JP3933072B2 - Wave compressor - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a waveform compression apparatus that compresses waveform data.
[0002]
[Prior art]
Conventionally, as one method of generating a musical sound signal in an electronic musical instrument, the instantaneous value of a musical sound waveform of a natural musical instrument is sequentially sampled and stored in advance as digital sample waveform data, and this sample is generated when a musical sound is generated. There is a PCM system in which waveform data is read to generate a musical sound signal. This PCM method is excellent in that it can generate musical sounds close to natural instruments, but the capacity of the memory for storing waveform data has become enormous. Therefore, in order to reduce the amount of waveform data, the waveform data is compressed, the compressed waveform data is stored in the memory, and the compressed waveform data is expanded during reproduction to generate a musical signal. An apparatus has been proposed (see, for example, Patent Document 1).
Also, a waveform memory sound source using a waveform memory is known in which waveform data is linearly predicted and compressed by a waveform compression device to obtain compressed waveform data, and the compressed waveform data is stored in a waveform memory. (For example, refer to Patent Document 2). In this waveform compression apparatus, compressed waveform data is stored as variable bit length in units of frames. Further, level downshift processing is performed before or after the compressed waveform data processing for obtaining compressed waveform data, and the compressed waveform data after the level downshift processing is densely stored in the memory.
[0003]
[Problems to be solved by the invention]
In such a waveform compression apparatus, the data is compressed and stored in order to reduce the data amount of the compressed waveform data, but the shift down amount is controlled and expanded based on the limit bit length. There is a problem that the degree of deterioration of the quality of the waveform data is not taken into consideration. This shift-down amount is based on the residual data peak in the frame. If pulse-like noise enters the frame, the frame shift amount becomes unreasonably large due to the influence of quantization noise. There has been a problem that increases. Further, since the search for the shift data is performed using a predetermined value not related to the characteristics of the waveform data as an initial value, there is a problem that the search takes time.
[0004]
Therefore, an object of the present invention is to provide a waveform compression apparatus that can compress waveform data using a scale factor that gives quantization accuracy.
[0005]
[Patent Document 1]
Patent No. 2605434
[Patent Document 2]
Japanese Patent No. 2727798
[0006]
[Means for Solving the Problems]
In order to achieve the above object, the waveform compression apparatus according to the present invention, when performing linear predictive compression on waveform data in a predetermined section, a scale factor that is a constant value in the predetermined section is a residual. A scale factor whose noise level based on the code sample is smaller than the threshold It is said that. Thereby, even if pulse noise enters in a predetermined section, it is possible to prevent the quantization noise in the predetermined section from increasing due to the influence. Also, Threshold Is changed according to the value of the scale factor. Conversion An error level can be set. Furthermore, since the initial scale factor calculation means is provided to calculate the initial scale factor corresponding to the waveform data, the time for searching for the scale factor can be shortened.
[0008]
DETAILED DESCRIPTION OF THE INVENTION
FIG. 1 shows a block diagram of a musical tone generation apparatus including a storage means in which waveform data compressed by the waveform compression apparatus according to the present invention is stored in units of frames. In the musical sound generating apparatus 1 shown in FIG. 1, the CPU 10 is a central processing unit that controls the operation of musical sound generation in the musical sound generating apparatus 1 by executing various programs related to musical sound generation. That is, when a sound generation start instruction (note-on) is generated by operating the performance operator, automatic performance, input from the communication I / O, or the like, generation of a tone corresponding to the sound generation start instruction is started to the sound source unit 30. I am instructing. A ROM (Read Only Memory) 11 is, for example, a flash ROM, and stores a program for musical tone generation processing executed by the CPU 10 and various data. A RAM (Random Access Memory) 12 is a main memory in the musical sound generation apparatus 1 and includes a waveform storage area 12a in which waveform data compressed by the waveform compression apparatus according to the present invention is stored in units of frames. This is a rewritable storage means in which an area such as a work area of the CPU 10 is set. The waveform storage area 12a can store a plurality of pieces of compressed waveform data.
