JP2900078B2 - Waveform recording / playback method and waveform playback device - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a technique for recording and reproducing an audio signal such as a natural musical instrument sound as a digital waveform data sequence.

[Prior art and its problems] The sound source of the PCM system has a built-in waveform memory for storing musical tones as a sequence of waveform data (PCM data). In operation,
The PCM tone generator generates a tone signal of a desired pitch by reading waveform data from a waveform memory at a speed corresponding to a required pitch.

A drawback of the PCM type sound source is that a memory with a large storage capacity is required as a waveform memory.

In order to avoid this disadvantage, there is a differential PCM method. In the differential PCM method, the number of bits required to represent a difference value between adjacent waveform data is expected to be smaller than the number of bits required to represent waveform data. A data string of difference values (differential data string) is recorded in the. At the time of reproduction, the differential PCM sound source reproduces the waveform data by accumulating the differential data read from the memory of the differential data string.

Unfortunately, the conventional differential PCM method has a problem that the accuracy of reproduction varies depending on the timbre of the original sound.

In the conventional differential PCM recording method, a sequence {x (m)} of waveform data having a certain number of bits (eg, 16 bits) is converted into a sequence {d (n)} of differential data having a smaller number of bits (eg, 8 bits). Convert to The basic principle of the conversion is as follows. First, adjacent waveform data x (k), x (k−
Calculate the difference of 1).

x (k) -x (k-1) = Dx (Equation 1) Here, the difference Dx is difference data represented by 16 bits.
When the size of the difference data Dx is a size that can be expressed by 8 bits (128 to 127), the lower 8 bits of the 16-bit difference data Dx are set as the k-th 8-bit difference data d (k).

(Note that a two's complement is assumed here).

However, if the magnitude of Dx indicates a value that cannot be represented by 8 bits, d (k) is clipped to the maximum value that can be represented by 8 bits.

Clipped to

Note that, in order to reduce the accumulation of errors due to this clip, instead of (Equation 1), Is calculated, and the difference data dx in the 16-bit expression is converted into 8-bit difference data d (k) as described above.

By the above-mentioned clip, the sequence of difference data ｛d
(N) The waveform reproduced from 波形 is distorted. Even more unfortunately, the frequency of occurrence of clips, which is a cause of deterioration of the S / N ratio, depends on the sequence {x (n)} of the waveform data, and therefore the spectrum or timbre of the original sound to be recorded.
As a result, the conventional differential PCM method uses S
The N ratio varies and the desired reproduction accuracy cannot be guaranteed.

The above-described differential PCM method is called a linear differential PCM method. The occurrence of clips seen in the linear differential PCM method can be eliminated by using a technique called a nonlinear differential PCM method. In the non-linear difference PCM method, when recording,
The waveform data sequence is converted into a non-linear differential data sequence having a limited number of bits according to a predetermined non-linear function (for example, a logarithmic function). At the time of reproduction, the non-linear differential data sequence is converted according to an inverse function of the non-linear function (for example, an exponential function). Play the waveform data string. If this kind of nonlinear difference PCM method is used, even when the linear difference value of the waveform data sequence is relatively large, nonlinear difference data with a small number of effective bits can be obtained by nonlinear conversion.

Unfortunately, even in the case of the nonlinear differential PCM method, the SN ratio or reproducibility varies depending on the type of the original sound. This is the type of original sound (the magnitude of fluctuation in the waveform data sequence)
Is caused by a change in the range of the nonlinear differential data sequence generated by the above. For example, in the logarithmic difference PCM method, the reproducibility in a portion where the fluctuation of the waveform data sequence is small is particularly affected by the type of the original sound.

[Object of the Invention] Accordingly, an object of the present invention is to record a waveform of an audio signal in such a manner that the waveform of an audio signal is not distorted as much as possible by a data sequence of a limited number of bits, and to obtain a high-quality signal with a small distortion from the recorded data sequence. An object of the present invention is to provide a waveform recording / reproducing method and a reproducing apparatus capable of reproducing a waveform.

According to the present invention, a preparation step of preparing a waveform data sequence representing an audio signal, a step of calculating the magnitude of the variation of the waveform data sequence, and a magnitude of the variation include: Calculating a compression ratio that is inversely proportional to the magnitude of the variation;
A compressed non-linear difference data sequence corresponding to a result obtained by multiplying a sequence of differences between adjacent waveform data in the waveform data sequence by the compression ratio by a predetermined non-linear function, or data of adjacent waveforms in the waveform data sequence Compressed nonlinear differential data sequence generating step of generating a compressed nonlinear differential data sequence corresponding to a product obtained by multiplying the difference sequence by the nonlinear function by the compression ratio; and data storing the compressed nonlinear differential data sequence. A column storing step, a decompression rate storing step of storing a decompression rate related to the compression rate, and a waveform from the stored compressed non-linear difference data sequence based on an inverse function of the non-linear function and the stored decompression rate. A reproducing step of reproducing a data sequence.

The basic feature of the present invention is that, while utilizing the nonlinear difference technique in the recording and reproduction of the waveform, the compression rate and the expansion rate are considered, and the magnitudes of the compression rate and the expansion rate are determined by the waveform data of the waveform data sequence. The point is that it is determined according to the fluctuation and thus the spectrum or timbre of the audio signal. The conventional differential PCM method, whether it is a linear difference method or a nonlinear difference method, is basically recorded as difference data without compressing the difference between the adjacent waveform data (linear difference, nonlinear difference) at the time of recording, This is a method in which difference data is converted into waveform data without expansion at the time of reproduction, and does not have a concept of compressing a difference or a concept of expanding. For this reason, various audio signals cannot be supported, and the reproducibility thereof varies.

On the other hand, according to the present invention, if the fluctuation of the waveform data sequence is large, a compression ratio corresponding to the fluctuation is set, and a sequence of nonlinear difference data compressed at the compression ratio is recorded. Then, at the time of reproduction, the accumulated output of the nonlinear difference data expanded at the expansion rate related to the compression rate (the reciprocal of the compression rate or a value close thereto) is reproduced as waveform data.

The magnitude of the fluctuation of the waveform data sequence can be evaluated in various ways. For example, the magnitude of the fluctuation of the waveform data sequence can be evaluated based on the maximum value of the difference between adjacent waveform data. The compression ratio can be determined based on the maximum value of the difference. However, the reference compression rate does not always give the highest sound quality when the waveform data is reproduced. Therefore, it is preferable to select the compression ratio and expansion ratio that are optimal for the audibility by recording and reproducing the waveform at several compression ratios and expansion ratios based on the reference compression ratio. Subjective assessment).

