JP2002287757A - Sound data transfer method and sound data transfer apparatus, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method and a device for sound data transfer and a program, which have a high use efficiency of a RAM and can generate a musical sound signal in real time, in spite of using a hard disk or the like. SOLUTION: When program change of playing data is supplied, candidates of sound data in an attack part corresponding to this program change are forecasted and read out (step S30). When note-on is next supplied and waveform synthesis in the attack part is started, candidates of sound data in a following body part are forecasted and read out (S31). Sound data which will be used next is successively forecasted and is read into a cache memory in advance in this manner when sound data concerned with present waveform synthesis is settled. Consequently, waveform synthesis can be quickly executed when sound data is actually settled.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a sound data transfer method for generating a waveform of a musical tone, voice, or any other sound based on reading of waveform data from a relatively low-speed storage medium such as a hard disk, and the like. The present invention relates to a transfer device and a program, and more particularly to a transfer device and a program for generating a waveform that faithfully expresses a timbre change caused by various playing techniques or articulations performed by a player. The present invention is applicable to not only electronic musical instruments but also apparatuses, apparatuses, or methods in all fields having a function of generating musical sounds, voices, or other arbitrary sounds, such as automatic performance devices, computers, electronic game devices, and other multimedia devices. It can be widely applied. In this specification, when referring to a musical sound waveform,
The waveform is not limited to a musical sound waveform, but may be a sound or any other sound waveform.

[0002]

2. Description of the Related Art PCM (pulse code modulation), DPCM (differential PCM) or ADPC
M (adaptive differential PCM) or the like, waveform data (that is, waveform sample data) encoded by an arbitrary encoding method is stored and read out in correspondence with a desired music pitch to form a musical tone waveform. The so-called "waveform memory readout" technique is already known, and various types of "waveform memory readout technique" are known. Most of the conventionally known "waveform memory reading method" techniques generate a waveform of one sound from the start to the end of sounding. As an example, there is a method of storing waveform data of the entire waveform of one sound from the start to the end of sounding. As another example, there is a method of storing waveform data of all the waveforms of an attack portion having a complicated change and the like, and storing a predetermined loop waveform for a sustain portion having little change. In this specification,
The “loop waveform” is used to mean a waveform that is repeatedly read (loop read).

[0003] As means for storing waveform data, ROM, RAM, hard disk, CD-ROM and the like are known. Hard disks, CD-ROMs, and the like are inexpensive per unit storage capacity and are suitable for large-capacity data storage. However, since the access time of a hard disk or the like is slow and not constant, it is impossible to immediately read necessary waveform data at the timing of outputting a tone signal. For this reason, various techniques have been proposed as follows. First, Japanese Patent Laying-Open No. 6-308964 discloses a technique in which head portions of a plurality of waveform data stored in a hard disk or the like are transferred to a RAM in advance. That is, when the sounding instruction is supplied, the reading of the subsequent portion of the waveform data from the hard disk or the like is started, and the leading portion stored in the RAM in advance is reproduced. Then, after the reproduction of the head portion is completed, the waveform data of the subsequent portion is reproduced.

Japanese Patent Application Laid-Open No. 63-181188 discloses a technique in which, when waveform data to be sequentially emitted is defined in advance as sequence data, each waveform data is reproduced while sequentially reading the data. In this technique, the time at which the reading of each waveform data is started is determined in advance in accordance with the sequence timing such that the time at which reading of each waveform data is started is earlier than the sounding timing.

[0005]

However, Japanese Patent Laid-Open No. 6-3 / 1994
According to the technique disclosed in Japanese Patent Application Laid-Open No. 08964, there is a problem that the use efficiency of the RAM is reduced because the leading portion of all the waveform data must be transferred to the RAM in advance. The technique disclosed in Japanese Patent Application Laid-Open No. 63-181188 is based on the premise that a tone signal is generated in non-real time, and cannot be applied when sequence data is supplied in real time. The present invention has been made in view of the above circumstances, and has a high use efficiency of a RAM, and a sound data transfer method, a sound data transfer apparatus, and a program capable of generating a tone signal in real time while using a hard disk or the like. It is intended to provide.

[0006]

Means for Solving the Problems In order to solve the above problems, the present invention is characterized by having the following constitution. Note that the contents in parentheses are examples. In the configuration according to the first aspect, a low-speed storage device (hard disk 109) for storing musical sound waveform sound data (vector data) and a high-speed storage device (cache memory 44) for caching the sound data are used. A sound data transfer method for transferring a part of sound data stored in the low-speed storage device to the high-speed storage device, wherein a tone color designation (program change) related to the sound data to be transferred to the high-speed storage device is provided. )
, A sound data predicting step of predicting a sound data candidate to be specified later based on the received sound color specification (step S30), and storing the predicted sound data candidate at the low speed. Transferring from the device to the high-speed storage device (step S43). In the sound data transfer method according to the second aspect, the sound data stored in the low-speed storage device may be a first sound data corresponding to a sound generation start unit (attack unit). And second sound data (vector data of the body part) corresponding to a portion other than the sound generation start portion, and the sound data predicted according to the tone color designation is the first sound data. A first sound data designation receiving step of receiving designation of sound data, and a second sound data prediction step of predicting a second sound data candidate to be designated later based on the designation of the first sound data (step S30) When,
Transferring the predicted second sound data from the low-speed storage device to the high-speed storage device (step S43). Further, in the configuration according to claim 3, in the sound data transfer method according to claim 2, the sound data stored in the low-speed storage device is used after the first and second sound data. A second sound data designation receiving step of further including three sound data (a release portion or a joint portion) for receiving designation of any of the second sound data, and later designated based on the designation of the second sound data. A third sound data prediction step of predicting third sound data candidates (step S33); and a step of transferring the predicted third sound data from the low-speed storage device to the high-speed storage device (step S43). It is characterized by the following. Further, in the configuration of claim 4, claim 1
4. In the sound data transfer method according to any one of the above-described items, when the designation of any one of the sound data is received, the release of the sound data candidates that are stored in the high-speed storage device and that are not predicted is released. The method further comprises the step of adding identification information indicating that the process is possible (setting the member dwStatus to 'USED'). According to a fifth aspect of the present invention, in the sound data transfer method according to any one of the first to fourth aspects, the basic sound data (alternate page) is provided for each tone before the tone designation receiving step. ) Is transferred to the high-speed storage device in advance, and if the sound data designated in the first, second, or third sound data designation receiving step has not been transferred to the high-speed storage device, The basic sound data is used in place of the designated sound data. According to a sixth aspect of the present invention, the method according to any one of the first to fifth aspects is performed. According to a seventh aspect of the present invention, the method according to any one of the first to fifth aspects is performed.

[0007]

BEST MODE FOR CARRYING OUT THE INVENTION 1. Hardware Configuration of Embodiment Hereinafter, an embodiment of the present invention will be described in detail with reference to the accompanying drawings. FIG. 1 is a block diagram showing an example of a hardware configuration of a waveform generation device according to the present invention. The hardware configuration example shown here is configured using a computer, and in the waveform generation processing, the computer executes a predetermined program (software) that implements the waveform generation processing according to the present invention. It is implemented by. Of course, this waveform generation processing is not limited to the form of computer software, but can also be carried out in the form of a microprogram processed by a DSP (digital signal processor). The present invention may be implemented in the form of a dedicated hardware device including a discrete circuit or an integrated circuit or a large-scale integrated circuit. Further, the waveform generating device may take any product application form such as an electronic musical instrument, a karaoke device, an electronic game device, other multimedia devices, or a personal computer.

In the example of the hardware configuration shown in FIG. 1, a CPU 10 as a main control unit of a computer is used.
1, a read-only memory (ROM) 10 via a bus line BL (data or address bus, etc.).
2. Random access memory (RAM) 103, panel switch 104, panel display 105, drive 10
6. The waveform acquisition unit 107, the waveform output unit 108, the hard disk 109, and the communication interface 111 are connected to each other. The CPU 101 executes processes such as “creation of a waveform database” and “sound synthesis (software sound source) based on the created database”, which will be described later, based on a predetermined program. These programs are supplied from a network via the communication interface 111 or an external storage medium 106A such as a CD or MO mounted on the drive 106 and stored in the hard disk 109. Then, at the time of execution, R
Loaded to AM103. Alternatively, the program may be stored in the ROM 102. ROM 102 is
It stores various programs and various data executed or referred to by the CPU 101. RAM10
Numeral 3 is used as a working memory for temporarily storing various information relating to the performance and various data generated when the CPU 101 executes the program, or as a memory for storing the program currently being executed and data related thereto. A predetermined address area of the RAM 103 is assigned to each function and used as a register, a flag, a table, a memory, and the like. Panel switch 10
Reference numeral 4 denotes various controls for editing a musical tone, editing sampled waveform data, inputting various information, and the like. For example, a numeric keypad for inputting numerical data, a keyboard for inputting character data, a panel switch, or the like. In addition, various controls for selecting, setting, and controlling a pitch, a tone, an effect, and the like may be included. The panel display 105 is a display such as a liquid crystal display panel (LCD) or a CRT that displays various information input by the panel switch 104, sampled waveform data, and the like.

The waveform acquisition unit 107 has a built-in A / D converter, converts (samples) an analog musical tone signal input from an external waveform (for example, input from a microphone or the like) to digital data, and stores the converted digital tone signal in the RAM 103 or the hard disk 109. The digital waveform data is converted to original waveform data (waveform data that is the material of the waveform data to be generated)
It is taken as. In the “waveform database creation” process executed by the CPU 101, a “waveform database” according to the present invention is created based on the fetched original waveform data. In the "sound synthesis based on database" process executed by the CPU 101, waveform data of an arbitrary tone signal corresponding to performance information is generated using the "waveform database". Of course, simultaneous generation of a plurality of tone signals is possible. The waveform data of the generated tone signal is supplied to the waveform output unit 108 via the bus line BL, and is appropriately buffer-stored. The waveform output unit 108 outputs the buffered waveform data according to a predetermined output sampling frequency,
A converted and sent to the sound system 108A. Thus, the tone signal output from the waveform output unit 108 is generated via the sound system 108A. The hard disk 109 stores a plurality of types of data related to performance, such as waveform data, data for synthesizing a waveform according to a playing style (data of a playing style table, a code book, etc., which will be described later), and tone data including various tone parameters. The CPU 101 stores data relating to control of various programs executed by the CPU 101 and the like.

The drive 106 includes performance data such as waveform data and data for synthesizing a waveform corresponding to a playing style (various data such as a playing style table and a code book to be described later), and timbre data including various timbre parameters. It drives a removable disk (external storage medium 106A) for storing a plurality of types of data related to the above, and storing data related to control of various programs executed by the CPU 101 and the like. The external storage medium 1 driven by the drive 106
06A is a compact disk (CD-ROM / CD-RA) in addition to a floppy (registered trademark) disk (FD).
M), magneto-optical disk (MO), or DVD (Di)
The storage medium may be various types of removable storage media such as a digital versatile disk. The external storage medium 106A storing the control program is set in the drive 106, and its contents (control program)
May be directly loaded into the RAM 103 without dropping it on the hard disk 109. The external storage medium 106
The method of providing the control program using A or via a network is convenient because the control program can be easily added or upgraded.

The communication interface 111 is, for example, LA
N, the Internet, a telephone line, or other communication network (not shown), and connected to a server computer or the like (not shown) via the communication network. Alternatively, it is for taking performance information and the like into the waveform generating device side. That is, when the control program and various data are not stored in the ROM 102 or the hard disk 109, the control program and various data are used for downloading from the server computer. The waveform generation device serving as a client transmits a control program and a command for requesting download of various data to the server computer via the communication interface 111. The server computer receives the command and stores the requested control program, data, and the like in the hard disk 109 via the communication interface 111, thereby completing the download. Furthermore, MI
Of course, a MIDI interface may be included to receive MIDI performance information. Needless to say, a music performance keyboard or performance operation device may be connected to the bus line BL to supply performance information by real-time performance. Of course, the performance information may be supplied using the external storage medium 106A storing the performance information of the desired music piece.

