JP4102931B2 - Sound waveform synthesizer - Google Patents

Description

  The present invention relates to a sound waveform synthesizer that synthesizes waveform data to generate a sound waveform such as a musical sound waveform or a human voice waveform.

2. Description of the Related Art There is known an electronic music apparatus of a waveform memory sound source method (waveform table sound source method) that synthesizes a musical sound waveform by processing waveform data as a material in accordance with performance data.
FIG. 4 is a block diagram showing a waveform memory sound source type electronic music apparatus.
4A, a CPU (Central Processing Unit) 102 for sound source control, a work RAM (Random Access Memory) 103, and a sound source LSI (Large Scale Integrated Circuit) 104 are connected to the CPU bus 101. . A waveform ROM (Read Only Memory) 106 is connected to the tone generator LSI 104 via a waveform memory bus 105.

  The work RAM 103 is loaded with a processing program for operating the CPU 102. On the other hand, performance data (musical tone waveform control information) is input to the work RAM 103 via the CPU bus 101 from a keyboard (not shown) or the like. The CPU 102 creates sound source parameters (control parameters) based on the performance data and outputs them to the sound source LSI 104. The tone generator LSI 104 reads waveform data as a material from a waveform ROM (Read Only Memory) 106 in accordance with the tone generator parameters, processes the data, and synthesizes a musical sound waveform. The synthesized musical sound waveform is converted into an analog signal by a CODEC (coder / decoder) (not shown) to generate a musical sound signal.

As shown in FIG. 4B, there is a type using a DSP (Digital Signal Processor) 107 for a sound source instead of the sound source LSI.
The DSP is a microprocessor dedicated to digital signal processing. Multiply-and-accumulate operations can be performed at high speed. Further, if the microprogram code is changed, the sound source method can be replaced with a different one, or the sound source method can be upgraded. The musical sound waveform synthesized by the DSP 107 is converted into an analog signal and output by a CODEC (not shown).
The illustrated DSP 107 is of a type in which processing data and a signal processing program to be subjected to signal processing share a bus (DSP bus 108) and work RAM 110.

Here, since both the CPU 102 and the DSP 107 have a function of assigning the right to use the bus, that is, a so-called bus arbitration function, the bus cannot be shared.
In general, it is known to connect two buses with a shared memory (see Patent Document 1). Therefore, also in FIG. 4B, the shared memory 109 is provided and the work RAM 110 is provided on the DSP bus 108.

On the other hand, the waveform data to be stored in the waveform ROM 106 in order to reproduce a high-quality musical sound waveform tends to increase in capacity.
However, storing waveform data of various instrument sounds in a large-capacity waveform ROM has a cost limitation.
Therefore, as shown in FIG. 4C, a waveform RAM 111 is used, and an HDD (Hard Disk Drive) 113 is connected to the CPU bus 101 via an HDC (Hard Disk Controller) 112.

The CPU 102 once transfers the waveform data from the HDD (Hard Disk Drive) 113 to the work RAM 103 before starting the performance, and transfers it to the waveform RAM 111 for storage.
Therefore, it is only necessary to transfer only the waveform data of the required instrument tone color from the HDD 113 to the waveform RAM 111 via the work RAM 103.
However, when switching instrument timbres, the waveform data of the switched instrument timbres must be transferred to the waveform RAM 111, so there is a problem that switching takes time.
There is a limit to the data size that can be stored in the waveform RAM 111 at a reasonable cost.

For this reason, after the performance is started, during the musical tone synthesis process, the waveform data necessary for the musical tone synthesis process is supplied from the HDD 113 to the waveform RAM 111 via the work RAM 103 each time. According to this method, virtually unlimited waveform data can be used.
This method is realized by a “sampler” program executed on a personal computer.
However, in the HDD 113, the delay time (latency) from when the waveform data is requested to read until the waveform data is read cannot be ignored. Therefore, it is necessary to pre-read and store a large amount of waveform data necessary for tone synthesis in the work RAM 103.

The processing of the CPU bus 101 is mainly sound source parameter synthesis processing. This processing is a relatively large-scale program having a complicated processing algorithm, and the processing load is very heavy.
However, the CPU bus for embedded devices is not as fast as the CPU bus of personal computers. The CPU bus speed is a fraction of the internal operation speed. When the CPU 102 performs the busiest processing, the worst value is designed to use about 70 to 80% of the data transfer capacity in the instruction fetch.
Therefore, if the data transfer capability of the CPU bus 101 is consumed for the transfer of waveform data from the HDC 112 to the work RAM 103 and the transfer of waveform data from the work RAM 103, the calculation capability of the CPU 102 cannot be fully exhibited.

By the way, a musical sound data recording / reproducing device, in which the HDD is connected to a bus on the playback side or a bus different from the CPU bus, and transfers waveform data read from the HDD to the bus on the playback side via a shared buffer Is known from US Pat.
FIG. 5 is a block diagram showing such a conventional musical sound data recording / reproducing apparatus. However, the configuration on the HDD recording side is not shown.
The personal computer 121 and the playback control device 122 are connected by a serial cable. The personal computer 121, the reproduction control device 122, and the HDD 123 are connected by a SCSI (Small Computer System Interface) bus.
The personal computer 121 performs a file name selection / setting operation, and sends the area instruction information of the data file to the reproduction control device 122 using a serial cable.
The reproduction control device 122 reads the musical sound waveform data stored in the HDD 123 via the SCSI bus and reproduces the musical sound waveform data.

In the reproduction control device 122, a serial input / output interface 124, a SCSI interface 129, a reproduction / transfer control unit 130, and a reproduction sound generation unit 131 are connected to the CPU bus 125 together with a microcomputer including a CPU 126, a RAM 127, and a ROM 128. These operations are controlled by the CPU 126.
The reproduction and transfer control unit 130 includes a capture buffer 134, which is connected to the SCSI interface 129 via the waveform data bus 132 and is connected to the reproduction buffer 135 via the bus 133.

The musical sound waveform data read from the HDD 123 is stored in the capture buffer 134. The capture buffer 134 removes the header part and rearranges the data, and transfers the sound waveform data for one block (16 kilowords) suitable for reproduction and reading to the reproduction buffer (4 kilowords × 2) 135 Is done. Writing / reading to / from the capture buffer 134 is switched depending on whether the system clock pulse is 1 or 0.
The reproduction sound generator 131 reads out the musical sound waveform data for one block from the reproduction buffer 135 in accordance with the sampling clock and outputs it. At that time, the pitch of the readout waveform can be controlled.

As disclosed in Patent Document 2 described above, according to the configuration in which the capture buffer 134 is provided between the waveform data bus 132 and the waveform reproduction bus 133, the transfer of musical sound waveforms in each bus can be performed independently.
However, the transfer function disclosed in Patent Document 2 is to read and reproduce the waveform data stored in the HDD 123 by a predetermined block size.
Accordingly, when the amount of waveform data necessary for the musical sound waveform synthesis processing varies according to the time variation factor such as performance data (musical sound waveform control information), the waveform data is read from a storage device having a large latency such as the HDD 123. At this time, it is not described how the transfer efficiency is improved by performing the transfer in each bus.
JP-A-6-59678 JP-A-6-51776

  The present invention has been made to solve the above-described problems, and an object of the present invention is to provide a sound waveform synthesizer that can read waveform data from a storage medium and efficiently transfer the waveform data to waveform data processing means. It is what.

In the invention according to claim 1, in the sound waveform synthesizer that inputs sound waveform control information from the outside and performs sound wave synthesis based on waveform data stored in a large-capacity storage medium, and data processing means, and the waveform data supply means, and the waveform buffer, a control data processing unit, the waveform buffer is connected between the waveform data processing section and the waveform data supplying means, before Symbol control data The processing means inputs the sound waveform control information from the outside , and stores the waveform data predicted to be newly required in the subsequent sound waveform synthesis in the waveform data processing means in accordance with the sound waveform control information. Create a read request for reading from the medium, notify the read request to the waveform data supply means, and from the waveform data processing means at the present time Receiving notification of the progress of sound waveform synthesis representing the waveform position where signal processing for waveform synthesis is performed, the waveform data newly required for sound waveform synthesis based on the progress of the sound waveform synthesis A first transfer request is generated to request transfer to the buffer, the first transfer request is notified to the waveform data supply means, and from the waveform data processing means, at the present time, for sound waveform synthesis notified of sound waveform synthesis state representing the contents of which signal processing, control to create a control parameter based the sound waveform control information input from the outside to the sound waveform synthesis conditions, the control parameters, the first a second transfer request according to the transfer request, which notifies the waveform data processing section, the waveform data supplying means, the notification of the first transfer request from the control data processor Only by, a plurality of samples as one frame as a unit frame period, the waveform data read from the storage medium is intended to be transferred to the waveform buffer, the waveform data processing section, from the control data processor In response to the notification of the second transfer request, the waveform data is transferred to itself from the waveform buffer in units of the frame period, and transferred from the control parameter and the waveform buffer acquired from the control data processing means. Based on the obtained waveform data, the sound waveform for one frame is synthesized for each frame, and the control data processing means is notified of the sound waveform synthesis state and the progress of the sound waveform synthesis .
Transfer to the means can be easily realized.

  Therefore, by using the waveform buffer, the sound waveform synthesis processing by the waveform data processing means and the reading of the waveform data from the storage medium by the waveform data supply means are controlled independently even during the sound waveform synthesis processing. The As a result, the waveform data processing means can transfer the waveform data from the waveform buffer at a convenient timing, so that the influence of reading the waveform data from the storage medium on the sound waveform synthesis process can be reduced.

Since the sound waveform data is synthesized in units of a plurality of samples for one frame, the required amount of processing of the waveform data by the waveform data processing means is made uniform in time and does not change much every frame period. It can be operated stably.
Since the waveform data transfer by the waveform data processing means and the waveform data transfer by the waveform data supply means are in units of one frame period, the amount of transfer at one time is a certain amount, so there is little overhead. Efficient transfer is possible.

According to a second aspect of the present invention, in the sound waveform synthesizer according to the first aspect, the waveform data processing means includes a first bus, a signal processing means connected to the first bus, and a first bus. The waveform data supply means includes a second bus, a storage medium reading means connected to the second bus, and a second storage means, and the waveform buffer includes the second buffer Connected between one bus and the second bus, the waveform data stored in the second storage means is supplied to the first storage means.
Therefore, the waveform data can be easily transferred from the waveform data supply means to the waveform data processing means via the waveform buffer.

According to the present invention, the waveform buffer can be used to read waveform data from a large-capacity storage device having a large latency such as a hard disk and transfer it to the waveform data processing means even during a sound generation operation, so that high quality even in real time. There is an effect that can generate the synthesized sound.
At that time, since the waveform data is transferred in units of one frame period, the data transfer capability can be used efficiently.
As a result, it is possible to prevent a decrease in processing capability of the waveform data processing means and to obtain a practically sufficient number of simultaneous pronunciations.

FIG. 1 is a block diagram illustrating the functional configuration of an embodiment of the present invention.
In a sound waveform synthesizing apparatus such as an electronic music apparatus that performs sound wave synthesizing processing based on sound waveform control information and waveform data stored in a storage medium, 1 is a control data processing unit, 2 is a waveform data processing unit, 3 Is a waveform data supply unit, and 5 is an interface unit.
A waveform data storage device 4 such as an HDD is connected to the waveform data supply unit 3.

The interface unit 5 connects the control data processing unit 1, the waveform data processing unit 2, and the waveform data supply unit 3 to each other. The waveform buffer 6 connects the waveform data processing unit 2 and the waveform data supply unit 3, and performs waveform data transfer buffer relay, thereby synthesizing the sound waveform by the waveform data processing unit 2 and the waveform data supply unit 3. The reading of the waveform data can be individually processed without interfering with each other.
On the other hand, since various methods can be adopted for the interface between the control data processing unit 1 and the waveform data processing unit 2 and between the control data processing unit 1 and the waveform data supply unit 3, in the illustrated example, Directly connected.
The control data processing unit 1, the waveform data processing unit 2, and the waveform data supply unit 3 autonomously perform each internal process independently of the internal processes of other processing units. However, in the illustrated example, the overall integration operation is performed in units of frame periods corresponding to a predetermined plurality of sample periods.

The waveform data processing unit 2 includes a signal processing bus (first bus) 20, and a signal processing unit 21 and a storage unit (first storage unit) 22 are connected to the signal processing bus 20. The signal processing unit 21 further includes a transfer control unit (first transfer control unit) 23. The output of the signal processing unit 21 is connected to a sound waveform output unit (not shown).
The waveform data processing unit 2 synthesizes a sound waveform for one frame for each frame based on the acquired control parameter and the transferred waveform data, and the sound waveforms of the synthesized samples are output as a sound waveform. The sound waveform is output one sample at every sampling period.
In order to perform sound waveform synthesis for each frame, a signal synchronized with a frame corresponding to a predetermined plurality of sampling clocks may be supplied from a sound waveform output unit (not shown).

The waveform data processing unit 2 acquires from the control data processing unit 1 control parameters as much as necessary to synthesize a sound waveform in at least the current frame period. At the same time, the transfer control unit 23 transfers waveform data from the waveform buffer 6 to the storage unit 22 as much as necessary to synthesize a sound waveform in the current frame period.
The acquisition of the control parameters and the transfer of the waveform data described above can be performed in units of frame periods in accordance with the sound waveform synthesis process.
The unit of frame period means a time interval that is an integral multiple of one frame period. From time to time, the value of an integer multiple may change.

Note that the waveform data processing unit 2 may perform sound waveform synthesis for each sample of sound waves. However, when synthesizing sound waveforms, the number of waveform data samples to be used varies greatly depending on the situation such as pitch. For this reason, there is a case where the waveform data to be temporarily supplied is insufficient and noise is generated in the synthesized waveform. Therefore, it is preferable for the signal processing unit 21 to perform processing in units of one frame of a uniform size because fluctuations in necessary waveform data are alleviated.
In addition, when sound waveform synthesis is performed for each sample of a sound waveform, waveform data may have to be transferred one sample at a time. Then, overhead is required for each transfer of one sample. On the other hand, if sound waveform synthesis is performed for each frame, the waveform data can be transferred on the signal processing bus 20 and the waveform data bus 30 in units of frame periods, so that it can be transferred efficiently at high speed. Become.
Since the waveform data processing unit 2 synthesizes the sound waveform every frame, a delay of one frame occurs. However, there is no practical problem unless one frame period is extremely long.

A specific configuration of the waveform data processing unit 2 will be described.
The signal processing unit 21 synthesizes the sound waveform for each frame based on the control parameter and waveform data stored in the storage unit 22, and the control data processing unit 1 indicates the sound waveform synthesis state and the progress of sound waveform synthesis. Notice.
In response to the notification of the waveform data transfer request (second transfer request) from the control data processing unit 1, the transfer control unit 23 transfers the waveform data from the waveform buffer 6 to the storage unit 22.

The signal processing unit 21 can perform processing inside the signal processing unit even while the waveform data is being transferred on the signal processing bus 20.
The sound wave forms of the plurality of samples synthesized by the signal processing unit 21 are supplied to a sound wave form output unit (not shown), and one sample is output for each sampling period.
The notification of the sound waveform synthesis state and the progress of sound waveform synthesis can be performed in units of frame periods in conjunction with the sound waveform synthesis processing.

The control data processing unit 1 has a control parameter creation unit 11 and a request creation unit 12, and inputs sound waveform control information. In the case of an electronic music apparatus, the input sound waveform control information is performance information, which is output from the keyboard or read from the song data storage unit.
In the waveform data processing unit 2 described above, the waveform data transferred from the waveform buffer 6 is sequentially used as the sound waveform synthesis of one sound progresses. Therefore, the waveform data of the subsequent waveform position of one sound is newly required sequentially. And Further, when the pronunciation of a different sound is instructed, the waveform data that has been used until then becomes unnecessary, and transfer of the waveform data of a new sound is required.

Therefore, the request creation unit 12 of the control data processing unit 1 receives a notification of the progress status of the sound waveform synthesis from the signal processing unit 21 and is acquired by the waveform data processing unit 2 based on the progress status of the sound waveform synthesis. Then, the remaining amount (remaining amount) of the waveform data necessary for the synthesis of the sound waveform stored in the storage unit 22 is calculated. In response to the calculation result, the waveform data supply unit 3 is notified of a first transfer request for requesting waveform data newly required for synthesis of the sound waveform. This first transfer request includes information specifying the portion of the waveform data to be supplemented.
In response to the first transfer request, the request creating unit 12 notifies the waveform data processing unit 2 of the second transfer request, thereby causing the signal processing unit 21 to transfer the waveform data from the waveform buffer 6.

The notification of the first transfer request and the second transfer request can be performed in units of frame periods in conjunction with the sound waveform synthesis process. The waveform data processing unit 2 may receive a signal synchronized with the frame.
When the waveform data processing unit 2 does not need to transfer new waveform data, the first transfer request and the second transfer request are not notified.
The first transfer request and the second transfer request are notified in accordance with the remaining amount of waveform data necessary for sound waveform synthesis in the storage unit 22 being, for example, a predetermined amount or less. Also good.
Further, instead of the remaining amount of waveform data described above, the first transfer request and the second transfer request may be notified according to the consumption amount of the waveform data consumed by the synthesis of the sound waveform.

The request creation unit 12 stores the waveform data predicted to be newly required in the subsequent sound waveform synthesis based on the input sound waveform control information so as to be read from the waveform data storage device 4 in advance. A read request is notified to the medium read control unit 31. The waveform data that can be read is less than the free space after the waveform data is transferred from the storage unit 32 to the waveform buffer 6.
On the other hand, the control parameter creation unit 11 creates a control parameter based on the sound waveform control information and the sound waveform synthesis state, and notifies the waveform data processing unit 2 of the control parameter. Notification of control parameters can also be performed in units of frame periods in conjunction with sound waveform synthesis processing.

In response to the notification of the first transfer request, the waveform data supply unit 3 transfers the waveform data newly required for the synthesis of the sound waveform from the storage unit 32 to the waveform buffer 6.
The waveform data processing unit 2 receives the notification of the second transfer request and transfers the waveform data from the waveform buffer 6 to the storage unit 22.
A specific configuration of the waveform data supply unit 3 will be described.
The waveform data supply unit 3 includes a waveform data bus (second bus) 30. The waveform data bus 30 includes a storage medium read control unit 31, a storage unit (second storage unit) 32, and transfer control. Unit (second transfer control unit) 33 is connected.

In response to the notification of the first transfer request described above, the transfer control unit 33 transfers at least part of the waveform data stored in the storage unit 32 from the storage unit 32 to the waveform buffer 6.
The storage medium read control unit 31 described above receives the notification of the read request described above, reads the waveform data from the storage medium of the waveform data storage device 4, and transfers it to the storage unit 32.
In the waveform data bus 30, the waveform data transferred from the storage unit 32 is transferred to the storage unit 32 so that the waveform data to be transferred to the waveform data processing unit 2 via the waveform buffer 6 is not deficient in the sound waveform synthesis processing. The bus arbitration is performed so that it is performed with a higher priority than the transfer of the waveform data to the.
Also in the waveform data bus 30, if the waveform data is transferred in units of frames, burst mode transfer can be performed, so that transfer can be efficiently performed at high speed. Since the first transfer request described above is notified in units of frames, it is sufficient to respond immediately. Alternatively, a signal synchronized with the frame may be received from the waveform data processing unit 2 or the like.

The control data processing unit 1 notifies the storage medium read control unit 31 of a read request before starting the input of the above-described sound waveform control information, so that the sound waveform control is performed from the storage medium of the waveform data storage device 4. Waveform data predicted to be required at the start of information input may be transferred to the storage unit 32 in advance.
When the input of the sound waveform control information is started, the waveform data necessary for the sound waveform synthesis after the start can be transferred from the storage unit 32 to the storage unit 22 via the waveform buffer 6 in a short time.

  When the waveform data processing unit 2 performs sound waveform synthesis simultaneously for a plurality of channels, the control data processing unit 1 is notified of the sound waveform synthesis state and the progress status of the sound waveform synthesis for each channel. . The control parameter creation unit 11 notifies the waveform data processing unit of the control parameters for each channel, and the request creation unit 12 also notifies the first request and the second request for each channel. The waveform data supply unit 3 reads the waveform data for each channel and supplies the waveform data to the waveform buffer 6 for each channel.

In the above description, the control parameter, the second transfer request, the sound waveform synthesis state, the progress status of the sound waveform synthesis, and the first transfer request are notified in units of the frame period in conjunction with the sound waveform synthesis process. Explained. However, it is not always necessary to notify the frame period as a unit.
On the other hand, regarding the transfer of the waveform data from the waveform data supply unit 3 to the waveform buffer 6 and the transfer of the waveform data from the waveform buffer 6 to the waveform data processing unit 2, it is preferable to transfer in units of frame periods. . When the transfer is performed in units of frame periods, the waveform data transferred at one time has a certain size, so that the overhead is relatively small and burst mode transfer is possible, so that transfer efficiency is improved.

FIG. 2 is a block diagram showing a specific example in which the embodiment shown in FIG. 1 is applied to a waveform memory sound source type electronic music apparatus. In the figure, the same parts as those in FIG.
This is an electronic music apparatus having a three-bus configuration comprising a control data processing CPU bus 41, a DSP bus 42 corresponding to the signal processing bus 20 of FIG. 1, and an HDD bus 43 corresponding to the waveform data bus 30 of FIG. By transferring data between the buses via the shared memory 45 of the interface unit 5, the occurrence of access contention on one bus is eliminated and the congestion of the bus is alleviated.

The control data processing unit 1, the waveform data processing unit 2, and the waveform data supply unit 3 are operated by the interface unit 5 in units of frame periods.
The interface unit 5 includes a shared memory 45 and a hardware interrupt line for IRQ # 2. On the other hand, IRQ # 1 is a hardware interrupt from the CODEC 52 to the DSP 50 and is generated every 64 samples (one frame). IRQ # 1 is a trigger for DSP 50 frame processing. On the other hand, IRQ # 2 is a hardware interrupt from the DSP 50 to the CPU 46 and occurs every frame, but occurs at a timing different from that of IRQ # 1 and serves as a trigger for frame processing of the CPU 46.

The control data processing unit 1 includes a CPU bus 41, which includes the CPU 46, work RAM 47, performance information (sound waveform control information) for realizing the functions of the control parameter creation unit 11 and the request creation unit 12 shown in FIG. Is connected to a MIDI (Musical Instrument Digital Interface) interface 48, a boot ROM 49 for loading a CPU control program into the work RAM 47 when the power is turned on. If the CPU 46 has a MIDI interface, it can be replaced with a MIDI interface 48.
Blocks such as an input operator, a display, an input / output interface, and an effector LSI that are not directly related to the present invention are not shown.

The CPU 46 reads the processing program from the work RAM 47, reads / writes the work RAM 47 for storing temporary data, and reads / writes the shared memory 45 as necessary. The CPU 46 occupies the CPU bus 41.
The data transfer capability of the CPU bus 41 is mainly used for reading (instruction fetch) into the CPU 46 when a processing program on the work RAM 47 is executed.

The waveform data processing unit 2 is connected to the DSP bus 42 with a DSP 50 that functions as the signal processing unit 50 in FIG. 1, a work RAM 51 that functions as the storage unit 22, a boot ROM 53 that loads a control program into the work RAM 51 when the power is turned on, and the like. ing. Connected to the DSP 50 is a CODEC 52 that functions as a sound waveform output unit.
The DSP 50 reads the control program from the work RAM 51, reads / writes the work RAM 51 for temporary data storage, and reads / writes the shared memory 45 as necessary. The DSP 50 occupies the DSP bus 42.

The musical sound waveform synthesis processing by the DSP 50 has a relatively simple processing algorithm based on repetitive processing (loop processing). In particular, since a program corresponding to repetitive processing is a small program that can be accommodated in an instruction cache in the DSP 50, the instruction fetch hardly uses the bandwidth of the DSP bus 42.
Most of the transfer capability is used to transfer waveform data from the waveform buffer shared memory 58 to the work RAM 51 and to transfer processing data between the DSP 50 and the work RAM 51.

The work RAM 51 stores the number of temporary waveform data corresponding to the maximum number of simultaneous sound generation channels for each channel. Each channel is configured by a buffer capable of storing “management information” + “partial waveform for 4096 samples”.
Digital output is also possible without going through the CODEC 52. When the effector LSI is connected to the CPU bus 41 and controlled by the CPU 46, the musical sound waveform data synthesized by the DSP 50 is supplied to the input terminal of the effector LSI, and after the effect is added, it is D / A converted. Is output.

On the other hand, the waveform data supply unit 3 provides the HDD bus 43 with the functions of the HDC 59 for realizing the function of the recording medium reading control unit 31 shown in FIG. 1, the work RAM 60 for realizing the function of the storage unit 32, and the transfer control unit 33. A dynamic memory access controller (DMAC) 61 to be realized is connected.
As the waveform data storage device 4 shown in FIG. 1, an HDD 44 having a large delay time (latency) but capable of storing a large amount of waveform data is used.
The HDD 44 can store not only large-volume waveform data but also high-speed data transfer. A recent HDD 44 generally has a performance of about 30 to 40 Mbyte / sec at the time of continuous reading (sequential access) and about 10 to 20 Mbyte / sec at the time of discontinuous reading (random access).
However, the delay time from when the read is requested to when the read is started is very large, about several tens of ms. However, in order to generate a musical sound in real time, it is necessary to keep the delay time until the sound is generated within about 10 ms.

Therefore, the waveform data is read from the HDD 44 and temporarily stored in the work RAM 60.
The transfer on the HDD bus 43 is mainly a transfer of waveform data from the HDD 44 to the work RAM 60 and a transfer of waveform data from the work RAM 60 to the waveform buffer 58.
At least a part of the waveform data is transferred twice on the HDD bus 43 with different transfer periods.
Since what is transferred from the HDD 44 to the work RAM 60 is waveform data that is predicted to be newly required, not all of them are transferred from the work RAM 60 to the waveform buffer 58. Therefore, when the transfer amount per unit time (bus bandwidth usage amount) is compared, the transfer amount per unit time of the waveform data from the work RAM 60 to the waveform buffer 58 is larger than that of the waveform data from the HDD 44 to the work RAM 60. It is a small one.
In addition, the number of transfers from the work RAM 60 to the waveform buffer 58 per unit time (for example, once per frame) is greater than the number of transfers from the HDD 44 to the work RAM 60 per unit time. The transfer amount is small, and the transfer is small.
As a result, by providing the DSP bus 42 and the HDD bus 43, the burden on the DSP bus 42 for waveform data transfer is reduced.

The work RAM 60, for example, at the time of system start-up, for example, reads the head part of a musical sound waveform that may be sounded (for example, 0.5 seconds), all key codes of all tones (all on the keyboard to be played) This is realized by a number of ring buffers corresponding to the maximum number of simultaneous sounding channels, together with a memory that stores statically for each key pitch). Note that the memory for static storage may be common to each channel.
The storage capacity of the ring buffer is constant, and the read address and write address move cyclically. Old waveform data is overwritten in order by newly written waveform data. Each ring buffer stores, for example, 1.5 seconds of waveform data.
The management information of the work RAM 60 is held by the CPU 46 on the control data processing unit 1 side.

The shared memory 45 is connected between two buses, logically couples the two buses and is accessible from either bus, and serves as a window for transferring data. Asynchronous writing / reading can be performed on the shared memory 45 at any timing from each bus.
To realize shared memory, use dual port RAM, use FIFO (First In First Out), or connect RAM to two buses alternately by high-speed clock. You may use time division by two buses. The shared memory 45 is realized by dedicated hardware. The shared memory 45 can also realize a function with an FPGA (Field Programmable Gate Array). At this time, the function of the DMAC 61 can be incorporated.

In addition to the transfer of the waveform data by the waveform buffer 6, the interface unit 5 is a shared memory for the transfer of the sound source parameters and the sound source state, the transfer of the read request and the transfer request, and the waveform buffer shared memory 58 and the sound source parameter shared memory 54. The sound source state shared memory 55, the read request shared memory 56, and the transfer request shared memory 57 are implemented.
The sound source parameter shared memory 54 and the sound source state shared memory 55 are connected to the CPU bus 41 and the DSP bus 42, and the read request shared memory 56 and the transfer request shared memory 57 are connected to the CPU bus 41 and the HDD bus 43 with waveforms. The buffer shared memory 58 is connected to the HDD bus 43 and the DSP bus 42.
The waveform buffer shared memory 58 is realized by a ring buffer and holds a part of the waveform data stored in the work RAM 60.
The management information of the waveform buffer shared memory 58 is held by the CPU 46. Only waveform data is stored on the ring buffer.

Here, the sound source parameters will be described.
“Sound source parameter” in FIG. 2 is “normal transfer source parameter” (control parameter in FIG. 1) and “transfer from waveform buffer shared memory 58 to work RAM 51” (second transfer in FIG. 1). Parameter) indicating request).
“Normal sound source parameters” include key-on, note number, envelope level, and the like.
The sound source parameters are rewritten in units of frame periods, and are written in the format of “packet size + packet body (16 bits × n)” (n is a positive integer) from the head of the sound source parameter sharing memory 54. If there is no sound source parameter to be transferred in a certain frame period, the packet size is written as 0. A plurality of sound source parameters can be stored in the packet body described above.

The “transfer request” will be described.
When a transfer from the work RAM 60 to the waveform buffer shared memory 58 is requested, this “transfer request” is notified.
It consists of “transfer start address on work RAM 60” + “transfer size” + “transfer destination sound channel”.

The “sound source state” will be described.
The “sound source state” as shown in FIG. 2 includes “normal sound source state” (sound waveform synthesis state as shown in FIG. 1) of each tone generation channel, and “progression state of musical sound waveform synthesis” (as shown in FIG. Information indicating the progress of shape synthesis).
The “normal sound source state” refers to a key-on state, a current state of an amplitude envelope, and the like as defined in a normal sound source device.
On the other hand, the information indicating the “progress status of the musical sound waveform synthesis” is, for example, data on the waveform position currently processed in the musical sound synthesis. The waveform position is a sound generation position with the position at the time of note-on as the reference point 0. For example, it is expressed by the cumulative number of samples up to the present time.
The sound source state sharing memory 55 is a map showing the state of each sound source for each sound generation channel that can be sounded simultaneously, and the memory address is determined in advance.

FIG. 3 is a sequence diagram illustrating an example of processing timing of each unit in the specific example illustrated in FIG.
It can be divided into processing synchronized with MIDI events, processing related to HDD reading, processing synchronized with IRQ # 2, processing synchronized with IRQ # 1, and CODEC processing.
In the figure, processes synchronized with MIDI events, HDD-related processes, and other processes have different time units.
Musical sound waveform synthesis in the DSP 50 performs signal processing in units of frame periods. If the sampling clock is 44.1 kHz and 64 samples are one frame, one frame period is about 1.45 msec.

It is assumed that programs and data necessary for the operation of the CPU 46 are stored in the Boot ROM 49, and are transferred to the work RAM 47 and started to be executed when the system is activated.
The same applies to the DSP 50. Programs and data necessary for the operation of the DSP 50 are stored in the Boot ROM 53, transferred to the work RAM 51 and started to be executed when the system is activated.
When generating a musical sound in real time, a read request is notified from the CPU 46 in advance at the time of system startup or the like, so that the waveform of the head portion of all waveform data that may be used for sound generation is obtained from the HDD 44. The waveform data is transferred to the RAM 60 via the HDC 59, and these waveform data are held in the RAM 60 during the musical tone synthesis process.
From the silent state, the performance information “Note On” is received and sound generation for one sound is started. During the middle, the performance information “Expression” is received and the volume changes. Finally, the performance information “Note Off” An example of simple performance information in which the pronunciation stops upon reception of a song will be described.

As shown in FIG. 3A, performance information (sound waveform control information) 71, 75, 77 is received as MIDI data and input to the work RAM 47.
As shown in FIG. 3B, the CPU 46 performs sound source parameter creation processing using events such as MIDI note-on, note-off, control change, and pitch bend as triggers.
Although illustration is omitted, the sound source parameter is also generated by a timer interrupt of 5 to 20 msec during sound generation.
The sound source parameters are created based on performance data such as MIDI, and the “normal sound source state” (sound waveform synthesis state) such as the key-on state and the current state of the amplitude envelope already described.
The CPU 46 creates sound source parameters 72, 76, and 78 on the work RAM 47. The time required for this creation process is not constant.
The CPU 46 performs a read control process of the HDD 44. Based on the received MIDI data, the waveform data newly required for the subsequent musical sound waveform synthesis is predicted.

As shown in FIG. 3C, a read request 73 for reading the waveform data predicted to be newly required from the HDD 44 to the work RAM 60 is transferred from the work RAM 47 to the read request shared memory 56.
For example, if the sound generation of a certain pitch is started by receiving MIDI data, the waveform data of the sound waveform of the generated pitch, and the pitch bend operation may be performed from this pitch and the pitch may change. Musical tone waveform data of all pitches in a certain range (for example, from the pitch that is currently sounding to 2 octaves lower to 2 octaves higher) are sequentially read out from the HDD 44 and stored in the work RAM 60 in advance.

As shown in FIG. 3D, the HDC 59 accesses the read request shared memory 56 and reads the stored read request notification 74.
As shown in FIG. 3E, the HDC 59 reads waveform data based on the read request from the HDD 44 and transfers it to the work RAM 60.
Until the request 79 is completed, the task of the CPU 46 shown in FIG. 3C is blocked. Therefore, when the reading of the waveform data is completed, the CPU 46 can transfer the next read request 80 to the read request shared memory 56.

Next, processing synchronized with IRQ # 2 will be described.
As shown in FIG. 3 (k), an interrupt signal (IRQ # 2) 91 is notified from the DSP 50 to the CPU 46.
As shown in FIG. 3 (f), the CPU 46 sends the sound source state (“normal sound source state” and musical tone waveform in the current frame (1) from the sound source state shared memory 55 at the timing of the interrupt signal (IRQ # 2) 91. The synthesis progress status) 92 is read out and written into the work RAM 47.
The CPU 46 calculates the waveform data necessary for the DSP 50 to synthesize the musical sound waveform in the frames after the next frame (2) based on the sound source state (musical sound waveform synthesis progress status) 92 in the current frame (1). The amount of data (remaining amount) remaining in the work RAM 51 is calculated, and if there is newly required waveform data, it is determined to transfer this.

As shown in FIG. 3G, the CPU 46 notifies the transfer request shared memory 57 of a “transfer request” 93 that requests transfer of waveform data from the work RAM 60 to the waveform buffer 58. The waveform data to be transferred corresponds to the waveform data to be transferred to the work RAM 51 later.
As shown in FIG. 3H, the DMAC 61 on the waveform data supply unit 3 side reads the transfer request shared memory 57 and acquires a “transfer request” 94. In general, the DMAC itself does not access the memory and read data, but here, using a specialized DMAC 61 or by providing a function for notifying the DMAC 61 on the shared memory 45 side. As a result, the DMAC 61 obtains a “transfer request”.

As shown in FIG. 3I, the DMAC 61 transfers the waveform data from the work RAM 60 to the waveform buffer shared memory 58. Until the transfer notified by the “transfer request” is completed, the transfer request task shown in FIG. 3G is blocked.
As shown in FIG. 3 (j), the CPU 46 creates sound source parameters based on the performance information and “normal sound source state” (sound waveform synthesis state), and uses the sound source parameters stored in the work RAM 47 as the sound source parameters. Transfer to the shared memory 54.

Next, processing synchronized with IRQ # 1 will be described.
As shown in FIG. 3 (p), the interrupt signals (IRQ # 1) 81, 86,... Are generated every frame and supplied to the DSP 50.
As shown in FIG. 3L, the DSP 50 transfers the sound source parameters 82, 87,... Transferred to the sound source parameter sharing memory 54 to the work RAM 51.
The sound source parameters may be transferred to the internal RAM of the DSP 50 instead of being transferred to the work RAM 51 because the data size is small.

As shown in FIG. 3 (m), the DSP 50 transfers the latest sound source states 83, 88,... In the current frame (1) from the work RAM 51 to the sound source state shared memory 55. As described above, the transferred sound source state is transferred to the RAM 47 on the control data processing unit 1 side at the timing shown in FIG.
Since the sound source state has a small data size, it may be transferred from the internal RAM of the DSP 50 instead of the work RAM 51.
As shown in FIG. 3 (n), the waveform data 84 and 89 stored in the waveform buffer shared memory 58 are transferred to the work RAM 51. This transfer control may be directly executed by the main body of the DSP 50, or when the DMAC (transfer control unit 23) is built in the DSP 50 as in the present embodiment, this built-in DMAC is connected to the DSP 50. Transfer control may be performed according to an instruction from the main body.

Here, the waveform data stored in the waveform buffer shared memory 58 is the waveform data shown in FIG. 3E read to the work RAM 60 in response to the read request 73 shown in FIG. (However, in FIG. 3, the waveform data by reading (not shown) before the time of frame (1)) is stored in the waveform buffer shared memory 58 at the timing of the waveform data 95 in FIG. 3 (i). is there.
That is, of the waveform data stored in advance in the work RAM 60, the waveform data necessary for the next frame (2) and thereafter is the latest sound source in the current frame (1) shown in FIG. It is determined based on the state (progress status of musical sound waveform synthesis) 92 and stored in the waveform buffer shared memory 58.

As a result, in FIG. 3 (n), the waveform data 84 transferred to the work RAM 51 is the waveform data required after the next frame (2).
In addition, during the period of frame (2), it was determined that transfer of new waveform data was not necessary, so that no waveform data transfer request was created as shown in FIG. Therefore, the waveform data is not transferred to the waveform buffer shared memory 58 as shown in FIG. 3 (i), and the waveform data is transferred to the work RAM 51 as shown in FIG. 3 (n) during the period of the frame (3). Not.

As shown in FIG. 3 (o), the DSP 50 reads the sound source parameters (normal sound source parameters) 82 and the waveform data 84 stored in the work RAM 51, and on the RAM 51 based on these, the next frame (2) is read out. Performs musical tone synthesis for 64 samples.
The musical tone synthesis process for one frame is performed in order for each of the sound generation channels that can be produced simultaneously, and the musical tone waveform of the newly synthesized musical tone channel is added to the accumulated waveform of the channel that has already been synthesized.
The CODEC 52 D / A converts the musical tone synthesis waveform for one frame stored in the work RAM 51 one sample at a time and outputs a synthesized tone.

  In the above description, on the DSP bus 42, as shown in FIGS. 3 (l) and 3 (n), the waveform data is transferred from the waveform buffer shared memory 58 to the work RAM 51, and the DSP 50 is connected to the work RAM 51. The period during which data for musical tone waveform synthesis is transferred between the DSPs 50 is sequentially performed by the bus arbitration by the DSP 50, and the transfer periods do not overlap.

On the other hand, on the HDD bus, the transfer from the work RAM 60 to the waveform buffer shared memory 58 by the DMAC 61 shown in FIG. 3 (i) is the waveform data from the HDC 59 to the work RAM 60 shown in FIG. 3 (e). Control is performed so that it is performed with higher priority than transfer.
As a result, when there is a conflict, the bus arbitration control is performed so that the waveform data transfer from the HDC 59 to the work RAM 60 is interrupted and the transfer from the work RAM 60 to the waveform buffer shared memory 58 is performed.
This priority control prevents the reading process from the HDD 44 from affecting the performance of the musical tone waveform synthesis process by the waveform data processing unit 2.

In FIG. 2, if the waveform buffer shared memory 58 is not provided and the HDD bus 43 is directly connected to the DSP bus 42 and bus arbitration is performed by the DSP 50, the predicted waveform data transferred from the HDC 59 to the work RAM 60 Among them, at least a part of the waveform data actually required is transferred from the work RAM 60 to the work RAM 51 again. Therefore, waveform data that is twice as large as the predetermined transfer amount flows through the DSP bus 42. Since the DSP bus 42 originally transfers a considerable amount of data, the transfer on the DSP bus 42 is congested, and the tone waveform synthesis processing of the DSP 50 is hindered by the transfer delay.
On the other hand, in the configuration provided with the waveform buffer shared memory 58 as shown in FIG. 2, the transfer capability of the DSP bus 42 is not impaired.

In the above description, the readout request shown in FIG. 3C is made in synchronization with the completion of the waveform data readout processing shown in FIG. Instead, a shared memory for storing a read completion notification is newly provided in the shared memory 45, and a read completion notification is notified from the HDC 59 to the CPU 46 via this shared memory, whereby the read processing and the read request are asynchronously performed. It may be operated with.
Similarly, for the waveform transfer request shown in FIG. 3G, the shared memory 45 is newly provided with a shared memory for storing the waveform transfer completion notification, and the waveform transfer completion notification is sent from the HDC 59 via this shared memory to the CPU 46. In this case, the waveform transfer process and the waveform transfer request may be operated asynchronously.

In the above description, as shown in FIGS. 3C and 3D, the read request shared memory 56 is used. Instead, an interrupt signal may be used to notify the HDC 59 of a read request from the CPU 46.
Similarly, as shown in FIGS. 3G and 3H, instead of using the transfer request shared memory 57, an interrupt signal may be used to notify a transfer request.

In the above description, the shared memory is used for notification of the sound source parameters and the sound source state. Instead, if the DSP 42 is provided with an I / O port connected to the CPU bus 41 as in the reproduced sound generating unit 131 shown in FIG. Since they are connected to both, the sound source parameter sharing memory 54 and the sound source state sharing memory 55 are not required.
As for the notification of the read request and the transfer request, if the HDC 59 and the DMAC 61 are provided with an I / O port connected to the CPU bus 41 as in the SCSI interface 129 shown in FIG. Does not require memory.

In the above description, the processing program necessary for the operation of the CPU 46 and the processing program necessary for the operation of the DSP 50 are stored in the boot ROMs 49 and 53, respectively, and are loaded into the work RAMs 47 and 51 when the system is activated. explained.
Instead, only the boot loader may be stored in the boot ROMs 49 and 53, and the processing program may be stored in the HDD 44. At the time of system startup, the boot loader may read a program from the HDD 44 and load it into the work RAMs 47 and 51.
In order to load a program from the HDD 44, a data buffer for program transfer may be provided in the shared memory 45.

In the above description, the HDD has been exemplified as the waveform data storage device. However, even a storage medium that takes time to read may be a large-capacity storage device. For example, a playback device for a storage medium such as a CD-ROM (Compact Disc Read Only Memory), an MO (Magneto Optical Disk), a DVD (Digital Versatile Disk), or a low-speed RAM with a large access delay time may be used. USB2.0 Flash Memory may be used.
In the above description, a specific example has been described in which musical tone information is synthesized by processing performance information such as MIDI messages input in real time.

However, it is also possible to synthesize musical tone waveforms by playing back music data such as SMF (Standard MIDI File) stored in the storage device. In this case, once converted into real-time MIDI data according to time information (duration information) included in the music data, a musical sound waveform can be synthesized as in the above-described specific example.
Further, the DSP 50 may be notified as a sound source parameter (control parameter) including time information, and the DSP 50 may synthesize a musical sound waveform by interpreting the time information.
In this case, since the pronunciation delay can be specified in advance, the capacity of the work RAM 60 can be reduced.

The above description is based on the assumption that a musical sound waveform is generated by the waveform memory sound source method. However, any sound source method can be applied as long as waveform data is used as a material.
For example, it can be applied to high-quality sound source (AEM: Articulation Element Modeling).
Further, not only a musical sound waveform that simulates a musical sound of a natural musical instrument, but a musical sound waveform that is artificially created may be used. In addition, when synthesizing human voices and singing lyrics or making voice announcements, any method that synthesizes sound waveforms can be similarly applied.
That is, the present invention can be applied when synthesizing general sounds including musical tone signals, human voice sounds, and the like.

It is a block diagram explaining the function structure about one Embodiment of this invention. It is a block diagram which shows the specific example which applied one Embodiment shown in FIG. 1 to the electronic music apparatus of a waveform memory sound source system. It is a sequence diagram explaining an example of the processing timing of each part in the specific example shown in FIG. It is a block diagram which shows the electronic music apparatus of a waveform memory sound source system. It is a block diagram which shows the conventional musical sound data recording / reproducing apparatus.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Control data processing part, 2 ... Waveform data processing part, 3 ... Waveform data supply part, 4 ... Waveform data storage device, 5 ... Interface part, 6 ... Waveform buffer

Claims (2)

In a sound waveform synthesizer that inputs sound waveform control information from the outside and performs sound waveform synthesis based on waveform data stored in a large-capacity storage medium.
A waveform data processing means, a waveform data supply means, a waveform buffer, and a control data processing means;
The waveform buffer is connected between the waveform data processing means and the waveform data supply means ,
Before Symbol control data processing means,
The sound waveform control information is input from the outside , and waveform data predicted to be newly required in subsequent sound waveform synthesis in the waveform data processing means is read from the storage medium in accordance with the sound waveform control information. Creating a read request to notify the waveform data supply means of the read request, and
The waveform data processing means is notified of the progress status of the sound waveform synthesis representing the waveform position that is currently being processed for sound waveform synthesis, and the sound waveform synthesis is performed based on the progress status of the sound waveform synthesis. Creating a first transfer request for requesting transfer of newly required waveform data to the waveform buffer, notifying the waveform data supply means of the first transfer request; and
Wherein the waveform data processing section receives a notification of the sound waveform synthesis state representing the content of signal processing which is due to the sound waveform synthesis at the present time, on the basis of the sound waveform synthesis conditions and sound waveform control information input from the external create a control parameter, which the control parameter, the second transfer request corresponding to the first transfer request, notifying the waveform data processing section,
The waveform data supply means includes
In response to the notification of the first transfer request from the control data processing means, the waveform data read from the storage medium is transferred to the waveform buffer with a plurality of samples as one frame and a frame period as a unit. Yes ,
The waveform data processing means includes
Upon receiving the notification of the second transfer request from the control data processor, a unit of the frame period, and transfers the waveform data itself from the waveform buffer,
Based on the control parameter acquired from the control data processing means and the waveform data transferred from the waveform buffer, the sound waveform for one frame is synthesized for each frame, and the sound waveform synthesis state and the sound waveform are synthesized. Notifying the control data processing means of the progress of synthesis,
A sound wave synthesizer characterized by the above. The waveform data processing means includes a first bus, signal processing means connected to the first bus, and first storage means,
The waveform data supply means has a second bus, a storage medium reading means connected to the second bus, and a second storage means,
The waveform buffer is connected between the first bus and the second bus, and supplies the waveform data stored in the second storage means to the first storage means.
The sound waveform synthesizer according to claim 1.

Images