In addition, it is possible to objectively evaluate the magnitude of the fluctuation of the waveform data sequence by statistically processing the sequence of the difference between each adjacent waveform data in the waveform data sequence and obtaining a parameter characterizing the fluctuation.

As a result, according to the waveform recording / reproducing method of the present invention, the information of the original waveform can be compressed and recorded as faithfully as possible by using a data string having a limited number of bits, and a high-quality waveform can be reproduced.

The above-described compression and decompression processes are performed in the data space before performing the non-linear function transformation (for convenience, referred to as a linear space) and the data space after performing the inverse function transformation (in the same manner, in the linear space). To be called). In this case, the compressed nonlinear difference data string generating step includes a step of compressing the waveform data string according to the compression ratio to generate a compressed waveform data string with a standardized range, and the compressed waveform data string according to the nonlinear function. Converting the compressed non-linear difference data sequence into a reproduced compressed waveform data sequence in accordance with the inverse function, and Expanding the reproduced compressed waveform data sequence according to the expansion ratio to generate a reproduced waveform data sequence.

According to this configuration, the entire region (full range) of the non-linear function conversion and the inverse function conversion can be effectively used regardless of the type of the waveform, thereby increasing the reproducibility of the waveform.

Alternatively, the compression and decompression processes can be performed in the data space (in the non-linear space) between the non-linear function conversion and the inverse function conversion. In this case, the compressed nonlinear difference data string generating step includes a step of changing the waveform data string into a nonlinear difference data string according to the nonlinear function, and a method of compressing the nonlinear difference data string according to the compression ratio, and a limited number of bits. Generating the compressed non-linear difference data sequence standardized with respect to, wherein the reproducing step expands the compressed non-linear difference data sequence according to the expansion rate to generate a reproduced non-linear difference data sequence, Converting the reproduced nonlinear differential data sequence into a reproduced waveform data sequence according to the inverse function.

In the case of this configuration, the information of the nonlinear difference data string converted by the nonlinear function is lost as much as possible by the compression processing using the compression rate dependent on the waveform data string and the expansion processing using the expansion rate depending on the waveform data string. Since the data is input to the inverse function transformation processing in a non-conformal manner, higher reproducibility than the conventional nonlinear differential PCM method can be obtained.

Further, according to the present invention, as a waveform reproducing apparatus that performs the above-described waveform reproducing method, a sequence of differences between adjacent waveform data of a waveform data sequence representing an audio signal is inversely proportional to a magnitude of a fluctuation of the waveform data sequence. Compressed non-linear difference data sequence corresponding to a product obtained by multiplying a compression ratio of a value to be converted by a predetermined non-linear function, or equivalent to a product obtained by converting the difference sequence by the non-linear function and a product obtained by multiplying the compression ratio. Compressed nonlinear difference data string storage means for storing a compressed nonlinear difference data string; expansion rate storage means for storing an expansion rate related to the compression rate; and the compression based on an inverse function of the nonlinear function and the expansion rate. A reproducing unit for reproducing a waveform data sequence representing an audio signal from the non-linear difference data sequence.

The reproducing means can perform the decompression process in a linear space or a non-linear space.

Reproducing means for performing the expansion process in the linear space, converting means for converting the compressed non-linear difference data sequence into a reproduced compressed waveform data sequence according to the inverse function, and reproducing and reproducing the reproduced compressed waveform data sequence according to the expansion ratio And a decompression means for generating a waveform data string. On the other hand, the reproducing means for performing the decompression process in the non-linear space includes: decompression means for decompressing the compressed non-linear differential data sequence according to the decompression ratio to generate a non-linear differential data sequence; And a conversion means for converting the data into a data string.

Further, the waveform recording / reproducing method of the present invention can include an audio signal envelope extraction process and a normalization process (envelope removal process) using the extracted envelope as preprocessing of the recording system. Such a waveform recording / reproducing method includes a preparation step of preparing a waveform data string representing an original audio signal, an envelope extracting step of extracting an envelope from the waveform data string, and normalizing the waveform data string by the extracted envelope. A normalizing step of generating a normalized waveform data sequence from which the envelope has been removed, a step of calculating the magnitude of the variation of the normalized waveform data sequence, and the magnitude of the variation is inversely proportional to the magnitude of the variation. Calculating a value compression ratio, and a compressed non-linear difference data sequence corresponding to a value obtained by multiplying a sequence of differences between adjacent waveform data of the normalized waveform data sequence by the compression ratio and converting the result by a predetermined non-linear function. Alternatively, a sequence obtained by converting a sequence of differences between adjacent waveform data of the normalized waveform data sequence by the non-linear function and a value obtained by multiplying the result by the compression ratio is used. A compressed non-linear difference data sequence generating step of generating a compressed non-linear difference data sequence, a data sequence storing step of storing the compressed non-linear difference data sequence, and a decompression ratio storing step of storing a decompression ratio related to the compression ratio. An envelope generating step of generating a predetermined envelope; and a reproduction of generating a waveform data stream from the stored compressed non-linear difference data stream based on the inverse function of the nonlinear function, the stored expansion rate, and the predetermined envelope. And a step.

By performing the above-described envelope extraction and normalization (envelope removal) of the waveform data string based on the extracted envelope, it is possible to improve the S / N ratio in a portion having a small volume (a rising portion or an attenuation portion of a sound).

One mode of a waveform reproducing apparatus to which such a waveform recording / reproducing method is applied is that a fluctuation of the waveform data sequence is added to a sequence of a difference between adjacent waveform data of a normalized waveform data sequence representing an audio signal from which an envelope has been removed. A product obtained by multiplying a value obtained by multiplying a compression ratio having a value inversely proportional to the above by a predetermined nonlinear function and a compressed non-linear difference data sequence corresponding to a value obtained by converting the difference sequence with the non-linear function, and multiplying the result by the compression ratio. Compression nonlinear difference data string storage means for storing a compressed nonlinear difference data string corresponding to: expansion rate storage means for storing an expansion rate related to the compression rate; envelope generation means for generating a predetermined envelope; Reproduction for reproducing a waveform data sequence from the compressed non-linear difference data sequence based on an inverse function of a non-linear function, the expansion ratio, and the predetermined envelope. And having a stage, a.

In this embodiment, the reproducing means requires two multiplications to reproduce one waveform data sample, ie, multiplication of the expansion rate and multiplication of the envelope.