2. Waveform database creation processing 2.1. FIG. 2 is a flowchart showing one embodiment of a “waveform database creation process” executed in the above-described waveform generation device. This process is a process for creating vector data using waveforms of performance sounds performed in various playing styles (or articulations) as materials in order to support various playing styles (or articulations). In step S1, a database for storing a rendition style table and a code book to be described later is prepared. As the medium serving as the database, for example, the hard disk 109 is used. Then, waveform data of various performance modes of various natural musical instruments are collected (step S2). That is, various actual performance sounds of various natural musical instruments are captured from an external waveform input (for example, a microphone or the like) via the waveform capturing unit 107, and the waveform data (original waveform data) of the performance sounds is stored in the hard disk 109. It is stored in a predetermined area. The waveform data of the performance sound taken in at this time may be the waveform data of the entire performance, or only a certain phrase, a single sound, or only a part of the performance data having a characteristic feature such as an attack portion and a release portion. Is also good.

Next, the obtained waveform data of the performance sound in various performance modes unique to the natural musical instrument is divided into characteristic portions, and tuning and file naming are performed (step S3). That is, the acquired original waveform data is converted into a partial waveform (for example, an attack portion waveform, a body portion waveform, a release portion waveform,
(For example, joint part waveforms)) (separation), determine the pitch of the separated waveform data of one or more cycles (tuning), and further assign a file name to the separated waveform data. Assign (file naming). However, when waveform data of a part of a performance such as an attack portion and a release portion is captured, such separation (separation) of the waveform can be omitted. Next, component separation is performed by frequency analysis (step S4). That is, a part of the waveform data separated and generated in step S3 is subjected to FFT (Fast Fourier Transform) analysis and separated into a plurality of components (in this embodiment, separated into a harmonic component and a non-harmonic component). Performs feature extraction for each element of waveform, pitch, and amplitude from harmonic components, non-harmonic components, etc., that is, feature separation (however, when separated into harmonic components and non-harmonic components, non-harmonic components have no pitch) Therefore, it is not necessary to perform pitch separation for nonharmonic components). For example, "waveform" (Timbre)
The elements are obtained by extracting features only in a waveform shape in which pitch and amplitude are normalized. The “pitch” (Pitch) element is obtained by extracting pitch fluctuation characteristics with respect to a reference pitch. The “amplitude” (Amplitude) element is obtained by extracting an amplitude envelope characteristic.

In step S5, vector data is created. That is, each separated component (harmonic component,
Extract multiple sample values for each element of the waveform (Timbre), pitch (Pitch), and amplitude (Amplitude) of the non-harmonic component etc. in a dispersed manner or continuously as necessary, and extract the sample value sequence. Then, different vector IDs (identification information) are given and stored in the code book together with the data of the time position of the sample value (hereinafter, such sample data is referred to as vector data). In this embodiment, the vector data and the pitch (Pi) of the waveform (Timbre) element of the harmonic component
The vector data of the (tch) element, the vector data of the amplitude (Amplitude) element, the vector data of the non-harmonic component waveform (Timbre) element, and the vector data of the amplitude (Amplitude) element are respectively created. Thus, the vector data for each of these component elements is data that can change as the time axis progresses. Next, data of the performance style module (details will be described later) is created, and the performance style module is stored in the performance style table. The rendition style module and vector data thus created are written in the rendition style table and codebook in the database (step S6), and the data is stored in the database. As described above, the vector data is not the captured original waveform data as it is, but is data obtained by separating a waveform representing the shape of the captured original waveform for each element. Is the unit of data. in this way,
In the code book, a part of the extracted waveform data representing the change of the waveform shape is stored in a compressed form. On the other hand, the rendition style table stores various rendition style data (that is, various kinds of data necessary for returning vector data stored in a compressed form to waveform data of an original waveform shape, and vector data stored in a code book). ID for specifying data
Data) is stored (details will be described later).

At the time of the above-described feature separation (see step S4), feature extraction is performed using time as an element in addition to amplitude, pitch, and waveform elements. Data)). For this time element, the time length of the original waveform data in the time section of some of the separated and generated waveform data is used as it is. Therefore, if the original time length (variable value) of the time section is indicated by the ratio “1”, it is not necessary to analyze and measure this time length during the “waveform database creation processing”. In that case,
Since the data of the time element (ie, “time vector data”) has the same value “1” in any time section, it is not necessary to store this in the codebook. Of course, not limited to this,
A modified example in which the measurement is performed and this is stored in the code book as “time vector data” is also possible.

Then, it is determined whether the database has been sufficiently created (step S7). That is, it is determined whether or not the original waveform data of the performance sounds of various natural musical instruments obtained from the external waveform input in various performance modes has been sufficiently collected to sufficiently obtain the data and vector data of various performance style modules. judge. This determination is not limited to the automatic determination, and may be performed in accordance with a processing continuation availability instruction based on a user's switch input operation. If it is determined that the collection of the original waveform data and the creation of the vector data based on the original data have been sufficiently performed (Step S7)
YES), the process ends. Subsequently, when the original waveform data is collected and the vector data is created based on the original data (NO in step S7), the process returns to step S2, and the above-described processes (steps S2 to S7) are repeatedly executed. Judgment of "whether or not vector data has been sufficiently created" (step S7)
May be performed by "creating a musical tone by actually using the created vector data". That is,
In step S7, "vector data has been sufficiently created"
After deciding (YES in step S7) and exiting the flow shown in FIG. 2, "playing a tone using the created vector data was not satisfactory. And add vector data ". That is, the process of “creating vector data and adding it to the database” is performed as needed. In the above-mentioned "waveform database creation processing", a rendition style module may be arbitrarily added / deleted, or data of the rendition style module may be edited.

2.2. Data Configuration of Performance Style Module Here, the data of the performance style module will be specifically described. The rendition style module is stored in a rendition style table configured as a database on the hard disk 109,
One rendition style module can be designated by a combination of “reproduction style ID” and “reproduction style parameter”. The “playing style ID” includes instrument information and module part names therein. For example, “performance style ID” is defined as follows. For example, one “playing style ID” is 32
If it is represented by a sequence of bits (0th to 31st bits), the musical instrument information is represented using 6 bits. For example, if the 6-bit string is “000000”, it indicates AltoSax, and “00100
If it is "0", it is instrument information indicating Violin (violin). The musical instrument information may use, for example, the upper 3 bit strings of the 6-bit string for the large classification of the instrument type, and the lower 3 bit strings for the small classification of the instrument type. Also, 3
The module part name is expressed using another 6 bits of the 2-bit string. For example, if the 6-bit string is “00000”
If "0", NormalAttack, if "000001",
BendAttack, GraceNoteAttac if "000010"
k, if “001000”, NormalShortBody, “00
1001 ”, VibBody,“ 001010 ”, NormalLongBody,“ 010000 ”, NormalRe
lease, if "011000", NormalJoint, "01
“1001” is a module part name indicating GraceNoteJoint. Of course, it is needless to say that the configuration is not limited to the above.

As described above, each performance style module is specified by a combination of the above-mentioned "performance style ID" and "performance style parameter". That is, a predetermined performance style module is specified according to the “performance style ID”, and the content thereof is variably set according to the “performance style parameter”. The “reproduction style parameters” are parameters for characterizing or controlling the waveform data corresponding to the rendition style module, and a predetermined type of “reproduction style parameter” exists for each rendition style module. For example, in the case of the AltoSax [NormalAttack] module, types of playing style parameters such as an absolute pitch immediately after Attack and a volume immediately after Attack may be given, and an AltoSax [B
endUpAttack] module, the absolute pitch at the end of BendUpAttack, the initial value of Bend depth at BendUpAttack, Bend
A performance style parameter such as the time from the start (note-on timing) to the end of the UpAttack, the volume immediately after the Attack, or the temporal expansion and contraction of the default curve during the BendUpAttack may be given. Also, AltoSax [NormalShortBo
dy] module, the absolute pitch of the module, N
normalShortBody end time-start time, NormalShortBod
Types of performance parameters such as the dynamics at the start of y and the dynamics at the end of NormalShortBody may be given. Note that the performance style module does not necessarily have data (element data described later) corresponding to all possible values of the “performance style parameter”. In some cases, data corresponding to only some of the discrete values of the "reproduction parameter" may be stored. That is, for example, in the case of the AltoSax [NormalAttack] module, data corresponding to only some data may be stored instead of all values of the absolute pitch immediately after Attack or the volume immediately after Attack. As described above, by enabling the performance style module to be designated by the “performance style ID” and the “performance style parameter”, for example, in the case of AltoSax [NormalAttack], a plurality of data (element data described later) indicating the normal attack part of the alto saxophone is displayed. Data can be specified according to the desired playing style parameters from among them, and Violin [Ben
dAttack], it is possible to specify data corresponding to a desired playing style parameter from a plurality of data (element data described later) indicating a bent attack portion of the violin.

2.3. Data structure of the rendition style table In the rendition style table, for each rendition style module, data necessary for generating a waveform corresponding to the rendition style module, for example, vector data for each component element (waveform element, pitch element (pitch envelope)) , An amplitude element (amplitude envelope, etc.), a representative point value sequence (data indicating representative sample points for correction in a plurality of sample sequences) or vector data for each component element ( Information such as a start time position and an end time position of a waveform element, a pitch element (pitch envelope), and an amplitude element (amplitude envelope) is stored. That is, it stores various data necessary for reproducing a waveform of a normal shape from a waveform stored in the database in a compressed form called vector data (hereinafter, such data is also referred to as “element data”). Call). In the rendition style table, an example of specific data stored corresponding to one rendition style module is AltoSax [NormalAttac
The case of the [k] module will be described as follows. Data 1: Sample length of playing style module. Data 2: Note-on timing position. Data 3: Vector ID and representative point value sequence of the amplitude element of the harmonic component. Data 4: Vector ID of a pitch element (Pitch) of a harmonic component and a representative point value sequence. Data 5: Harmonic component waveform (Timbre) element vector I
D. Data 6: A vector ID of a non-harmonic component amplitude element and a representative point value sequence. Data 7: Vector ID of waveform (Timbre) element of nonharmonic component. Data 8: Start position of the lump of the harmonic component waveform (Timbre) element. Data 9: The end position of the block of the harmonic component waveform (Timbre) element (the start position of the loop portion of the harmonic component waveform (Timbre) element). Data 10: Starting position of the lump of the waveform element (Timbre) of the nonharmonic component. Data 11: the end position of the lump of the non-harmonic component waveform (Timbre) element (the start position of the loop part of the non-harmonic component waveform (Timbre) element). Data 12: end position of loop part of waveform (Timbre) element of nonharmonic component.

The data 1 to 12 will be described with reference to FIG. FIG. 3 is a diagram schematically illustrating an example of each component and an element constituting an actual waveform section corresponding to the rendition style module. Pitch) element,
Harmonic component waveform (Timbre) element, non-harmonic component amplitude (Am
3 shows an example of a (plitter) element and a waveform (Timbre) element of a non-harmonic component. It is to be noted that the numbers shown in the figure are attached so as to correspond to the numbers of the respective data. 1 is a sample length (waveform section length) of a waveform corresponding to the rendition style module. For example, it corresponds to the entire time length of the original waveform data on which the performance style module is based. Reference numeral 2 denotes a note-on timing position, which can be variably set at any time position of the performance style module.
In principle, the sound of the performance sound starts in accordance with the waveform from the position of the note-on timing.However, depending on the playing technique such as a bend attack, the rising start point of the waveform component may precede the note-on timing. is there.