[0009]
The operation element 13 is a performance operation element such as a keyboard or a panel switch for performing various settings, and the display unit 14 is a display unit including a liquid crystal or the like for displaying various types of information when a musical sound is generated. The communication I / O 15 is a network interface for connecting to a server computer via a communication network such as a LAN (Local Area Network), the Internet, or a telephone line. Via this communication I / O 15, it is possible to send out a MIDI message created inside the musical tone generating device 1 and to receive a MIDI message from the outside. The control register 20 is a register in which the sound generation parameters of each sound generation channel are written from the CPU 10. The sound source unit 30 includes a decoder that performs decompression processing of waveform data that is compressed in units of frames. Based on the control of the CPU 10, the compressed waveform necessary for generating musical sounds from the waveform storage area 12a of the RAM 12 is provided. Data is read for each small frame, which will be described later, and the read waveform data is expanded. Then, processing such as interpolation of the decoded waveform data, envelope addition, channel accumulation (mixing), and effect (effect) addition are performed and output as musical sound waveform data. The musical sound waveform data output from the sound source unit 30 is converted into an analog signal and emitted from the sound system 40. Each unit is connected via a bus line 16.
[0010]
Next, before describing the sound source unit 30, the data structure of the compressed waveform data stored in the waveform storage area 12a of the RAM 12 will be described with reference to FIG.
FIG. 2 shows waveform data for one piece of compressed waveform data stored in the waveform storage area 12a. The waveform data shown in FIG. 2 includes header information in which information related to the entire waveform data is written, and frames 1 to n in which compressed waveform data is accommodated. The header information includes the number of bits of the residual code sample, the reading start address, the reading end address, the loop address, the prediction coefficient and scale factor of the first frame, and other data. The data structure of each frame in which the residual code sample, which is compressed waveform data, is written is the same. FIG. 2 shows the data structure of frame 2 as an example. For example, one frame has a data amount corresponding to 10 addresses from “00” to “09” as shown in the figure, and the data width corresponding to one address can be defined as a small frame. That is, one frame shown in FIG. 2 is composed of ten small frames. This small frame consists of sub information and a residual code.
[0011]
In this case, the number of bits of the sub information is a predetermined number of bits, and the number of bits of one sample of the residual code is determined for each waveform data so as to be suitable for the properties of the original waveform data before compression. The number of bits is fixed. When the data width of the small frame is 16 bits as shown in FIG. 2 and the number of bits of one sample of the residual code is fixed to 3 bits, for example, a 4-bit data area is allocated to the sub information. A 12-bit data area is allocated to the residual code, and the number of samples in one small frame is 4. When the number of bits of one sample of the residual code is fixed to 6 bits, the number of samples of one small frame is 2 as shown in the lower stage of the residual code in FIG. Similarly, if the number of bits of one sample of the residual code that is fixed is 4 bits, the number of samples of one small frame is 3 as shown in the lower stage of the residual code in FIG. If the number of bits of one sample of the residual code is 2 bits, the number of samples of one small frame is 6 as shown in the lower stage of the residual code in FIG. When the data width of a small frame is 16 bits, the number of bits constituting one frame is 160 bits, of which 40-bit data area is sub-information area and the remaining 120-bit data area is remaining. This is the difference code area. As described above, the number of samples of the residual code in one frame or one small frame is changed according to the number of bits of one sample of the residual code fixed for each music piece.
[0012]
By the way, the sub information includes prediction coefficient information which is a linear prediction coefficient used for decompressing the residual code of the next frame, a scale factor used for decompressing the residual code of the next frame, and other information. The scale factor is a factor for denormalizing a plurality of residual codes constituting one frame. In this case, if the scale factor is stored not as an absolute value but as a ratio between frames or a difference in logarithmic scale, the efficiency of the information amount with respect to the number of bits can be improved. When the scale factor is a ratio between frames or a difference in logarithmic scale, it is converted into an absolute value at the time of inverse quantization and multiplied by a residual code. Other information may be information such as volume information and loop address of waveform data.