Another aspect of the waveform reproducing apparatus is that a compression ratio of a value inversely proportional to the magnitude of the fluctuation of the waveform data sequence is stored in a sequence of differences between adjacent waveform data sequences of a normalized waveform data sequence representing an audio signal from which an envelope has been removed. A compressed non-linear difference data sequence corresponding to a product obtained by converting the product by a predetermined non-linear function or a compressed non-linear difference data sequence corresponding to a product obtained by multiplying the result obtained by converting the difference sequence by the non-linear function by the compression ratio. A compressed nonlinear difference data string storing means for storing, and a waveform reproducing means for reproducing a waveform data string from the compressed nonlinear differential data string, wherein the waveform reproducing means performs the compression nonlinearity according to an inverse function of the nonlinear function. Converting means for converting the differential data sequence into a compressed normalized waveform data sequence; and an envelope generating an envelope expanded at an expansion ratio related to the compression ratio. Characterized in that it has a rope generating means, and multiplying means for multiplying the envelope and the decompression to the compressed normalized waveform data sequence, the.

In this embodiment, since the envelope generating means generates an envelope in consideration of the expansion rate, one multiplication (envelope multiplication) is performed for one waveform data sample reproduction.
And the number of multiplications can be reduced.

Further, the present invention can be applied not only to the reproduction of the waveform by the above-described reproducing means, but also to a waveform generating apparatus including a digital filter means for further processing the reproduced waveform.

Embodiment An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows the overall configuration of the waveform recording apparatus. The digital audio tape 1 is a master tape on which a natural musical instrument sound or the like is recorded.
Is transferred to the computer 4 via the. Computer 4
Then, a process described later is performed on the transferred waveform data sequence to generate a compressed nonlinear difference data sequence and various parameters. The compressed non-linear difference data string is written to the PROM 6 by the PROM writer 5. The written PROM 6 becomes the difference waveform memory 11 of the waveform reproducing device (FIG. 5). Further, various parameters (expansion rate, envelope parameter) are used as control information of the waveform reproducing apparatus.

FIG. 2 is a flowchart of generation of the compressed nonlinear differential data sequence in the computer 4. First, in S1, the parameter a
Is set, and the waveform data sequence is loaded into the array {x (n)}. Here, the parameter a is a coefficient used for generating a compression ratio (a compression scale factor) described later. In the next step S2, the waveform data sequence {x (n)}, that is, x (0),
x (1)... x (n)...
(N) Extract｝. Several methods are available for envelope extraction (for example, Japanese Patent Application No. 61-264205,
See Japanese Patent Application Nos. 61-264206 and 62-264207). For example, information indicating the size of the waveform data for each predetermined section of the waveform data string, for example, the absolute value | x (n) |
, The peak-to-peak magnitude of the waveform data, or the power of the waveform data (eg, the square root of the sum of squares of the waveform data) is determined as the envelope. As the predetermined section, for example, the basic cycle of the waveform data sequence is selected. In FIG. 3, | x (n) |
The maximum value is extracted, and obtains it as the envelope level ELk, the envelope rate ER _k at a predetermined interval X ER _k =
An envelope value e (n) corresponding to each waveform data d (n) is obtained based on (EL _k −EL _k−1 ) / X and based on these parameters. Extracted envelope @e
(N)} is used to normalize the waveform data sequence {x (n)} in the next S3. That is, each waveform data x (n) is divided by the envelope value e (n). As a result, the waveform data sequence ｛x with the envelope removed (normalized)
₀ (n)｝ is obtained. Eliminating the envelope improves reproducibility in low volume areas.

Next, in S4, in accordance with _{x 1 (n) ← (A} / x 0 max) · x 0 (n), to normalize the waveform data sequence {x ₁ (n)}. Here, x ₀ max is the absolute value of the largest waveform data in the waveform data sequence {x ₀ (n)}. For example, if A = 32767.0 is selected as A, the waveform data sequence {x ₁ (n)} is normalized to ± 15 bits.
That is, the maximum value in the waveform data sequence {x ₁ (n)} is +32
Expressed as a binary number indicating 767 or -32767. The reason for the normalization is to make the reproduction levels uniform when reproducing the waveform data sequence {x ₁ (n)} for various waveforms in the reproducing apparatus.

Then in step S5 (this is actually a step that is shown for explanation) To obtain the nonlinear difference data sequence {d ₁ (n)}.

Where f (x) is a non-linear function, Is a logarithmic function represented by In this (Equation 4),
Xmax is the maximum value of the input signal x, and Ymax is the output signal y (= f
The maximum value of (n)), M is a conversion characteristic parameter. Assuming that Xmax = 32767.0 and Ymax = 127.0, the input signal x of ± 15 bits is converted into an output signal y of ± 7 bits by this (Equation 4).

In the conventional non-linear difference method, a non-linear difference data sequence {d ₁ (n)} obtained by (Equation 3) is recorded as final waveform information. Here, the input signal x referred to in (Equation 4) is expressed in [] in (Equation 3). Is a linear difference of the waveform data sequence {x ₁ (n)}. The maximum value of the linear difference depends on the waveform data sequence, and can take various values depending on the type of the waveform data sequence (and thus the type of the original sound). Therefore, Xmax = 32767.0 described in (Equation 3) generally does not hold, and 3max
Takes various values smaller than 2767.0. Assuming that the nonlinear function f (x) shown in (Equation 4) is a 16 / 8-bit logarithmic conversion function, the whole area (full range) of the conversion is −3276
7.0 to +32767.0, and -127.0 to +127.0 on the output side. In the case of the conventional nonlinear difference method, the maximum value of the linear difference of the waveform data sequence {x ₁ (n)} is, for example, ±
If it is 4000, the input full range is -32767.0 to +3
Only a partial range of -4000 to 4000 out of 2767.0 will be used. As described above, in the conventional nonlinear difference method, a part of the nonlinear conversion is not used, so that sufficient reproducibility cannot be obtained. Further, in the conventional non-linear difference method, there is a problem that a region in which non-linear conversion is used differs depending on the type of waveform, and the reproducibility fluctuates accordingly.

FIG. 4 illustrates SN characteristics when a logarithmic function for (Equation 4) is used as the nonlinear function f (x).
When a 16-bit input signal (difference in a waveform data string) is passed through a 16 / 8-bit logarithmic function f (), and further passed through an 8 / 16-bit exponential function f ⁻¹ () which is an inverse function of the f (). , The value of the 16-bit input signal to the logarithmic function f () and the exponential function f ^-1
The values of the 16-bit output signal from () generally do not match,
An input signal in the range of S-e1 to S + e2 is converted to an output signal of value S. Here, for convenience of explanation, it is assumed that (e1 + e2) / 2 = N (noise) and N / S is defined as a noise-to-signal ratio (a noise-to-signal ratio through a nonlinear function and its inverse function). 4th
The figure shows the N / S (SN) for the magnitude of the input signal (which may be regarded as the output signal) when μ = 200 in (Equation 4).
(Reciprocal of the ratio). As can be understood from FIG. 4, when the signal is passed through the logarithmic function f (x) and its inverse function, the N / S decreases as the signal value increases (SN
Ratio is better). Now, 4000 Datosuruto as example the maximum value of the linear difference data string of the waveform data sequence, which is about the input full range value X _F (= 32767) in Figure 4 1 /
8, i.e. substantially corresponding to the X _F / 2 ^3. Such a linear difference data sequence is composed of data (elements) having various values in the partial range of -4000 to 4000. Here, if each element of the linear difference data sequence is changed to a large value in accordance with a compression ratio described later according to the present invention, for example, if the range to be used is changed to a full range,
The SN ratio is improved from the SN ratio before the change. In particular, in the case of the characteristic as shown in FIG. 4, the SN ratio for an element having a small value (a portion where the fluctuation of the waveform data is small) is greatly improved.