Reference numeral 3 denotes a vector ID and a representative point value sequence for indicating the vector data of the amplitude element of the harmonic component stored in the codebook (in the figure, two points indicated by black squares are representative). Point). Reference numeral 4 denotes a vector ID and a representative point value sequence for indicating vector data of a pitch element of the harmonic component. Reference numeral 6 denotes a vector ID and a representative point value sequence for indicating vector data of an amplitude (Amplitude) element of a nonharmonic component. The representative point value sequence data is data for changing and controlling the vector data (composed of a plurality of sample sequences) indicated by the vector ID, and indicates (specifies) some representative sample points. By changing or correcting the time position (horizontal axis) and the level axis (vertical axis) of the identified representative sample point,
The other remaining sample points are also changed in conjunction, thereby changing the shape of the vector. For example, the data is data indicating a smaller number of dispersed samples than the number of samples. Of course, the data is not limited thereto, and the representative point value sequence data may be data at an intermediate position between the samples, or (A plurality of continuous samples). Further, data such as a difference and a multiplier may be used instead of the sample value itself. By moving the representative point on the horizontal axis and / or the vertical axis (time axis), the shape of each vector data can be changed. That is, the shape of the envelope waveform can be changed.

Reference numeral 5 denotes a vector ID for pointing the vector data of the waveform (Timbre) element of the harmonic component. 7
Is a vector ID for indicating vector data of a waveform (Timbre) element of a non-harmonic component. Numeral 8 is the start position of the block of the waveform of the waveform (Timbre) element of the harmonic component.
Reference numeral 9 denotes the end position of the block of the waveform of the harmonic component waveform (Timbre) element (or the start position of the loop portion of the waveform of the harmonic component waveform (Timbre) element). That is, a triangle starting from 8 indicates a non-loop waveform portion in which characteristic waveform shapes are continuously stored, and a subsequent rectangle starting from 9 indicates a loop waveform portion that can be repeatedly read. The non-loop waveform is a high-quality waveform having characteristics such as a rendition style (or articulation). The loop waveform is a unit waveform of a relatively monotonous sound portion consisting of a waveform for one cycle or an appropriate plurality of cycles. Numeral 10 is the start position of the block of the waveform of the non-harmonic component (Timbre) element. Numeral 11 denotes the end position of the lump of the waveform of the non-harmonic component (Timbre) element (or the start position of the loop part of the waveform of the non-harmonic component waveform (Timbre) element). Reference numeral 12 denotes the end position of the loop portion of the waveform of the non-harmonic component (Timbre) element. The above data 3 ~
Data 7 is data of identification information for indicating vector data stored in a code book for each component element. Data 2 and data 8 to data 12 represent original (before separation) waveforms from vector data. Time information data for assembling.

As described above, the data of the rendition style module is composed of the data for indicating the vector data and the data of the time information. By using the data of the rendition style module stored in such a rendition style table, it is possible to freely assemble the waveform using the waveform material (vector data) stored in the code book. That is, the rendition style module is data representing the behavior of a waveform generated according to the rendition style (or articulation). Note that the type and number of data of the performance style modules may be different for each performance style module. Further, other information may be provided in addition to the data described above. For example, it may have data for controlling the expansion / compression of the time axis of the waveform.

In the above-described example, in order to make the description easy to understand, one rendition style module uses all the elements (waveform, pitch, amplitude) of the harmonic component and all the elements (waveform, amplitude) of the nonharmonic component. Although the example provided is described, the present invention is not limited to this.
Amplitude). For example, the playing style module includes a harmonic component waveform (Timbre) element, a harmonic component pitch (Pitch) element, a harmonic component amplitude (Amplitude) element, a non-harmonic component waveform (Timbre) element, and a non-harmonic component amplitude (Amplitude). One of the elements
It may consist of three elements. This is preferable because the rendition style modules can be freely combined and used for each component.

As described above, the waveform data of the performance sounds of various natural musical instruments in various performance modes are not included in the entire waveform data, but some of the waveforms required for changing the waveform shape (for example, an attack portion waveform, Body part waveform, release part waveform,
Waveform data is stored in the hard disk 109 in a compressed form using a hierarchical compression method such as components, elements, and representative points. The required storage capacity of the hard disk 109 can be reduced.

3. Waveform Synthesis Process In the waveform generation device shown in FIG. 1, the synthesis of the waveform is performed by a computer executing a predetermined program (software) for realizing the waveform synthesis process according to the present invention. FIG. 4 shows an embodiment of a flowchart of a predetermined program for realizing the waveform synthesizing process ("musical tone synthesizing process based on database"). However, in the present embodiment, since various functions realized by the program are mutually independent, the operation will be described based on the functional block diagram shown in FIG. Also, the waveform synthesis processing is not limited to this type of program, and may be performed in the form of a dedicated hardware device. In this case, FIG. 5 is a block diagram in the case where the same waveform synthesis processing as in FIG. 4 is configured in the form of a dedicated hardware device. Therefore, hereinafter, the description will be mainly made with reference to FIG.
For, the corresponding steps are shown in parentheses.

3.1. Music Data Reproduction Unit 101A The music data reproduction unit 101A performs reproduction processing of music data with performance style symbols (step S11). First, the music data reproduction unit 101A receives music data with performance symbols (performance information). Musical symbols such as dynamic symbols (crescendo, decrescendo, etc.), tempo symbols (allegro, ritardand, etc.), slur symbols, tenuto symbols, accent symbols, etc., which do not directly become MIDI data, are attached to ordinary music scores. . Therefore, these symbols are converted into data as “performance symbols”, and MIDI music data including the “performance symbols” is “music data with performance symbols”.
The “rendering style symbol” includes a chart ID and a chart parameter. The chart ID is an ID indicating a music symbol described in the musical score, and the chart parameter is a chart I.
A parameter indicating the degree of the content of the music symbol indicated by D. For example, when the chart ID indicates “vibrato”, the speed or depth of the vibrato is given as a chart parameter, and when the chart ID indicates “crescendo”, the volume at the start of the crescendo and the end at the end of the crescendo. The volume, the length of time during which the volume changes, and the like are given as chart parameters.

3.2. Music score interpretation unit 101B The music score interpretation unit (player) 101B performs a music score interpretation process (step S12). More specifically, the MIDI data included in the music data and the above-described “reproduction style symbol” (chart ID and chart parameter) are stored in the rendition style specification information (reproduction style I).
D and performance parameter) and outputs it to the performance synthesis section (articulator) 101C together with the time information.
In general, even for the same musical symbol, the interpretation of the symbol differs depending on the performer, and a performance may be performed in a different play style (that is, a playing style or articulation) for each performer. Alternatively, a performance may be performed in a different performance method for each musician depending on the arrangement of notes and the like. Thus, the score interpreting unit 101B is a system in which knowledge for interpreting such symbols (music symbols, arrangement of musical notes, and the like) on the score is converted into an expert system.

The following are examples of criteria for interpreting symbols on a musical score in the musical score interpretation unit 101B. For example, vibrato can only be played with eighth notes or more. In staccato, the dynamics naturally increase. The note attenuation rate is determined by the degree of tenuto. Legato does not decay in one note. The eighth note vibrato speed is largely determined by the note value. The dynamics differ depending on the pitch. Furthermore, changes in dynamics due to the rise or fall of the pitch in one phrase, attenuation dynamics: db linear, change in note length according to tenuto or staccato, etc. There are various interpretation criteria, such as width and curve. The musical score interpretation unit 101B converts the musical score into performance data by interpreting the musical score according to such criteria. Further, the musical score interpretation unit 101B performs the above-described musical score interpretation processing in accordance with the player's designation from the user, that is, the user's designation of the performance (performance style). The score interpretation unit 101B interprets the score by changing the interpretation method of the score according to the player's designation. For example, different musical score interpretation methods corresponding to the plurality of players are stored in a database, and the musical score interpretation unit 101B selectively interprets the musical score by changing the musical score interpretation method according to the player's designation from the user.

The music data (performance information) may be configured to include data indicating the interpretation result of the musical score in advance. Needless to say, when the music data including the data as a result of the interpretation of the musical score is input, the above-described processing is not required. Further, the interpretation process of the score in the score interpretation unit 101B (step S12) may be performed fully automatically, or may be performed with an artificial input operation by the user as appropriate.

Here, the contents of the performance data created in the musical score interpretation section 101B will be described with reference to FIG. The performance data is composed of a header and a plurality of tracks, similarly to a normal SMF (standard MIDI format). Each track includes program change and event data (note-on, note-off, etc.). In the present embodiment, a module specifying unit for specifying an attack unit, a body unit, a joint unit, and a release unit. (Reproduction style designation information including a rendition style ID and a rendition style parameter) is included. These module designation units actually use undefined system exclusive messages, meta events, 14-bit control changes, and the like. Further, each track includes these event data or time difference data for specifying a time difference between the module specifying units.

If each track contains a program change that specifies a tone, that is, a program change that specifies a new database (reproduction style table + codebook), the musical score interpretation unit 101B predicts the program change. Waveform synthesis unit 101D as data
To supply. This is because the waveform synthesis unit 101D specifies a program change in order to narrow down the prediction range to some extent in order to predict the vector data to be used next and read it from the codebook (details will be described later).
Note that the prediction data includes various other data.
In the music score interpretation unit 101B, based on the music data,
"Data indicating subsequent performance style designation" is supplied to waveform synthesizing section 101D as prediction data. The program change is included as one of the prediction data.

3.3. The rendition style synthesizing section (articulator) 101C refers to the rendition style table based on the rendition style specification (reproduction style ID + reproduction style parameter) converted by the score interpretation section (player) 101B, and specifies the rendition style (reproduction style ID + reproduction style parameter). And a vector parameter related to the stream corresponding to the rendition style parameter.
1D (step S13). The data supplied to the waveform synthesizing unit 101D as a packet stream includes packet time information, a vector ID, a representative point value sequence, etc. for a pitch (Pitch) element and an amplitude (Amplitude) element, and data for a waveform (Timbre) element. Vector ID,
Time information and the like (details will be described later).

Next, the waveform synthesizing unit 101D extracts vector data from the codebook according to the packet stream, modifies the vector data according to the vector parameters, and synthesizes a waveform based on the deformed vector data (step S14). ). Then, a waveform generation process for another part is performed (step S15). Here, the other part means that the performance synthesis processing is not performed among the plurality of performance parts.
This is a performance part to which normal tone waveform synthesis processing is applied. For example, these other parts generate musical tones using a normal waveform memory sound source system. This “other part waveform generation processing” may be performed by a dedicated hardware sound source (an external sound source unit or a sound source card attachable to a computer). For simplicity of explanation, in this embodiment, it is assumed that a musical tone is generated in accordance with a playing style (or articulation) for only one part. Of course, the playing style may be reproduced in a plurality of parts.

FIG. 6 is a block diagram for explaining the flow of the rendition style synthesizing process in the rendition style synthesizing section 101C. Although FIG. 6 shows that the rendition style module and the code book are separately stored, actually both are stored in the database of the hard disk 109. The rendition style synthesizing unit 101C includes a musical score interpretation unit 101B.
A variety of packet streams to be supplied to the waveform synthesizing unit 101D are created based on the performance style designation (performance style ID + performance style parameter) and time information data. Performance synthesis section 101
The playing style module used for each tone in C is not fixed, and the user may add a new playing style module to the currently used playing style module or use a part of the used playing style module. Can be stopped. Also, the rendition style synthesizing unit 101C performs processing for creating correction information for correcting a deviation between the selected element data and the rendition style parameter value, and connection for smoothly connecting waveform characteristics of the preceding and following rendition style modules. Processing such as smoothing of the section is also performed (details will be described later). Note that data is normally given from the score interpretation unit 101B to the rendition style synthesizing unit 101C, but the present invention is not limited to this. The music score may be interpreted to prepare music data with performance style designation to which the performance style ID and the performance parameter are added, and the reproduced data may be supplied to the performance style synthesis unit 101C.