[0013]
Next, the sound generator 30 will be described. The address generator 32 generates a sample counter for accumulating frequency information (F number) and a read address for reading waveform data from the waveform storage area 12a in the RAM 12 for each small frame. A memory counter is provided. The sample counter accumulates the F number which is the pitch shift amount of the waveform data corresponding to the specified pitch, generates a sample address consisting of an integer part and a decimal part, and uses the integer part as a residual code cache The remaining fractional parts are supplied to the unit 33 to the interpolation unit 36, respectively. A request pulse is output every time the integer part of the sample address reaches the number of samples in the small frame. Note that the number of samples in the small frame differs depending on the number of bits of one sample of the residual code as described above with reference to FIG. The memory counter of the address generator 32 generates a read address for reading out a small frame by incrementing the memory address by one each time a request pulse is input from the sample counter.
[0014]
Each time a request pulse output from the sample counter of the address generator 32 is input, the frame reading unit 31 is a small frame indicated by a memory address supplied from the waveform storage area 12a of the RAM 12 from the memory counter of the address generator 32. Read the data. The sub information in the read small frame data is supplied to the sub information decoding unit 34, and the residual code sample of the read small frame data is supplied to the residual code cache unit 33. In the sub information decoding unit 34, sub information of each small frame supplied from the frame reading unit 31 is sequentially collected in a period of one frame, and each data of the sub information is decoded. The decoded prediction coefficient information and scale factor are supplied to the decoder 35 and other data are supplied to each block of the sound source unit 30 during the next frame period of the frame. That is, the sub information of each frame is not sub frame but used in the next frame.
[0015]
In the residual code cache unit 33, the latest three small frames among the read small frames are held in the cache. Then, according to the integer part of the sample address from the address generation unit 32, a number of residual code samples corresponding to the amount of progress of the integer part are taken out from the three cached small frames, and the taken out residual is obtained. Code samples are sent to the decoder 35. In the decoder (cache) 35, every time a residual code sample is sent from the residual code cache unit 33, the residual code sample is decoded by fourth-order linear prediction expansion to obtain a restored waveform sample. . For example, 4 samples of the restored waveform samples are stored in the waveform data cache in the decoder 35.
[0016]
Here, when a sound generation start instruction (note-on) is generated by operation of a performance operator, automatic performance, input from the communication I / O 15 or the like, the CPU 10 sends a tone corresponding to the sound generation start instruction to the sound source unit 30. The generation start is instructed. The procedure in this case is as follows. Depending on the note-on (part PT, pitch N, intensity V)
(1) One of a plurality of sound generation channels of the sound source unit 30 is assigned to the generation of the musical sound.
(2) One of the waveform data stored in the waveform storage area 12a is selected based on the timbre data (on the RAM 12) currently selected in the part PT, and the pitch shift amount, volume EG parameter, LFO parameter, The output level and the like are set in the tone generation channel area assigned by the control register 20.
(3) Read the header of the selected waveform data, and read the number of bits of the residual code, the read start address, the read end address, the loop address, the prediction coefficient of the first frame, the scale factor and other data, Set to area. Each address in this case may be an address in units of frames.
(4) Write note-on in the sounding channel area.
As a result, the generation of the musical tone is started in the sound source unit 30.
[0017]
Next, an example of the configuration of the decoder 35 in the sound source unit 30 is shown in FIG. Residual code sample L supplied from the residual code cache unit 33 _n Is the sample L based on the bit number information of the residual code sample in the inverse quantization unit 71 in the decoder 35. _n Each sample L based on a scale factor SF that is extracted every time and gives quantization accuracy _n Are denormalized and dequantized. As a result, the residual code sample L from the inverse quantization unit 71. _n A decoded waveform sample qn decoded every time is output. The decoded waveform sample qn is supplied to the adder 72 and output from the linear prediction calculation unit 73. _n-1 Is added to restore waveform sample ◇ X _n Is restored. Restored waveform sample ◇ X _n Is cached in the waveform data cache unit 74 and decompressed restored waveform sample ◇ X _n Is output as
[0018]
The waveform data cache unit 74 has four restored waveform samples ◇ X from the present to the past. _n ~ ◇ X _n-3 (D1 to D4) are cached, and the four restored waveform samples D1 to D4 cached are supplied to the linear prediction calculation unit 73. The linear prediction calculation unit 73 performs the fourth-order linear prediction by multiplying the four restored waveform samples D1 to D4 by the linear prediction coefficients P of the respective orders and adds them, and the next restored waveform sample ◇ X _{n + 1} Linear prediction sample to obtain ◇ S _n Is generated. The bit number information of the residual code sample Ln is supplied from the sub information decoding unit 34 by decoding the header of the compressed waveform data in the sub information decoding unit 34, and is supplied with the scale factor SF and linearity. As described above, the prediction coefficient P is decoded and supplied by the sub information decoding unit 34.
[0019]
Returning to FIG. 1, the decompressed restored waveform data output from the decoder 35 is supplied to the interpolation unit 36. In this case, the four restored waveform samples D1 to D4 cached in the waveform data cache unit 74 in the decoder 35 are supplied to the interpolation unit 36. Based on the fractional part of the sample address from the address generation unit 32, the interpolation unit 36 interpolates the supplied four restored waveform samples D1 to D4, for example, to generate an interpolation sample. The volume of the interpolated sample output from the interpolation unit 36 is controlled by the volume EG unit 37 based on the volume EG parameter set in the control register 20 when the note is turned on, and is supplied to the mixer 38. In the mixer 38, waveform samples in all sound generation channels are accumulated, and an effect is given and output at every reproduction timing. The output from the mixer 38 is supplied to a digital-analog converter (DAC) 39 to be converted into an analog signal and emitted from the sound system 40.
[0020]
As described above, in the musical sound generating apparatus 1 according to the embodiment of the present invention, the waveform data is stored in the waveform storage area 12a in the format shown in FIG. 2, and the sound source unit 30 reproduces the residual code. Therefore, by sequentially reading out the small frames constituting the frame, the residual code samples necessary for the reproduction of each frame are sequentially taken out, and at the same time, sub information for expanding the residual code of the next frame is provided. It comes to be taken out. That is, the sound source unit 30 does not require a dedicated circuit for taking out the sub information in a timely manner, and the sub information of each subframe and the number of bits of the residual code are fixed. A configuration for extracting information and a residual code can also be made extremely simple.
[0021]
Next, FIG. 4 and FIG. 5 show the configuration of the waveform compression apparatus of the present invention that performs compression processing of waveform data using linear prediction.
The waveform compression apparatus according to the present invention includes an initial SF calculation unit 50 shown in FIG. 4 to be described later, a coder 60 and a frame filling unit 67 shown in FIG. That is, in the waveform compression apparatus, the original waveform data is compressed to the number of bits k, and is framed and output. At this time, the scale factor search is performed in the waveform compression apparatus. In the search for the scale factor, the coder 60 sequentially supplies a plurality of samples of the original waveform data in the section of one frame to the quantizing unit 62 while changing the value of the scale factor, and the scale factor set at that time is used. A search is made for a scale factor value at which each level of the obtained residual sample dn of one frame hardly exceeds the allowable quantization error level. In this case, the scale factor is gradually increased to search for a suitable scale factor. However, since the scale factor is repeatedly set, the search time becomes longer. Thus, the initial scale factor corresponding to the nature of the original waveform data is calculated first so that the scale factor search time is shortened. The configuration for calculating the initial scale factor is the initial SF calculation unit 50 shown in FIG.
[0022]
The initial SF calculation unit 50 shown in FIG. 4 includes a sample (hereinafter referred to as “original waveform sample”) S of original waveform data selected from the waveform storage area 12a. _n Is supplied for one frame. Then, in the prediction coefficient calculation unit 51, an original waveform sample S for one frame. _n The linear prediction coefficient P is calculated based on the above, and the calculated linear prediction coefficient P is supplied to the linear prediction unit 52. The linear prediction unit 52 multiplies and adds the linear prediction coefficient P corresponding to the order to, for example, four samples from the present to the past, which are delayed by the delay means, and adds them. _n-1 Has been generated. This linear prediction sample ◇ S _n-1 Is supplied to the subtractor 53 and the original waveform sample S _n The subtractor 53 outputs a residual sample dn. The residual sample dn is supplied to the peak detection and scale factor calculation unit 54 to which the bit number k of the residual code sample is given, and the peak of the residual sample dn is obtained. That is, the original waveform sample S for one frame _n Are sequentially supplied, and when the residual sample dn for one frame is supplied to the peak detection and scale factor calculation unit 54, the peak value of the residual sample dn is detected, and the residual of the detected peak value is detected. An ideal scale factor SFP that is large enough to normalize and quantize the difference sample dn without overflowing with a given number of bits k is calculated. The constant D is subtracted from the ideal scale factor SFP thus calculated, and the initial scale factor (initial SF) is output from the initial SF calculator 50.
[0023]
The initial scale factor (initial SF) and the prediction coefficient P calculated in this way are set in the prediction coefficient & scale factor generation unit 63 of the coder 60 shown in FIG. 5 in the waveform compression apparatus. And the original waveform sample S before compression processing _n Is input to the subtractor 61, and this original waveform sample S _n Linear prediction sample ◇ S output from the linear prediction unit 64 _n-1 Are subtracted and the residual sample dn (= S _n -◇ S _n-1 ) Is output. The linear prediction coefficient P set in the prediction coefficient & scale factor generation unit 63 is supplied to the linear prediction unit 64, and the initial scale factor (initial SF) is supplied as the scale factor SF to the quantization unit 62 and the inverse quantization unit 66. The Here, the residual sample dn output from the subtractor 61 is normalized by the quantization unit 62 based on the supplied scale factor SF, and then is quantized and scaled to the set number of bits k. , A residual code sample Ln is generated.
[0024]
The residual code sample Ln is output as compressed waveform data from the coder 60 and supplied to the frame stuffing unit 67, where a frame consisting of small frames is generated and output. Further, the residual code sample Ln is supplied to the inverse quantization unit 66, and is subjected to inverse normalization and inverse quantization to output a decoded waveform sample qn obtained by decoding the residual code sample Ln. The decoded waveform sample qn is added to the linear prediction sample ◇ S output from the linear prediction unit 64 in the adder 65. _n-1 Is added to restore waveform sample ◇ X _n Is restored and supplied to the linear prediction unit 64. The linear prediction unit 64 performs fourth-order linear prediction by multiplying and adding, for example, the four restored waveform samples from the present to the past, which are delayed by the delay means, with the respective linear prediction coefficients P. The next original waveform sample S _n Next linear prediction sample for compressing ◇ S _n Is generated. The frame stuffing portion 67 includes a residual code sample L _n In addition, the linear prediction coefficient P and the scale factor SF and other information are supplied. Then, a linear prediction coefficient P and a scale factor SF and other information are stored in the sub information area, and a residual code sample is stored in the data area, and a small frame is formed. As shown in FIG. 1, one frame is configured.
[0025]
A method for obtaining a residual code sample without feedback as shown in FIG. 4 is defined here as AbS (Analysys by Synthesys), and the residual code sample is obtained by feedback as shown in FIG. Define the method as AbS.
[0026]
Next, FIG. 6 shows a flowchart of the SF search process for one frame executed in the waveform compression apparatus of the present invention.
In the SF search process for one frame shown in FIG. 6, when the search for the scale factor is started, the sample of waveform data (original waveform sample) selected from the waveform storage area 12a is extracted for one frame in step S10. Next, the sample for one frame taken out in step S11 is supplied to the initial SF calculation unit 50 shown in FIG. 4, and the linear prediction coefficient P is calculated. In step S12, there is no AbS, that is, the original waveform samples for one frame are sequentially supplied to the initial SF calculation unit 50 shown in FIG. 4, and the residual sample dn for one frame is supplied by the peak detection and scale factor calculation unit 54. Is required. This residual sample dn corresponds to the error of the prediction signal, and the maximum scale of the residual sample dn for one frame is detected in step S13, and an ideal scale that can make the detected maximum amplitude equal to or smaller than a predetermined size. A factor SFP is determined. In step S14, an initial scale factor (initial SF) is calculated by subtracting a constant D from the ideal scale factor SFP. Here, the reason for subtracting the constant D is to make sure that the initial scale factor (initial SF) is smaller than the ideal scale factor SFP. However, since the scale factor is expressed in logarithm, the ideal scale factor SFP is divided by the constant D in step S14.
[0027]
Next, AbS is present based on the linear prediction coefficient P and the initial scale factor (initial SF) calculated in step S15, that is, the residual code sample Ln is obtained in the coder 60 shown in FIG. Then, a noise level that is an error from the original waveform is obtained in step S16 based on the residual code sample Ln obtained in step S15. In this case, the linear prediction coefficient P calculated in step S11 is set in the linear prediction unit 64 in the coder 60, and the initial scale factor (initial SF) calculated in step S14 is the quantization factor 62 as the scale factor SF. And the inverse quantization unit 66 is set. When the noise level is obtained in step S16, it is determined in step S17 whether or not the scale factor SF set in the coder 60 has reached the maximum value. If it is determined that the scale factor SF has not reached the maximum value, the process proceeds to step S18.
[0028]
In step S18, it is determined whether or not the noise level obtained in step S16 is smaller than the threshold value. If it is determined that the noise level is larger than the threshold value, the scale factor is not an appropriate size, so that step S19 is performed. Branch to In step S19, the scale factor and the threshold value are changed to a value larger by one step, and the process returns to step S15. The process of step S15 to step S18 is repeated by reflecting this new scale factor value. Then, when it is determined in step S18 that the noise level is smaller than the threshold value, the SF search process for one frame ends. As described above, when the noise level in one frame is less than the threshold, it is determined that the scale factor is suitable. However, if the noise level instantaneously exceeds the threshold, this is ignored and the noise level is ignored. It may be determined that the scale factor is suitable when the average value is less than the threshold value. In addition, since the threshold value is changed according to the scale factor, the noise level can be allowed according to the scale factor.
[0029]
If the scale factor SF has reached the maximum value in step S17, the SF search process for one frame is finished as it is. The framed waveform data composed of the residual code samples obtained by compressing the original waveform data using the scale factor SF searched in this way is stored in the storage means. In step S18, it is determined that the noise level is smaller than the threshold value in the following case.
(1) When the effective level in one frame period of the residual sample dn (waveform) is smaller than the allowable quantization error instead of the peak value of the residual sample dn.
(2) A case where a predetermined ratio (for example, 90%) or more of the residual samples dn in one frame section is included within the allowable quantization error.
(3) When the integrated value over one frame section of the amount (or its square) that each residual sample dn exceeds the allowable quantization error is smaller than a predetermined reference value.
[0030]
In the SF search process for one frame executed in the waveform compression apparatus of the present invention, as described above, when calculating the initial scale factor, the sample for one frame of the waveform data is first supplied to the initial SF calculation unit 50. Thus, the prediction coefficient P and the initial scale factor are calculated. Then, the calculated prediction coefficient P and the initial scale factor are set in the coder 60, and the threshold at which the noise level, which is the error between the original waveform sample and the linear prediction sample, is changed to the allowable quantization error level by changing the scale factor. The original waveform sample for one frame is repeatedly supplied to the coder 60 until it becomes smaller. As described above, in the waveform compression apparatus, compression processing suitable for the waveform data is performed by repeatedly reading and processing the sample for one frame of the waveform data. Thus, since the initial scale factor is calculated in the initial SF calculation unit 50, the number of times the coder 60 searches for the scale factor can be significantly reduced. As a result, the compression process can be performed at high speed.
[0031]
In the above description, the number of bits of the residual code sample is fixed in the waveform data for each song, but the number of bits of the residual code sample is variable for each frame, and the number of bits is specified by the sub information. You may make it do. In this case, information on the number of bits in the sub information is supplied from the sub information decoding unit 34 to the address generation unit 32 and the residual code cache unit 33.
In the waveform compression apparatus of the present invention described above, compression / expansion is performed by linear prediction, but compression / expansion may be performed by a method other than linear prediction such as adaptive difference.
[0032]
【The invention's effect】
As described above, according to the present invention, when linear prediction compression is performed on waveform data in a predetermined section, the scale factor that is set to a constant value in the predetermined section indicates that each level of the residual data has an allowable quantization error level. It is assumed that the value will not exceed. Thereby, even if pulse noise enters in a predetermined section, it is possible to prevent the quantization noise in the predetermined section from increasing due to the influence. Further, since the allowable quantization error level is changed according to the value of the scale factor, an appropriate allowable quantization error level can be set. Furthermore, since the initial scale factor calculation means is provided to calculate the initial scale factor corresponding to the waveform data, the time for searching for the scale factor can be shortened.
[Brief description of the drawings]
FIG. 1 is a block diagram of a musical sound generating apparatus including storage means in which waveform data compressed by a waveform compression apparatus according to the present invention is stored in units of frames.
FIG. 2 is a diagram showing a data structure of waveform data compressed by the waveform compression apparatus according to the present invention.
FIG. 3 is a diagram showing a detailed configuration of a decoder included in the musical sound generating device shown in FIG. 1;
FIG. 4 is a diagram illustrating a configuration of an initial SF calculation unit included in the waveform compression device according to the embodiment of the present invention.
FIG. 5 is a diagram showing a configuration of a coder included in the waveform compression device according to the embodiment of the present invention.
FIG. 6 is a flowchart of SF search processing for one frame executed by the waveform compression apparatus according to the embodiment of the present invention.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 Musical sound production | generation apparatus, 10 CPU, 11 ROM, 12 RAM, 12a Waveform storage area, 13 Operator, 14 Display, 15 Communication I / O, 16 Bus line, 20 Control register, 30 Sound source part, 31 Frame reading part, 32 Address generator, 33 Residual code cache, 34 Sub information decoder, 35 Decoder, 36 Interpolator, 37 Volume EG, 38 Mixer, 39 DAC, 40 Sound system, 50 Initial SF calculator, 51 Prediction coefficient calculation Unit, 52 linear prediction unit, 53 subtractor, 54 peak detection and scale factor calculation unit, 60 coder, 61 subtractor, 62 quantization unit, 63 prediction coefficient & scale factor generation unit, 64 linear prediction unit, 65 adder, 66 Inverse quantization unit, 67 Frame filling unit, 71 Inverse quantization unit, 72 Adder, 73 Linear prediction operation unit, 74 waves Data cache unit

Claims (4)

Supply means for supplying waveform data to be compressed;
Prediction coefficient determination means for determining a prediction coefficient for linear prediction compression from a sample of waveform data supplied from the supply means;
A residual sample, which is a residual between a sample of waveform data supplied from the supply means and a prediction sample, is normalized based on a scale factor that provides quantization accuracy, and the normalized residual sample is encoded. Quantizing means for outputting as a residual code sample;
Dequantizing means for decoding the residual code sample output from the quantizing means, denormalizing based on the scale factor, and outputting as a decoded residual sample;
The decoded sample generated by adding the decoded residual sample and the prediction sample output from the inverse quantization means is linearly predicted based on the prediction coefficient, and used for linear prediction compression of the next waveform data. Linear prediction means for outputting a prediction sample,
The scale factor is set to a constant value in the predetermined section when the waveform data sample in the predetermined section is subjected to linear predictive compression, and the initial scale factor corresponding to the waveform data is an initial scale factor calculating means. Is calculated by
Using the initial scale factor calculated by the initial scale factor calculating means as an initial value, sequentially supplying a plurality of waveform data samples in the predetermined section to the quantizing means while changing the scale factor value. The scale factor of which the noise level based on the sample of the residual code for each of the scale factors obtained by the above is searched for is smaller than a threshold value, and the scale factor is used for compressing the waveform data in a predetermined section A waveform compression apparatus characterized by being determined as a factor. 2. The waveform compression apparatus according to claim 1, wherein the threshold value is changed in accordance with the set scale factor. The initial scale factor calculation means, an initial prediction coefficient determining means for determining an initial prediction coefficients for linear prediction compressed from the waveform data supplied from said supply means, said initial estimate determined by the initial prediction coefficient determining means An initial linear prediction unit that outputs an initial prediction sample by performing linear prediction based on a coefficient; and a residual obtained by quantizing a residual between a sample of waveform data in a predetermined section supplied from the supply unit and the initial prediction sample. 2. The waveform compression according to claim 1, further comprising initial scale factor determination means for sequentially obtaining differential code samples and calculating the initial scale factor in accordance with a peak value of the residual code samples. apparatus.   The initial scale factor calculated by the initial scale factor determining means is a scale factor having a magnitude obtained by subtracting a predetermined value from a scale factor corresponding to a peak value in the residual data. The waveform compression apparatus according to claim 3.

Images