Even in the case of other nonlinear functions, the characteristic that the N / S ratio (the ratio of noise in the signal) caused by the combination of the nonlinear function and the inverse function decreases or increases as the signal increases (full-range conversion) Therefore, the overall SN ratio can be improved by the above-described method as long as the characteristic is such that the overall SN ratio is minimized.

In order to realize this method, in S6 of FIG. 2, the absolute value of the linear difference is calculated from the linear difference data sequence of the waveform data sequence {x ₁ (n)}. Seeking the maximum dmax of X ₁ (n-1) can be used instead of The compression ratio a · (B / dmax) is calculated from the maximum value dmax of the linear difference, the input full range value B (for example, 32767.0), and the parameter a. In S7, the waveform data sequence {x ₁ (n)} is compressed using this compressed value a · (B / dmax) to generate a compressed waveform data sequence {x
₂ (n) is formed. If a = 1, the maximum value dmax of the linear difference sequence of the waveform data sequence {x ₁ (n)} becomes the input full range value B in the linear difference sequence of the compressed waveform data sequence {x ₂ (n)}, and the compressed waveform The range of the linear difference sequence (input of the logarithmic function f (x) for the non-linear conversion) of the data sequence {x ₂ (n)} is a full range (for example, ± 15 bits). Actually, several values of a are tried on the basis of a = 1, and a value that is most reproducibly auditory is selected upon reproduction. As described later, if a> 1, a part of the difference data (large difference data) is clipped to the full range value, but the SN ratio of the small difference data is further improved.
If a ≦ 1, no clip occurs.

Next, in S8, using a logarithmic function f (x), From the compressed waveform data string {x ₂ (n)} of M bits (for example, 16 bits) to the compressed non-linear difference data string {d ₂ (n)} of bits less than M bits (for example, 8 bits)
To form Here, x ₂ (n) represents the n-th compressed waveform data in the recording system, Represents the (n-1) th compressed waveform data to be reproduced by the reproduction system. Shown in [] in the calculation of (Equation 5) Is greater than B, the value is clipped to a positive input full-range value B, and if the value is less than -B-1, it is clipped to a negative input full-range value -B-1. Such a clip occurs when a> 1 and does not occur when a ≦ 1.

Next, in S9, the reciprocal dmax / a · B of the compression ratio is stored as a scale factor (extension ratio) norm in the waveform data reproduction system. As described later, the reproduction system multiplies the waveform data obtained by performing the inverse function conversion on the compressed nonlinear difference data by the norm value. As a result, the waveform data has a substantially constant numerical range regardless of the type of the waveform data (for example, -32767 to 327).
67). An example of the norm value is 5 to 15 for piano waveform data sampled at a sampling frequency of 32 KHz.

In the last S10, the calculated compressed difference data sequence ｛d
₂ (n) Save｝ to a file.

By the above processing, the compressed nonlinear differential data sequence ｛d
₂ (n)}, scale factor (expansion rate) norm, and envelope data sequence {e (n)} were obtained.

The waveform recording apparatus performs the above-described processing (FIG. 2) for some values of the parameter a, reproduces the data, and obtains the data with the best reproducibility in a hearing experiment (scale factor, compressed non-linear difference data sequence, envelope data sequence).
Is stored, and the compressed nonlinear differential data sequence is written to the PROM 6 to create the differential waveform memory 11.

Next, a waveform reproducing apparatus will be described. FIG. 5 is a block diagram of an electronic musical instrument incorporating a waveform reproducing device. The microcomputer 8 scans the switch 9 to transfer data corresponding to the selected tone color to the tone generator 10 in a well-known manner, and scans the keyboard 7 to transfer information on a pressed key to the tone generator 10. . In accordance with the waveform reproducing method of the present invention, the tone generator 10 generates (reproduces) the musical tone waveform W by calculating the compressed nonlinear differential data of the differential waveform memory 11, and emits the tone waveform W via the DAC 12, the amplifier 13, and the speaker 14.

FIG. 6 shows the configuration of the tone generator 10. The interface 24 is for writing data from the microcomputer 8 of FIG. 5 to the tone generator 10, and the envelope data memory 25 has an envelope level data EL.
k and the envelope plate data ERk are stored in the scale factor memory 27 by the scale factor (expansion rate) norm described above.
Is written to the address generator 28 in accordance with the pitch. The envelope generator 26 receives the envelope level data ELk and the envelope plate data ERk from the envelope data memory 25, and generates an envelope E. The address generator 28 generates an address A that changes at a speed corresponding to the frequency data, and generates an increment signal i every time the address A changes.
Send to The waveform generator 29 is means for reproducing a waveform in accordance with the waveform reproducing method of the present invention. The waveform generator 29 responds to the increment signal i from the address generator 28, and outputs the compressed nonlinear differential data D from the differential waveform memory 11 and the scale factor memory 27. The tone waveform W is generated using the scale factor norm and the envelope E from the envelope generator 26.

FIG. 7 shows the function of the waveform generator 29. The conversion memory 30 stores an inverse function f ⁻¹ () of the nonlinear function f () used by the waveform recording device (FIG. 1) for recording the waveform. For convenience of explanation, it is assumed that the non-linear conversion processing S8 in FIG. 2 is processing for converting 16-bit compressed linear differential data into 8-bit compressed non-linear differential data by the logarithmic function shown in (Equation 4). In this case, the conversion memory 30 stores the stored inverse function f ^-1
() Provides means for inversely converting the 8-bit compressed nonlinear difference data from the difference waveform memory 11 into 16-bit compressed linear difference data. The conversion memory 30 includes a gate responsive to the signal i, and only when the signal i is 1, that is, when the address A is updated, the (new) 8-bit compressed nonlinear differential signal from the differential waveform memory 11 is stored. 16 for the data
Output bit-compressed linear difference data. The 16-bit compressed linear difference data from the conversion memory 30 is input to an accumulator (accumulation means) including an adder 31 and FF32. The accumulator accumulates the compressed linear difference data whose range is standardized to ± 15 bits to reproduce compressed waveform data (cumulative calculation power). This cumulative power is multiplied by the scale factor norm from the scale factor memory 27 by the multiplier 33, whereby the cumulative power having a magnitude different depending on the type of waveform is normalized to 16-bit reproduced waveform data. The multiplier 34 multiplies the normalized reproduced waveform data by the envelope E, and outputs final waveform data W. As a result, the output W of the waveform generator 29 becomes a maximum of 16 with an envelope added.
The tone waveform has a bit range.