FIG. 7 is a flowchart showing one embodiment of the rendition style synthesizing process in detail. Performance synthesis section 101C
Selects a rendition style module from the rendition style table according to the rendition style ID and the rendition style parameter (step S21).
That is, the rendition style ID transmitted from the score interpretation unit 101B.
One playing style module is selected according to (instrument information + module part name) and playing style parameters. At this time, before interpreting the score, the score interpreting unit 101B checks the database to check in advance what kind of module parts are present in the rendition style table corresponding to the tone indicated by the instrument information, and exists. Designate a rendition style ID within the range of parts. When a non-existent part is designated, a rendition style ID having similar characteristics may be selected instead. Next, a plurality of element data are selected according to the specified rendition style ID and rendition style parameter (step S22). That is, by referring to the rendition style table based on the specified rendition style ID and rendition style parameter, the rendition style module is specified, and a plurality of element data corresponding to the rendition style parameter is selected from the module. At this time, if there is no element data that completely matches the rendition style parameter in the rendition style module, the element data corresponding to the rendition style parameter close to the value is selected.

Next, the time at each position in the element data is calculated according to the time information (step S23). That is,
Each element data is arranged at an absolute time position based on the time information. Specifically, based on the time information, a corresponding absolute time is calculated from the element data indicating each relative time position. Thus, the timing of each element data is determined (see FIG. 3). Then, the value of each element data is corrected according to the performance style parameter (step S24). That is, the deviation between the selected element data and the value of the rendition style parameter is corrected. For example, the Attack of the AltoSax [NormalAttack] module transmitted from the score interpretation unit 101B
When the volume immediately after the attack (reproduction style parameter) is “95” and the volume immediately after the Attack of the AltoSax [NormalAttack] module existing in the performance style table is “100”, the volume immediately after the Attack is set to “100” by the rendition style synthesis unit 101C. Alto
Sax Select element data of [NormalAttack] module. However, in this state, the volume immediately after Attack is "10
Since the value remains “0”, the volume immediately after Attack is corrected to “95” by performing correction on the representative point of the selected element data. As described above, the correction is performed so that the value of the selected element data is closer to the value of the transmitted performance parameter. Further, correction is performed according to the set value of the micro-tuning (tuning of the musical instrument), and the volume is corrected according to the volume change characteristic of the musical instrument. These corrections are performed by changing the representative point value of each element data, and the representative point value may change greatly. That is,
The data necessary and sufficient for the correction is the representative point, and various corrections are performed by controlling the representative point.

In step S23, the time position indicated by the time information may be corrected using correction information such as the performance style parameter. For example, when the time position obtained based on the performance data does not match the time position indicated by the time information, time information indicating a time position close to the time position obtained based on the performance data is selected and acquired there. By correcting the obtained time position information according to the performance data, the time position information intended by the performance data can be obtained. When the performance data includes a variable control factor such as touch or velocity, the time position information can be variably controlled by correcting the time position information according to the variable control factor. it can. The correction information includes information for performing such time position correction.

Further, a link process is performed to adjust each element data and smooth the connection between adjacent performance style modules (step S25). That is, by connecting the representative points of the connection portions in the front and rear performance style modules close to each other, the waveform characteristics of the front and rear performance style modules are made smooth. Such connection or link processing includes a harmonic component waveform (Timbre) and amplitude (Amplitud).
e), each of the elements such as pitch, or each of the non-harmonic component waveforms (Timbre) and amplitude (Amplitude).

At this time, the adjustment is performed in the range from the “link start point” of the preceding rendition style module to the “link end point” of the subsequent rendition style module. That is, the representative point within the range from the "link start point" to the "link end point" is adjusted based on the "rate from step". The “rate from step” is a parameter for controlling how far the connection is made from the preceding and later playing style modules, and is determined according to the combination of the preceding and following playing style modules as described later. . If the connection of the waveforms is not successful when the preceding and following rendition style modules are connected, the connection is smoothed by thinning out the vector ID of the waveform characteristic in one of the front and rear rendition style modules. In order to realize this thinning, a "performance style module combination table", a "thinning execution parameter range table" to be referred to hereafter, and a "thinning time table" to be referred to hereafter are prepared.

In addition to the above, the score interpreting section 10 as follows.
Waveform characteristics can be smoothly connected by the link processing in 1B. For example, discontinuous portions of performance style parameters (dynamic values, pitch parameter values, etc.) are connected smoothly regardless of the performance style module. Or,
Make a smooth connection by reducing vibrato early when going from vibrato to release.

Here, the above link processing will be described in detail. That is, adjustment of each element data for smoothing the connection part of the preceding and succeeding rendition style modules (see step S25) will be briefly described. First, referring to FIG. 8, the playing style module determines whether an amplitude element or a pitch (Pit
A description will be given of the link processing in the case of corresponding to the (ch) element.

If there is a step at the connection point between the preceding performance module and the subsequent performance module due to discontinuity in the value of the representative point at the connection point, first, a dynamics connection point (in the case of Amplitude) or a pitch connection The “rate from step” of the index for determining whether the target value of the point (in the case of Pitch) is closer to the value of the performance module before or after the pitch is determined. In the present embodiment, it is assumed that the “rate from step” is given by a table as shown. For example, the “rate from step” when the vector ID of the previous performance module is “3” and the vector ID of the subsequent performance module is “7”
Is determined as “30” from the table. The envelope shape is gradually deformed toward the target value from the “link start point” of the previous rendition style module to the “end point of the rendition style module” according to the “rate from step” thus determined. In addition, the envelope shape is gradually changed from the “link end point” of the later performance style module to the “reproduction style module start point” toward the target value. For example, if the “rate from step” is determined to be “30”, the target value for the previous rendition style module is “30”, and the previous rendition style module performs “30”% step forward on the subsequent rendition style module side ( In this embodiment,
The last representative point in the previous playing style module is "3
0 "% step).

On the other hand, the subsequent rendition style module performs "70" (100-30)% steps ahead of the previous rendition style module (in the present embodiment, the first representative point in the rear rendition style module is "70" upward). % Step). In addition, a plurality of representative points of the rendition style modules existing before and after the link start point to the link end point perform up and down steps along with the above steps. In this way, the performance is performed at a plurality of representative points of the rendition style module before and after the step. Note that the link start point and the link end point may be appropriately determined. However, if the link start point and the link end point are set to the same point as the desired representative point, the link start point and the link end point as shown in FIG. This is desirable because the envelope shape does not bend. Of course, even when the link start point and the link end point are not set to the same point as the desired representative point, it is needless to say that the step may be performed so that the envelope shape is not bent.

It should be noted that the determination of the "rate from step" is not limited to the above example. For example, it may be determined based on performance style parameters specified before and after the connection point. Alternatively, the determination may be made based on the performance data before the performance style ID or the performance style parameters. Or you may determine based on the combination of those data. Also, in the above example, the number of representative points to be taken from the step based on the “rate from the step” is one, and the other representative points are to be taken a proper amount along the step. The “ratio from step” may be determined separately, and a plurality of representative points may be respectively deviated from the step by the “ratio than step”.

Next, the link processing in the case where the rendition style module is a waveform (Timbre) element will be described. FIG. 9 to FIG. 12 are conceptual diagrams for explaining the link processing when the rendition style module is a waveform (Timbre) element. FIG. 9 is a conceptual diagram for explaining thinning of a waveform when an attack part waveform and a body part waveform are connected, and FIG. 10 is an explanatory view of a waveform thinning when a body part waveform and a release part waveform are connected. It is a conceptual diagram for performing. In FIG.
The body part waveform is composed of five loop waveforms L1 to L5,
It is assumed that loop playback is performed in a predetermined time range. Similarly, the body part waveform of FIG. 10 has six loop waveforms L1 '.
~ L6 '.

Adjustment of the element data relating to the waveform (that is,
There are various methods for linking waveforms). One example is the connection between the playing style module of the attack part or the joint part and the playing style module of the body part (or the playing style module of the body part and the release part or the joint part). We propose a method for smooth connection by partially thinning out waveforms. It is well known that cross-fade synthesis is performed when connecting waveforms. However, when the time t from the connection time to the start position of the first loop waveform L1 is short, as in the case of the example of FIG. 9, a sudden crossfade synthesis must be performed within the short time t. When such a steep cross-fade waveform synthesis, that is, when a time between a connected waveform and a waveform is very close, a cross-fade waveform synthesis is performed between the waveforms, a large noise is generated accordingly. A waveform is generated, which is not preferable.

Therefore, by thinning (deleting) a part of the waveform and widening the time interval between the connected waveforms,
Avoid steep crossfade waveform synthesis. In this case, the waveform in the attack portion, the release portion, or the joint portion is one lump, and the waveform cannot be thinned out. In this case, the loop waveform on the body portion side is thinned out. In FIGS. 9 and 10, the loop waveforms L1 and L6 'indicated by black rectangles are thinned out.
For example, in FIG. 9, the time difference from the connection time is relatively long.
Cross-fade synthesis is performed between the second loop waveform L2 and the tail waveform of the attack portion waveform, and the first loop waveform L1 is not used. Similarly, in FIG. 10, crossfade synthesis is performed between the loop waveform L5 'and the release portion waveform, and the waveform L6'
Is not used. It should be noted that the joint portion is a waveform section that connects between sounds (or between sound portions) by an arbitrary playing style.

Further, the connection between the rendition style module in the attack section and the rendition style module in the release section or the joint section is smoothed by waveform thinning. 11 and 12
FIG. 3 is a conceptual diagram for explaining waveform thinning in a case where an attack part waveform and a release part waveform are connected.
In this case, there are cases where the rendition style module such as the attack unit or the release unit can thin out the waveform, and cases where it cannot. An example in which the rendition style module of the attack section can thin out the waveform is a bend attack section (having several loop waveforms in the latter half). Also, in the case of a release section having several loop waveforms in the first half, waveform thinning can be performed. In this manner, the waveform of the rendition style module on which the waveform can be thinned is thinned. For example, when connecting the bend attack part and the release part, the loop waveform on the bend attack part side is thinned out as shown in FIG. One waveform is thinned out). When connecting a normal attack section and a release section having a loop waveform, FIG.
As shown in FIG. 2, the loop waveform on the release section side is thinned out (in FIG. 12, one loop waveform shown by a black rectangle on the release section side is thinned out). The loop waveform to be thinned is not limited to the loop waveform closest to the connection between the rendition style module and the rendition style module (the first or last loop waveform). The target loop waveform may be specified.

As described above, in a certain combination of performance style modules, waveforms are thinned out when connection is not successful within a certain performance style parameter range. To realize this, for example, a “performance style module combination table”
A “thinning execution parameter range table” to be referred to hereafter and a “thinning time table” to be referred to hereafter are prepared. The “performance style module combination table” is a table for determining a predetermined parameter based on a combination of performance style modules before and after connection. The “thinning execution parameter range table” is a table for determining a time range in which thinning is performed for each parameter. The “thinning time table” is a table for determining a thinning time. When the time difference (time t shown in FIGS. 9 to 12) between the connection time and the first (or last) loop waveform L1 (or L6 ′) is shorter than the reference thinning time, the loop waveform is thinned.

Further, a description will be given of the waveform connection in the case where the sample length of the rendition style module is short and ends before the rendition style module following the rendition style module starts. However, in FIG. 13, A.Sax [BendUpAttack], A.Sax
[NormalShortBody], A.Sax [VibratoBody], A.Sax [Norma
[Release], a packet stream of a waveform (Timbre) element is formed. The sample length (section length) of each playing style module is
It can be represented by the length indicated by “length”. In FIG.
“Note on” and “note off” described at the top are event timings of MIDI data.
A.Sax [BendUpAttack] and the like described in the middle row are the timings at which the rendition style IDs are generated.
mics, depth, and the like each indicate the timing of generation of the performance style parameter.

The A.Sax [BendUpAttack] module starts at time t0. Time t1 is a note-on timing in the module, and is adjusted to the designated note-on timing. In addition, the contents of the packet stream of the relevant module are the above notes, dynamics, dept
It is controlled based on performance parameters such as h. A.Sax [Nor
malShortBody] module is started at time t2 immediately after the attack module. Time t3 is a timing at which the vibrato performance starts in the middle of the connection. This timing is determined based on, for example, the start timing of the vibrato symbol added to the music data. Time t5 is a note-off timing in the A.Sax [NormalRelease] module, and is adjusted to the specified note-off timing. A.Sax [NormalReleas
e] The start time t4 of the module is specified accordingly.

That is, since the note-on is performed at time t1 and the note-off is performed at time t5, the time during which the sound is actually generated according to the waveform generated from the packet stream is the time from time t1 to time t5. In the case of such a packet stream, at time t2
Time from time t4 to A.Sax [NormalRe
In many cases, the sum of the sample lengths of the [lease] module and the A.Sax [VibratoBody] module does not match, and it is necessary to take appropriate measures. In such a case, the total of the sample length length is adjusted to the time length by repeating the same rendition style module, or the sample length of the rendition style module is adjusted to the time length, or the both are used in combination. Adjust the length of time. In this way, the waveform connection is performed by adjusting between the modules. In the above example, A.Sax [NormalShortB
ody] module, so that A.
Wave connection with Sax [VibratoBody] module. Similarly, by repeating the A.Sax [VibratoBody] module, the waveform connection with the subsequent A.Sax [NormalRelease] module is performed.

As described above, when performing the waveform connection by repeating the rendition style module a plurality of times, the time length of the rendition style module on the repetition side is varied. The variable control of the time length is performed by moving the representative point in the A.Sax [NormalShortBody] module or the A.Sax [VibratoBody] module in the example described above. That is, A.Sax [No
rmalShortBody] module or A.Sax [VibratoBody]
This is realized by an appropriate method such as changing the time of cross-fade connection between a plurality of loop waveforms constituting a module. In the case of a loop waveform, the time length of the entire loop reproduction waveform can be relatively easily controlled by varying the number of loops or the loop duration.
On the other hand, in the case of a non-loop waveform, it is not so easy to variably control its existence length on the time axis. Therefore, as described above, in a series of sound waveforms composed of a non-loop waveform and a loop waveform, the entire sounding time length is variably controlled by variably controlling the time axis of the waveform data in the loop readout section. It is very preferable to perform time expansion / contraction control. For this purpose, in the Japanese Patent Application Laid-Open No. 10-307586, "Ti
It is preferable to use “me Stretch & Compress” control (abbreviated as “TSC control”). In particular, in order to vary the length of the time axis in non-loop waveforms corresponding to special playing styles,
The above “TSC control” can be preferably applied.

FIG. 14 conceptually shows an example of the packet stream created in this way. FIG. 14 shows the packet streams in the amplitude (Amplitude) element of the harmonic component, the pitch (Pitch) element, the waveform (Timbre) element, the amplitude (Amplitude) element of the non-harmonic component, and the waveform (Timbre) element in order from the top. In the amplitude (Amplitude) element, the pitch (Pitch) element, and the amplitude (Amplitude) element of the nonharmonic component, the black squares are representative points, and the curve connecting these is the packet in the packet stream. Included vector ID
Shows the shape of the vector indicated by. Harmonic component waveform (Ti
In the mbre) element and the non-harmonic component waveform (Timbre) element, an outlined rectangle L indicates a loop waveform,
Other rectangles NL indicate non-loop waveforms. It should be noted that, among the non-loop waveforms, those shaded with oblique lines indicate non-loop waveforms that are particularly characteristic. Furthermore, in this embodiment, the waveform (Timbre) elements of the harmonic component and the non-harmonic component in the NormalAttack module are each composed of two vectors, and the amplitude (Amplitud) of the harmonic component is used.
e) The element, the pitch element, and the amplitude element of the out-of-harmonic component are each composed of one vector.

In this embodiment, for both the harmonic component and the non-harmonic component, the amplitude (Amplitude) element and the pitch (P
itch) The element has no vector. However, even when the waveform (Timbre) element is a non-loop waveform, the amplitude (Amplitude) element and the pitch (Pitc
h) The generated waveform may be controlled by having a vector of elements. In the VibratoBody module, the waveform (Timbre) element of the harmonic component is composed of five vectors, and the amplitude (Amplitude) and pitch (Pitch) elements of the harmonic component and the waveform (Timbre) element and the amplitude (Amplitude) of the nonharmonic component are used. Each element is composed of one vector. here,
Note that the VibratoBody module is repeated three times, but each iteration has a different shape for each vector. This is because different rendition style parameters are specified for each repetition. Different element data is selected according to different rendition style parameters, and different level control and time axis control are performed according to different rendition style parameters. In the NormalJoint module, the waveform (Timbre) elements of the harmonic component and the non-harmonic component are respectively composed of three vectors, and the amplitude (Amplitude) and pitch (Pitch) elements of the harmonic component and the amplitude (Amplitude) of the non-harmonic component
Each element is composed of two vectors. In addition,
The description of the NormalBody module is omitted.

As described above, the rendition style synthesizing section 101C converts the packet stream into each component (harmonic component and non-harmonic component).
Generated every time. Each of these packet streams is composed of a plurality of packets, and each packet has a vector ID and time information of the packet. In addition, the amplitude (Amplitude) element and pitch (Pitc
h) In the case of an element or an amplitude element of a non-harmonic component, a definite value of each representative point is provided. Of course, the present invention is not limited to this, and other information may be provided in addition to the vector ID and the packet time information. A packet stream is formed for each component according to the content of each such packet. As described above, the packet stream includes a plurality of packets and time information (start time) of each packet. It goes without saying that the number of packet streams may differ depending on the type of musical instrument and the like.

3.4. Waveform synthesis section 101D 3.4.1. General Operation of Waveform Synthesizing Unit 101D The waveform synthesizing unit 101D converts a waveform based on a packet stream (a sequence of a plurality of packets including vector ID, time information, correction information, and the like) for each component supplied from the rendition style synthesizing unit 101C. Combine. FIG. 15 is a conceptual diagram showing the overall configuration for explaining the operation in waveform synthesizing section 101D. FIG. 16 to FIG. 19, FIG. 22, and FIG. 23 are drawings for describing each operation in the waveform synthesis unit 101D in detail. FIG. 16 is a block diagram simply showing the overall flow of the waveform synthesis. FIG. 17 is a block diagram for explaining the vector loader. FIG. 18 is a block diagram for explaining a vector operator. FIG. 19 is a block diagram for explaining a vector decoder. FIG.
FIG. 2 shows the timing at which each packet is supplied from the rendition style synthesizing unit 101C to the waveform synthesizing unit 101D.
3 is a block diagram for explaining the cache control unit 40.

In FIG. 15, the packet stream for each component element created by the rendition style synthesizing section (articulator) 101C is provided in predetermined packet queue buffers 21 to 21 provided corresponding to each component element in the waveform synthesizing section 101D. The packets are sequentially input to the P.25 (that is, input in packet units). The input packets are accumulated in the packet queue buffers 21 to 25 and are sequentially sent to the vector loader 20 in a predetermined order. Vector loader 2
At 0, the original vector data corresponding to the vector ID is read out (loaded) from the code book 26 via the cache control unit 40 with reference to the vector ID in the packet. The read vector data is sent to predetermined vector decoders 31 to 35 provided for each component element, and the vector decoders 31 to 35 generate a waveform for each component element.

Further, the vector decoders 31 to 35 generate the waveforms for each component (harmonic component and non-harmonic component) while synchronizing the generated waveform for each component element among the vector decoders 31 to 35. The thus generated waveform for each component is sent to the mixer 38. The rendition style synthesizing section (articulator) 101C inputs packets to the packet queue buffers 21 to 25,
Various controls such as stream management (management of individual vector data generation or deletion or connection between vector data) and playback control (control of execution of desired waveform generation or playback / stop of generated desired waveform) Is performed on the waveform synthesis unit 101D.

As described above, the packets constituting the packet stream stored in the packet queue buffer 21 are sequentially input to the vector loader 20, and the vector loader 20 performs the processing based on the vector ID in each packet.
The vector data corresponding to the vector ID is read from the code book 26 via the cache control unit 40,
The read vector data is sent to the vector decoder 31 (see FIG. 16). At this time, correction information (for example, correction information on a representative point) may be added to each read packet. In such a case, the vector loader 20 deforms the read original vector data in accordance with the correction information, and generates a packet (hereinafter, referred to as vector information data to distinguish this from the original vector data) as information. This is called a vector packet to distinguish it from the packet input from the rendition style synthesizing unit 101C).
Output to 1 to 35. Thus, the vector loader 20
Reads the original vector data from the codebook 26 based on the vector ID of the packet input from the rendition style synthesizing section (articulator) 101C, corrects the vector data with the correction information as needed, and then performs the vector decoder 31- The vector packet is passed to 35 (see FIG. 17). The correction information on the representative point of the vector data as described above may include various correction information such as correction information for shifting time information based on a random number.

As shown in FIG. 18, the vector decoder 3
Reference numerals 1 to 35 denote generation and discard of vector operators for processing input vector packets, connection / synchronization management between vector operators, time management, and conversion setting of input from other vector ID streams into parameters for each vector operator. And various other types of operator operation management. The vector operators 36 and 37 read the vector information data, or control the read position of the vector information data (Speed input) or the gain control (Gai
n input) is performed. This vector operator 36
And 37 are managed by the vector decoders 31 to 35. Vector decoders 31 to 35 are provided so as to correspond to the respective component elements. The vector decoders 31 to 35 read out the vector information data in the vector packet and generate a desired waveform in time series.

For example, as shown in FIG. 19, the vector decoder 31 generates an envelope waveform of the amplitude component of the harmonic component, the vector decoder 32 generates an envelope waveform of the pitch component of the harmonic component, The vector decoder 33 generates a harmonic component waveform (Timbre).
The vector decoder 34 generates an envelope waveform of the amplitude component of the non-harmonic component (Amplitude), and the vector decoder 35 generates a waveform of the non-harmonic component (Timbr
e) Generate an envelope waveform for the element. The vector decoder 33 generates a harmonic component waveform to which the envelope waveform of the amplitude component of the harmonic component and the envelope waveform of the pitch component of the harmonic component generated by the vector decoders 31 and 32 are added, and a mixer 38. Output to That is, as a vector operator for performing gain control (Gain input) on the envelope waveform of the amplitude component (Amplitude) element of the harmonic component, the vector waveform of the envelope waveform of the pitch element (Pitch) of the harmonic component is given to the vector information data. Read position control (Sp
eed input) as a vector operator. In the vector decoder 35, the amplitude (Amplitude) of the nonharmonic component generated by the vector decoder 34 is used.
The waveform of the non-harmonic component to which the envelope waveform of the element is added is generated and output to the mixer 38. That is, for the vector decoder 35, the amplitude (Amplitud
e) Gain control of element envelope waveform (Gain input)
Is input as a control command for performing

Further, when the waveforms of the components and elements are generated in time series, the waveforms are generated while synchronizing the waveforms among the vector decoders 31 to 35. For example,
Vector packet and amplitude (Amplit) of waveform (Timbre) element
If a vector packet of ude) element is input, the waveform (T
imbre) The amplitude (Amplitude) is synchronized with the time of generating the waveform combination based on the vector packet of the element.
Generate an amplitude waveform based on the element's vector packet.
The amplitude of the waveform generated based on the vector packet of the waveform (Timbre) element is controlled by the amplitude waveform. Waveform (Timbre) element vector packet and pitch (Pitc
h) When a vector packet of elements is input, the waveform (Tim
bre) A pitch waveform based on a pitch (Pitch) element vector packet is synthesized in synchronization with the waveform generation time based on the vector packet of the element.

The pitch of the waveform generated based on the vector packet of the waveform (Timbre) element is controlled by the pitch waveform. When the vector packet of the harmonic component waveform (Timbre) element and the vector packet of the non-harmonic component waveform (Timbre) element are input, the harmonic component waveform (Timbre)
Based on the time of harmonic component synthesis based on the element vector packet, the waveform of the non-harmonic component (Timbr
e) Out-of-harmonic components based on vector packets of elements are synthesized. By mixing the waveforms of the synthesized harmonic component and the non-harmonic component, a desired musical sound waveform is generated. In this embodiment, the harmonic component and the non-harmonic component are configured to be able to select synchronous or asynchronous, and only when synchronous is selected, the waveform of the above-described harmonic component (Timbre)
Based on the time of the waveform synthesis of the harmonic component generated based on the vector packet of the element, the waveform of the non-harmonic component generated based on the vector packet of the non-harmonic component (Timbre) element is synthesized in synchronization with the time. You may make it.

As described above, the packet stream is composed of a plurality of packet sequences, and in the case of a vector bucket packet stream, each packet contains vector data. That is, the packet stream is a vector data in which vector data is arranged in the time direction, and the vector data of the amplitude element is also pitch (Pitc).
h) The vector data of the element and the vector data of the waveform (Timbre) element have different data structures and meanings, but are basically the same when viewed from the vector operators 36 and 37.

3.4.2. Structure of Vector Data FIG. 20 is a conceptual diagram conceptually showing an embodiment of a data structure of vector data. For example, when the unit of the reading position of the vector data is [SEC] and the reading speed is constant, one sample on the vector data coincides with one sample of the output waveform. Also, the unit of the reading speed is 1/1200 [cent] (= 2 to the nth power). If it is -1.0, it becomes 0.5 times (for example, in the case of a waveform (Timbre) element, it becomes one octave lower) (see the upper diagram in FIG. 20). The codebook 26 stores actual vector data. For example, vector data of an amplitude element or vector data of a pitch element include an array of VECTORPOINT structures and representative point data.

In the array of the VECTORPOINT structure, sample positions and values of respective points are sequentially entered.
The value of the vector data of the ude) element is in units of [db], and the value of the vector data of the pitch element is in units of 1/1200 [cent] when the MIDI note number 0 is 0.0. The representative point data is a DWORD type array, and stores the index number of the array of the VECTORPOINT structure serving as the representative point (see the lower part of FIG. 20). Of course, it is needless to say that the present invention is not limited to the above example.

3.4.3. Details of Cache Control Unit 40 (1) Overall Configuration of Cache Control Unit 40 Next, the overall configuration of the cache control unit 40 will be described with reference to FIG. . Since the code book 26 is stored in the hard disk 109, vector data required by the vector decoders 31 to 35 is read from the hard disk 109. However, since the access time of a hard disk or the like is slow and not constant, it is impossible to immediately read the vector data at the timing when the vector decoders 31 to 35 process the vector data. Therefore, in the present embodiment, the waveform synthesizing unit 1
A cache control unit 40 is provided in 01D, and vector data to be used in the future (or predicted to be used) is pre-loaded into the cache memory.

In FIG. 23, reference numeral 42 denotes a look-ahead prefetch unit, which includes a rendition style synthesizing unit 101C to a waveform synthesizing unit 101D.
, A vector ID is extracted from the packet supplied to the hard disk 109, and read control is performed on the hard disk 109 so as to prefetch (prefetch) the vector data corresponding to the vector ID from the codebook 26. As described above, these packets constitute a packet stream in the packet queue buffers 21 to 25, and are read out by the vector loader 20 at a later time. Prefetch processing is performed in parallel with this.

Reference numeral 41 denotes a prediction control unit, and the score interpretation unit 10
1B, and predicts vector data likely to be used in the future based on the prediction data (such as a program change) supplied from the prefetch unit 42 and the prediction conditions supplied from the prefetching prefetch unit 42. The ID is supplied to the prefetch prefetch unit 42. Here, the “prediction condition” refers to the vector I determined to be used (that is, supplied from the rendition style synthesizing unit 101C).
D and the like. Thus, in the prefetching prefetch unit 42, the vector ID is supplied from both the rendition style synthesizing unit 101C and the prediction control unit 41. Then, the prefetching prefetch unit 42 gives priority to loading of vector data determined to be used (that is, loading of vector data designated by the rendition style synthesizing unit 101C),
Prefetch processing is performed on vector data related to both vector IDs. Hereinafter, the loading of vector data that is determined to be used is referred to as “designated loading”,
Loading vector data that has not been determined to be used (just predicted) is referred to as “prediction loading”.
A cache memory 44 stores prefetched vector data. 43 is a read control unit,
When receiving the vector ID from the vector loader 20, the vector data corresponding to the vector ID is mainly read from the cache memory 44 and supplied to the vector loader 20. A time management unit 45 controls timing such as prefetch.

(2) Operation of Prediction Control Unit 41 Next, the processing contents of the prediction control unit 41 will be described with reference to the state transition diagram of FIG. The state of the prediction control unit 41 is determined depending on whether or not the waveform synthesis is performed in the vector decoders 31 to 35 and, if the synthesis is performed, which module is involved. In the initial state, since no waveform synthesis is performed in the vector decoders 31 to 35, the process of step S30 is performed in the prediction control unit 41. here,
The vector data candidates of the attack unit are predicted, and the predicted vector IDs are sequentially supplied to the prefetching prefetch unit 42. However, in step S30, the prediction of the attack portion is not executed unless “program change” is supplied as the prediction data from the musical score interpretation unit 101B. This is because if the program change is undecided, the vector data of the attack portion that may be specified will be enormous. Therefore, in the initial state, when a program change is supplied from the musical score interpretation unit 101B, the predictive loading of the attack section vector data according to the program change is started immediately.

For example, if "piano" is designated as the program change, and if there are 100 types of vector data of the attack portion relating to "piano", the predictive loading of 100 types of vector data is started. When the packet related to the attack section is determined, the vector loader 20 and the vector decoders 31 to 35 will start the waveform synthesis processing of the attack section. At that time, the state of the prediction control unit 41 transitions to step S31.

In step S31, predictive loading is performed on the vector data candidates of the body part based on the determined attack part vector ID. That is, since the pitch of the attack portion is known, the candidate vector data is limited to those corresponding to the same pitch as the attack portion. Further, the vector data of the body part that may be connected to the determined envelope waveform of the attack part is further limited. In step S31, candidates narrowed down according to such conditions are predicted and loaded. Here, the packet related to the body part is actually
When the packet of the body part is supplied from C and is determined, the state of the prediction control unit 41 transitions to step S33.

In step S33, prediction loading is performed for the next vector data candidate based on the determined vector ID of the body part. As mentioned above, the module connected to the body part is the other body part,
Either a joint part or a release part.
Therefore, the prediction control section 41 predictively loads these vector data. It should be noted that the vector data candidates to be predicted loaded are narrowed down in accordance with the determined pitch of the body part, the envelope waveform, and the like, as in step S31.

When a packet is actually supplied from the rendition style synthesizing unit 101C during the execution of step S33, the state of the prediction control unit 41 changes according to the packet. First, if the supplied packet is a packet of the body part, the state remains at step S33, and the prediction load of another body part, joint part or release part is executed again based on the body part. . If the supplied packet is a packet of the joint unit, the state of the prediction control unit 41 transitions to step S32.

In step S32, vector data candidates of the body part are predicted and loaded. Here, the candidates for the vector data to be predicted loaded are narrowed down to the vector data of the body part which may be connected to the determined joint part. Here, when the packet relating to the next body part is supplied from the rendition style synthesizing part 101C, the state of the prediction control part 41 returns to step S33 again, and the vector data of another body part, joint part or release part is loaded with the prediction load. Is done. Here, when the packet of the release unit is supplied from the rendition style synthesizing unit 101C, the prediction control unit 4
The state of 1 makes a transition to step S30.

In step S30, as described above, candidates for the vector data of the attack portion are predicted and loaded. However, if the pitch of the immediately preceding release portion has already been determined, it is highly possible that the pitch of the next attack portion does not deviate so much. Therefore, during the synthesis of the release portion, the vector data of the attack portion to be predicted and loaded is located around the pitch of the release portion (for example, ±
(Corresponding to one octave).

(3) Operation of Prefetch Prefetch Unit 42 (3.1) Load Processing (FIG. 25) Next, the operation of the prefetch prefetch unit 42 will be described. First, in the prefetch prefetch unit 42, a load process shown in FIG. 25 is executed at predetermined intervals. In the figure, when the process proceeds to step S41, it is determined whether or not a designated load request that has not been executed has been received. If "YES" is determined here, the process proceeds to step S42, and the designated load of the requested vector data is executed. As a result, after a while, the vector data is read from the hard disk 109 and stored in the cache memory 44.

On the other hand, if "NO" is determined in the step S41, the process proceeds to a step S43 to execute a prediction load. In other words, of the vector data for which a prediction load request has been issued, the vector data for which a read command has been issued to the cache memory 44 (already the cache memory 44).
, And those which are scheduled to be loaded in response to another load request) and those for which a read command has not yet been issued to the hard disk 109, and only the latter is read to the hard disk 109. Commands are supplied. If there is a large amount of vector data to be predicted loaded, this routine may be called again during the predicted loading. In such a case as well, as long as the requested designated load exists, the predicted load is interrupted, and the designated load is executed via step S42. As described above, in the prefetch prefetch unit 42, the designated load is executed with higher priority than the predicted load.

(3.2) Packet Receiving Process (FIG. 26) The prefetching prefetch unit 42 executes the packet receiving process shown in FIG. 25 every time a packet is supplied from the rendition style synthesizing unit 101C. In the figure, when the process proceeds to step S51, a vector ID is extracted from the supplied packet, and it is determined whether or not vector data corresponding to the vector ID is hit. here,
"If hit" means (1) the cache memory 4
If any of the vector data predicted and loaded in 4 hits, (2) if the vector data has not yet been loaded into the cache memory 44, but if the predicted load of the vector data is scheduled, (3) Either the loaded and used (USED state) vector data can be used as it is.

In the present embodiment, the used vector data is not immediately released, but is used in the "USED
Cache memory 4 for the time being as vector data of "state"
4 is left. When the storage capacity for storing other vector data becomes insufficient, the area of the vector data in the USED state is released, and new vector data is stored. The details will be described later.

When there is no hit vector data, "NO" is determined in the step S51, and the process proceeds to a step S53. Here, a designated load request is generated for the vector ID included in the packet. Therefore, the next time the above-described packet reception processing (FIG. 26) is executed, the vector data is specified and loaded.

On the other hand, if there is hit vector data, "YES" is determined and the process proceeds to step S52. Here, if the predicted load for the vector data has not been completed, the predicted load request is changed to a designated load request. This is because the packet is preferentially loaded when the packet receiving process is performed thereafter. Next, when the process proceeds to step S54, the rendition style synthesizing unit 101
For C, the handle of the vector data requested to be specified is returned. This handle uniquely corresponds to the vector data. If the vector data requested to be specified is already stored in the cache memory 44, the handle of the vector data is returned. Otherwise, a new handle is generated and returned to the rendition style synthesis unit 101C. .

Next, when the processing proceeds to step S55, any of the predicted load requests generated earlier that have been missed are canceled. Here, the specific contents of “cancel” will be described later. Next, when the process proceeds to step S56,
The prediction condition in the prediction control unit 41 is changed. That is, according to the packet supplied from the rendition style synthesizing unit 101C, the candidates for the prediction load in each state in FIG.

This packet reception processing is performed by the
From the 1C side, the “GetVector command” shown in FIG. 28 is obtained. That is, the rendition style synthesizing unit 101C is
The operation of supplying a packet to D is nothing less than transmitting a GetVector command meaning "Get vector data" based on the vector ID included in the packet. This vector data is later stored in the vector loader 20.
Must be prepared before the data is read out, but at that time, a handle is required to specify required vector data from a large number of vector data.

(4) Data structure in the cache memory 44 (4.1) Cache page (FIG. 27) Of the vector data stored in the cache memory 44, those used simultaneously are packed and packed into one file (or vector). Is converted to multiple files of less than the total number of data). This conversion is automatically executed prior to the performance in accordance with a user's instruction or created music data. An example is shown in FIG. In the figure, of the vector data read out from the codebook 26, the one that is designated and loaded includes time information. Therefore, a plurality of vector data used simultaneously can be extracted. Each vector data read from the code book 26 individually has a header. Vector data used at the same time is put together in, for example, one file, and a common header is given to this file. As a result, the time for the vector loader 20 to access each vector data can be reduced.

The following information is stored in the header of this file. Data ID: A 4-byte identification character “PACK” that identifies the type of file is stored here. Data Size: Indicates the data size of the file. VQ Type: Indicates the type of the stored vector data. Version: Indicates the version of the file format. A similar header is provided for vector data before packing, but the data ID is “PACK”.
Instead, a data ID for identifying the type of each vector data is assigned.

The waveform generating apparatus of this embodiment is realized by an application program on a personal computer, and its cache exists on a system memory. When vector data is stored in a cache, cache control is performed in units of a cache page of a predetermined size, instead of in units of one vector data (uneven size). In other words, one cache data is divided into multiple cache pages,
It is stored and managed in the cache on a cache page basis. The system memory stores cache management data (page header) corresponding to each cache page. The data storage on the hard disk is
Usually, this is done in fixed-size cluster units,
The size of the cache page may be equal to or an integer multiple of that size. Considering the capacity of vector data, the size of a cache page is 1 k to 1
Preferably, it is 0 kbytes.

(4.2) Data Structure of Page Header Next, the data structure of the header added to each cache page will be described. Each header is “VDDLCSPAG
The member of the structure is as follows. dwPage: a page number uniquely assigned to the cache page. dwID: Vector ID of vector data included in this cache page. dwSize: The data size of this cache page. dwCount: Indicates the number of prefetches. This member dwCount
Is "1" each time it is prefetched by the specified load
It is incremented by one, and is decremented by "1" each time it is read by the vector loader 20. dwStatus: Indicates the status of the cache page. Cache page status is FREE, ALLOCATED, USED, FILLE
Either D or LOCKED. lpBuf: A pointer indicating the start address of the entity of vector data (other than the header) in the cache page. lpForward / lpBackward: In the present embodiment, a plurality of VDDLCSPAGE structures form a bidirectional linked list. The member lpForward is a pointer to another VDDLCSPAGE structure existing in the forward direction of the link.
lpBackward: is another VDDLCS that exists in the backward direction of the link
Pointer to a PAGE structure. lpNext: As described above, in a case where the waveform generating apparatus is realized on dedicated hardware, a plurality of vector data used simultaneously is divided into a plurality of cache pages. These plurality of cache pages constitute a “group”, and the member lpNext is a pointer that sequentially points to another VDDLCSPAGE structure belonging to the same group.

Here, the role of the member dwCount will be described. For example, if the same vector data is used twice in a series of packet streams,
Two designated load requests are generated for the vector data. However, if the cache page is set to the USED state at the time of the first reading by the vector loader 20, the second reading cannot be performed.
Therefore, this problem is prevented by counting how many times the cache page is read. By the way, when the vector data is used a plurality of times, the same vector data is used a plurality of times in the waveform synthesis processing for the same event data, and the same vector data is used a plurality of times in the waveform synthesis processing for the plurality of event data. It is considered that it is used. In this embodiment, in any case, the remaining number of uses is determined by the member dwCount, so that the cache pages in the cache memory 44 can be handled uniformly.

(4.3) Link Structure of Page Header (FIG. 29) Next, FIG. 29 shows the structure of a bidirectional linked list realized by the pointer lpForward and the pointer lpBackward. In the present embodiment, a list structure as shown in the figure is adopted in order to freely rearrange and add / delete data at high speed. In the figure, A-1 is a page header located at the head of the bidirectional linked list, the head of which is indicated by a predetermined pointer lpTop, and the end of which is indicated by a pointer lpTail. A-2, A-
Reference numeral 3 denotes a page header belonging to the same group as the page header A-1. That is, the “group” corresponds to data of each cache page formed by dividing one file. The start address of the page header A-2 is indicated by the member lpNext of the page header A-1, and the start address of the page header A-3 is indicated by the member lpNext of the page header A-2.

B-1 is a page header linked to the next stage of the page header A-1. The head address of the page header B-1 is a member lp of the page header A-1.
Indexed by Forward. The head address of the page header B-2 belonging to the same group is indicated by the member lpNext of the page header B-1. The head address of the page header C-1 linked to the next stage is indexed by the member lpForward of the page header B-1, and the page headers C-1, C-2
Page headers C-2 and C-2 respectively by the member lpNext of
The start address of -3 is indicated. Here, since the FREE cache pages also constitute one group, for example, the groups A and B are set to the FILLED state, and the group C is set to the FREE state. When caching new vector data, a cache page is obtained from the group in the FREE state or the USED state, and the cache page is set as the cache page of the new vector data.

Further, a new cache page and a corresponding page header may be created on the system memory, and this may be used as a new vector data cache page. However, in that case, it is preferable to limit the total amount of cache pages to a specified value so as not to consume too much memory resources. This designated value may be automatically set according to the system memory amount, or may be manually set by the user. In addition, if control is performed so that a certain fixed value is always secured as the memory capacity of the group in the FREE state, the allocation of new vector data to the cache page can be speeded up.

According to the bidirectional linked list in FIG. 29,
By referring to the member lpForward, the first page headers A-1 and B- of each group constituting the list are displayed.
1, C-1 can be traced in order. Also, member lp
By referring to Backward, the bidirectional linked list can be traced in the reverse direction. Therefore, when investigating whether or not certain vector data is cached, the first page header constituting the list is sequentially traced, and the member dwID in the page header matches the same vector ID as the vector data. It is good to judge whether or not. If there is a match, the cache page related to the page header becomes the first cache page of the group caching the vector data. If all vector data used at the same time is always stored in one page, the bidirectional linked list includes only page headers A-1, B-1, and C-1. Null data is stored in the lpNext member of these page headers.

When the states indicated by the member dwStatus are arranged in the order of importance, "LOCKED ≧ FILLED> ALLOCA
TED>USED> FREE. Therefore, if the order of the cache pages is set in this order of importance, the allocation of cache pages to new vector data can be speeded up. Further, the value of the member dwCount may be directly reflected in the importance. That is, member dwCo
The larger the unt, the higher the importance. USED
Between the cache pages in the state, it is preferable to set so that the transition to the USED state first becomes less important.

(5) Operation in Read Control Unit 43 Next, the operation of the read control unit 43 will be described with reference to FIG. 28 again. In the vector loader 20, a command is transmitted to the read control unit 43 to read the cache memory 44 based on the contents of the packet stream, that is, the vector ID in each packet. Change this command to Lo
Called the ckVector command. This LockVector directive includes Ge
The handle returned in response to the tVector command is attached. In a normal operation state, the vector data should have been stored in the cache memory 44. Therefore, the pointer of the head address of the cache page related to the vector data is returned to the vector loader 20.

As a result, the contents of the cache memory 44 are appropriately read out by the vector loader 20 and the vector decoders 31 to 35, and the waveform synthesizing process is executed. As described above, the cache page that can be read by the vector loader 20 or the like is set to the LOCKED state (details will be described later), and the content is set so as not to be altered by other processing.

Incidentally, depending on the situation, when the LockVector command is received from the vector loader 20, the required vector data may not be loaded on the cache memory 44 yet. In particular, when the waveform generating apparatus of the present embodiment is realized by an application program of a personal computer, the hard disk 109 may be occupied for a relatively long time due to the operating system of the personal computer, and such a situation may often occur. . In this case, different processing is executed depending on the operation mode of the waveform generation device.

First, when waveform synthesis is performed in non-real time, it is preferable to stop the subsequent processing until the designated vector data is loaded. Therefore, in the read control unit 43, the hard disk 10
9 is executed synchronously. This synchronous reading means that no other processing is performed until the desired data is read.

On the other hand, when the waveform generating apparatus is operating in real time, executing the synchronous readout may result in interruption of the musical sound. Therefore, the pointer of the substitute page is returned to the vector loader 20. In the present embodiment, various types of vector data are used to faithfully represent various timbre changes. If the vector is limited, the capacity is not so large. Such a vector (plural) is stored in the RAM 103 in advance as a default vector of the tone color.
And is used as a substitute for vector data that could not be prepared in the cache memory 44.

Even in the case of performing real-time synthesis, a sufficient time delay is ensured in the waveform synthesis processing S14 (waveform synthesis unit 101D), and necessary vector loading and waveform synthesis can be performed within the range. In this case, as in the case of performing non-real-time waveform synthesis, when it is determined that necessary vector data has not been loaded, the vector data can be urgently read to perform waveform synthesis. For the cache page that has been used by the vector loader 20 and the vector decoders 31 to 35, a release command with a handle is supplied from the vector loader 20 to the read control unit 43. When this release command is supplied, the L of the cache page is
The OCKED state is released and the handle is released.

(6) State Transition Operation for Cache Pages Next, with reference to the state transition diagram of FIG. 30, an operation of transitioning the state of each cache page will be described. First, in the initial state, all cache pages are set to the FREE state (S61). That is, the member dwStatus of the header of the cache page is set to 'FREE'.
This means that the contents are unused and the contents are not guaranteed. When vector data is loaded into this cache page, the cache page is set to the ALLOCATED state (S62). ALLOCAT
The ED state means that data storage has not been completed yet, but is reserved for reading data from the codebook 26. At this time, “1” is stored in the member dwCount.

Thereafter, when the vector data is stored in the page, the state of the cache page transits to the FILLED state. In the ALLOCATED state or the FILLED state, when a designated load request is issued again for the same vector data, the member dwCount is incremented by "1" each time. When the predicted load request is canceled in the ALLOCATED state or the FILLED state (step S55 of the load processing (FIG. 25)), the member dwCount is decremented by "1". When the value of the member dwCount becomes "0" in these states, the cache page is set to the USED state (S63).

After the cache page transits to the FILLED state, when a LockVector command is supplied from the vector loader 20, the cache page is set to the LOCKED state (S65). When the use of the cache page by the vector loader 20 and the vector decoders 31 to 35 is completed and a Release command is supplied, the LOCKED state is canceled (UnLocked) and the cache page is filled again.
The state is set to the ED state (S64). At this time, the member dwCount of the cache page is decremented by “1”. At this time, if the member dwCount becomes “0”, the state of the cache page is immediately changed to the USED state (S6).
Set to 3).

After the cache page is set to the USED state, when a predicted load request or a designated load request (prefetch) is generated again for the cache page, the cache page is returned to the FILLED state again.
As the above processing continues, the number of cache pages in the USED state increases in the cache memory 44 and the number of cache pages in the FREE state gradually decreases. Then, when the prefetch of new vector data occurs after the cache page in the FREE state disappears, the oldest US page in the cache page in the USED state
Cancellation (SwapOut) is performed in order from the ED state, and the state is returned to the FREE state (S61). And this FREE
The cache page returned to the state is set to the ALLOCATED state for loading the new vector data.

Here, as a method for quickly specifying the oldest cache page in the USED state, the bidirectional linked list described above with reference to FIG. 29 may be used. That is, any cache page is USED
When set to the state, the cache page is added to the head of the linked list, and if there are not enough free cache pages, the cache pages at the end of the linked list are sequentially removed from the linked list, and It is good to change to.

(7) Time Management Unit 45 The time management unit 45 controls the timing of the entire waveform synthesizing unit 101D. Here, the outline of the timing control in the present embodiment will be described with reference to FIG. In the figure, it is assumed that the reproduction start is instructed at time t30 (for example, the reproduction button is pressed by the user), and the time at which the musical sound output is actually started is t40. The time from time t30 to t40 is called latency β. The latency β is set, for example, to about 2000 msec, but need not be set.

When a reproduction start is instructed, a reproduction thread is generated in the CPU 101 at predetermined time intervals. In the reproduction thread, music data is read, and a prefetch process is performed on the hard disk 109 or the like according to the content. Time t at which certain song data is read
The time from 32 to the actual sounding start time t40 is called the advance time γ. The reference value of the advance time γ is, for example, 4000 msec.
About 1000 to 10,000 msec depending on the processing situation.
It is preferable to be able to expand and contract within a range of about c.

As described above, by securing the advance time γ to some extent, the reproduction thread is generated intermittently, and the reading of the music data is performed intermittently (to some extent collectively).
It is possible to do. The interval at which a playback thread occurs is called a playback thread occurrence interval ε. The reference value of the reproduction thread generation interval ε may be set to, for example, 20 msec, and may be expanded or contracted in a range of about 5 to 100 msec according to the processing situation.

When the music data is read out at time t32, part of the music data is extracted in step S11 as described above, and the score is interpreted in step S12.
Thereafter, in step S13, the rendition style synthesis is executed. As a result, vector data is prefetched at time t34. The time from the prefetch time t34 to the sound generation start time t40 is called a prefetch time α.

The vector data read from the hard disk 109 is written to the cache memory 44 soon. Thereafter, at time t36, the vector loader 2
When 0, the vector data is read from the cache memory 44 via the read control unit 43, and the waveform synthesis is started. The time from the waveform synthesis start time t36 to the tone generation start time t40 is called an output latency δ. The reference value of the output latency δ may be set to, for example, about 300 msec, and may be expanded or contracted in a range of about 10 to 1000 msec according to the processing situation.

It is preferable that the waveform synthesizing process is performed intermittently in a certain time range as in the case of the reproduction thread. The reference value of the activation frequency interval of the waveform synthesizing process is set to, for example, 50 msec, and may be expanded or contracted in a range of about 10 to 500 msec according to the processing status.

As apparent from FIG. 31, the advance time γ, the prefetch time α, and the output latency δ include:
There is a relationship of “γ>α> δ”. Here, the time interval from the prefetch time t34 to the waveform synthesis start time t36 "α-
δ ”needs to be set to such an extent that vector data can be loaded from the hard disk 109. When the peak value of the load amount of the vector data is high, generation of noise or the like can be prevented by increasing the time interval “α−δ”.

4. Modifications The present invention is not limited to the embodiments described above,
For example, various modifications are possible as follows. (1) When the above-described waveform generating apparatus is used for an electronic musical instrument, the electronic musical instrument is not limited to a keyboard instrument, but may be a string instrument, a wind instrument, or a percussion instrument. In this case, the music data reproducing unit 101A, the musical score interpreting unit 101B, the playing style synthesizing unit 101C, the waveform synthesizing unit 10
It is not limited to the one in which the 1D or the like is built in one electronic musical instrument main body, but also the one in which each is separately configured and configured to connect each component using communication means such as a MIDI interface or various networks. It goes without saying that the same can be applied. Further, the configuration may be a configuration of a personal computer and application software. In this case, the processing program may be stored and supplied to a recording medium such as a magnetic disk, an optical disk, or a semiconductor memory, or may be supplied via a network. . Further, the present invention may be applied to an automatic performance device such as an automatic performance piano.

(2) In the above embodiment, a plurality of vector data are stored in one cache page.
It goes without saying that one vector page data may be stored in one cache page.

(3) In the above embodiment, it is determined that a transition to the USED state is made by incrementing or decrementing the member dwCount of the header of the cache page. However, in this method, when the number of unused cache pages increases, the capacity of the cache memory 44 may become insufficient. In such a case, “priority” may be defined for the cache page in the FILLED state, and the cache page in the FILLED state having a lower priority may be sequentially shifted to the USED state and the FREE state. In this case, when the LockVector command is received from the vector loader 20 for the cache page that has transitioned to the FREE state, the above-described alternative page is used.
As a specific example of “priority”, the value of the member dwCount may be used as it is (the higher the value, the higher the priority). Further, the number of times the cache page has been used in the past and the maximum value of the member dwCount (the maximum value in the past history) may be added.

(4) In the above embodiment, the cache pages in the USED state are linked by the bidirectional linked list, and when the cache pages in the FREE state run short, the cache page at the end of the linked list is deleted from the I removed them sequentially (from the oldest one) and changed them to FREE status. However, the order of transition to the FREE state is not limited to this. For example, for a cache page with a large cumulative number of times used, a cache page with a large maximum value in the history of the member dwCount, or vector data used at the beginning of each module,
It may be set so as not to transition to the FREE state as much as possible, in other words, to preferentially leave in the cache memory 44. Alternatively, the used vector data that has been actually used may be preferentially left over the vector data that has not been actually used (the prediction is incorrect).

Also, by leaving the vector data used at the head of the module preferentially, the margin for the operation of the hard disk 109 is increased and it is possible to reduce the situation where the substitute page must be used. it can. Specifically, when a cache page to be preferentially left is shifted to the USED state, the cache page is added to the head of the linked list, and when the other cache pages are shifted to the USED state, the cache page is changed. It is good to insert in the middle part of a linked list.

(5) In the above embodiment, an example using vector data has been described as a specific example of sound data.
The sound data is not limited to vector data, and it goes without saying that various waveform data or parameters defining the musical tone waveform may be used as sound data.

(6) In the above embodiment, as described with reference to FIGS.
The packet stream from C is a prefetch prefetch unit 4
2 was directly entered. However, instead of such an operation, the prefetching prefetch unit 42 may read out the packet stream once written in the packet queue buffers 21 to 25. In this case, the prefetching prefetch unit 42 may read the packet from the packet queue buffer and process the packet at a predetermined time before the timing at which the vector loader 20 reads and processes the packet stored in the packet queue buffer. It is suitable.

[0122]

As described above, according to the present invention, when the tone color designation or the first to third tone data designations are received,
Since the sound data candidates specified thereafter are predicted and transferred in advance from the low-speed storage device to the high-speed storage device, a large amount of sound data is stored in the low-speed storage device while the first to third sound data are specified. Can respond quickly. Thus, sound data for performing expressive waveform synthesis can be reproduced in real time.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an example of a hardware configuration of a waveform generation device according to the present invention.

FIG. 2 is a flowchart illustrating one embodiment of a “waveform database creation process” executed in the waveform generation device.

FIG. 3 is a diagram schematically showing an example of each component and element constituting an actual waveform section corresponding to a rendition style module.

FIG. 4 is a flowchart showing an embodiment of “musical sound synthesis processing based on a database”.

FIG. 5 is a block diagram showing an embodiment in which the same waveform synthesis processing as in FIG. 4 is configured in the form of dedicated hardware.

FIG. 6 is a block diagram for explaining a flow of a rendition style synthesis process in the rendition style synthesis section described above.

FIG. 7 is a flowchart showing in detail one embodiment of a rendition style synthesizing process performed by the rendition style synthesizing unit.

FIG. 8 is a conceptual diagram for explaining link processing in a case where a rendition style module corresponds to an amplitude element or a pitch element.

FIG. 9 is a conceptual diagram for explaining thinning of a waveform when an attack waveform and a body waveform are connected.

FIG. 10 is a conceptual diagram illustrating thinning out of a waveform when a body waveform and a release waveform are connected.

FIG. 11 is a conceptual diagram for explaining waveform thinning when a bend attack waveform and a release waveform are connected.

FIG. 12 is a conceptual diagram for explaining thinning out of a waveform when a normal attack waveform and a release waveform having a loop portion are connected.

FIG. 13 is a conceptual diagram for explaining a waveform connection in a case where a rendition style module before a subsequent rendition style module is started ends.

FIG. 14 is a conceptual diagram for explaining a packet stream generated by a rendition style synthesizing unit.

FIG. 15 is a conceptual diagram showing an embodiment of the overall configuration for explaining the operation of the waveform synthesizing unit.

FIG. 16 is a block diagram simply showing the overall flow of waveform synthesis.

FIG. 17 is a block diagram illustrating a vector loader.

FIG. 18 is a block diagram for explaining a vector operator.

FIG. 19 is a block diagram for explaining a vector recorder.

FIG. 20 is a conceptual diagram conceptually showing one embodiment of a data structure of vector data.

FIG. 21 is a diagram showing the contents of performance data created by a music score interpretation unit 101B.

FIG. 22 shows a rendition style synthesis section 101C to a waveform synthesis section 101.
The figure which shows the timing at which each packet is supplied to D.

FIG. 23 is a block diagram showing the overall configuration of a cache control unit 40.

FIG. 24 is a state transition diagram of a prediction operation of the prediction control unit 41.

FIG. 25 is a flowchart of a load process executed in a prefetch prefetch unit.

FIG. 26 is a flowchart of a packet receiving process executed in the prefetch prefetch unit.

FIG. 27 is an operation explanatory view showing a method of configuring a cache page.

FIG. 28 is a rendition style synthesizing unit 101C and a waveform synthesizing unit 10.
Signal flow diagram during 1D.

FIG. 29 is a diagram showing a link structure of a page header in the cache memory 44.

30 is a state transition diagram of each page in the cache memory 44. FIG.

FIG. 31 is a timing chart showing an outline of the timing control of the embodiment.

[Explanation of symbols]

1 to 12 data, 20 vector loaders, 21 to
25 packet queue buffer, 26 codebook, 31-35 vector decoder, 36, 37
Vector operator, 38, mixer, 40, cache control unit, 41, prediction control unit, 42, prefetch prefetch unit, 43, read control unit, 44, cache memory (high-speed storage device), 45. Time management unit, 50
… Attack part playing style module, 101… CPU, 1
01A: music data reproducing unit; 101B: music score interpreting unit
101C ... rendition style synthesis unit, 101D ... waveform synthesis unit, 1
02 ROM, 103 RAM, 104 panel switch, 105 panel display, 106 drive, 106 drive, 106A external storage medium, 107 waveform capture unit, 108 ... Waveform output unit, 108A sound system, 109 hard disk (low-speed storage device), 111 communication interface.

Claims (7)

[Claims] 1. Using a low-speed storage device for storing musical sound waveform sound data and a high-speed storage device for caching the sound data, a part of the sound data stored in the low-speed storage device is stored in the high-speed storage device. A sound data transfer method for transferring to a storage device, wherein a sound color specification receiving step of receiving a sound color specification relating to the sound data to be transferred to the high-speed storage device; A sound data transfer method, comprising: a sound data prediction step of predicting sound data candidates; and a step of transferring the predicted sound data candidates from the low-speed storage device to the high-speed storage device. 2. The sound data stored in the low-speed storage device includes first sound data corresponding to a sound generation start portion and second sound data corresponding to a portion other than the sound generation start portion. The sound data predicted is the first sound data, a first sound data specification receiving step of receiving the specification of any of the first sound data, and a later specified sound data based on the specification of the first sound data. Second
The method according to claim 1, further comprising: a second sound data prediction step of predicting sound data candidates; and a step of transferring the predicted second sound data from the low-speed storage device to the high-speed storage device. Sound data transfer method. 3. The sound data stored in the low-speed storage device further includes third sound data used after the first and second sound data, and receives designation of any one of the second sound data. A second sound data designation receiving step, and a third designation specified later based on the designation of the second sound data
The method according to claim 2, further comprising: a third sound data prediction step of predicting sound data candidates; and a step of transferring the predicted third sound data from the low-speed storage device to the high-speed storage device. Sound data transfer method. 4. Receiving the designation of any one of the sound data, the identification information stored in the high-speed storage device and indicating that release of the sound data candidate that is not predicted is possible is possible. 4. The sound data transfer method according to claim 1, further comprising a step of providing. 5. The method according to claim 1, further comprising: before the tone color designation receiving step, transferring basic tone data for each tone color to the high-speed storage device in advance, wherein the first, second or third tone data designation reception is performed. If the designated sound data is not transferred to the high-speed storage device in the process, instead of the designated sound data,
2. The method according to claim 1, wherein the basic sound data is used.
5. The sound data transfer method according to any one of claims 1 to 4. 6. A sound data transfer apparatus for performing the method according to claim 1. 7. A program for executing the method according to claim 1. Description: