JPH05265462A - Musical tone synthesizing device - Google Patents

Abstract

PURPOSE:To solve the problem point that two waveform data need to be read out so as to calculate one interpolated musical tone waveform value corresponding to the pitch of a musical instrument to be outputted. CONSTITUTION:One word of data in a waveform memory 11 consists of a combination of difference data, composed of an exponent part and a mantissa part varying in the number of bits relating to each other, and waveform data, and the difference data mantissa part is extracted by a bit extractor 13 according to a mode value decided by a mode deciding unit 12 from the difference data exponent part and then has its bits shifted by a shifter 14; and an offset generator 15 outputs an offset, but the respective outputs are added by an adder 16, the addition result is multiplied by an address decimal part through a multiplier 17, and the multiplication result is added by an adder 18, so the interpolated musical tone waveform value can be calculated by reading data out of the one waveform memory only once.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a musical tone synthesizer for controlling a pitch by interpolating a waveform in a so-called electronic musical instrument.

[0002]

2. Description of the Related Art In recent musical tone synthesizers, as a method of obtaining a musical tone waveform of a pitch to be output, a method of only skipping the waveform data stored in a waveform memory or a method of skipping after reading the waveform data is performed. , A method of obtaining by interpolating between waveform data.

As a conventional musical tone synthesizer, for example, Japanese Patent Publication No. 56-52317 and Japanese Patent Publication No. 53-30015.
It is shown in the publication.

FIG. 5 shows the configuration of a conventional tone synthesizer.
The explanation will be given. Here, an example is shown in which the waveform data stored in the waveform memory is linearly interpolated in order to obtain a musical tone waveform having a desired pitch.

In FIG. 5, 51 is an address integer part Ai.
Is an address generator for outputting an address formed from the address fractional part Af. An adder 52 increments the address integer part Ai by 1. Reference numerals 53 and 54 are waveform memories in which the same waveform data is stored.
A subtracter 55 subtracts the waveform data read from the waveform memories 53 and 54. A multiplier 56 multiplies the output of the subtractor 55 and the address fractional part Af. 5
An adder 7 adds the waveform data output from the waveform memory 53 and the output from the multiplier 56. The operation of the musical tone synthesizer configured as above will be described below.

First, when the pitch information of the musical instrument to be output is input to the address generator 51, the address formed by the address integer part Ai and the address decimal part Af is output.
The pitch information input to the address generator 51 is determined based on the pitch of the musical instrument to be output,
Pitch F (Hz), number of samples of one cycle waveform stored in the waveform memory 53 is WS (sample), predetermined timing, that is, sampling frequency is Fs (Hz)
Then, the pitch information S is represented by (Equation 1).

[0007]

[Equation 1]

Here, the configuration of the address generator 51 is shown in FIG.
Shown in. In FIG. 6, 61 is an adder and 62 is a register. The register 62 is a register that stores an address, and the storage area is 20 bits. If the number of samples WS of one cycle waveform stored in the waveform memory is 1024, the address integer part Ai is 10 bits and the address fraction part Af is 10 bits, as shown in FIG.

The operation of the address integer part Ai when the sampling frequency Fs is 32 KHz will be described below.

When the pitch information S is 1, the pitch information 1 is added to the contents of the register 62 for each sampling frequency by the adder 61. Therefore, assuming that the value currently stored in the register 62 is 0, the address Integer part Ai is address 0 to 1
The 1024 points of waveform data stored up to 023 are read from the waveform memory 53 in the order of addresses 0, 1, 2, 3, .... Since the address integer part Ai is 10 bits, when the value of the address integer part Ai exceeds the address 1023, it starts from the address 0 again, so that 1024 points of waveform data stored in the waveform memory 53 are repeatedly output. It will be. The pitch output at this time is 31.25 Hz because it is obtained by modifying (Equation 1).

Similarly, when the pitch information S is 2, if the value currently stored in the register 62 is 0, the address integer part Ai addresses the waveform data 1024 points stored in the addresses 0 to 1023. .. are read out from the waveform memory 53 in the order of 0, 2, 4, 6, ..., As a result, a tone having a pitch of 62.5 Hz is obtained. There is a one-octave relationship between when the pitch information S is 1 and when the pitch information S is 2, and the pitch information S takes a decimal value to form another pitch within the one octave. The integer part of the pitch information S is 2
Bit and fractional part are 10 bits, and pitch information S is 1 to 2
If a value of is taken, it becomes possible to form a sound having a pitch within the octave of 31.25 Hz to 62.5 Hz.

When the pitch information S is 1.5, the waveform data output from the waveform memory 53 as the waveform data of the first point is the address 1 or the address 2 obtained by rounding off the address 1.5. If the stored waveform data is output, the output waveform has large distortion. Therefore, the distortion can be reduced by calculating the waveform data corresponding to the address 1.5. The operation of performing linear interpolation for such calculation will be described below. The waveform memory 54 has 1024 samples of one cycle waveform, and is the same as the waveform memory 53.

The waveform data of the waveform memory 53 is sent out in correspondence with the address integer part Ai, while the waveform of the waveform memory 54 corresponds to the address integer part (Ai + 1) incremented by 1 in the adder 52. Data is sent out. The difference Δ between the waveform data output from the waveform memories 53 and 54 is obtained by the subtractor 55, and the subtracter 5
The five outputs Δ and the address fractional part Af are multiplied by the multiplier 56. In the adder 57, the multiplication result is added to the waveform data P which is the output of the waveform memory 53 to obtain the tone waveform interpolation value. The waveform memory 5
3, 54 may be one memory, and the waveform data may be read twice from one memory.

[0014]

However, in the above conventional configuration, two waveform data must be read in order to calculate one interpolated musical tone waveform value corresponding to the pitch of the musical instrument to be output. Therefore, there is a problem in that two identical waveform memories are required, or even if there is only one waveform memory, it is necessary to read the waveform data twice.

The present invention solves the above-mentioned conventional problems, and a tone synthesizer capable of calculating an interpolated tone waveform value corresponding to the pitch of a musical instrument desired to be output by reading one piece of waveform data. The purpose is to provide.

[0016]

In order to achieve this object, the musical tone synthesizer of the first invention is formed of an address integer part and an address decimal part which change corresponding to the pitch of the musical instrument to be output. A predetermined number n of bits including an address generator that outputs an address, and a differential data exponent part and a differential data mantissa part that change the number of bits in relation to each other.
A waveform that stores a predetermined number of bits (m + n) bits as one word data by combining bit difference data and m-bit waveform data, and outputs data corresponding to the address integer part output from the address generator. A memory, a mode determiner for determining a mode value from the differential data exponent of the data read from the waveform memory, and a differential data mantissa having a bit number according to the output of the mode determiner for the data read from the waveform memory. A bit extractor for extracting, a shifter for bit-shifting the differential data mantissa output from the bit extractor according to the output of the mode determiner, and a positive / negative code identical to the differential data mantissa corresponding to the mode determiner output. The offset generator that generates an offset having the difference between the shifter output and the offset generator output is added. A first adder for outputting a value, the first
A multiplier for multiplying the output of the adder and the address fractional part output from the address generator, and a second for adding the multiplier output and the waveform data of the data read from the waveform memory
And the adder of.

The tone synthesizer of the second aspect of the present invention further comprises an address generator for outputting an address formed of an address integer part and an address decimal part which change corresponding to the pitch of the musical instrument desired to be output. The bit number that changes in association with the difference data exponent part and the differential data mantissa part in which the bit number changes and the n-bit difference data and the m-bit waveform data are combined to set a predetermined bit number (m + n) bits to 1 A waveform memory that stores as word data and outputs data corresponding to the address integer part output from the address generator; a mode determiner that determines a mode value from the differential data exponent part of the data read from the waveform memory; The first bit for extracting the waveform data having the number of bits according to the output from the mode determiner with respect to the data read from the waveform memory. And extractor, a first shifter for outputting a waveform value after bit shift of the waveform data outputted from the first bit extractor according to the mode decision output,
A second bit extractor for extracting a differential data mantissa having a bit number according to the mode determiner output for the data read from the waveform memory, and an output from the second bit extractor according to the mode determiner output. Of the differential data mantissa, a second shifter for bit-shifting the differential data mantissa, an offset generator for generating an offset having the same positive / negative sign as the difference data mantissa corresponding to the output of the mode determiner, and the differential data mantissa A first adder that adds a second shifter output that performs bit shift and an offset generator output and outputs a difference value, and the first adder output and the address fractional part output from the address generator are multiplied. And a second adder that adds the multiplier output and the waveform value output from the first shifter.

[0018]

According to the first aspect of the present invention, with the above-mentioned structure, the address formed by the address integer part and the address decimal part, which change corresponding to the pitch of the musical instrument desired to be output by the address generator, is output, and each address is predetermined. The 1-word (m + n) -bit data in which the m-bit waveform data having the number of bits and the corresponding n-bit difference data are combined one by one is waveform corresponding to the address integer part output from the address generator. Read from memory. here,
The difference data is composed of a difference data exponent part and a difference data mantissa part in which the number of bits changes in association with each other.
The mode value is determined and output from the differential data exponent part of the data read from the waveform memory in the mode determiner, and the differential data mantissa part having the number of bits according to the mode value output from the mode determiner in the bit extractor is output. Bit shift of the differential data mantissa, which is extracted from the data read from the waveform memory and output from the bit extractor according to the mode value output from the mode determiner in the shifter, is output from the mode determiner in the offset generator. Corresponding to the mode value, an offset having the same positive / negative sign as the difference data mantissa is generated, the offset generator output and the shifter output are added in the adder, the difference value is output, and the difference value is multiplied. And the address fractional part output from the address generator is multiplied. Since the multiplier output is added to the waveform data of the data read from the waveform memory in the adder, the interpolated tone waveform value will be obtained.

According to the second aspect of the present invention, with the above-mentioned configuration, an address formed by the address integer part and the address decimal part, which change corresponding to the pitch of the musical instrument desired to be output by the address generator, is output and the address integer part is output. Based on this, 1-word (m + n) -bit data in which m-bit waveform data whose bit number changes in relation to each other and corresponding n-bit difference data are combined one by one is read from the waveform memory. Here, the difference data is composed of a difference data exponent part and a difference data mantissa part whose bit numbers change in association with each other. Waveform data having a bit number corresponding to the mode value output from the mode determiner in the first bit extractor after the mode value is determined and output from the difference data exponent part of the data read from the waveform memory in the mode determiner Is extracted from the data read from the waveform memory, and the waveform data output from the bit extractor is bit-shifted according to the mode value output from the mode determiner in the first shifter, and the waveform value is output. On the other hand, the differential data mantissa having the number of bits corresponding to the mode value output from the mode determiner in the second bit extractor is extracted from the data read from the waveform memory, and output from the mode determiner in the second shifter. The differential data mantissa part output from the bit extractor is bit-shifted according to the mode value, and the same positive / negative sign as that of the difference data mantissa part corresponding to the mode value output from the mode determiner in the offset generator is performed. An offset having a sign is generated, the offset generator output and the second shifter output are added in an adder to output a difference value, and the difference value is multiplied by the address fractional part in the multiplier. Since the waveform value and the multiplier output are added in the adder, the interpolated tone waveform value is obtained.

[0020]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 shows the configuration of a musical sound synthesizer according to the first embodiment of the present invention. In FIG. 1, 11 is a waveform memory, 12 is a mode determiner, 13 is a bit extractor, 14 is a shifter, 15 is an offset generator, 16 is an adder, 17 is a multiplier, and 18 is an adder. Note that 51
Is an address generator, which has the same configuration as the conventional example. The operation of the musical tone synthesizing apparatus of this embodiment having the above-described configuration will be described below.

The operation of the address generator 51 is the same as that of the conventional example. When the pitch information of the musical instrument to be output is input to the address generator 51, the address integer part Ai and the address decimal part Af are entered.
And are divided and output, the address integer part Ai is sent to the waveform memory 11, and the address fraction part Af is sent to the multiplier 17.

By inputting the address integer part Ai, the waveform memory 11 outputs data (m + n) bits obtained by combining the waveform data m bits and the difference data n bits in the corresponding waveform memory.

Generally, 16 bits are often used as the number of bits of waveform data in digital audio products such as a digital audio tape recorder (DAT) and a compact disc player (CD).
The number of bits of 16 bits does not always correspond to human hearing, and is because one unit of the semiconductor memory used as the waveform memory, that is, one word constitutes a power of two. The number of bits used for one word should be changed according to the desired S / N, and does not necessarily have to be a power of 2, and can be further reduced. It is also possible to adopt a configuration in which matching waveform data and difference data are combined.

Here, the number of bits of one word is Nw, and the number of bits required for human hearing as waveform data is Np (<
Nw), the number of bits used for difference data between adjacent waveform data is Nd, both the waveform data and the difference data are used in fixed-point notation, and the difference data stored in the waveform memory is directly expressed as the actual difference value. Therefore, Nd> Nw-Np often occurs especially in a musical tone (violin, trumpet, etc.) having many harmonic components and large waveform changes. Therefore, using a floating point display in which a large difference value has a large resolution and a small difference value has a small difference value, a difference data exponent part and a difference data mantissa part are provided, and By controlling the resolution in the mode, the range of the difference value can be widened, and it is possible to deal with a musical sound with a large waveform change.

On the other hand, since a tone having a smooth waveform with a small change (flute, clarinet, etc.) does not require a mode with a large difference value such that the resolution is large, the bit of the mantissa of the difference data is small in the mode with a small resolution. If the number is increased, the range of difference values with small resolution can be further widened, and small and smooth waveform changes can be accurately taken. That is, the difference data is a combination of the difference data exponent part and the difference data mantissa part, and by providing a bit that becomes the difference data mantissa part when the difference value is small and the difference data exponent part when the difference value is large. , The difference data exponent part and the difference data mantissa part are associated with each other depending on the mode, and the number of bits is changed,
It is possible to widen the range of difference values while accurately taking small movements of a musical tone waveform.

Therefore, the data stored in the waveform memory 11 is 1 word (m + n) as shown in FIG.
= 16 bits, the configuration is waveform data m = 10 bits (D6 to D15), difference data n = 6 bits (D0 to
D5). The area (D4 to D5) that can be taken as the differential data exponent part is always the differential data exponent part even if the mode is changed.
5) and a differential data exponent part mantissa shared bit 1 bit (D4) which becomes a differential data exponent part or a differential data mantissa part depending on the mode. Then, three modes are provided from the smallest difference value range. When D5 is 0, the mode value is 0, when D5 is 1 and D4 is 0, the mode value is 1, and when D5 is 1 and D4 is 1, the mode is set. If the value is 2, the difference data n = 6 bits (D0-
D5) has a difference data mantissa part 5 bits (D0 to D4) and a difference data exponent part 1 bit (D5) when the mode value is 0, and a difference data mantissa part 4 when the mode value is 1 or 2. Bits (D0 to D3) and differential data exponent 2 bits (D4 to D5).

The mode discriminator 12 calculates the difference data exponent (D5 or D5) of the data sent from the waveform memory 11.
5 to D4), the mode value corresponding to the difference data exponent part as shown in FIG. 2 is determined and output. Here, the mode value is determined to be 0 when D5 is 0, the mode value is 1 when D5 is 1 and D4 is 0, and the mode value is 2 when D5 is 1 and D4 is 1. It is assumed that the number of bits of data input to the mode determiner 12 is the maximum area of 2 bits (D4 to D5) that can be taken as the differential data exponent part.
And

The bit extractor 13 inputs a mode value to the data (D0 to D15) read from the waveform memory 11 and outputs the corresponding differential data mantissa (D0 to D4 when the mode value is 0, the mode value is 0). D0 when 1 or 2
~ D3) bit extraction is performed. Here, the bit extractor 1
The number of bits of the data read from the waveform memory 11 input to 3 is n of the difference data, that is, 6 bits (D0 to D5).

The shifter 14 bit-shifts the differential data mantissa obtained by the bit extractor 13 according to the shift amount (corresponding to the resolution of the differential data mantissa) corresponding to each mode value, and outputs it.

The offset generator 15 sets the absolute value of the offset corresponding to each mode value to the positive / negative sign of the differential data mantissa (D4 when the mode value is 0, D3 when the mode value is 1 or 2). Convert it to the corresponding value and output it.

The adder 16 adds the output of the shifter 14 and the output of the offset generator 15 and outputs a difference value Δ.

Here, when the mode value is 0, the shift amount in the shifter 14 is 0 bit and the offset output from the offset generator 15 is 0 so that the resolution of the differential data mantissa is 1. Further, when the mode value is 1, the shift amount in the shifter 14 is 1 bit and the offset output from the offset generator 15 corresponds to the positive and negative signs of the differential data mantissa so that the resolution of the differential data mantissa is 2. 16 or -16. When the mode value is 2, the shift amount in the shifter 14 is 2 bits and the offset output from the offset generator 15 corresponds to the positive and negative signs of the differential data mantissa so that the resolution of the differential data mantissa is 4. 32 or -32. At this time, the waveform data and the differential data mantissa are used as the two's complement, and the bit extractor 13, the shifter 14, the offset generator 1
5, when the operation in the adder 16 is performed in correspondence with 2's complement, the range of the difference value in each mode is −16 to 15 (resolution 1) when the mode value is 0, and − when the mode value is 1. 32 to -18 and 16 to 30 (resolution 2), when the mode value is 2, it is -64 to -36 and 32 to 60 (resolution 4), and the range of -64 to 60 (7 bits) should be taken as a whole. Since the difference data is wider than the difference value range of 32 to 31 (6 bits) when the difference data of 6 bits is fixed point representation (resolution 1), the number of bits of the difference data mantissa part when the mode value is 0 Is 5 bits instead of 4 bits, so -16 to 15 (5
The difference value can be taken with a resolution of 1 up to (bit).

The multiplier 17 multiplies the difference value Δ (7 bits) expressed by the integer value output from the adder 16 by the address fractional part Af (10 bits). Since it is a small decimal number, in the adder 18, the waveform data (m = 10 bits) of the data read from the waveform memory 11 is directly used as the waveform value P, and the waveform value P (10 bits) is multiplied by the multiplication result of the multiplier 17. If the upper 7 bits are added, the interpolated tone waveform value (1
(0 bit) will be obtained.

A combination of the number of bits in one word (16 bits) of the waveform data (m bits) and the difference data (n bits) is previously set to m = 10 bits, n =
Although it is defined as 6 bits, for example, m = 11 bits, n =
A combination of 5 bits or a combination of m = 9 bits and n = 7 bits may be used.

Here, the differential data exponent part peculiar bit is 1 bit and the differential data exponent part mantissa shared bit is 1
Although there are three types of difference value modes as bits, it is possible to increase the range of the difference value by increasing one or both of the bits and setting more modes.

Here, when the mode value output from the mode determiner corresponding to the differential data exponent (D5 or D4 to D5) is 0, the shift amount of the differential data mantissa is 0,
When the offset is 0 and the mode value is 1, the shift amount of the differential data mantissa is 1, and the offset is 16 or -16 corresponding to the positive / negative sign of the differential data mantissa, or
When the mode value is 2, the shift amount is 2 and the offset is 32
Alternatively, the value is −32, but for example, the shift amount and offset amount in each mode value are increased by 1 bit (2
However, in this case, the difference value is −128 to 120.
It becomes possible to take the range of (8 bits).

As described above, according to this embodiment, one word of the waveform memory is composed of 10 bits of waveform data and 6 bits of difference data between adjacent waveform data. The configuration 16 bits can be used efficiently. Further, as a configuration in which the differential data exponent part and the differential data mantissa part in which the difference data 6 bits are changed by 1 bit in relation to each other are combined, a total of three modes (resolution 1 or 2 of the differential data mantissa part or By taking 4), the waveform data 10
With respect to the bit range, it is possible to widen the range of the total difference value to 7 bits while keeping the range of the difference value with resolution of 1 to 5 bits. Therefore, the difference value can be accurately measured in the musical tone waveform with a small waveform change. Therefore, the range of the difference value can be widened until it can be applied to a musical tone waveform having a relatively large waveform change.

FIG. 3 shows the configuration of a musical sound synthesizer in the second embodiment of the present invention. In FIG. 3, 30
1 is a waveform memory, 302 is a mode determiner, and 303 is a first
Bit extractor, 304 is the first shifter, and 305 is the second
Bit extractor, 306 is a second shifter, 307 is an offset generator, 308 is an adder, 309 is a multiplier, 31
0 is an adder. Incidentally, 51 is an address generator, which has the same configuration as the conventional example. The operation of the musical tone synthesizing apparatus of this embodiment having the above-described configuration will be described below.

The operation of the address generator 51 is the same as that of the conventional example. When the pitch information of the musical instrument to be output is input to the address generator 51, the address integer part Ai and the address decimal part Af are entered.
And are divided and output, the address integer part Ai is sent to the waveform memory 301, and the address fraction part Af is sent to the multiplier 309.

The waveform memory 301 has an address integer part Ai.
Is input, the data (m + n) bits in which the waveform data m bits and the difference data n bits in the corresponding waveform memory are combined are output.

Generally, as the number of bits of waveform data, 16 bits are often used in digital audio products such as a digital audio tape recorder (DAT) and a compact disc player (CD).
The number of bits of 16 bits does not always match the human hearing, and is because the unit of one data of the semiconductor memory used as the waveform memory, that is, one word is a power of two. The number of bits of the waveform data used for one word should be changed according to the desired S / N, and does not necessarily have to be a power of 2, and can be further reduced. And the difference data between adjacent waveform data may be combined. Here, the number of bits of one word is Nw, the number of bits required for human hearing as waveform data is Np (<Nw), and the number of bits used for difference data between adjacent waveform data is Nd. If both and are used in fixed-point notation and the difference data stored in the waveform memory is directly represented as the actual difference value, Nd> especially for musical sounds with a large number of harmonic components and large waveform changes (violin, trumpet, etc.). It often becomes Nw-Np. Therefore, a floating point display is used in which a large difference value has a large resolution and a small difference value has a small difference value. The range of the difference value can be widened by controlling the resolution in the mode.

By the way, the musical instrument waveform is composed of (1) a portion having a particularly large waveform change at the start of sound generation (transient portion) and (2) a portion having a relatively small waveform change (steady portion) that persists after the transition portion. It can be divided into two parts. Even if the number of bits is set so that the range of the difference value corresponding to the magnitude of the waveform change in the steady part can be obtained by the floating point display, the range of the difference value in the transient part where the waveform change is particularly large cannot be supported. Therefore, a bit that becomes waveform data or a difference data exponent part depending on the range of the difference value is provided, and when the difference value is particularly large, the bit is set as the difference data exponent part and the number of bits of the entire difference data exponent part is set. By setting the increasing mode, that is, by changing the number of bits in relation to the waveform data and the difference data depending on the mode, it becomes possible to apply even a large difference value in a transient portion where the waveform change is particularly large. ..

On the other hand, a musical tone having a smooth waveform with a small change (flute, clarinet, etc.) does not require a mode having a large difference value such that the resolution becomes large,
Rather, in order to accurately take small and smooth waveform changes,
By increasing the number of bits of the mantissa of the difference data in the mode of low resolution, it is possible to further widen the range of the difference value of low resolution. To this end, the difference data exponent part and the difference data mantissa part are associated with each other depending on the mode and the number of bits is changed, and when the difference value is small, the number of bits of the difference data mantissa part is increased, so that a small waveform movement occurs. Can be taken accurately.

FIG. 4 shows an example of the structure in which the data stored in the waveform memory 301 is 16 bits per word. Area that can be used as the difference data exponent (D4 to D
5) is the differential data exponent part which is always the differential data exponent part even if the mode is changed. The unique bit 1 bit (D5) and the differential data exponent part which becomes the differential data exponent part or the differential data mantissa part depending on the mode. It consists of a mantissa part shared bit 1 bit (D4) and a waveform data difference data exponent part shared bit 1 bit (D6) which becomes waveform data or a difference data exponent part depending on the mode. Then, four modes are provided from the one having the smallest difference value range. When D5 is 0, the mode value is 0, when D5 is 1 and D4 is 0, the mode value is 1, and D5 is 1 and D4 is 1.
When D6 is 0, the mode value is 2, and when D5 is 1 and D4 is 1 and D6 is 1, the mode value is 3, and the configuration of 1 word (m + n) = 16 bits in each mode is Is 0, the differential data mantissa part is 5 bits (D0 to D4), the differential data exponent part is 1 bit (D5), and the differential data n = 6 bits (D0 to D5), and the waveform data m = 10 bits (D6 to D15). ). When the mode value is 1, the difference data mantissa part 4 bits (D0 to D3) and the difference data exponent part 2 bits (D4 to D5)
The difference data n = 6 bits (D0 to D5), and the waveform data m = 10 bits (D6 to D15). When the mode value is 2 or 3, the difference data mantissa part is 4 bits (D0 to D3), the difference data exponent part is 3 bits (D4 to D6), and the difference data n = 7 bits (D0 to D).
6), and the combination of the waveform data m = 9 bits (D7 to D15).

The mode determiner 302 is the waveform memory 301.
From the differential data exponent part (D5 or D4 to D5 or D4 to D6) of the data sent from, the mode value corresponding to the differential data exponent part as shown in FIG. 4 is determined and output.
Here, when D5 is 0, the mode value is determined to be 0, and D5
When D4 is 0, the mode value is 1, when D5 is 1, when D4 is 1, and when D6 is 0, the mode value is 2, D
When 5 is 1, when D4 is 1 and when D6 is 1, the mode value is 3, and the mode determiner 3
The number of bits of data input to 02 is 3 bits (D4 to D6), which is the maximum area that can be taken as the differential data exponent.

The bit extractor 303 is a waveform memory 301.
By inputting the mode value to the data (D0 to D15) read from, the corresponding waveform data (D6 to D15 when the mode value is 0 or 1 and D7 to D15 when the mode value is 2 or 3) Extract bits. Here, the number of bits of the data read from the waveform memory 301 input to the bit extractor 303 is 10 bits (D6 to D15) that can be taken as the waveform data.

The shifter 304 bit-shifts the waveform data obtained by the bit extractor 303 according to the shift amount corresponding to each mode value, and outputs it as the waveform value P. Here, since the shift amount is 0 when the mode value is 0 or 1, and the shift amount is 1 when the mode value is 2 or 3, the waveform value P is output as 10 bits.

The bit extractor 305 has a waveform memory 301.
By inputting the mode value to the data (D0 to D15) read from the corresponding differential data mantissa (D0 to D4 when the mode value is 0, the mode value is 1 or 2).
Alternatively, when it is 3, the bits of D0 to D3) are extracted. Here, the number of bits of the data read from the waveform memory 301 input to the bit extractor 305 is set to 6 bits (D0 to D5) excluding the area that the waveform data can take.

The shifter 306 bit-shifts the differential data mantissa obtained by the bit extractor 305 according to the shift amount (corresponding to the resolution of the differential data mantissa) corresponding to each mode value, and outputs it.

The offset generator 307 determines the absolute value of the offset corresponding to each mode value as the positive or negative sign of the differential data mantissa (D4 when the mode value is 0, and 1 when the mode value is 1).
Alternatively, when it is 2 or 3, it is converted into a value corresponding to D3) and output.

The adder 308 adds the shifter 306 output and the offset generator 307 output, and outputs the difference value Δ.

Here, when the mode value is 0, the shift amount in the shifter 306 is 0 bit and the offset output from the offset generator 307 is 0 so that the resolution of the differential data mantissa is 1. Further, when the mode value is 1, the shifter 3 is used to set the resolution of the differential data mantissa to 2.
The shift amount in 06 is 1 bit, and the offset generator 3
The offset output from 07 is set to 16 or -16 corresponding to the positive / negative sign of the difference data mantissa. Also,
When the mode value is 2, the shift amount in the shifter 306 is 2 bits in order to set the resolution of the differential data mantissa to 4, and the offset output from the offset generator 307 corresponds to the positive / negative sign of the differential data mantissa 32. Or -3
Set to 2. Further, when the mode value is 3, the shift amount in the shifter 306 is 3 bits and the offset output from the offset generator 307 corresponds to the positive and negative signs of the differential data mantissa so that the resolution of the differential data mantissa is 8. 64 or −64. At this time, when the waveform data and the differential data mantissa are expressed in the 2's complement format and the operations of the bit extractor 303, the shifter 306, the offset generator 307, and the adder 308 are performed in the 2's complement correspondence,
The range of the difference value in each mode is -16 to 15 (resolution 1) when the mode value is 0, and -3 when the mode value is 1.
2 to -18 and 16 to 30 (resolution 2), mode value is 2
Is -64 to -36 and 32 to 60 (resolution 4), and the mode value is 3 is -72 to -128 and 64 to 120.
(Resolution is 8), and as a whole, -128 to 120 (8
Since it can take a range of (bits), it is much wider than the difference value range of 32 to 31 (6 bits) when the difference data of 6 bits is fixed point representation (resolution 1), and the mode value is Since the number of bits of the differential data mantissa at 0 is 5 bits instead of 4 bits, the difference value can be taken with a resolution of 1 from -16 to 15 (5 bits).

The multiplier 309 multiplies the difference value Δ (8 bits) represented by the integer value output from the adder 308 by the address fraction part Af (10 bits), but the address fraction part Af is 1 or more. Adder 31 because it is a small decimal
In 0, by adding the upper 8 bits of the multiplication result of the multiplier 309 and the waveform value P (10 bits) sent from the shifter 304, an interpolated tone waveform value (10 bits) can be obtained. Become.

Here, the differential data exponent part peculiar bit is 1 bit, and the differential data exponent part mantissa part shared bit is 1
Bit and waveform data Difference data The common bit for the exponent part was set to 1 bit, and the difference value mode was set to 4 types, but one or more of these bits were increased to set more modes and a wider difference value range was set. It is also possible.

Here, when the mode value output by the mode determiner corresponding to the differential data exponent part (D5 or D4 to D5 or D4 to D6) is 0, the shift amount of the differential data mantissa part is 0, When the offset is 0 and the mode value is 1, the shift amount of the differential data mantissa is 1, the offset is 16 or -16 corresponding to the positive / negative sign of the differential data mantissa, and the mode value is 2. , The shift amount is 2, the offset is 32 or -32, and when the mode value is 3, the shift amount is 3, the offset is 64 or -64.
However, for example, the shift amount and the offset amount in each mode value may be increased (doubled) by 1 bit, and in that case, the difference value is -256 to 240 (9 bits).
Will be able to take the range of.

Here, 10 bits (D6 to D15) of the data read from the waveform memory 301 are input to the bit extractor 303, and 10 bits (D6 to D15) or 9 bits according to the mode value are used as the waveform data. (D7 to D15) are extracted and output, and the bit extractor 305 inputs 6 bits (D0 to D5) of the data read from the waveform memory 301 and follows the mode value as the differential data mantissa. 5 bits (D0-4)
Alternatively, the configuration is such that 4 bits (D0 to D3) are extracted and output. For example, the inputs of the bit extractors 303 and 304 are both data 1 read from the waveform memory 301.
It is also possible to use 6 bits (D0 to D15) so as to obtain the same output as that of the embodiment.

As described above, according to this embodiment, one word in the waveform memory has 6 bits of adjacent waveform data when the waveform data has 10 bits, and 7 bits of adjacent difference data when the waveform data has 9 bits. Since the configuration is such that the bits are combined, 16-bit one word configuration of the waveform memory can be efficiently used. Also, when the combination of waveform data 10 bits and difference data 6 bits, the number of bits of the difference data exponent part and the difference data mantissa part are associated with each other by 1
Two modes (resolution 1 or 2 of the mantissa of the differential data) are set by changing the bit, and the number of bits of the waveform data and the differential data is changed by 1 bit in association with each other to change the waveform data 9 bits and the differential data 7 When a combination of bits is set, two modes (resolution of mantissa of difference data 4 or 8) are set, and a total of 4 modes are set, so that the difference of resolution is 1 with respect to the range of 10-bit waveform data. It is possible to extend the range of the total difference value to 8 bits while keeping the value range to 5 bits.
In the case of a musical tone waveform having a small waveform change, the waveform change, that is, the difference value can be accurately obtained, and the difference value range can be widened until it can be applied to a considerably large waveform change such as a transient portion.

[0059]

As described above, according to the present invention, one word of data in the waveform memory is converted into one waveform data and the corresponding difference data.
By combining them one by one, the interpolated tone waveform value can be calculated by reading the data from the waveform memory only once, and the differential data is provided as a floating point display with a mantissa part and an exponent part. By doing so, the range of difference values can be widened,
It is possible to calculate the interpolated tone waveform value even for a tone having a large waveform change, which has an excellent effect in practical use.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a musical sound synthesizer according to a first embodiment of the present invention.

FIG. 2 is a bit configuration diagram of one word of waveform data stored in a waveform memory according to the first embodiment.

FIG. 3 is a block diagram showing a configuration of a musical sound synthesizer in a second embodiment of the present invention.

FIG. 4 is a bit configuration diagram of one word of waveform data stored in a waveform memory according to the second embodiment.

FIG. 5 is a block diagram showing a configuration of a conventional musical sound synthesizer.

FIG. 6 is a block diagram showing a configuration of an address generator in the conventional example.

FIG. 7 is a bit configuration diagram of an address in the conventional example.

[Explanation of symbols]

 11,301 Waveform memory 12,302 Mode determiner 13,303,305 Bit extractor 14,304,306 Shifter 15,307 Offset generator 16,18,308,310 Adder 17,309 Multiplier 51 Address generator

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Masahiro Nakanishi 1006 Kadoma, Kadoma City, Osaka Prefecture Matsushita Electric Industrial Co., Ltd.

Claims (2)

[Claims] 1. An address generator for outputting an address formed from an address integer part and an address decimal part which change corresponding to the pitch of an instrument to be output, and a differential data exponent in which the number of bits changes in relation to each other. Part and difference data mantissa part, a predetermined number of bits n bits of difference data and m bits of waveform data are combined to store a predetermined number of bits (m + n) bits as 1-word data, and the address is generated. A waveform memory that outputs data corresponding to the address integer part output from the waveform unit, a mode determiner that determines a mode value from the differential data exponent part of the data read from the waveform memory, and data read from the waveform memory Bit extractor for extracting the differential data mantissa having the number of bits according to the output of the mode determiner A shifter for bit-shifting the differential data mantissa output from the bit extractor according to the mode determiner output, and a positive / negative code identical to the difference data mantissa corresponding to the mode determiner output An offset generator for generating an offset having the following: a first adder for adding the shifter output and the offset generator output to output a difference value; and a first adder output and an output from the address generator. And a second adder for adding the output of the multiplier and the waveform data of the data read out from the waveform memory. .. 2. An address generator for outputting an address formed from an address integer part and an address fraction part which change corresponding to the pitch of a musical instrument to be output, and a differential data exponent whose bit number changes in relation to each other. Part and the differential data mantissa, the changing bit number n bits of difference data and m bits of waveform data are combined to store a predetermined number of bits (m + n) bits as 1 word data, and the address generator A waveform memory that outputs data corresponding to the address integer part output from the waveform memory; a mode determiner that determines a mode value from the differential data exponent part of the data read from the waveform memory; and a data read from the waveform memory. On the other hand, a first bit extractor for extracting the waveform data having the number of bits according to the output of the mode determiner, A first shifter for bit-shifting the waveform data output from the first bit extractor according to the output of the mode determiner and outputting a waveform value; and a mode for the data read from the waveform memory. A second bit extractor for extracting the differential data mantissa having a bit number according to the decision device output; and a difference data mantissa output from the second bit extractor according to the mode determiner output. A second shifter for performing a bit shift; an offset generator for generating an offset having the same positive / negative sign as the difference data mantissa corresponding to the mode determiner output; the second shifter output and the offset; A first adder for adding a generator output and outputting a difference value; and an adder output from the first adder output and the address generator. And a second adder for adding the multiplier output and the first shifter output, and the waveform data and the corresponding difference data are associated with each other. A musical tone synthesizer characterized in that the number of bits changes.