Note that the envelope level and rate data set in the envelope data memory 25 are stored in S2 (second
It can be determined based on the envelope extracted in FIG. Since the number of envelope segments (attack segments, etc.) generated by the envelope generator 26 is generally limited, the extracted envelope should be approximated by a limited number of segment envelopes (envelope level and rate for each segment). become. For such an envelope approximation, for example, a technique disclosed in Japanese Patent Application No. 61-264205 can be used. Basically, an appropriate number of segments is set, and the error between the polygonal envelope (for example, a characteristic switching point, ie, a folding point is selected as a point on the extraction envelope) and the extraction envelope at the number of segments is evaluated. However, optimal approximation is possible by selecting a polygonal envelope that minimizes the error. The result (envelope level and rate information for each segment that defines the selected folded line envelope) is stored in the microcomputer 8 of the electronic musical instrument as reference envelope parameters. In operation, the microcomputer 8
Changes the reference envelope parameter depending on key touch or the like, and stores the result in the envelope data memory 25.
Will be written to.

However, if desired, the envelope may be generated independently of the extracted envelope.

In the case of the configuration shown in FIG. 7, the waveform generator 29 must execute multiplication (norm multiplication and envelope E multiplication) twice each time a musical tone waveform W sample is generated (the function shown in FIG. 7). Time division multiplexing (TDM)
In this case, one multiplier hardware multiplies norm by the cumulative calculation power in one time slot within the sampling period, and multiplies the multiplication result by the envelope E in another time slot.) If the number of polyphonics of the waveform generator 29 is N, the waveform generator is
29 must perform 2N synthesis multiplications. Therefore, reducing the number of multiplications in order to reduce the processing amount of the waveform generator 29 is significant.

The configurations of the tone generator 10M and the waveform generator 29M in which the number of times of multiplication is reduced are shown in FIGS. 8 and 9, respectively. In this configuration, the tone generator 10M does not require the scale factor memory 27, and the microcomputer 8 stores the scale factor n in the envelope data memory 25M.
Write the envelope level data ELk 'and the envelope plate data ERk' to which orm has been added.

As a result, the envelope generator 26M generates an envelope E 'expanded by a scale factor (expansion rate) norm. That is, the envelope E 'corresponds to the envelope E shown in FIGS. 6 and 7 multiplied by the scale factor norm. Therefore, this envelope E ′ is
As shown in the figure, the accumulator (3
By multiplying the cumulative calculation power of (1, 32), the final music waveform W can be obtained immediately.

With this configuration, the waveform generator 29M needs to execute only one multiplication to generate one musical tone waveform sample, and the number of necessary multiplications is reduced.

In the case of electronic musical instruments, in addition to simply playing back the waveform with a waveform generator such as 29, to process the reproduced waveform,
Digital filtering is often performed.
Such a digital filter is, for example, a multiplier shown in FIG.
An output of 33, that is, reproduced waveform data whose range is standardized by the expansion rate norm is processed as an input, and an output is provided to the envelope multiplier. FIG. 10 illustrates the function of this type of digital filter. The digital filter shown is a first-order IIR digital filter including a filter gain multiplier 38, an adder 39, a delay element (D) 40, a feedback multiplier 41, and an adder 42. Of the desired digital filters can be used. The filter gain coefficient K used in the filter gain multiplier 38 and the feedback coefficient B used in the feedback multiplier 41 are provided from an appropriate scale factor memory (not shown) in which data is set by the microcomputer 8.

When the function of the digital filter 37 as exemplified in FIG. 10 is incorporated in the waveform generator 29 of FIG. 7, it is necessary to separately perform the multiplication of the expansion rate norm and the multiplication of the filter gain K (separate hardware). A multiplication circuit is used, or the multiplication of norm and the multiplication of K are separately performed by a shared hardware multiplication circuit). Fig. 11 shows the standardized digital filter 37M.
By using a coefficient K ′ corresponding to (K × norm) as the scale factor in the multiplier 38M, it is possible to simultaneously achieve the multiplication of norm and the multiplication of K by one multiplication. It is possible to reduce the number of multiplications in a waveform generator requiring high-speed processing. In this case, multiplication 38
For example, M multiplies the cumulative calculation power (compressed waveform data) from the adder 31 in FIG. 7 by a coefficient K.

Generally, multiple multiplications performed on the direct path of a linear system can be combined into one multiplication. Therefore, if desired, the multiplication of the envelope E, the multiplication of the expansion rate norm, and the multiplication of the filter gain K in the derivative generator may be simultaneously achieved by one multiplication (from the envelope generator, E × norm × By setting data in the envelope data memory 25 so that an envelope corresponding to K is generated).

The compression processing and the expansion processing described in FIGS. 2 and 7 are performed in a linear space. That is, in the recording system (FIG. 2), the compression process S7 is performed on the waveform data sequence before performing the nonlinear conversion S8, and in the reproduction system (FIG. 7), the inverse conversion of the nonlinear conversion S8 in the recording system ( At the stage after the second non-linear conversion is performed by the conversion memory 30, the expansion process is performed on the compressed waveform data sequence by the multiplier 33.

Instead, the compression processing and the expansion processing may be performed in a non-linear space. That is, in the recording system, the compression process is performed after performing the nonlinear conversion, and in the reproduction system, the expansion process is performed after performing the inverse conversion.

FIG. 12 shows a main part of a waveform recording / reproducing method in which compression processing and expansion processing are performed in a nonlinear space. In step T1 of the recording system, for example, a linear difference sequence from a linear waveform data sequence normalized to a range of ± 15 bits (corresponding to the data sequence {x ₁ (n)} obtained in the normalization processing S4 in FIG. 2) ｛L
₁ (n) is formed. This is obtained by l ₁ (n) ← x ₁ (n) −x ₁ (n−1).

Next, in step T2, a relatively high-accuracy (higher-precision than the conversion in S8 in FIG. 2) nonlinear conversion is performed to obtain a nonlinear difference sequence ｛d _H
(N) Obtain｝. This is obtained by d _H (n) ← f _H (l ₁ (n)). Here, f _H () represents a relatively high-precision nonlinear conversion function. This transformation moves the data space from a linear space to a non-linear space.

