sound source system such as the ADPCM system has become possible.

[0006]

Here, when the ADPCM method is adopted as the sound source processing method, the adaptively quantized difference data is stored in a waveform memory or the like, and inverse quantization is performed while reading the data. Thus, the difference value of the tone signal is decoded, and the difference value is accumulated at each timing to generate a tone signal.

The above-mentioned adaptively quantized difference data is represented by P
It is obtained by calculating a difference signal with respect to the CM tone signal and performing adaptive quantization such that the amplitude level of the difference signal is compressed and quantized.

In the process of generating the adaptively quantized difference data, if the amplitude level of the difference signal is compressed and quantized, a quantization error having a maximum half width of each quantization width is generated. I do. Then, in the sound source processing, when the inverse quantization is performed based on the difference data including such a quantization error, and the difference value obtained thereby is accumulated to generate a tone signal, the obtained tone signal is In some cases, the amplitude value slightly exceeds the maximum or minimum amplitude value that can be represented by the number of bits of the digital signal assigned to the musical tone signal by an amount corresponding to the quantization error.

In such a case, if no measures are taken,
Even if the amplitude value slightly exceeds the maximum or minimum amplitude value, in the case of a digital signal, sign inversion or the like occurs. That is,
Even if the amplitude value becomes slightly larger than the positive maximum value, the digital amplitude value is folded back to the negative amplitude value side. Conversely, if the amplitude value becomes slightly smaller than the negative minimum value, the digital amplitude value will turn back to the positive amplitude value side.

When such a situation occurs, the waveform of the obtained musical tone signal becomes a waveform far apart from the change of the actual waveform, and if the musical tone signal is produced as it is, a large noise is generated. Therefore, there is a problem that the tone quality of the musical tone signal is remarkably deteriorated.

The present invention relates to an ADPCM as a sound source processing system.
It is an object of the present invention to provide a musical sound waveform compression technique capable of minimizing a noise component generated in a musical sound signal in a sound source processing process due to a quantization error when a system is adopted.

[0012]

SUMMARY OF THE INVENTION The present invention provides a tone waveform compression method for generating adaptively quantized difference data used in a tone waveform generator for generating a tone signal by an adaptive difference pulse code modulation method. Assuming that the following steps are executed.

First, a difference calculation step of calculating the difference between the value of the digital musical tone signal at the current timing and the accumulated value of a reproduced difference value to be described later up to the previous timing to obtain a difference value at the current timing is described. Be executed.

Next, an adaptive quantization step of performing adaptive quantization while compressing the difference value of the current timing obtained in the difference calculation process to obtain adaptively quantized difference data of the current timing is executed. .

Further, the adaptively quantized difference data at the current timing obtained in the adaptive quantization process is inversely quantized,
An inverse quantization process for calculating the reproduced difference value at the current timing is performed.

Subsequently, the reproduced difference value at the current timing obtained in the inverse quantization process is accumulated to the accumulated value of the reproduced difference values up to the previous timing, and the reproduced value up to the current timing is reproduced. An accumulation process for obtaining an accumulated value of the difference values is performed.

Also, whether or not the accumulated value of the reproduced difference values up to the current timing obtained in the accumulation process exceeds the maximum change width of the number of bits allocated to the tone signal for generating the tone signal. Is determined.

Further, when it is determined by the determining step that the accumulated value of the reproduced difference values up to the current timing does not exceed the maximum change width, the current timing obtained in the adaptive quantization step is obtained. An output process of outputting the adaptively quantized difference data of the above is performed.

In addition, if it is determined in the determining step that the accumulated value of the reproduced difference values up to the above-mentioned present timing exceeds the above-mentioned maximum change width, the current value obtained in the adaptive quantization step is obtained. The value of the adaptively quantized difference data at the timing of the above is changed so that the accumulated value of the reproduced difference values up to the current timing obtained in the accumulation process based on the value does not exceed the maximum change width. Then, a difference data changing step of repeating the above-described inverse quantization step, accumulation step and determination step is executed. The changing operation in this case is, for example, an operation of changing the absolute value of the adaptively quantized difference data at the current timing to a data value smaller by one data value.

Then, a repetition process of repeating each process up to this point for each tone signal generation timing is executed.

[0021]

According to the present invention, in the inverse quantization step and the accumulation step, the reproduction processing similar to the sound source processing is performed on the adaptively quantized difference data at the current timing obtained in the adaptive quantization step. Then, the accumulated value of the reproduced difference value at the current timing similar to the peak value of the musical sound signal which will be obtained by the sound source processing is obtained.

Then, by using the accumulated value corresponding to the reproducing side to execute the above-described difference calculation process,
In the operation of calculating the difference value of the tone signal, an error caused by performing the adaptive quantization can be absorbed.

In particular, in the present invention, it is obtained in the adaptive quantization step by determining whether or not the above-mentioned accumulated value corresponding to the reproducing side exceeds the maximum change width of the number of bits allocated to the tone signal. When the sound source processing is performed based on the adaptively quantized difference data at the current timing,
It can be determined whether or not the sign of the output value has been inverted.

If it is determined in this determination step that the accumulated value has exceeded the above-mentioned maximum change width, the difference data change step includes the adaptive quantization of the current timing obtained in the adaptive quantization step. The value of the obtained difference data is changed so that the accumulated value obtained based on the value does not exceed the maximum change width, and the above-described inverse quantization step, accumulation step, determination step and output step are repeated. Let

By outputting the adaptively quantized difference data of the current timing changed in this way, it is possible to prevent in advance the occurrence of the sign inversion of the output value when the sound source processing is executed. And the occurrence of noise can be minimized.

[0026]

Embodiments of the present invention will be described below with reference to the drawings. Diagram 1 of the present embodiment is an overall configuration diagram of an embodiment of the present invention.

In the figure, first, the entire apparatus is controlled by a microcomputer 101. In particular, not only the processing of control input of the musical instrument but also the processing of generating musical tones are executed by the microcomputer 101, and a tone generator for generating musical sounds is not required.

A switch section 104 comprising a keyboard 102 and function keys 103 is an operation input portion of the musical instrument. Performance information input from the switch section 104 is processed by the microcomputer 101.

The tone signal after analog conversion generated by the microcomputer 101 is smoothed by a low-pass filter 105, amplified by an amplifier 106,
7 is emitted. The power supply circuit 108 supplies necessary power to the microcomputer 101, the low-pass filter 105, and the amplifier 106.

Next, FIG.
FIG. 2 is a block diagram showing an internal configuration of the device. The control data / waveform ROM 212 stores tone control parameters such as a target value of an envelope value to be described later, tone difference waveform data in ADPCM (adaptive differential pulse code modulation), and quantized data. Then, the command analysis unit 207 sequentially analyzes the contents of the program in the control ROM 201 while controlling the control data / waveform ROM 212.
Each of the above data is accessed to perform sound source processing by software.

The control ROM 201 stores a tone control program to be described later, and sequentially outputs a program word (instruction) at a designated address from the ROM address control unit 205 via the ROM address decoder 202. Specifically, the word length of each program word is, for example, 28
And a part of the program word is designated as RO as the lower part (in-page address) of the address to be read next.
The next address method is input to the M address control unit 205. Of course, it may be constituted by a CPU of an ordinary program counter system.

The command analysis unit 207 includes a control ROM 2
It analyzes the operation code of the instruction output from 01 and sends a control signal to each part of the circuit to execute the specified operation.

The RAM address control unit 204 includes a control R
When the operand of the instruction from the OM 201 specifies a register, the address of the corresponding register in the RAM 206 is specified. 9 and 10 are stored in the RAM 206.
For example, various tone control data described later are stored for eight sounding channels, and various buffers and the like described later are stored and used for sound source processing described later.

When the instruction from the control ROM 31 is an operation instruction, the ALU unit 208 and the multiplier 209 execute addition / subtraction and logical operation, and the latter execute multiplication based on an instruction from the command analysis unit 207.

The interrupt control unit 203, based on an internal hard timer (not shown), executes
ROM address control unit 205 and D / A converter unit 213
Supply an interrupt signal.

The switch unit 104 of FIG. 1 is connected to the input port 210 and the output port 211. RO for control
Various data read from the M201 or the RAM 206 are transmitted to the ROM address control unit 205 and the ALU via the bus.
Section 208, multiplier 209, control data / waveform ROM2
12, the D / A converter section 213, the input port 210, and the output port 211. Also, the ALU unit 208,
Each output of the multiplier 209 and the control data / waveform ROM 212 is supplied to the RAM 206 via the bus.

FIG. 4 shows the D / A converter 21 of FIG.
3 shows an internal configuration of the latch 3 via a data bus.
01 is input. When the tone generator processing end signal is input from the command analyzer 207 of FIG. 2 to the clock input of the latch 301, the musical tone data for one sample on the data bus is latched by the latch 301 as shown in FIG. .

Here, the time required for the sound source processing is as follows.
Since the sound source processing varies depending on the execution condition of the software for the sound source processing, the timing at which the sound source processing ends and the tone data is latched in the latch 301 is not constant. Therefore, as shown in FIG. 3, the output of the latch 301 is directly used as the D / A converter
03 cannot be entered.

Therefore, in this embodiment, as shown in FIG. 4, the output of the latch 301 is further latched by the latch 401, and the tone signal is latched by an interrupt signal output from the interrupt control unit 203 shown in FIG. The data is latched at 401 and output to the D / A converter 303 at regular intervals.

As described above, the use of two latches absorbs a change in the processing time in the tone generator system, so that a complicated timing control program for outputting musical sound data to the D / A converter is unnecessary. Next, the overall operation of the present embodiment will be described.

In this embodiment, the microcomputer 10
As shown in the main flowchart of FIG.
₆₀₂ repeatedly performs a series of processes to S _610. The actual sound source processing is performed by interrupt processing. More specifically, the program executed as the main flowchart of FIG. 6 is interrupted at certain intervals, and a sound source processing program for generating an 8-channel tone signal is executed based on the interrupt. When the processing is completed, the tone waveforms for eight channels are added and output from the D / A converter unit 213 connected to the microcomputer 101. Thereafter, the process returns from the interrupt state to the main flow. Note that the above-described interrupt is periodically performed based on a hard timer in the interrupt control unit 203 in FIG. This period is equal to the sampling period at the time of outputting the musical sound.

The above is the schematic operation of this embodiment. Next, the overall operation of this embodiment will be described in detail with reference to FIGS. The main flowchart of FIG. 6 shows a flow of processing other than the sound source processing, which is executed by the microcomputer 101 in a state where no interrupt is issued from the interrupt control unit 203.

First, the power is turned on, and initial settings such as the contents of the RAM 206 (see FIG. 2) in the microcomputer 101 are performed (S ₆₀₁ ). Next, each switch of the function key 103 (see FIG. 1) connected to the outside of the microcomputer 101 is scanned ( _S602 ), and the state of each switch is taken into the key buffer area in the RAM 206 from the input port 210. As a result of the scanning, the function key whose state has changed is identified, and the processing of the corresponding function is performed ( _S603 ). For example, setting of a musical tone number, setting of an envelope number, and setting of a rhythm number if the additional function has rhythm performance.

Thereafter, the keyboard key pressed on the keyboard 102 in FIG. 1 is captured in the same manner as in the case of the function key ( _S604 ), and a key assignment process is performed by identifying the changed key ( _S604 ). _S605 ).

Next, when a demo performance key (not shown) is pressed with the function key 103 (FIG. 1), demo performance data (sequencer data) is sequentially read from the control data / waveform ROM 212 of FIG. Then, key assignment processing and the like are performed ( _S606 ). When the rhythm start key is pressed, the rhythm data is stored in the control data / waveform R
The data is sequentially read from the OM 212, and key assignment processing and the like are performed ( _S607 ).

Thereafter, a timer process described below is performed ( _S608 ). That is, the time value of the incremented time data is determined in an interrupt timer process ( _S702 ) described later, and the sequencer data for time control sequentially read out for demo performance control or the time control for time control read out for rhythm performance control By performing comparison with the rhythm data of step _S606 , time control for performing the demonstration performance of _S606 or the rhythm performance of _S607 is performed.

Further, in the sound generation processing _S609 , a pitch envelope processing or the like is performed in which an envelope is added to the pitch of the musical sound to be sound-processed, and pitch data is set in the corresponding sound generation channel.

Further, a one-round flow preparation process is executed.
( _S610 ). In this process, alter or change the state of the sound channel of the note number that became a key-start in keyboard key processing of S ₆₀₅ in the key depression, the state of the sound channel of the note number that became a key release in the silencer Are performed.

Next, the interrupt processing shown in FIG. 7 will be described. 6 by the interrupt control unit 203 of FIG.
When the program corresponding to the main flow is interrupted, the processing of the program is interrupted, and the execution of the interrupt processing program of FIG. 7 is started. In this case, in the interrupt processing program, control is performed so that the contents of registers and the like to which writing is performed in the main flow program of FIG. 6 are not performed. Therefore, it is unnecessary to save and restore the registers at the start and end of the normal interrupt processing.
As a result, the transition between the processing of the main flowchart in FIG. 6 and the interrupt processing is quickly performed.

Subsequently, sound source processing is started in the interrupt processing ( _S701 ). This sound source processing is shown in FIG. As a result, tone waveform data in which eight sounding channels are accumulated is obtained in a buffer B described later in the RAM 206 of FIG.

[0051] In addition, in S _702, interrupt timer processing is performed. Here, the value of the time data (not shown) on the RAM 206 (FIG. 2) is incremented by utilizing the fact that the interrupt process of FIG. 7 is executed at a constant sampling period. That is, the value of the time data indicates the passage of time. The time data thus obtained is used for time control in the timer process _S608 of the main flow of FIG. 6, as described above.

Then, in _S703 , the contents of the buffer area are latched by the latch 301 (FIG. 4) of the D / A converter section 213. Next, the operation of the sound source processing executed in step _S701 of the interrupt processing will be described with reference to the flowchart of FIG.

First, an area for adding waveform data in the RAM 206 is cleared (S ₈₀₁ ). Then, tone generator processing is performed for each channel of the sound channel (S ₈₀₂ ~S _809),
Finally, when the sound source processing for the eighth channel is completed, waveform data obtained by adding the data for the eight channels to the predetermined buffer area B is obtained. Details of these processes will be described later.

FIG. 9 is a flow chart conceptually showing the relationship between the processes in the flowcharts of FIGS. 6 and 7 described above. First, a certain process A (hereinafter, the same applies to B, C,..., F) is performed (S ₉₀₁ ). This “processing” corresponds to, for example, “function key processing” or “keyboard key processing” in the main flowchart of FIG. Thereafter, an interrupt process is started, and a sound source process is started ( _S902 ). This allows
A tone signal in which eight sounding channels for one sample are collected is obtained and output to the D / A converter unit 213. After that, the process returns to some process B in the main flowchart. Above operation is, the sound source processing for all eight sound channel is repeated while made (S ₉₀₄ ~S _911). This repetition processing is continued while the musical sound is being generated. Data structure of tone generator processing Next, a specific example of tone generator processing executed in S ₇₀₁ of FIG.

In this embodiment, the microcomputer 10
As described above, 1 shares the sound source processing for eight sounding channels. The data for sound source processing for these eight channels has a data format as shown in FIG.
This is set in an area for each sounding channel in 206.

In this RAM 206, a buffer B for waveform accumulation as shown in FIG. In this case, various control data for sound source processing based on the ADPCM method are stored in each sounding channel area in FIG.
In the figure, A is represents the address specified when adaptive quantized differential data at the sound source processing (described later) is read, an integer part of A _I the current address, control data and the waveform ROM 212 (Fig. 2 ) Directly corresponds to the address where the adaptive quantization difference data is stored. Also, A _F
Is a decimal part of the current address, which is used for interpolation of the dequantized difference value read from the ROM 412. The integer part of the next P _I is the pitch data, P _F represents the fractional part of the pitch data. For example, P _I = 1, P _F = 0 indicates the pitch of the original sound, P _I = 2, P _F = 0 indicates the pitch one octave higher, and P _I = 0, P _F = 0.5. Represents a pitch one octave below. Other various control data will be described in detail when the ADPCM system is described later.

In this embodiment, when the main flow of FIG. 6 is executed, control data necessary for sound source processing, such as pitch data and envelope data, are set in the corresponding sounding channel area. Then, in the sound source processing corresponding to each channel shown in FIG. 8 which is executed as the sound source processing in the interrupt processing shown in FIG. 7, the tone generation processing is executed while using the various control data set in the sounding channel area. You. As described above, data communication between the main flow program and the sound source processing program is performed by RA
This is performed via the control data (musical tone generation data) of the tone generation channel area on M206, and the access to the tone generation channel area in each program may be performed irrespective of the execution state of the other program. An independent module configuration can be used, and a simple and efficient program structure can be obtained.

Furthermore, ROM 212 for control data and waveform
17 stores adaptive quantized difference data (described later) in a data format (not shown).
(a) and an expansion table as shown in FIG. 18 (b) are stored.
This table is data used to perform inverse quantization of the adaptive quantization difference data read from the control data / waveform amount ROM 212 as described later. In addition, the ROM 212 stores control data for musical tone generation corresponding to each tone, and when a certain tone is set by a player, the ROM 212 stores the RAM 206
The above-mentioned control data is transferred to each sounding channel area of the above,
Is set. The principle of the sound source processing by the ADPCM method will be described below.
The sound source processing of the ADPCM method, which is the sound source processing for any one channel (any of S _{802 to} S ₈₀₉ ), will be described. This sound source processing is performed by the microcomputer 10
The first command analysis unit 207 is realized by interpreting and executing a sound source processing program stored in the control ROM 201. In the following, unless otherwise noted,
It is assumed that processing is performed on this premise.

First, an outline of the operation principle of the ADPCM system will be described. First, in FIG. 12, the sample data X _P corresponding to the address A _I for control data and the waveform ROM 212 (Fig. 2) is, prior to one of the addresses A _I, the address (A _I -1) which is not shown in FIG. This is a value obtained from the difference value with the corresponding sample data.

At the address A _I of the control data / waveform ROM 212, adaptive quantization difference data for obtaining a difference value D from the next sample data is written, and the difference value D is obtained from this data. Thus, the sample data of the next address Motomari in X _P + D, which replaces the new sample data X _P.

In this case, assuming that the current address is A _F as shown in the figure, the sample data corresponding to the current address A _F can be obtained by X _P + D × A _F. Thus, AD
In the PCM method, adaptive quantization difference data for obtaining a difference value D between a current address and sample data corresponding to the next address is stored in the control data / waveform ROM 212.
(FIG. 2), the difference value D is calculated based on the calculated value, added to the current sample data, and the next sample data is obtained, thereby sequentially generating waveform data.

When such an ADPCM system is adopted,
When quantizing a waveform such as a voice or a musical tone in which the difference value between adjacent samples is generally small, it is apparent that the quantization can be performed with a smaller number of bits as compared with the ordinary PCM method.

Further, in the ADPCM method, the following principle of adaptive quantization is adopted.
Now, for example, when storing a difference value of a musical tone waveform in a memory, n is quantized to a difference value that can take an amplitude value (eg, a voltage value) of a maximum ± E while securing a constant S / N.
It is assumed that a bit quantization bit number is required. On the other hand, when only a data amount of m bits smaller than n bits per sample data is allocated due to a limitation of a memory capacity or the like, the difference value must be somehow quantized by m bits per sample.

For this reason, the range of the amplitude value of ± E is usually divided into 2 ^m, but this increases the quantization width and degrades the S / N. Therefore, in the ADPCM method, when storing a difference value of a musical tone waveform or the like in a memory, a constant S / N can be secured, and the absolute value of an amplitude value that can be expressed by m bits is smaller than | ± E |. The range of | ± e | is determined. Then, the difference value whose absolute value of the amplitude value is larger than | ± e | is divided by one or more normalization coefficients, compressed so as to fall within the range, and then quantized with m bits.
It is stored in a memory. This makes it possible to reduce the number of quantization bits per sample data while securing a constant S / N, and reduce the memory capacity.
Such an operation is called adaptive quantization.

Here, in order to reproduce the original difference value from the difference data adaptively quantized as described above, the above-mentioned normalization coefficient at the time of performing the adaptive quantization of each sample data is also stored. In addition, when the adaptively quantized difference data is read from the memory, the corresponding normalization coefficient must also be read out, and the original difference value must be reproduced by multiplying the adaptively quantized difference data. . This operation is called inverse quantization.

However, storing the above-described normalization coefficient for each sample data requires a storage capacity equivalent to that obtained by quantizing a difference value having an amplitude value of ± E in n bits. Therefore, data compression cannot be realized.

An embodiment of the sound source processing according to the ADPCM method which can suppress an increase in storage capacity will be described below. Embodiment of the sound source processing by the ADPCM method This embodiment utilizes the fact that the amplitude distribution of the difference value of the musical tone waveform becomes a distribution specific to the musical tone waveform in advance, and uses each amplitude of the difference value instead of the normalization coefficient. 17 (a) and 18 (a), a table of compression characteristics, which will be described later, is used to determine which value is to be compressed. When the difference value is stored in the memory (212 in FIG. 2), Adaptively quantized while compressing the data and storing it. At the time of sound source processing, the original difference value is reproduced by performing inverse quantization while expanding the difference value adaptively quantized by the expansion characteristic, which is the inverse characteristic of the compression characteristic, and is accumulated to obtain the tone waveform. Can be played. Therefore, the control data / waveform ROM 212
(FIG. 2) shows FIG. 17 for performing expansion with the above expansion characteristics.
(b), an expansion table as described later with reference to FIG. 18 (b) is stored.

The operation of the embodiment of the sound source processing of the ADPCM system will be described with reference to the operation flowcharts of FIGS. Each variable in the flow is secured in the corresponding sounding channel area on the RAM 206 in the microcomputer 101 of FIG. 2 in the data format of FIG.

The sound source processing based on FIGS. 13 to 15 is roughly divided into an envelope processing (S _{1301 to} S ₁₃₀₇ ) and a waveform processing (S _{1308 to} S ₁₃₂₄ ). First, FIG.
Before the waveform processing based on the principle of the ADPCM method shown in FIG. 13, the envelope processing will be described.

FIG. 16 is a diagram showing an envelope generated by the envelope processing. Envelope value E that is generated by the processing of S ₁₃₀₁ to S _1307, by being multiplied by the tone waveform output O in step S ₁₃₂₃ to be described later, the envelope is added to each sample data of the tone waveform.

The envelope added to each tone waveform data is composed of several steps (segments) in time. In the figure, an example of four steps is shown. In the figure, Δx is the sampling period of the envelope, and Δy is the change width of the envelope value.

In the envelope processing S _{1301 to} S ₁₃₀₇ , calculation of the envelope value E for each sampling timing,
A check is made as to whether the value has reached the target envelope value OE of the current step. Then, when E reaches OE, it is detected in the sound generation processing of _S609 in the main flow of FIG. 6, and the data (Δx, Δy and target envelope value OE) for the envelope of the next step are shown in FIG. 2 control data and waveform ROM2
12 are set in the corresponding sounding channel area (see FIG. 10) on the RAM 206 (FIG. 2).

Specifically, in S ₁₃₀₁ , the timer value Δx _t for comparison with the envelope calculation period Δx is incremented at each interrupt timing. Next, S
In _1302, whether [Delta] x matches the [Delta] x _t is determined.

If they do not match, no envelope processing is performed. In S _1302, if the [Delta] x is determined to match the [Delta] x _t, in S _1303, the sign bit of variation Δy of the envelope values are determined.

If the sign bit is positive and the determination in S ₁₃₀₃ is YES, that is, if the envelope is rising, the change width Δy is added to the current envelope value E in step S ₁₃₀₄ .

Conversely, if the sign bit is negative and the determination in S ₁₃₀₃ is N
O, see i.e., envelope in the case of descending, in step S _1305, the variation width Δy from a current envelope value E is subtracted.

Thereafter, in _S1306 , it is determined whether or not the current envelope value E has become equal to or larger than the target envelope value OE. When E becomes OE or more, the current envelope value E is replaced with the target envelope value OE.

As described above, this is detected in the sound generation processing of _S609 in the main flow of FIG. 6, and data for the envelope of the next step is set on the RAM 206. If 0 is detected as the current envelope value E in the sound generation processing, the processing is performed as the end of sound generation.

Next, the waveform processing of S _{1308 to} S ₁₃₂₄ based on the principle of the ADPCM system shown in FIG. 12 will be described. Of the addresses at which the adaptive quantization difference data is stored in the control data / waveform ROM 212 (FIG. 2), the addresses at which the data to be processed at present are stored are indicated by (A _I, A in FIG. 10). _F ).

First, pitch data (P _I, P _F ) is added to the current address (A _I, A _F ) (S ₁₃₀₈ ). This pitch data corresponds to a key number of a key pressed on the keyboard 102 of FIG.

Then, the integer part A _I of the added address
It is determined whether or not there is a change ( _S1309 ). Judgment is N
If it is O, the interpolation data value O corresponding to the decimal part A _F of the address is calculated by the calculation process of D × A _F using the difference value D at the address A _I in FIG. 12 (S ₁₃₂₁ ). Note that the difference value D is obtained by the sound source processing at the interrupt timing before this time (see _S1314 or _S1317 described later).

Next, the sample data X _P corresponding to the integer part A _I address to the interpolation data value O is added, the current address (A _I, A _F) new sample data corresponding to O (in FIG. 12 X (Corresponding to _Q ) is obtained ( _S1322 ).

Thereafter, the sample data is multiplied by the envelope value E obtained by the above-described envelope processing (S ₁₃₂₃ ), and the contents of the obtained O are stored in the waveform data buffer B (FIG. 11) in the RAM 206 (FIG. 2). (See S ₁₃₂₄ ). The buffer B, tone generator processing (FIG 8S ₈₀₂ to S ₈₀₉₎ tone waveform output produced by for the other sound channel are accumulated, finally as 8 channels are accumulated data, one sample Is generated.

Thereafter, returning to the main flow of FIG.
FIG. 13 shows an interrupt in the sampling period of FIG.
-The operation flow chart of the sound source processing of FIG.
The current address (A_I,A_F) To pitch data
(P_I,P_F) Is added (S ₁₃₀₈).

The above operation is repeated until a change occurs in the integer part A _{I of the} address. During this time, the sample data _XP and the difference value D are not updated, only the interpolation data O is updated according to the address _AF, and new sample data _XQ is obtained each time.

[0086] Next, in _S1308 , the current address (A
_If the integer part A _I of the current address changes as a result of adding the pitch data (P _I, P _F ) to the current address ( _I, A _F ) (S ₁₃₀₉ ), the address A _I reaches or exceeds the end address A _E. It is determined whether or not it is ( _S1310 ).

If the determination is NO, the following S _{1311 to} S ₁₃₁₄
, The sample data corresponding to the integer part A _I of the current address is calculated. That is, first, the value before the change of the integer part A _I of the current address is stored in the variable A ₁ (see FIG. 10). this is,
Is realized by repeating the processing of S ₁₃₁₃ or S ₁₃₂₀ will be described later.

While the value of the old A _I is sequentially incremented in S ₁₃₁₃ , the adaptive quantization difference data on the control data / waveform ROM 212 (FIG. 2) designated by the old A _I is read in S ₁₃₁₄ . , The difference value D is calculated by the inverse quantization process. As described above, this inverse quantization is performed by inputting the adaptive quantization difference data to the ROM 2.
17 (b) and 18 (b) stored on the memory 12 and directly expanded by an expansion table having characteristics as shown in FIGS. 17 (b) and 18 (b), whereby the difference value D is calculated.

In the case of a musical tone waveform signal, since its waveform characteristics are known in advance, FIG.
By preparing a plurality of sets of expansion tables as shown in (a) or FIG. 18 (b) and storing them in the control data / waveform ROM 212, based on the size of the current adaptive quantized difference data, It can be directly dequantized by a decompression table corresponding to the tone to be generated. For this reason, it is possible to perform much more accurate adaptive quantization than when applied to signals whose average characteristics are known for a long time, such as a normal communication voice signal. Recording / reproduction of a musical tone signal can be performed at a desired rate.

In this case, table selection data TBL indicating which expansion table is to be used for expansion in accordance with the setting of the tone color by the player, etc., is stored in the RAM 206 (see FIG. 2) Figure 10 above
Are set in each sounding channel area shown in FIG. And
At the time of the above expansion processing, the expansion table corresponding to the table selection data TBL is stored in the control data / waveform ROM 212.
Is read from.

As described above, the difference value D is calculated in step _S1314 . Then, the difference value D is accumulated in the sequential sample data X _P at step S _1312.
The results above operation is repeated, when the value of the old A _I is equal to the integer portion A _I of the current address after the change, the sample data X _P value integer part A of the current address after change
_The value corresponds to _I.

Thus, the integer part A of the current address
If the sample data X _P corresponding to _I is obtained, the determination is YES in S _1311, proceeds to the processing of the aforementioned interpolated value (S _1321).

[0093] sound source processing described above is repeated for each interrupt timing, the determination of S ₁₃₁₀ is When changes to YES, starts the processing for the next loop reproduction. Loop reproduction is a process executed when a predetermined waveform region of a musical tone signal is repeatedly and continuously generated, and when a preset loop address A _L to an end address A _E is set to a mute instruction by a player. This is a process that is repeatedly sounded until it occurs. As a result, it is possible to generate sound for a long time without storing all the tone signals.

[0094] First, an end address A _E min addresses beyond the (A _I -A _E) is added to the loop address A _L, the resulting address is a new current address integer part A
_It is set to _I ( _S1315 ).

The operation of calculating and accumulating the difference value D is repeated several times depending on how much the address has advanced from the loop address A _L to correspond to the integer part A _I of the new current address. Sample data _XP is calculated.

[0096] That is, first, the value of the sample data X _PL (see FIG. 10) of the loop the address A _L that sample data X _P as the initial configuration is set in advance, the old A
_I is set to the value of the loop address A _L (S ₁₃₁₆ ).

A _L and X _PL are read in advance from the control data / waveform ROM 212 into each sounding channel area of the RAM 206 (see FIG. 10). It should be noted that these may be set by the player by means not specifically shown.

Subsequently, the following steps S _{1317 to} S ₁₃₂₀ are repeated. That is, while the value of the old A _I is sequentially incremented in S ₁₃₂₀ , the adaptive quantization difference data on the control data / waveform ROM 212 (FIG. 2) indicated by the old A _I is read in S ₁₃₁₇ , _The difference value D is calculated by the same inverse quantization (expansion) processing as in the case of ₁₃₁₄ .

[0099] Then, the difference value D is accumulated in the sequential sample data X _P in S _1319. As a result of repeating the above operation, when the value of the old A _I becomes equal to the integer part A _I of the changed current address, the sample data X _P
Is the integer part A _I of the new current address after loop processing.
Becomes the value corresponding to.

[0100] In this way, the sample data X _P corresponding to the integer part A _I of the new current address is obtained, S
The determination at ₁₃₁₈ is YES, and the process proceeds to the above-described interpolation value calculation processing ( _S1321 ).

As described above, A for one sounding channel
Waveform data according to the DPCM method is generated. Embodiment of Adaptive Quantization Processing of Musical Sound by ADPCM Method Next, in order to enable the above-described sound source processing, adaptive quantization (compression processing) is performed on a PCM musical sound signal to obtain adaptive quantization difference data. An example of the processing will be described.

In the adaptive quantization processing employed in this embodiment,
As described above, a compression table is determined to which value each amplitude of the difference value of the PCM tone signal is to be compressed, and the difference value is adaptively quantized and stored while being compressed by the compression table. At the time of sound source processing, the original difference value is reproduced by performing inverse quantization while expanding the difference value adaptively quantized with a decompression table having the inverse characteristic of the compression characteristic, and is accumulated to obtain the tone waveform. To play.

Therefore, for each characteristic of a tone signal, for example, for each type of tone such as flute, cymbal, shakuhachi, etc., FIG.
7 (a), a compression table as shown in FIG. 18 (a) is determined. In these figures, d (ADR) on the vertical axis is the difference value of the PCM tone signal, and G ｛d obtained by table conversion.
(ADR)｝ is adaptively quantized difference data.

FIG. 17 (a) shows a characteristic having a large compression ratio, and is used for compressing a tone signal such as a cymbal in which a PCM difference signal can have a large value. On the other hand, FIG. 18A has a characteristic that the compression ratio is small, and is used for compression of a tone signal in which the PCM difference signal cannot have a very large value such as a flute.

Also, decompression tables as shown in FIGS. 17 (b) and 18 (b) are determined in correspondence with the above-described compression tables. This decompression table has the inverse characteristics of the compression tables of FIGS. 17 (a) and 18 (a) and the like, and the adaptive quantization difference data rom (ADR) on the horizontal axis is subjected to expansion conversion to obtain the reproduction difference value J.
{Rom (ADR)} is obtained. This expansion table is, as described above, the control data and waveform ROM of FIG.
Although stored in 212 and used for sound source processing, it is also used in the following adaptive quantization processing.

After selecting the compression table as described above, the adaptive quantization processing shown in the operation flowchart of FIG. 19 is executed, and the adaptive quantization difference data is obtained.
Note that this flowchart shows the control data and waveform RO
Since the process is for obtaining the adaptive quantization difference data stored in M212 (FIG. 2), the process is executed not by the musical instrument itself but by a control device such as a general-purpose computer (not shown). The table is also prepared on a memory (not shown). It is also assumed that a PCM tone signal as shown in FIG. 20A is selectively stored in advance on a waveform memory (not shown). Further, for simplification of description, when a PCM tone signal of a certain musical instrument is selected, the signal is stored from address 0 on the waveform memory to FFF (hexadecimal representation). Furthermore, the symbol values used in the following description actually indicate the contents of variables provided on the RAM, but in the description, they will be described simply as symbols indicating values.

In FIG. 19, first, P
The storage address value ADR of the CM tone signal is initialized to the head address 0 of the tone signal ( _S1901 ). Then, while the address value ADR is incremented by +1 at step S _1910, until it is determined that the last address FFF (S _1911), PCM tone signal is read out sequentially, the following adaptive quantization steps S ₁₉₀₂ to S ₁₉₁₄ Is performed.

That is, the PCM tone signal value PCM (AD) read from the address corresponding to the address value ADR
R) is set as the input value H (ADR) (S ₁₉₀₂ ). Next, the accumulated peak value R (ADR-1) up to the previous time is subtracted from the value of the input value H (ADR), and the result is set as the current PCM difference value d (ADR) ( _S1903 ). Cumulative peak value R (A
DR-1) will be described later.

Next, the PCM difference value d (ADR) is
The compression conversion is performed by a compression table selected in advance from FIG. 17A and FIG. 18A, and the compression value G ｛d (AD
R)} is the current adaptive quantized difference data rom (AD) stored in the control data / waveform ROM 212 (FIG. 2).
R) ( _S1904 ).

Subsequently, the adaptive quantization difference data ro
m (ADR) is inverse quantized is decompression conversion by stretching table corresponding to the compression table, the conversion value is the restored difference value _{K (ADR) (S 1905)} . Then, this value is added to the accumulator peak value R (ADR-1) up to the previous, current accumulated peak value R (ADR) is calculated (S _1906).

In these processes, the above-described adaptive quantized difference data rom (ADR) is subjected to the same reproduction process as the sound source process, and the current accumulated peak value R ( (ADR). And
Using the accumulated peak value R (ADR) corresponding to the reproducing side, in the aforementioned step _S1903 , the next PCM difference value d
By calculating (ADR), it is possible to absorb an error caused by performing adaptive quantization.

The accumulated peak value R obtained as described above
In this embodiment, (ADR) is obtained as 12-bit data including positive and negative signs in the sound source processing. Then, it is determined whether or not this absolute value exceeds the maximum value 2047 (S ₁₉₀₇ ).

If not, it is further determined whether or not the current address ADR is equal to the previously set loop address A _L for loop reproduction (see FIG. 14) (S ₁₉₀₈ ).

If they are not equal, the address ADR is incremented by 1 (S ₁₉₁₀ ), and the process returns to step S 2002 through the determination in step S ₁₉₁₁ . Here, when it is determined that the current address ADR is equal to the loop address A _L (S ₁₉₀₈₎
Is YES), the accumulated peak value R (ADR) at that time
There is a loop peak value X _PL, it is stored in the control data and the waveform for ROM212 of Fig. 2 with the loop address A _L. These data are used in the above-described loop reproduction processing of FIG.

In step S ₁₉₀₇ , the accumulated peak value R
The absolute value of (ADR) is 20 which is the maximum value of 12 bits.
If it is determined that the difference exceeds 47, the value of the current adaptive quantization difference data rom (ADR) is reduced to the next lower quantization level. That is, if rom (ADR) is a positive value, -1 is set (S ₁₉₁₂ → S ₁₉₁₃ ), and if rom (ADR) is a negative value, +1 is set (S ₁₉₁₂ → S ₁₉₁₄ ).

As described above, the accumulated peak value R (ADR) is the actual tone signal that would be obtained by the sound source processing on the reproducing side (instrument side). When accumulation is performed while performing inverse quantization, the absolute value of the accumulated peak value becomes 1 of the specified number of bits on the reproduction side due to a quantization error.
It is possible that the maximum value of 2 bits is exceeded. When this occurs, sign inversion or the like occurs, and large noise occurs. Therefore, when the absolute value of the accumulated peak value R (ADR) that can be obtained on the reproducing side exceeds the above-mentioned maximum value, the current adaptive quantized difference data rom ( By lowering the quantization level of ADR, control is performed so that the absolute value of the accumulated peak value on the reproduction side does not exceed the maximum value.

After the above processing, the above-described step S is performed again.
₁₉₀₅ and S ₁₉₀₆ are recalculated, and the above processing is repeated until the determination in step S ₁₉₀₇ becomes NO. Normally, the absolute value of the accumulated peak value R (ADR) can be set to be equal to or less than the above-mentioned maximum value only by lowering the quantization level by one level.

By the above-described adaptive quantization processing, for example, a 12-bit PCM tone signal shown in FIG. 20A can be compressed into a 4-bit ADPCM tone signal.

[0119]

According to the present invention, the inverse quantization process and the accumulator accumulate the reproduced difference value at the current timing similar to the peak value of the tone signal that would be obtained by the sound source processing. By determining whether the accumulated value corresponding to the reproducing side exceeds the maximum change width of the number of bits allocated to the tone signal, it is possible to determine whether or not the current timing obtained in the adaptive quantization process is adaptive. When the sound source process is executed based on the quantized difference data, it is possible to determine whether or not the sign of the output value is inverted.

Then, based on the determination result of this determination step, the difference data changing step changes the value of the adaptively quantized difference data at the current timing, so that when the sound source processing is executed, the output value is changed. The occurrence of sign inversion can be prevented in advance, and the occurrence of noise can be minimized.

[Brief description of the drawings]

FIG. 1 is an overall configuration diagram of the present embodiment.

FIG. 2 is an internal configuration diagram of a microcomputer.

FIG. 3 is a configuration diagram of a conventional D / A converter unit.

FIG. 4 is a configuration diagram of a D / A converter unit according to the present embodiment.

FIG. 5 is a timing chart in D / A conversion.

FIG. 6 is an overall operation flowchart of the embodiment.

FIG. 7 is an operation flowchart of an interface process.

FIG. 8 is an operation flowchart of a sound source process.

FIG. 9 is a conceptual diagram showing a relationship between a main operation flowchart and interrupt processing.

FIG. 10 is a diagram showing a storage area for each sounding channel on a RAM.

FIG. 11 is a diagram showing a buffer area on a RAM.

12 is an explanatory view of the principle of the case of obtaining the interpolation value X _Q current difference value D by using the address A _F in ADPCM scheme.

FIG. 13 is an operation flowchart (part 1) of sound source processing by the ADPCM method.

FIG. 14 is an operation flowchart (part 2) of a sound source process according to the ADPCM method.

FIG. 15 is an operation flowchart (part 3) of the embodiment of the sound source processing according to the ADPCM method.

FIG. 16 is an explanatory diagram of envelope characteristics.

FIG. 17 is a diagram showing a first example of a compression table and a decompression table.

FIG. 18 is a diagram illustrating a second example of a compression table and a decompression table.

FIG. 19 is an operation flowchart of an adaptive quantization process.

FIG. 20 is a diagram showing an example of a PCM musical tone master and an ADPCM musical tone signal.

[Explanation of symbols]

 101 microcomputer 102 keyboard 103 function key 104 switch section 105 low-pass filter 106 amplifier 107 speaker 108 power supply circuit 201 control ROM 202 ROM address decoder 203 interrupt control section 204 RAM address control section 205 ROM address control section 206 RAM 207 command analysis section 208 ALU unit 209 Multiplier 210 Input port 211 Output port 212 Control data / waveform ROM 213 D / A converter unit 301, 401 Latch 303 D / A converter

────────────────────────────────────────────────── ─── Continued on the front page (72) Inventor Kosuke Shinami 3-2-1, Sakaemachi, Hamura-machi, Nishitama-gun, Tokyo Casio Computer Co., Ltd. Hamura Technical Center (72) Inventor Koichiro Taiki Hamura-cho, Nishitama-gun, Tokyo 3-2-1, Sakaemachi Casio Computer Co., Ltd. Hamura Technology Center (72) Inventor Kazuo Ogura 3-2-1, Sakaemachi, Hamura-machi, Nishitama-gun, Tokyo Casio Computer Co., Ltd. Hamura Technology Center (58) Field (Int.Cl. ⁷ , DB name) G10H 7/02

Claims (2)

(57) [Claims] 1. A musical tone waveform compression method for generating adaptively quantized difference data used in a musical tone waveform generator for generating a tone signal by an adaptive differential pulse code modulation method, wherein a digital tone signal at a current timing is provided. And the difference between the current timing obtained by the difference calculation process and the difference between the current timing obtained by the difference calculation process and the difference between the current timing and the accumulated value of the difference value reproduced up to the previous timing. An adaptive quantization process of performing adaptive quantization while compressing a value to obtain adaptively quantized difference data of a current timing; and an adaptively quantized difference data of a current timing obtained in the adaptive quantization process. Is inversely quantized to calculate a reproduced difference value of the current timing, and the reproduced difference value of the current timing obtained in the inverse quantization process is An accumulation process of accumulating the accumulated value of the reproduced difference values up to the timing and obtaining an accumulated value of the reproduced difference values up to the current timing; and A determining step of determining whether or not the accumulated value of the reproduced difference values has exceeded a maximum change width of the number of bits allocated to the tone signal for generating the tone signal; and determining the current timing by the determining step. If it is determined that the accumulated value of the reproduced difference values does not exceed the maximum change width, adaptively quantized difference data at the current timing obtained in the adaptive quantization process is output. An output step, and when it is determined in the determination step that the accumulated value of the reproduced difference values up to the current timing exceeds the maximum change width, the current timing obtained in the adaptive quantization step is obtained. Adaptive quantum The value of the obtained difference data is changed based on the value so that the accumulated value of the reproduced difference values obtained up to the current timing obtained in the accumulation process does not exceed the maximum change width, and the reverse is performed. A tone comprising: a difference data changing step of repeating a quantization step, the accumulation step, the determination step, and the output step; and a repeating step of repeating each of the above steps for each tone signal generation timing. Waveform compression method. 2. The adaptive quantization method according to claim 1, wherein the difference data changing step includes the step of determining whether the accumulated value of the reproduced difference values up to the current timing exceeds the maximum change width. Changing the absolute value of the adaptively quantized difference data of the current timing obtained in the process to a data value smaller by one data value, and repeating the inverse quantization process, the accumulation process, and the determination process. 2. A method for compressing a musical sound waveform according to claim 1, wherein: