EP0448010A2 - Method for generating sound with an electronic musical instrument and electronic musical instrument - Google Patents

Abstract

The invention relates to a method for generating sound with an electronic musical instrument and an electronic musical instrument, in which sound patterns are stored digitally in the form of samples in a preparation section and the samples are read out from the memory device (7) in a sound section, are digitally interpolated and converted from digital to analog in an interpolation filter, with digital interpolation allowing attenuation below the blocking frequency of the interpolation filter and in the preparatory section the earlier frequency components of the sound pattern that are attenuated in the interpolation filter are amplified before storage.

Description

The invention relates to a method for sound generation with an electronic musical instrument, an electronic musical instrument and a device for loading a storage device for an electronic musical instrument.

The musical instrument can be designed, for example, as a keyboard musical instrument (keybord) or as a pure sound generation module (expander), which can be controlled via external signals.

Before sounds are reproduced, that is to say, for example when manufacturing a corresponding musical instrument, sound patterns are stored digitally in the form of sampled values in a storage device. When the sounds are generated, for example during a lecture by a musician, the samples are read from the memory device, processed, converted to digital-analogue and reproduced as sounds via an audio device.

The problem arises that after the digitally stored samples have been converted into analog values, they must be converted back into digital values and back several times in subsequent steps in order to apply digitally calculated effects, for example a digital Hall. In order to avoid the multiple conversion of digital values into analog values and vice versa, all sample values in the system must be available at absolutely the same intervals. However, if all sound patterns are sampled and stored with the same sampling rate, the so-called system sampling rate, which applies to the entire musical instrument, these sound patterns cannot be reproduced at other frequencies. For example, the chamber tone a with 440 Hz at a sampling rate of 44.1 kHz has 100.2 samples per wave train. If you want to play this tone half a tone lower with the same sampling frequency, 106.2 samples per wave train are required. It is therefore necessary to convert the stored sampling rate to the system sampling rate specific to the musical instrument during playback. For this purpose, a method is used which is known under the name "Sample Rate Conversion" (see, for example, Chamberlin "Musical Applications of Microprocessors", pages 470 to 477, Haydn Book Company, Inc., 1980). In many cases, however, it is not possible to produce more than three adjacent semitones from a single "sample", ie sound pattern which is represented by the sequence of samples without major losses in quality. If you want to recreate a piano sound, ie the sound of a piano or grand piano with a length of 88 keys, about 30 samples or sound patterns are required. The sound of a piano is not only dependent on the pitch, but also on the velocity. Furthermore, lower tones decay much longer than high tones. Assume a division into eight dynamic levels and one average recording time of approx. 20 sec. per note, there is a memory requirement of approximately 420 megabytes, whereby only a single instrument, namely the piano, is covered. This does not take into account the fact that the sound of a piano can be further modified using the damper and sustain pedals, so that a multiple of the specified memory requirement is necessary for the complete reproduction of the piano sound. The average recording time of a sound can be reduced by looping, ie repetitive output of a section of the sound, while accepting a loss in quality, with an average recording time of approx. 3 seconds. however, a memory requirement of approximately 60 megabytes is still required. The memory requirement is higher or lower for other instruments. However, since you want to produce a multitude of sounds with a modern electronic musical instrument, the memory requirement grows practically immeasurably. Even taking into account the fact that storage media are becoming cheaper and cheaper, it is difficult to process such large amounts of data at the required high speed required for virtuoso playing of the electronic musical instrument.

The mathematical operation on which the sample rate conversion is based can be implemented by means of a digital interpolation filter, ie the interpolation in the time domain can also be viewed as filtering in the frequency domain. The most suitable filter function for this purpose is a low-pass filter, which allows everything to happen up to half the sampling frequency with which the sound pattern has been sampled, but beyond which it completely suppresses all frequency components. The cutting edge should therefore run practically perpendicular to the blocking frequency. As is well known, an ideal filter cannot be realized. The ideal However, low-pass filters can be approximated quite well if a filter with a large number of filter poles is used. The higher the number of filter poles, the better the approximation to the ideal filter characteristic. However, a large number of poles in the digital case has the disadvantage that a predetermined number of arithmetic operations is required per pole, that is to say, for example, an addition and a multiplication per filter pole, so that with many filter poles a correspondingly large number of mathematical operations must be performed by the musical instrument relatively slow despite a lot of effort. This disadvantage is particularly evident when the musical instrument is to be operated polyphonically, ie when a large number of different sound patterns are to be reproduced at the same time. In this case, the many mathematical operations need to be performed not only for one sound pattern, but in parallel for a whole series of sound patterns. Even when using extremely fast components, the musical instrument will sooner or later reach a limit beyond which an expansion of the variety of sounds is no longer possible, especially in view of the fact that very large memories have to be used.

It is therefore the object of the present invention to provide a method in which the sounds are reproduced as realistically as possible with little effort.

This object is achieved in a method for sound generation with an electronic musical instrument, in which sound patterns are stored digitally in the form of samples in a preparation section and the samples are read out from the memory device in a sound generation section are digitally interpolated and converted from digital to analog in an interpolation filter, whereby attenuation below the blocking frequency of the interpolation filter is permitted in the digital interpolation and the higher-frequency components of the sound patterns that are attenuated in the interpolation filter are amplified in the preparation section before storage will.

In this method, a compromise in the filter property is chosen, namely an attenuation is permitted in a range that is actually still the pass band. Frequency components that fall within this range are undesirably heavily damped. The damping becomes stronger the closer the frequencies come to the blocking frequency. For this reason, filters with a relatively steep cut-off characteristic have so far been chosen. However, the attenuation can be compensated for by a pre-emphasis, which increases or increases the higher-frequency components before the sound patterns are saved. During the interpolation, these parts are damped accordingly, so that despite the poor filter properties, a sound pattern is available at the output of the interpolation filter that practically corresponds to the original.

The amplification of the higher-frequency components is preferably carried out before storage with a frequency-dependent gain characteristic which is inversely proportional to the frequency-dependent pass characteristic in the pass band. In other words, the higher the frequency components are amplified depending on the frequency, the greater the attenuation in the interpolation filter. This allows an almost rectangular filter characteristic to be achieved in the frequency range, ie the cut-off edge of the filter is almost perpendicular to the blocking frequency of the filter. By preemphasis you of course take away the disadvantage of one the additional gain diminished controllability in purchase. However, this plays practically no role in the frequency ranges under consideration, since the energy component for the higher-frequency components is generally so small that they do not come into the modulation range anyway.

It is preferred that the frequency spectrum of the sound patterns is limited to frequencies below a cutoff frequency, the cutoff frequency being less than the cutoff frequency of the interpolation filter. For example, the sound patterns are limited to the audible spectrum, the so-called audio spectrum. By taking the double measure before saving the sound pattern, i.e. the preemphasis on the one hand and the limitation of the frequency spectrum on the other hand, after the reproduction, sound patterns are obtained which are practically unattenuated up to the cutoff frequency, but are practically completely suppressed.

In a preferred embodiment, the sound patterns are put together from individual sound features, which are stored digitally in the form of sample values, are read out in a controlled manner, are subjected to a sampling rate conversion according to which all sound features are available with a uniform system sampling rate, and are then combined, at least some sound features being components of different types Have frequency, all sound features at the start time of a sound have a predetermined phase relationship to one another and the components of a sound feature at this start time have a predetermined phase relationship to the other components of the same sound feature. This embodiment can also be implemented without the features of the preemphasis. A sound feature is therefore a frequency mix that is stored in the memory in the form of a predetermined number of samples. Would be a sound feature containing only a single frequency, one could assemble practically any tone from a multitude of sound characteristics, as is known from harmonic synthesis. However, a separate sine wave oscillator with its own amplitude control is required for each overtone. With an overtone-rich but deep tone of, for example, 40 Hz with overtones up to 10 kHz, approximately 250 oscillators come together. Each of these must be controlled and managed. According to the invention, not only sine vibrations are generated, but frequency mixtures that can already contain a large number of harmonics. One takes advantage of the fact that with the many sound patterns that are stored in the memory, certain similarities prevail which determine, for example, the characteristic sound pattern of an instrument. If you extract and save the "basic sound" of an instrument as a sound feature, you only need a few more sound features for the sound patterns that correspond to the individual tones of the instrument in order to reproduce the tones with high similarity. The astonishing thing is that the acoustic reproduction of the basic sound feature does not necessarily have to be similar to the sound of the instrument to be reproduced. Similarities that can be used to combine parts of different sound patterns into common sound characteristics exist not only between the tones of a single instrument, but also between sound patterns of different instruments, for example in the case of instrument families such as brass, woodwind or strings. However, you can also find cross-instrument sound characteristics that can be used to further reduce memory requirements. The sound characteristics can, for example, be found out manually, ie on the basis of audio samples. Another option is to look at wave trains, generated by the individual instruments. Here, too, certain similarities can be recognized, which can then be extracted in the form of sound characteristics. Finally, suitable calculation methods can be developed to extract the similarities between individual sounds as sound characteristics.

What is important in the composition of individual sound features to the desired sounds is the fact that the individual sound features at the start time, i.e. when the sound is first composed, have a predetermined phase relationship to one another. The same applies to the individual frequency components of a single sound feature. Only in this case does the superimposition of the individual sound characteristics lead to the desired synthesis, i.e. to the sound to be reproduced. If the individual sound features do not have this predetermined phase relationship, i.e. depending on the coincidence when the individual sound characteristics start, it can happen that individual overtones are amplified or extinguished and thus distort the sound image. The sound characteristics are not limited to a period of the fundamental wave. As with normal sampling, i.e. Sampling, sound patterns of practically any length can be saved.

All sound features preferably have a zero crossing at the start time. This facilitates the establishment of the desired phase relationship. In addition, this structure essentially corresponds to the model of a natural musical instrument. With conventional musical instruments, too, the tone begins with an amplitude value of zero.

It is also preferred that all components of a sound feature have a zero crossing at the start time. The amplitude of a component can develop positively or negatively after the start time. If the amplitude develops negatively in the time range after the start, i.e. the first derivative after the time is negative, this means that the relevant component of the same frequency, which initially assumes an amplitude greater than zero after the start time, is out of phase by 180 °. Such a component can be represented in the frequency spectrum with a negative amplitude. A sound feature with such a component will completely or partially cancel out the component of another sound feature with the same frequency. The "negative" amplitude is of course only used here as a calculation parameter, since the amount of the amplitude is decisive for the ear.

Each sound feature is preferably stored with a number of samples which has a predetermined relationship with the highest frequency occurring in the sound feature, the blocking frequency of the interpolation filter being selected depending on the sampling frequency with which the samples were generated from the sound patterns and preferably in on the order of 50% to 60% of the sampling frequency. In this area you have the greatest certainty that no aliasing will occur. It is advantageous that the cutoff frequency is selected in the order of 30% to 50% of the sampling frequency. In this way, practically any blocking or limit frequencies can be realized, so that the characteristics of the interpolation filter also influence the reproduction of the sound pattern is possible. In particular, the so-called alias effect can be suppressed by a clever choice of the blocking frequency of the interpolation filter or the cut-off frequency, which causes undesirable disturbances in the reproduction of the sound pattern, especially at high frequencies. It is therefore not necessary to sample all sound features with the same, high sampling frequency, that is to say the system sampling rate, and to store the resulting high number of sampling values. Rather, the number of samples is limited to what is absolutely necessary.

The sound features are advantageously subjected to an amplitude control before the assembly. For example, a sound feature that predominantly contains overtones can be amplified in one case to produce a sound that is rich in overtones, but in another case can only be used with a low amplification, so that the overtones are not so clearly audible. With the same sound characteristic, different sounds can already be generated by the amplitude control. This also applies to a single sound to show its change over time.

The amplitude control is preferably carried out with the aid of digitally stored amplitude envelopes. Changing frequency mixes can then also be realized in one sound. This can lead to interesting beat effects.

It is preferred that the amplitude envelopes are composed of envelope characteristics. The technology, the similarities in the individual curves or storing vibrations only once and using them multiple times can also be used for the envelopes.

Individual sound features are advantageously used to generate a plurality of sounds. This is also possible at the same time, since the sound characteristics are not lost when they are read out of the memory. Rather, a sound feature that is being read out can also be processed further in a plurality of processing sections arranged in parallel. A sound feature, of which some components have already been read out, can also be read out again from the beginning without the first reading of the sound feature being disturbed. Ultimately, this is just a question of memory management or readout control. This allows a polyphonic game to be realized.

By combining the sound features, both an amplification and an attenuation of frequency components in the resulting signal can preferably be achieved. This amplification or weakening can be achieved on the one hand by simply superimposing the sound characteristics. If a sound feature has one of the above-mentioned components with negative amplitude, the corresponding frequency, if it also occurs in the other sound feature, will be attenuated. The same effect, but not limited to a single frequency, can be achieved by providing the sound feature with a negative amplitude envelope. The term negative amplitude envelope is also used here for illustrative purposes only introduced. This is to express that sound features that have a negative amplitude envelope are not added to another sound feature when they are joined, but are subtracted from it.

In a particularly preferred embodiment, the reading out and the sampling rate conversion of a plurality of sound features are carried out by means which are common to a plurality of sound features and which process the sound features in series or in the multiplex method. This can be easily achieved due to the high processing frequencies in relation to the frequencies of the sound features. Strictly speaking, all processing only has to be completed within a period of the system sampling rate of, for example, 44.1 kHz, in order to provide the musical instrument with the instantaneous value of the sound to be reproduced. Assuming a system sampling rate of 44.1 kHz, this is more than 20 µs.

Advantageously, during the reading out of a sample of a sound feature from the memory, the sampling rate processing of a sample of another sound feature is carried out, which was read out in the previous access to the memory. The calculation of individual values thus takes place in succession, which means that a certain pipelining can be implemented.

The individual sound features are preferably assembled with the aid of an accumulator in which instantaneous values of the sound features provided with the uniform system sampling rate are added or subtracted serially, the content being read out at a predetermined point in time. For the "filling" of the battery stands a complete period of the system sampling rate is available. It is therefore not necessary that all sound features of a sound are processed at the same time. Rather, this preferred embodiment opens up the possibility of serial processing of individual sound features in succession, the accumulator making it possible to assemble the individual sound features.

In a particularly preferred embodiment, each sound is composed of a maximum of sixteen sound features. In many cases, two to eight sound characteristics will suffice. The decision on how many sound features to use is ultimately a question of hearing. The limitation to sixteen sound characteristics means that the amount of data to be stored and managed is limited.

In a further preferred embodiment, a plurality of sets of sound features are put together in parallel for a sound and are superimposed with opposite amplitude control curves. This allows the sound of an instrument to be simulated in a simple manner, which would otherwise be difficult to simulate even with a multitude of sound features. As an example, reference is made to a trumpet into which a damper is inserted during the blowing. Another example is a clarinet or saxophone, the players of which greatly change the tension of the lower lip when playing.

In a preferred embodiment, an interpolation filter is used, the cut-off frequency of which changes automatically as a function of the sampling frequency. This saves storage space in the samples for the sound pattern. If For example, the chamber tone a has only a low harmonic content, for example, if the sound pattern comes from a flute that essentially only has the third harmonic (1760 Hz), a sampling frequency of approximately 3.5 kHz is sufficient to generate the samples. So only about eight samples per wave train. In this example, this results in a memory saving of more than 90%. When reproducing with the system sampling rate of the musical instrument of 44.1 kHz, for example, 100.2 samples must of course then be generated or calculated again. In this context, the advantage of automatically adjusting the cut-off frequency of the digital interpolation filter is particularly evident. The limit frequency is set at a sampling frequency of 3.5 kHz, for example, to a value in the order of 1.4 kHz.

The invention is also based on the object of specifying an electronic musical instrument for generating sounds, in which the sounds are reproduced as faithfully as possible with little effort. This object is achieved in an electronic mustikinstrument for generating sounds with a storage device in which sound patterns are stored digitally in the form of samples, a readout device which reads the samples in a controlled manner from the storage device, a digital interpolation filter which carries out a sampling rate conversion and at its output provides the sound characteristics with a system sampling rate that is uniform for the entire musical instrument, and a digital / analog converter, which is connected to the output of the interpolation filter, in which the interpolation filter has a pronounced attenuation below the cutoff frequency, and the samples sound patterns correspond, the higher-frequency components that are attenuated in the interpolation filter are amplified.

The musical instrument therefore has the information it needs to generate the sounds stored in it. When a key is pressed or when a generation signal occurs, the readout device reads out the stored samples, which are then processed subsequently. Because the stored values have a different frequency spectrum than the values to be reproduced - namely, according to the invention they are amplified in the upper frequency ranges - the interpolation filter with the pronounced attenuation can be used without problems, without the output of the interpolation filter or the downstream digital / analog Unwanted distortion can be heard.

The musical instrument preferably has an accumulator, which is connected to the output of the interpolation filter and adds up values which have been read in serially and outputs the sum, the addition and output taking place within a period of the system sampling rate, and an audio device which produces the output signal of the digital / Makes the analog converter audible. In such a musical instrument, the individual sound features can be overlaid because the interpolation filter ensures a sampling rate conversion. The sound features are still available in digital form at the output of the interpolation filter, but they are all sampled there at the system sampling rate, i.e. within each period of the system sampling rate, an instantaneous value of each sound feature is stable and available for further processing.

The individual instantaneous values can therefore be superimposed in each period of the system sampling rate. The output signal then contains the addition or subtraction of the individual frequencies, which leads to the frequency spectrum of the sound to be reproduced.

The interpolation filter preferably has 32 or fewer poles. Compared to the commonly used interpolation filters, which have an order of magnitude of 100 or more poles, this saves a considerable amount of computation and processing time and enables the read-out values to be processed quickly, since a smaller number of poles means that fewer computation operations are necessary. In the event that the filter has to be available for several sounds at the same time, for example has to perform the arithmetic operations for a polyphonic sound, a correspondingly greater variety of sounds can be generated simultaneously with the 32 or fewer poles specified. Compared to the otherwise usual hundred or more-pole filters, the variety of sounds can be practically more than tripled. In a particularly preferred embodiment, the interpolation filter has eight poles. A relatively broad transition range is obtained here, i.e. damping sets in relatively early. However, this effect can be easily compensated for by the pre-emphasis provided.

An amplitude control device is preferably arranged between the interpolation filter and the accumulator. This amplitude control device ensures that individual sound features have a stronger or weaker influence on the composition of the sounds.

In a preferred embodiment, readout device, interpolation filter, accumulator and optionally amplitude control device are combined in one sound generation module and a plurality of sound generation modules are arranged in parallel. This extends the possibility of polyphony. A plurality of sound generation modules arranged in parallel can produce an incomparably larger number of sounds in parallel, without there being any restriction in terms of computing or processing time.

A further advantageous effect can be achieved if the individual sound features have a predetermined additional change in the phase shift with respect to one another. This makes it possible for the listener to feel that a large number of instruments are playing at the same time, like an orchestra.

The sound generating device preferably has at least one transputer for processing the sound features into sound signals. A transputer, such as that sold by the company Inmos, is a programmable circuit component that enables parallel processing of processes at high speed. "Transputer" is an artificial word that emerged from the contraction of "transistor" and "computer". A transputer thus behaves like a computer on the one hand, but also physically like a semiconductor circuit element. A transputer is a self-contained unit that manages processes internally and only leads the process inputs and outputs to the outside. Transputers allow the parallel processing of processes at high speed. It is irrelevant whether there are actually several units in the transputer that work in parallel, or whether in a special embodiment a unit, ie a single transputer, simulates the parallelism. This is due to the fact that each transputer has local memory that it can access. The transputer can therefore manage itself to a great extent without being dependent on the restrictions of a bus or a controller for an external memory, which may also have to be accessed by other processors. Another reason for the high speed in parallel processing is that the transputer has a very fast interface, a so-called link, through which it can communicate with other transputers, other interfaces or other components. In this case, only the end results of the processes, i.e. data processing, and the own data have to be conveyed to the outside or supplied from outside. Communication problems, such as occurred with the conventional use of processors as main and subprocessors or as parallel processors, do not occur with transputers. Since the processing of signals for sound generation can in most cases be divided into parallel processes, the performance of the musical instrument increases almost linearly with an increasing number of transputers. Especially when simulating sounds, a multitude of parallel processes take place, especially with polyphonic playing. In many cases, this corresponds to the original of the replica, for example if an orchestra with many instruments working in parallel or a piano that is played with full grip is to be replicated. The parallel processes are, for example, the generation of single tones that are generated parallel to one another. Beyond running but also in the production of a single tone a large number of parallel processes, for example when the tone is interpolated by a multi-pole digital filter. In extreme cases, each pole can be viewed as a separate process.

A plurality of sound generation modules is preferably provided, the communication between the reading device and the sound generation modules being carried out with the aid of at least one transputer. The advantages of a transputer can be seen here too. The communication between the data input device and the modules has always been limited by the performance of a processor that could either accept many input signals or manage many modules at the same time. By using transputers, the data flow between the data input device and the modules of the processing device can now be controlled much better without being limited to assigning certain signal paths from the data input device to certain modules in the processing device. Rather, all modules can be connected to all elements of the data input device in order to serve the sound generation.

The transputer preferably forms at least one digital interpolation filter in each module. With the help of the interpolation filter, the temporal amplitude sequences of each sound can be simulated from the sampled values, which serve here as support points. The use of the interpolation filter has the advantage that all sounds at the output of the interpolation filter with a uniform sampling rate for the entire musical instrument, the so-called system sampling rate. The further processing of the data still available in digital form can then also be carried out in digital form without the need for conversion to analog form and back again, possibly even several times. The interpolation thus carries out a sample rate conversion. The digital interpolation filter consists of a series connection of several adders and multipliers. The use of the transputer has the advantage that each addition and each multiplication can be understood and implemented as a separate process. Since the transputer can carry out several processes in parallel, the individual mathematical operations required for each pole can also be carried out in parallel. Successive values can then be calculated simultaneously in the form of a pipeline.

It is preferred that the transputer forms several parallel interpolation filters in each module, which process several sound patterns in parallel. Already with one module, several sounds can be generated in parallel without the need for time slice management, which assigns each interpolation filter only a fraction of the time available in a period of the system sampling rate.

It is also an object of the present invention to provide a device for loading a storage device for an electronic musical instrument, in which the sounds are reproduced as faithfully as possible with little effort.

To achieve this object, the device is provided with a recording device for generating sound patterns in the form of electrical or magnetic signals from sound waves, a scanning device that scans the sound patterns at a sampling frequency at predetermined times and generates samples, and with a storage device that stores the samples in inscribes the memory device, provided, a pre-emphasis device being arranged between the recording device and the scanning device, which amplifies the sound patterns in predetermined frequency ranges.

The loading device therefore ensures in advance that the damping carried out by the digital filter is compensated again.

It is preferred that the pre-emphasis device amplifies the sound patterns in a frequency range that is below 50% of the sampling frequency. The amplification therefore takes place in a relatively broad frequency band, so that the stored samples, if they are not read out by the electronic musical instrument, could probably not be recognized at all as belonging to the sound patterns.

The pre-emphasis device advantageously has a frequency-dependent gain characteristic which is essentially inversely proportional, in the best case even exactly inversely proportional, to the frequency-dependent pass characteristic of the interpolation filter. The greater the attenuation in the interpolation filter as a function of the frequency, the greater the amplification or the increase in the same frequencies in the pre-emphasis device. After the signal, ie the sampled values, have passed through the interpolation filter, the previously applied pre-emphasis has been compensated for by the attenuation of the filter.

It is also preferred that the pre-emphasis device attenuates frequency components above the cut-off frequency. The pre-emphasis device cuts off frequencies above the cut-off frequency or attenuates them considerably.

Fig. 1

eine schematische Darstellung der Komponenten eines Musikinstruments,

Fig. 2

eine schematische Darstellung eines Tonerzeugungsmoduls,

Fig. 3

Klangmerkmale und Klangmuster im Frequenzbereich,

Fig. 4

ein Klangmerkmal im Zeitbereich,

Fig. 5

eine schematische Darstellung einer Beschickungseinrichtung,

Fig. 6

eine schematische Darstellung eines Filters,

Fig. 7

eine schematische Darstellung eines Filterelements,

Fig. 8

Frequenzabhängigkeiten von Filterdämpfung und Preemphasis-Verstärker und

Fig. 9

eine nähere Darstellung der Verwendung von Transputern.

The invention is described below with reference to a preferred embodiment in conjunction with the drawing. In it show:

Fig. 1

1 shows a schematic representation of the components of a musical instrument,

Fig. 2

1 shows a schematic illustration of a sound generation module,

Fig. 3

Sound characteristics and sound patterns in the frequency domain,

Fig. 4

a sound feature in the time domain,

Fig. 5

a schematic representation of a loading device,

Fig. 6

a schematic representation of a filter,

Fig. 7

a schematic representation of a filter element,

Fig. 8

Frequency dependencies of filter attenuation and pre-emphasis amplifier and

Fig. 9

a more detailed description of the use of transputers.

An electronic musical instrument generally has a keyboard 1, operating elements 2 and interfaces 3. The keyboard, which comprises one or more manuals with a range of four to eight octaves and / or a pedal, is used by the practicing musician to generate tone sequences. When you press a key, you usually hear a tone. Depressing several keys simultaneously causes the polyphonic generation of several tones. The timbre can be set or changed using the controls 2. Signals from other devices, for example other electronic musical instruments, computers or storage media, can be coupled into the electronic musical instrument via the interfaces 3. Keyboard 1, control element 2 and interfaces 3 are connected to an interface processor 4, which in turn has a memory 5. The interface processor 4 manages the signals received from the keyboard 1, the operating elements 2 and the interfaces 3 and generates a suitable code which is passed on to a main processor 6. The main processor 6 has a memory 7 in which, among other things, the processing instructions for the signals received by the interface processor 4 are stored. After the main processor 6 has processed the signals received by the interface processor 4, it sends 18 addresses and / or commands via a control bus, with the aid of which tone generation modules can generate 8-10 tones. A variety of tone generating modules 8-10 can be provided. In principle, their number is only limited by the capacity of the main processor 6. Each tone generation module 8-10 is capable of generating one or more tones simultaneously. The tone is generated digitally, with each tone generation module 8-10 being connected to a control curve memory 11 and to a sample value memory 12, both of which are common to all tone generation modules 8-10 are able to access. The electronic musical instrument emulates other musical instruments, for example a piano, a string instrument or a wind instrument or also a drum set, by means of the tone generation modules 8-10. The information for the sound is stored in the control curve memory 11 and the sample value memory 12.

The sound generating modules 8-10 generate digital signals, which they put on an audio bus 17. Also connected to the audio bus 17 are effect processors 13-15 which, if desired, subject the digital output signals of the tone generation modules 8-10 to a digital effect treatment, for example the generation of a reverb, a distortion, a vibrato or other effects.

The structure of a sound generation module is shown schematically in FIG. 2. In principle, all sound generation modules 8-10 have the same structure. You can also use the same operating instructions, i.e. Programs in the individual processors can be controlled.

The sound generation module 8 has a control processor 19 with memory 20, which receives information from the main processor 6 via the control bus 18. The control processor 19 is connected on the one hand to a control curve processor 21 and on the other hand to a phase processor 22. Upon request from the control processor 19, the control curve processor 21 accesses the control curve memory common to all tone generation modules 8-10, determines the value of a control curve at the required point by interpolation and, if necessary, calculates a final value from a plurality of such values, which it returns to the control processor 19. The control curves are used to determine the frequency sequence, the volume sequence and the relative importance of individual components control the sounds to be generated for a variety of sound components. The individual values can be calculated one after the other, the separation of memory access and processing permitting the simultaneous execution of both functions for successive calls.

The control curve values can be stored in the control curve memory 11, for example, in the form of successive base values. If the ordinate distance of all base values is the same, only the abscissa values are saved. Otherwise, pairs of values from ordinates and abscissa can be saved.

The various tone generation modules 8-10 can access the control curve memory cyclically at specified time intervals in order to rule out access conflicts. In another embodiment, individual tone generation modules 8-10 may have different priorities, with the highest priority tone generation module having access to the cam memory 11 before anyone else, if need be.

The control processor 19 transfers its output data to a phase processor 22. The phase processor 22 has the task of tracking the phase for each of the sound components for a fixed time interval from the information received from the control processor 19. The phase processor calculates an address from the phase values and thus accesses the sample value memory 12. So-called sound features are stored in the form of sample values in the sample value memory 12. The samples can only include one period of a sound, so-called "waves", but they can also include sound features of an entire tone, ie the tone from the beginning to the end, so-called "samples". This can be useful, for example, if a tone changes irregularly from start to decay. This is the case, for example, with a drum cymbal sound. Another example is the blowing behavior of pipe organs, which have a much richer frequency spectrum at the moment of blowing than in the steady state.

The data read from the sample value memory 12 with the aid of the addresses generated by the phase processor 22 are forwarded directly to an interpolation filter 23. The calculation of the individual phase values can also take place here one after the other, with the separation of memory access by the phase processor 22 and processing by the interpolation filter 23 making it possible to carry out both functions simultaneously for successive values. The interpolation filter 23 calculates amplitude values from the sample values stored in the sample value memory 12 at a number of support points specified by the phase processor. The support points are determined by a system sampling rate. After passing through the interpolation filter, all tones are available with a uniform system sampling rate. At the times determined by the sampling rate, the instantaneous values of the sounds to be digitally reproduced are therefore always available in the entire musical instrument. They can then simply be added or subtracted without having to pay attention to different phase relationships.

An amplitude processor 24, which is also controlled by the control processor 19, is connected to the output of the interpolation filter 23. The amplitude processor 24 has the task of controlling the amplitude of the output signal of the interpolation filter 23. Since the output of the interpolation filter 23 is available in digital form, this means that the individual digital values are subjected to predetermined mathematical operations. The amplitude can be amplified, for example, by multiplication by a factor greater than 1. A weakening takes place by multiplying by a factor smaller than 1. It is also possible to form a "negative" amplitude by multiplying by a negative factor. A negative amplitude is, of course, only to be understood here as a calculation variable, since a difference between a positive and a negative amplitude cannot be made audible.

An accumulator is connected to the output of the amplitude processor 24 and can also be controlled by the control processor 19. The accumulator has the task of summing up digital signals which are supplied to it at successive points in time and passing them on to the audio bus 17 at a point in time following the individual summation points in time. Sounds can thus be “put together” in the accumulator 25. An accumulator for each tone generation module is shown. However, it is also possible to provide only one accumulator for the entire musical instrument. The accumulator can also be replaced by an adder, which adds up the digital quantities emerging at its inputs at predetermined times.

The sounds are stored in the form of sound features in the sample value memory 12. A sound is composed of a plurality of individual sound features in the accumulator 25. At least one sound feature thereof has components of different frequencies. As a rule, however, most or even all of the sound features will consist of a frequency mix. The individual sound features are stored in successive samples. A special feature is that all sound features are stored with a predetermined phase relationship to one another. In the present exemplary embodiment, all sound features begin with a zero crossing of the amplitude. All frequency components also have a predetermined phase relationship to one another within the sound feature. It is also preferred here that all frequency components have a zero crossing at the start time. With this special regulation it is possible to easily overlay the individual sound characteristics and to create targeted overlay effects. FIG. 3a shows a first sound feature with three frequencies f1, f3 and f7, each of which has the significance 80, 30 and 10. For example, this is a frequency spectrum with the fundamental wave f1 and the third and seventh harmonics f3 and f7. Such a sound feature does not necessarily have to evoke a memory of the instrument to be reproduced with the aid of this sound feature. 3b shows a further sound feature which, however, does not contain a fundamental wave, but only the third and seventh harmonics f3 and f7. It is remarkable that the seventh harmonic is shown with a negative amplitude. This means nothing else than that the seventh harmonic f7 has a negative slope at the start time, ie its amplitude immediately less than zero immediately after the start time. It has a phase shift of 180 ° compared to a seventh harmonic, which has a positive slope at the start.

3c shows the superposition of the two sound features from FIGS. 3a and 3b. The positive frequency component f7 in FIG. 3a is canceled out by the negative sound component f7 in FIG. 3b. The superimposition thus leads to a frequency spectrum which only contains the fundamental wave and the third harmonic, the amplitude of the third harmonic being the sum of the corresponding frequency components from the sound feature from FIG. 3a and the sound feature from FIG. 3b.

FIG. 3d shows a further composite frequency spectrum, in which only the sound features from FIGS. 3a and 3b have also been used. In contrast to FIG. 3c, where the two sound features of FIGS. 3a and b have been added, the sound feature of FIG. 3b was subtracted from the sound feature of FIG. 3a for the frequency spectrum from FIG. 3d. The third harmonic f3 is therefore only available with the difference. The seventh harmonic f7, on the other hand, is available with the sum of its amplitudes, since the subtracting corresponds to a negative quantity by the addition of its amount. The subtraction can take place, for example, by multiplying the output value of the interpolation filter 23 in the amplitude processor 24 by a factor (−1).

3e shows a third frequency spectrum, which has also been generated from the two sound features according to FIGS. 3a and b. 3a was read unchanged into the accumulator 25, the sound feature according to FIG. 3b was multiplied in the amplitude processor 24 by a factor of 0.5. As a result, the frequency components of the third and seventh harmonics are only increased or decreased by an amount attenuated by a factor of 0.5.

The use of individual sound features is not necessarily limited to the reproduction of a sound from a single instrument. Of course, it can be expected that all tones of an instrument to be reproduced have a certain commonality. However, the frequency spectrum will not be the same for all tones. In this case, for example, the sound feature common to all tones of an instrument can be extracted and, in other sound features, only the differences to the individual tones, which will be different over the range of the instrument, can be additionally stored. Surprisingly, however, it turned out that it is also possible to find out individual sound characteristics that are the same for a group of instruments. Furthermore, by cleverly selecting the sound characteristics, practically all sound characteristics can be used for several tones or even sounds. In the present case, only two sound features have been combined for the sake of simplicity. However, the special effect is better seen when a larger number of sound features, for example three or four, are combined with one another. However, using more than sixteen sound features to create a sound is not particularly advantageous in most cases, since on the one hand each sound feature requires additional storage space, on the other hand the differences that are to be created by adding a seventeenth sound feature are practically inaudible. If you were to make a single sound characteristic directly audible, it would probably be impossible to find any or only a few similarities with the instrument to be reproduced. The desired sound is only obtained by superimposing several sound characteristics.

The individual sound features are stored in the sample value memory 12 in the form of sample values. Sound characteristics with only a few overtones are also stored with only a relatively few samples. No information about the highest frequency occurring in the sound feature can be derived from the sound feature itself. The frequency with which the individual sound feature is reproduced results only from the sampling rate conversion in the interpolation filter 23. This should be made clear using an example shown in FIG. 4. Only the sample values marked with an x are stored in the sample value memory 12. The interpolation filter 23 calculates intermediate values at positions predetermined by the phase processor 22, which are identified by a vertical line. These intermediate values are read into the accumulator at the system sampling rate. Assuming, for example, a system sampling rate of 44.1 kHz, the accumulator 25 sums up all the values that are applied to the input at discrete times within 22.7 μs. At the end of these 22.7 µs, the totalized values are then read out. 4, the sound according to FIG. A is read out within 15 periods of the system sampling rate, in FIG. 4b, however, within 30 periods. The tone according to FIG. 4b thus sounds an octave lower than the tone according to FIG. 4a. In a scale representation, the vertical lines should be much denser. However, the overview would suffer from this. For example, for a representation of a tone with the frequency 440 Hz, approximately 100 periods of the system sampling rate would have to be represented in the figure.

A single sound feature can therefore be used for different pitches of a single sound. The frequencies stored in the sound feature only indicate the relationship of the frequencies, for example the relationship of a fundamental frequency to the harmonics. These relative frequencies are only converted into an absolute frequency spectrum by converting the sampling rate in the interpolation filter 23. The information that is stored in a sound feature is limited to the most necessary. The number of stored samples corresponds to twice the highest frequency occurring in the sound feature.

Sound patterns that change significantly over time can be stored as sound feature sets that are mixed differently, i.e. time-dependent, with the aid of the amplitude control. It is also possible to crossfade from one sound to another. The amplitude control curves are constructed so that their sum always remains constant. The individual amplitude control curves are therefore opposite.

The interpolation filter 23 carries out a "sample rate conversion", as is known, for example, from Chamberlin's "Musical Applications of Microprocessors". The interpolation filter 23 can also be connected to the interface processor 4 via a bus. This processor 4 supplies the interpolation filter 23 with information about the pitch to be generated, for example about the frequency with which the samples are to be reproduced from the memory device 7. The frequency information is transmitted to the interpolation filter 23 in the form of phase information, ie the interpolation filter 23 receives the information via the interface processor 4 which phase distance the individual support points should have from one another. The interpolation filter 23 carries out an interpolation between the individual samples according to its stored filter coefficients.

The finished digital signal is fed to a digital / analog converter 26, which converts the digital signal into an analog one. The analog signal is fed to an audio unit 16 which makes the analog but electrical signal audible, that is to say generates sound waves therefrom and couples it to the air.

FIG. 6 shows the schematic structure of the interpolation filter 23, eight poles 31, which are basically the same structure, being shown. Each pole receives the phase information via a bus 32, i.e. the information about the position of the support point within the sound pattern at the desired frequency at which the instantaneous value is to be calculated from the sample values. The sampled values read out from the memory device 7 with the aid of a readout device are fed to the digital interpolation filter 23 via the bus 33. The sample values which are supplied via the bus 33 are available to each filter pole, possibly with a time offset. The output of a filter pole is passed on to the input of the next filter pole.

Fig. 7 shows schematically the structure of a single filter pole. Filter coefficients are stored in a memory 38, which may be in the form of RAM or ROM, and can be read out under the control of the phase value supplied via bus 32. The filter coefficient memory 38 is connected via bus lines 39, 40 to an interpolator 41, which also receives the phase value information. The filter coefficient memory 38 provides two successive filter coefficients via the bus lines 39, 40, with the aid of which the interpolator 41 can, for example, carry out a linear interpolation.

The output of the interpolator 41 is connected to a multiplier 36 which multiplies the output of the interpolator 41 by the samples which are supplied via the bus 33. The output of the multiplier 36 is connected to an adder 37 which adds the output value of the multiplier 36 to the output value of the previous filter pole. For this purpose, the output of the previous filter pole is fed to the second input of the adder 37 via a bus 34. The output of the adder 37 is passed on to the next filter pole via a bus 35. At the last filter pole (filter pole 8), bus 35 corresponds to output 42.

The interpolation filter 23 is a filter with a relatively low order, ie it has only 32 or fewer, in the present case even only 8 poles. Filters with such a low order do not cut off sharply at their cut-off frequency f _A , but already have a considerable attenuation below it. This state of affairs is shown schematically in FIG. 8. The dependence of the amplitude A on the frequency f is shown in the upper part. It can clearly be seen that the present filter 23 is a low-pass filter which already exerts a maximum attenuation D on the input signal at the end of a pass band 43. The pass band ends here at the cut-off frequency f _G , ie the highest frequency occurring in the sound pattern. In a region 44 adjoining it, which for the sake of simplicity will be referred to in the following, the signal components with the corresponding frequencies are still allowed to pass through, but are attenuated more strongly than with the damping D. A blocking area 45 adjoins the transition area 44. Frequencies in this range are almost completely suppressed.

The purpose of the low-pass filter property of the interpolation filter 23 is to suppress disturbing frequency components which arise from the sampling of the original sound pattern. For this purpose, the cut-off frequency f _G must be set such that it lies at least as far below half the sampling frequency f _S / 2 of the original sound pattern as the cut-off frequency f _A above f _S / 2. The problem with the "cheap" design of the filter now arises that, in the case where the blocking region 45 is allowed to begin near the cut-off frequency f _G , the frequency components are damped too much, which is actually still completely in the sound pattern to be reproduced should be included. If, on the other hand, the pass band 44 is shifted further in the direction of a higher frequency, ie if the cut-off frequency f _{G is left} within the transition region or even on its left side, interfering frequencies are also let through, which audibly and disturbingly change the reproduction of the sound pattern.

However, the attenuation in the pass band 43, that is to say at frequencies below the cut-off frequency f _G , can be permitted if it is ensured that the corresponding frequencies of the sound patterns have been correspondingly raised or amplified before being stored. 5 shows a suitable arrangement for this. With the aid of a microphone 46, sound patterns, for example from a conventional musical instrument, are recorded and converted into electrical signals. The output of the microphone 46 is fed to a pre-emphasis device 47, which amplifies selected frequency components of the electrical signal generated by the microphone 46.

8 shows the dependence of the gain factor V on the frequency f. It can be seen that the greater the attenuation in the pass band 43, the greater the gain. The sound patterns changed in this way are fed to a sampling device 40, which sample the sound patterns at a sampling frequency which corresponds to twice the highest frequency occurring in the sound pattern. The sampled values sampled in this way are stored in the memory 7.

In the "Sample Rate Conversion" after reading out with the help of the interpolation filter 23, the higher-frequency components of the sound pattern are damped, but this damping is compensated for by the previous increase, so that a signal is available at the output of the interpolation filter that contains all frequencies up to the cut-off frequency f _G undamped, but contains higher frequencies almost completely damped.

The structure of the interpolation filter automatically adjusts the cut-off frequency f _G to the original sampling rate at which the sound pattern in the sampling device 48 has been sampled. The interpolation filter 23 interpolates according to the filter coefficients between two successive samples, it being irrelevant whether the samples originally followed one another closely in time or were further apart in time. The temporal relationship is only established by the phase information via line 32, with the aid of which the interpolation filter 23 calculates the necessary number of support points that are necessary for further processing at the system sampling rate. The low-pass filter property of the interpolation filter 23 results from the type of interpolation between the two successive ones Sampled values, ie the “relative” cutoff frequency, that is to say the cutoff frequency based on the original sampling rate, and is determined by the filter coefficients stored in the memory 38.

The main processor 6 consists of several transputers 50, 51, 52. The individual transputers of this main processor 6 process the incoming input data essentially in parallel, the processing of the individual input data in one of the transputers 50-52 being independent of the processing in the other transputers 50- 52 takes place. So only a very small amount of communication between the individual transputers 50-52 is necessary. After the main processor 6 has processed the signals received by the interface processor 4, it sends addresses and / or commands via the control bus 18, with the aid of which the tone generation modules 8-10, which are also connected to the control bus 18, can generate tones. A variety of tone generating modules 8-10 can be provided. Up to now, their capacity has been limited by the performance of the managing processor. By using the main processor 6, which can have any number of transputers 50-52, there are practically no more restrictions in the number of tone generation modules 8-10 to be managed.

Claims (26)

Method for sound generation with an electronic musical instrument, in which sound patterns are stored digitally in the form of sample values in a preparation section in a preparation section and the sample values are read out from the memory device (7) in a sound generation section, digitally interpolated in an interpolation filter (23) and are converted from digital to analog, with digital interpolation allowing attenuation below the blocking frequency of the interpolation filter (23) and in the preparation section the higher-frequency components of the sound patterns that are attenuated in the interpolation filter (23) are amplified before being stored. Method according to Claim 1, characterized in that the amplification of the higher-frequency components takes place with a frequency-dependent gain characteristic which is essentially inversely proportional to the frequency-dependent pass characteristic in the pass region (43). Method according to Claim 1 or 2, characterized in that the frequency spectrum of the sound patterns is limited to frequencies below a cut-off frequency, the cut-off frequency being lower than the cut-off frequency of the interpolation filter (23). Method for sound generation, in particular according to one of claims 1 to 3, characterized in that the sound patterns are composed of individual sound features which are digitally stored in the form of sample values, are read out in a controlled manner, are subjected to a sampling rate conversion, according to which all sound features are standardized System sampling rate are available, and then put together, wherein at least some sound features have components of different frequency, all sound features have a predetermined phase relationship to one another at the start time of a sound and the components of a sound feature have a predetermined phase relationship to the other components of the same sound feature at this start time. Method according to one of claims 1 to 4, characterized in that each sound feature is stored with a number of samples which has a predetermined relationship with the highest frequency present in the sound feature, the blocking frequency of the interpolation filter (4) being selected as a function of the sampling frequency with which the samples have been generated from the sound patterns and in particular is selected in the order of 50% to 60% of the sampling frequency. A method according to claim 5, characterized in that the cut-off frequency (f _G ) is selected in the order of 30% to 50% of the sampling frequency. Method according to one of Claims 1 to 6, characterized in that the sound features are subjected to an amplitude control before the assembly, in particular with the aid of digitally stored amplitude envelope curves which are composed of envelope curve features. Method according to one of claims 1 to 7, characterized in that individual sound features are used to generate a plurality of sounds. Method according to one of claims 1 to 8, characterized in that both the amplification and the attenuation of frequency components in the resulting signal can be achieved by combining the sound features. Method according to one of claims 1 to 9, characterized in that the reading out and the sampling rate conversion of a plurality of sound characteristics is carried out by means common to a plurality of sound characteristics in each case, which process the sound characteristics in series or in a multiplex process, in particular during the reading out of a sample value of a sound characteristic from the Memory the sampling rate processing of a sample of another sound feature is performed, which has been read out in the previous access to the memory. Method according to one of Claims 1 to 10, characterized in that the individual sound features are put together with the aid of an accumulator, in which instantaneous values of the sound features provided with the uniform system sampling rate are added or subtracted serially, the content being read out at a predetermined point in time and in particular each sound is composed of a maximum of sixteen sound characteristics. Method according to one of claims 1 to 11, characterized in that for a sound several sets of sound features are put together in parallel, which are superimposed with opposite amplitude control curves. Method according to one of claims 1 to 12, characterized in that an interpolation filter (23) is used, the blocking frequency (f _A ) of which changes automatically as a function of the sampling frequency (f _S ). Electronic musical instrument for generating sounds with a storage device (7), in which sound patterns are stored digitally in the form of samples, a readout device (6) which reads the samples in a controlled manner from the storage device (1), a digital interpolation filter (23), the one Performs sampling rate conversion and provides the sound features at its output with a system sampling rate that is uniform for the entire musical instrument, and a digital / analog converter (26) which is connected to the output of the interpolation filter, in particular for carrying out the method according to a of claims 1 to 13, wherein the interpolation filter (23) has a pronounced attenuation, and the samples correspond to sound patterns whose higher-frequency components which are attenuated in the interpolation filter (23) are amplified. Musical instrument according to claim 14, characterized by an accumulator (25) which is connected to the output of the interpolation filter (23) and adds up read values and outputs the sum, the adding and output taking place within a period of the system sampling rate, and an audio Device (16) which makes the output signal of the digital / analog converter (26) audible. Musical instrument according to claim 14 or 15, characterized in that the interpolation filter (23) has thirty-two or fewer poles (31), in particular eight poles (31). Musical instrument according to claim 15 or 16, characterized in that an amplitude control device (24) is arranged between the interpolation filter (23) and the accumulator (25). Musical instrument according to one of Claims 15 to 17, characterized in that readout device (22), interpolation filter (23), accumulator (25) and possibly amplitude control device (24) are combined in one sound generation module (8-10) and several sound generation modules (8-10 ) are arranged in parallel. Musical instrument according to claim 18, characterized in that the output signals of the individual sound generating modules (8-10) have a predetermined phase shift with respect to one another. Electronic musical instrument according to claim 18 or 19, characterized in that the sound generating modules (4-14) has at least one transputer (4-7) for processing the sound features into sound signals. Musical instrument according to claim 20, characterized in that a plurality of sound generating modules (8-10) are provided, the communication between the reading device (1-3) and the sound generating modules (8-10) using at least one transputer (5-7) he follows. Musical instrument according to claim 20 or 21, characterized in that the transputer (19, 21-24) in each module forms at least the digital interpolation filter (23), in particular a plurality of digital interpolation filters (23) arranged in parallel, which process several sound patterns in parallel. Device for loading a storage device for an electronic musical instrument according to one of claims 14 to 22, with a receiving device (24) for generating sound patterns in the form of electrical or magnetic signals from sound waves, a sampling device (26) which converts the sound patterns with a sampling frequency samples predetermined times and generates samples, and with a memory loading device which stores the samples in the memory device (1) inscribes, a pre-emphasis device (25) being arranged between the recording device (24) and the scanning device (26), which amplifies the sound patterns in predetermined frequency ranges. Device according to claim 23, characterized in that the pre-emphasis device (47) amplifies the sound patterns in a frequency range which is below 50% of the sampling frequency. Apparatus according to claim 23 or 24, characterized in that the pre-emphasis device has a frequency-dependent gain characteristic which is essentially inversely proportional to the frequency-dependent pass characteristic of the interpolation filter (23). Device according to one of claims 23 to 25, characterized in that the pre-emphasis device (47) attenuates frequency components above the cut-off frequency (f _G ).

Images