Next, in step T3, an appropriate compression ratio depending on the waveform data sequence (for example, a compression ratio such that the maximum value d _H max of | d _H (n) | And performs a compression process to form a compressed non-linear difference sequence {d _C (n)} with a limited number of bits. Assuming that B / d _H max is used as the compression ratio (this corresponds to the parameter a in FIG. 2)
, The compressed non-linear difference data d _C (n) is obtained by d _C (n) ← (B / d _H max) · d _H (n). Here, d _C (n) is represented by a smaller number of bits than d _H (n).

The compressed nonlinear difference sequence ｛d thus obtained
_C (n)} is stored in an appropriate memory in step T4.

In the reproducing system, in step T5, a decompression process is performed on the compressed nonlinear differential data sequence to obtain a nonlinear differential data sequence {d '
_H (n)} is reproduced. This expansion process can be performed by using the reciprocal of the compression ratio used in the recording system as the expansion ratio and multiplying it by each compressed nonlinear difference data d _C (n).
For example, if the compression ratio is B / d _H max, the expansion ratio is d _H max / B, and each reproduced nonlinear difference data d ′ _H (n) is d ′ _H (n) ← (d _H max / B) · d _C (N). Here, d ′ _H (n) is represented by a bit number larger than the bit number of d _C (n).

Next, at step T6, a relatively high accuracy (the conversion memory shown in FIG. 7)
The linear difference sequence {l ′ ₁ (n)} is reproduced by performing an inverse transform (higher accuracy than the inverse function of 30). By this inverse transformation, the data space returns from the nonlinear space to the linear space. If the data of this linear difference sequence {l ′ ₁ (n)} is accumulated as shown in step T7, a normalized linear waveform data sequence is reproduced.

As an example of numerical accuracy, the normalized linear waveform data string is ±
16-bit data sequence in 15-bit range, linear difference sequence ｛l
₁ (n)｝ is a 16-bit data sequence, nonlinear difference sequence {d
_H (n)｝ is a 16-bit data sequence (therefore, the nonlinear transformation T
2 is 16 / 16-bit precision), reproduced nonlinear difference sequence ｛d '
_H (n)} is a 16-bit data sequence, and a reproduced linear difference sequence {l '
₁ (n)} is a 16-bit data sequence (therefore, the inverse transform T6 is
16/16 bit accuracy), ± 1 for playback normalized linear waveform data string
This is a 16-bit data string in a 5-bit range.

Here, the conventional nonlinear difference method will be described with reference to the method of FIG. 12.
(16 / 16-bit non-linear conversion) and T3 (variable compression) processing are replaced with 16 / 8-bit non-linear conversion processing, and T5 in FIG. 12 is performed.
It can be said that this is a method of performing an 8 / 16-bit inverse conversion process instead of the (variable expansion) and T6 (16 / 16-bit inverse conversion) processes. The 16 / 8-bit non-linear conversion process can be considered to include a 16 / 16-bit non-linear conversion process and a fixed compression process that takes the upper 8 bits of the 16-bit non-linear data that is the result of the process. Similarly, the 8 / 16-bit inverse conversion process is a fixed decompression process that generates 16-bit nonlinear data using 8-bit nonlinear data as the upper 8 bits, and the 16 / 16-bit inverse data is input as 16-bit nonlinear data that is the processing result. It can be considered that it consists of an inverse conversion process for performing the conversion. Therefore, in the conventional nonlinear difference method, information of lower 8 bits of 16-bit nonlinear data is basically ignored regardless of the type of waveform data. The range of the 8-bit nonlinear difference sequence obtained by the fixed compression processing depends on the type of waveform data. For example, if the maximum value of the elements of such an 8-bit nonlinear difference sequence is 63,
The actual range of this 8-bit nonlinear difference sequence is ± 6 bits, and the effective bit length is only 7 bits instead of 8 bits. For example, an element having a size of 1/128 of the maximum value 63 originally has a value close to 0.5, but due to fixed compression processing, this element on an 8-bit nonlinear difference sequence is 0 (rounded down or Due to rounding). From the above description, it can be understood that the reproducibility of the conventional nonlinear difference method is not optimal, and that the reproducibility varies considerably depending on the type of waveform.

In contrast, the method of the present invention shown in FIG. 12 performs variable compression processing that uses a compression rate that depends on the type of waveform instead of fixed compression processing during recording, and performs playback instead of fixed decompression processing, instead of using fixed expansion processing. Perform variable decompression using a dependent decompression rate. Therefore, the range of the 8-bit nonlinear difference sequence output from the compression process T3 is independent of the type of the waveform,
It can be standardized to a range of ± 7 bits, and an effective bit length of 8 bits can be secured. Therefore, it is possible to obtain a reproduced waveform with higher reproducibility than the conventional nonlinear difference method. In particular, the reproducibility in a portion where the fluctuation of the waveform is small is improved.

The numerical accuracy described above with reference to FIG. 12 is merely exemplary.
Further, the number of input bits and the number of output bits of the nonlinear transform T2 do not need to be the same, and the number of input bits and the number of output bits of the inverse transform T6 do not need to be the same. In particular, for the latter inverse transform T6, it is common to use an inverse transform memory for high-speed processing. Therefore, the number of input bits should be as small as possible within a range where the effect of the present invention is not lost. Saves the storage capacity of the inverse transform memory.

For example, assume that the largest element of the 16-bit linear difference data sequence {l ₁ (n)} has a value of at least 1000 (about 1/32 of the full range). It is assumed that a logarithmic function (μ = 200) as shown in (Equation 4) is used as the nonlinear conversion.
First, the conversion precision of this logarithmic function is set to 16/16 bits. In this case, the value of the converted 16-bit nonlinear data is approximately 47 × 256 with respect to the value 1000 of the input data. Such a 16-bit nonlinear data sequence is subjected to variable compression processing T3.
Since the sequence becomes an 8-bit compressed nonlinear difference data sequence of at most ± 128 stages, the lower bits of the 16-bit nonlinear difference data do not actually affect the value of the 8-bit compressed nonlinear difference data. That is, the nonlinear difference data is not required up to 16 bits, and 11 bits is sufficient. In other words, as long as a waveform data sequence in which the maximum element of the 16-bit linear differential data sequence {l ₁ (n)} does not become smaller than 1000, the conversion accuracy of the non-linear conversion (logarithmic conversion) T2 is 16/11 bits. . Similarly, inverse conversion (exponential conversion) T6 processing is 11/16
This can be realized by using a bit conversion memory. Alternatively, up to the expansion process T5, the numerical precision described first may be used, the 16-bit data output from the expansion process T5 may be rounded to 11 bits, and the value may be used to access the 11 / 16-bit conversion memory. Good. Further, by using the sign bit of the 11-bit reproduced nonlinear data, a 10 / 15-bit conversion memory can be used. For example, when a two's complement format is used as a binary number, the lower 10 bits of the 11-bit reproduced nonlinear data are converted to the sign bit (MS bit).
(B bit) to perform 2's complement processing (for negative data) to convert to 10-bit data of positive expression, address the conversion memory with this 10-bit data to obtain positive 15-bit data, and obtain this 15-bit data Then, by performing reproduction and 2's complement processing using the sign bit and adding the sign bit (to the negative data), 16-bit linear difference data is reproduced.

The compression ratio a · depends on the waveform data sequence in the nonlinear space.
When the compression process is performed at (B / d _H max), a clip occurs when a> 1. In the compression processing with a clip, the reproducibility of a portion having a large waveform variation is deteriorated, but the reproducibility of a portion having a small waveform variation is further improved. Compressed nonlinear difference data d _C (n) for such a compression ratio is expressed as d _C (n) = a · (B / d _H max) · A (Equation 6) Is obtained by calculating However, if the value on the right side of (Equation 6) is smaller than −B−1, it is clipped to d _C (n) = − B−1, and if the value on the right side is larger than B, d _C (n) =
Clipped to B. When a ≦ 1, A = d _H (n). Note that in (Equation 7) Represents the n-th element x ₂ (n) of the linear waveform data sequence standardized by the recording system, Term represents the normalized linear waveform data sequence to be played by the playback system (n-1) th element x ₂ '(n-1). Therefore, (Equation 7) expresses x ₂ (n) and x ₂ ′ (n−
This is an equation of a process of performing a non-linear conversion of a linear difference from 1) according to a non-linear function f _H (). (Equation 6) is an equation of a process of multiplying the non-linear difference value obtained as a result of (Equation 7) by the compression rate a · (B / d _H max) and selectively clipping the result. For reference, Fig. 13 shows the compressed nonlinear differential data sequence ｛d
₄ shows an example of the flow of a compression process for forming _C (n)｝.
If a ≦ 1, no clip occurs, so e in the figure is always e = 0, and f _H ｛l ₁ (n) + e｝ = f _H ｛l ₁ (n)｝
= D _H (n). Similarly, FIG.
By executing (Equation 7), the normalized linear waveform data sequence ｛x
_{1 is} a functional block diagram for converting ₁ (n)} into a compressed nonlinear differential data sequence {d _C (n)}. Next, FIG. 15 shows a function for converting a normalized linear waveform data sequence {x ₁ (n)} into a compressed nonlinear differential data sequence {d ₂ (n)} in a waveform recording / reproducing method that performs companding in linear space. FIG. 2 shows a block diagram (corresponding to S7 and S8 in FIG. 2).

A waveform generator for executing the reproduction of the waveform shown in FIG. 12 is shown as 29N in FIG. This waveform generator 29N
The multiplier 63 multiplies the expansion nonlinearity norm (the reciprocal of the compression ratio in the recording system) to the compressed nonlinear difference data D from the difference waveform memory in which the waveform information is recorded in accordance with the waveform recording method described in FIG.
The linear difference data is reproduced by passing it through a conversion memory 64 that stores an inverse function f _H ⁻¹ () of the nonlinear interval at the time of recording, and is further passed through an accumulator including an adder 31 and an FF 32 to obtain a normalized waveform. The data is reproduced, and the multiplier 34 multiplies the normalized waveform data by the envelope E to generate a musical tone waveform W to which an envelope is added. The difference from the waveform generator 29 in FIG.
7 is performed on a direct path in a linear space in FIG. 7, whereas the multiplication (expansion processing) is performed on a direct path in a non-linear space in FIG.

The description of the embodiment is finished above, but various modifications and changes can be made within the scope of the present invention.

[Effects of the Invention] As described above in detail, according to the waveform recording / reproducing method of the present invention, a waveform data stream is recorded based on a compression rate dependent on the waveform data stream and a predetermined nonlinear function during recording. From the compressed non-linear difference data string with a limited number of bits as waveform information, and based on the expansion rate (reciprocal of the compression rate or a value close to it) and the inverse function of the nonlinear function that depend on the waveform data string during reproduction Thus, the waveform data sequence is reproduced from the compressed nonlinear difference data sequence, so that reproducibility that cannot be achieved by the conventional nonlinear difference technique can be obtained. Further, the reproducibility that fluctuates depending on the type of waveform can be suppressed to as small a fluctuation as possible.

Further, according to the waveform reproducing apparatus of the present invention, it is possible to reproduce a high-quality waveform regardless of the type of the waveform while saving the storage capacity of the waveform information memory.

[Brief description of the drawings]

FIG. 1 is an overall configuration diagram of a waveform recording apparatus using the waveform recording method according to the present invention, FIG. 2 is a flowchart of generation of a compressed nonlinear differential data sequence executed by the computer of FIG. 1, and FIG. 3 is FIG. FIG. 4 shows an SN characteristic relating to a specific non-linear function, and FIG. 4 shows an SN ratio improvement effect by a compression process in a linear space executed in the flowchart of FIG. FIG. 5 is an overall configuration diagram of an electronic musical instrument incorporating the waveform reproducing apparatus according to the present invention, FIG. 6 is a configuration diagram of the tone generator of FIG. 5, and FIG. 7 is a waveform of FIG. FIG. 8 is a block diagram showing a configuration example in which a decompression process is performed in a linear space as a generator. FIG. 8 is a block diagram of a modification of a tone generator. FIG. 9 is a waveform generator shown in FIG. And FIG. 10 is a block diagram showing a configuration example of a digital filter which can be incorporated in the waveform generator shown in FIG. 6. FIG. 11 is a block diagram of a configuration example of a digital filter capable of realizing filter gain processing and expansion processing by one multiplication. FIG. 12 performs compression processing and expansion processing in a nonlinear space. 4 is a flowchart of a waveform recording / reproducing method according to the present invention. FIG. 13 is a flowchart of a compression process with a clip. FIG. 14 is a waveform recording method for performing compression depending on a waveform data sequence in a non-linear space. FIG. 15 is a functional block diagram for converting into a sequence. FIG. 15 shows a waveform recording method for performing compression depending on a waveform data sequence in a linear space. FIG. 16 is a block diagram of a waveform generator using the waveform reproducing method shown in FIG. 4 Computer 29, 29M, 29N Waveform generator 11 Difference waveform memory 27 Scale factor memory 30, 64 Conversion memory 33, 63 (for expansion processing) Multiplier f (), f _H () ...... nonlinear function _{^{f - (), f H- (}} ) ...... inverse function x (n) ...... waveform data _{d 2 (n), d C} (n) ...... compressed nonlinear difference data d _max / ( a ・ B)… compression ratio norm… expansion ratio

──────────────────────────────────────────────────続 き Continued on front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G10H 7/00-7/12

Claims (9)

(57) [Claims] A step of preparing a waveform data string representing an audio signal; a step of calculating a magnitude of a variation in the waveform data string; and a value inversely proportional to the magnitude of the variation from the magnitude of the variation. A step of calculating a compression ratio of, and a compressed nonlinear difference data sequence corresponding to a sequence obtained by multiplying a sequence of differences between adjacent waveform data of the waveform data sequence by the compression ratio with a predetermined nonlinear function, or A compressed non-linear difference data sequence generating step of generating a compressed non-linear difference data sequence corresponding to a product obtained by multiplying a sequence of data differences between adjacent waveforms of the waveform data sequence by the non-linear function and the compression rate; A data string storing step of storing a compressed non-linear difference data string; a decompression rate storing step of storing a decompression rate related to the compression rate; an inverse function of the non-linear function; Waveform recording and reproducing method characterized by having a reproducing step of reproducing the waveform data sequence from the stored compressed nonlinear difference data string on the basis of the expansion ratio. 2. The waveform recording / reproducing method according to claim 1, wherein said compressed non-linear difference data sequence generating step comprises compressing said waveform data sequence according to said compression ratio to generate a compressed waveform data sequence having a range standardized. Generating, and converting the compressed waveform data sequence to the compressed non-linear difference data sequence of a limited number of bits according to the non-linear function, wherein the reproducing step comprises: Converting a data sequence into a reproduced compressed waveform data sequence; and expanding the reproduced compressed waveform data sequence in accordance with the expansion rate to generate a reproduced waveform data sequence. . 3. The waveform recording / reproducing method according to claim 1, wherein said compressed nonlinear differential data sequence generating step comprises: converting said waveform data sequence into a nonlinear differential data sequence according to said nonlinear function; Compressing the sequence in accordance with the compression ratio to generate the compressed nonlinear differential data sequence standardized for a limited number of bits, and wherein the reproducing step includes converting the compressed nonlinear differential data sequence to the expansion ratio. A waveform recording / reproducing method, comprising the steps of: generating a reproduced nonlinear differential data sequence by expanding the sequence according to the following formula: and converting the reproduced nonlinear differential data sequence into a reproduced waveform data sequence according to the inverse function. 4. A predetermined non-linear function, which is obtained by multiplying a sequence of differences between adjacent waveform data of a waveform data sequence representing an audio signal by a compression ratio having a value inversely proportional to the magnitude of the fluctuation of the waveform data sequence. Compressed nonlinear difference data sequence storage means for storing a compressed nonlinear difference data sequence corresponding to a product obtained by multiplying the compressed nonlinear difference data sequence or the difference sequence obtained by converting the difference sequence by the nonlinear function and the compression rate, Expansion rate storage means for storing an expansion rate related to the compression rate; and reproduction means for reproducing a waveform data sequence representing an audio signal from the compressed non-linear difference data sequence based on the inverse function of the nonlinear function and the expansion rate. A waveform reproducing apparatus comprising: 5. The waveform reproducing apparatus according to claim 4, wherein said reproducing means converts said compressed nonlinear differential data sequence into a reproduced compressed waveform data sequence in accordance with said inverse function, and converts said reproduced compressed waveform data sequence. And a decompression means for decompressing according to the decompression ratio to generate a reproduced waveform data sequence. 6. The waveform reproducing apparatus according to claim 4, wherein the reproducing means expands the compressed nonlinear differential data sequence according to the expansion rate to generate a nonlinear differential data sequence; Conversion means for converting a non-linear difference data sequence into a waveform data sequence; 7. A preparatory step for preparing a waveform data sequence representing an original audio signal; an envelope extracting step for extracting an envelope from the waveform data sequence; and normalizing the waveform data sequence by the extracted envelope to remove the envelope. A normalization step of generating a normalized waveform data sequence, a step of calculating the magnitude of the variation of the normalized waveform data sequence, and, from the magnitude of the variation, a compression ratio of a value inversely proportional to the magnitude of the variation. A calculating step, and a compressed non-linear difference data sequence corresponding to a sequence obtained by multiplying a sequence of differences between adjacent waveform data of the normalized waveform data sequence by the compression ratio by a predetermined non-linear function, or
A compressed non-linear difference data sequence generating step of generating a compressed non-linear difference data sequence corresponding to a result obtained by multiplying a sequence of differences between adjacent waveform data of the normalized waveform data sequence by the non-linear function and the compression rate; A data string storing step of storing the compressed non-linear differential data string; an expansion rate storing step of storing an expansion rate related to the compression rate; an envelope generating step of generating a predetermined envelope; and an inverse function of the nonlinear function And a reproducing step of reproducing a waveform data sequence from the stored compressed nonlinear differential data sequence based on the stored expansion ratio and the predetermined envelope. 8. A predetermined value obtained by multiplying a sequence of differences between adjacent waveform data of a normalized waveform data sequence representing an audio signal from which an envelope has been removed by a compression ratio having a value inversely proportional to the magnitude of fluctuation of the waveform data sequence. A compressed non-linear difference data sequence corresponding to a compressed non-linear difference data sequence corresponding to a result obtained by conversion of the non-linear function or a compressed non-linear difference data sequence corresponding to a result obtained by multiplying the result obtained by converting the difference sequence by the non-linear function by the compression ratio. Difference data string storage means, expansion rate storage means for storing an expansion rate related to the compression rate, envelope generation means for generating a predetermined envelope, an inverse function of the non-linear function, the expansion rate, and the predetermined envelope Reproduction means for reproducing a waveform data sequence from the compressed non-linear differential data sequence based on the waveform data. 9. A method in which a sequence of differences between adjacent waveform data of a normalized waveform data sequence representing an audio signal from which an envelope has been removed is multiplied by a compression ratio having a value inversely proportional to the magnitude of the fluctuation of the waveform data sequence. Compressed nonlinear difference data string corresponding to the compressed nonlinear difference data string corresponding to the one converted by the nonlinear function or the compressed nonlinear difference data string equivalent to the one obtained by converting the difference string by the nonlinear function and the compression rate Difference data string storage means, and waveform reproducing means for reproducing a waveform data string from the compressed nonlinear differential data string, wherein the waveform reproducing means compresses the compressed nonlinear differential data string according to an inverse function of the nonlinear function. Converting means for converting the data into a normalized waveform data sequence; and envelope generating means for generating an envelope expanded at an expansion rate related to the compression rate. , Waveform reproduction apparatus characterized by having a multiplication means for multiplying an envelope that the stretched to the compressed normalized waveform data sequence.