JPS6391695A - Instrument sound reproduction system - Google Patents

Abstract

(57) [Abstract] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a musical instrument sound reproduction method for a sampling synthesizer that is capable of reproducing sampled sounds at a pitch different from that at the time of sampling.

"Prior Art" FIG. 4 is a diagram for explaining a conventional sampling synthesizer.

A desired sound 2, for example, a musical instrument sound, is converted into an electrical signal by a microphone 1), and sampled at predetermined time intervals by a sample-and-hold circuit 14 via an amplifier 13 according to a timing command from a control device 12. The sampled analog signal is converted into a digital signal by an A/D converter 15, and is stored in a waveform memory 16 as waveform data of the musical instrument sound.
is stored in The waveform data collected in this way is read out from the waveform memory 16 to the bumper register 18 by a keystroke signal applied from the keyboard 17, for example, and is read out from the waveform memory 16 to the bumper register 18.
It is converted back into an analog signal by the A converter 19 and sent to the amplifier 21.
After being amplified, the sound is emitted from the speaker 22 as a musical instrument sound.

In this way, when playing a collected instrument sound, if you want to play it back at the pitch at which it was collected, the captured waveform data will be played back to the original waveform as it is and the sound will be emitted. To emit a unique musical instrument sound, change the signal frequency of the read waveform data and emit the sound.For example, you can shorten the sampling time interval to half of the sampling time interval and emit the waveform data one after another. When a sound is made, the frequencies of the fundamental tone signal and its multiple harmonic signals included in the waveform data are also doubled, making it possible to generate an instrument sound with a pitch one octave higher. Alternatively, an instrument sound with a pitch one octave higher can be obtained by thinning out the collected waveform data and reading every other data without changing the reading speed from the waveform memory 16.

``Problems to be solved by the invention'' The sounds that people hear as musical instrument sounds are not directly generated by the sound source of the musical instrument.

When a sound source 1 of a musical instrument, for example, a string, vibrates, the main body of the musical instrument acts as a resonator for the vibration, and a vibration wave with a superimposed resonance effect is emitted and audible. When changing the pitch of a musical instrument's sound, the frequency of sound vibration of the sound source is changed. For example, the sound source vibrates at twice the frequency, and the instrument body resonates with the vibrations at twice the height, emitting an instrument sound that is one octave higher than the previous sound. At this time, the resonator characteristics of the instrument body hardly change due to vibrations of different frequencies generated from the sound source.

However, when conventional sampling synthesizers emit a musical tone with a pitch one octave higher, the collected and stored waveform data is emitted with the time interval reduced to half, so The resonant effect portion of the instrument itself will also be reduced in time to one-half, resulting in the generation of musical instrument sounds with different resonance effects depending on the pitch. In other words, for example, one formant due to the resonance characteristics of the musical instrument is present in the sampled waveform data as shown at point A of waveform A in Figure 5.
Hz, if you change the pitch by simply changing the readout time as in the conventional method, the formant representing the resonance characteristics will also move as the pitch changes. For example, for a pitch one octave higher, as shown in Figure 5. As shown by the four points in waveform B, the formant appears at 2 KHz. This is equivalent to making the resonator smaller. Therefore, when you play with a conventional sampling synthesizer, the size of the resonator changes depending on the pitch, but the size of the resonator does not change during the performance of an actual musical instrument, as shown in Figure 4. When musical instrument sounds are reproduced using such conventional musical instrument sound reproduction methods, it is inevitable that they will sound unnatural to the auditory sense.

``Means for solving the problem'' The signal of the musical tone to be reproduced is captured and the characteristic parameters that determine the timbre of that musical tone are determined. Then, an inverse filter having a characteristic opposite to the characteristic based on the characteristic parameter is constructed, and a second filter having the same characteristic as the characteristic according to the characteristic parameter is constructed.

In this invention, a musical tone signal to be reproduced is supplied to this inverse filter, and a residual signal in which the characteristics of the musical tone signal have been lost is extracted from the inverse filter and stored.

Furthermore, in this invention, this stored residual signal is read out,
The residual signal is subjected to pitch conversion according to the desired pitch,
supplying this pinch-transformed signal to a second filter;
By D/A converting the output of the second filter, an instrument sound of a desired pitch is obtained.

``Operation of the invention'' The frequency pinch of the residual signal related to the pitch of the musical instrument is converted according to the pitch to be reproduced, but the spectrum envelope characteristics separated from the residual signal, that is, the formant characteristics due to resonance, are It does not change depending on the pitch. Therefore, even if the pitch changes, the resonance characteristics of the musical instrument as a resonator do not change.

``Embodiment'' In this invention, first, the musical instrument sound that is actually perceived is converted into an electrical signal, and its formant characteristics are extracted. In addition to configuring an inverse filter with inverse characteristics, the residual signal containing only the sopifian component contained in the musical instrument sound is separated and stored.

There are several methods for determining the resonance characteristics contained in an auditory signal, depending on the target sound source.For example, (1), when a signal with a flat spectrum signal is obtained as an auditory signal.

(2) When the sound source can be regarded as a pseudo-impulse signal or a noise-like signal with a known spectrum, such as in a percussion instrument.

(3) In the case of a musical tone signal with a structure close to a line spectrum, such as a bowed string instrument.

Examples include.

In the case of terms (1) and (2), linear prediction method (L P
G) to find the autocorrelation or covariance of the auditory signal,
By solving the Yule-Walker equation to obtain linear prediction coefficients, it is possible to directly obtain the coefficients of a digital filter having a pulse transfer characteristic of resonance characteristics without explicitly obtaining resonance characteristics. Furthermore, if you want to know the spectrum envelope, you can also find it by FFT of linear prediction coefficients.

In the case of term (3), the spectrum of the sound source signal has a line spectrum (overtone structure) consisting of frequencies that are integral multiples of the fundamental frequency, so this must be removed using a comb filter or cepstral method. . In this case, the estimation error becomes large, so it is preferable to sample musical instrument sounds of a plurality of different pitches and average them.

Here, a case using the linear prediction method (LPG) will be explained.

Estimation based on the linear prediction method approximates resonance characteristics using an all-pole model, and musical instrument sounds are collected and converted into an X series of sampling values. In the linear prediction method, for example, m past sample values X1-8. The current value is estimated by the value obtained by multiplying and adding the coefficient sequences a I + a Z ~ a, which consist of X2゜~X, -1, respectively, and the linear predicted value YR is calculated as Yn'' (at -z +
~ +anXn-m).

When the error e7 between X and the predicted value Yfi is defined as e, Tomishima-Yll = Prediction coefficient al+ a! ~a, is determined.

The autocorrelation function R of the sampling value series X is R1=ΣX1
lX1). It is known that the k-th linear prediction coefficient a can be obtained by solving the Yule-Walker equation shown in Equation (1).

At this time, the spectrum envelope of the musical instrument sound X(kT) is approximated by 1/(1+Σa, exp(jkωT)).

FIG. 1 is a block diagram showing an example of implementing the musical instrument sound reproduction method according to the present invention.

A desired musical instrument sound is sampled in a conventional manner, and the waveform data is sequentially stored in the memory 31. After the sampling of the musical instrument sound is completed, the present invention extracts the spectrum envelope characteristic, that is, the formant characteristic, of the musical instrument sound. In this embodiment, the waveform data in the memory 31 is supplied to the linear predictive analysis section 32, and the characteristics of the spectrum envelope of the musical instrument sound are extracted. In this linear prediction analysis unit 32, as explained earlier, Y ule −W
Set up the alker equation and calculate the linear prediction coefficient a l + a
Find Z~a7.

From these linear prediction coefficients al+al~a, it is possible to construct a filter 33 having the spectrum envelope characteristic of the musical instrument sound, that is, the formant characteristic of the musical instrument, and an inverse filter 34 having the opposite characteristic.

FIG. 2 is a diagram showing an example of a filter 33 configured using n unit delay elements 35 and n coefficient units 36. n
unit delay elements 35 are arranged in series, the signal given to the first delay element 35A is delayed by the unit delay amount,
The signal is applied to the second delay element 35B. In this way, the signal is sequentially delayed by a unit delay amount by each of the delay elements 35A and 35B, and is finally applied to the n-th delay element 35N.

The delay signals output from each of these delay elements 35A, 35B to 35N are also input to a coefficient multiplier 36A.

36B to 36N. Each coefficient unit 36A.

36B to 36N are a I + a Z to a respectively
It has an amplification factor of m times, and the outputs of the coefficient multipliers are each supplied to an adder 37. Filter 33
The input signal to, that is, the residual signal is supplied from the residual signal input terminal 33A to the adder 37, and the coefficient units 36A, 36B~
36N, and is output from the output end 33B of the filter 33 as a signal having the formant characteristics of the musical instrument, and is also supplied as an input signal to the first delay element 35A.

Further, an inverse filter 34 having a characteristic opposite to the formant characteristic is constructed using the linear prediction coefficients al+ax~a1. FIG. 3 is a diagram showing an example of the configuration of an inverse filter.

In this inverse filter 34, the n unit delay elements 35 are arranged in the reverse order to the arrangement of the unit delay elements in the filter of FIG. That is, the input signal to the inverse filter 34, that is, the data string obtained by sampling the musical instrument sound, is supplied to the nth delay element 35N, and its output is supplied to the (n-1)th delay element 35M, where it is sequentially delayed until the first delay. Element 3
It reaches 5A.

Each delay output of these delay elements 35N, 35M to 35A is supplied to coefficient multipliers 36N and 36M to 36A, respectively.
The coefficients are multiplied by a1, a1-5 to a1, and then supplied to an adder 38, where they are added together and output.

After configuring the filter 33 and the inverse filter 34 in this way, the waveform data in the memory 31 is transferred to the inverse filter 3.
Supply to 4. The output of the inverse filter 34 is a signal from which formant characteristics have been removed from the spectrum, ideally resulting in a uniform spectrum. However, the inverse filter 34 actually obtained through the process from the procedure for determining formant characteristics to the construction of the inverse filter 34 is not an ideal inverse filter 34. Therefore, the residual signal obtained from the inverse filter 34 does not completely lose formant characteristics and contains only pitch signals, but has formant characteristics corresponding to the extent to which it deviates from the ideal inverse characteristic of the inverse filter 34. A remaining residual signal is obtained. The residual signal, which is thus made to contain only the Zopic signal, is stored in the memory 39. This memory 39 is a memory 31 in which waveform data sampled from musical instrument sounds is stored.
may also be used.

To reproduce the musical instrument sound, the residual signal stored in this memory 39 is read out and supplied to the filter 33. The residual signal, which has almost lost the formant characteristics of the musical instrument, is given formant characteristics by the filter 33, and is reproduced into the original musical instrument sound and output as a sound.

By the way, the sampling synthesizer not only needs to reproduce the original musical instrument sound using the signal read from the memory 39, but also needs to change the pitch of the musical instrument sound to a desired pitch and output it. In this case, the pinch component of the instrument sound is converted to the pinch component of the residual signal that remains as it is. That is, in the present invention, the pitch of the musical instrument tone is converted by pitch-converting the residual signal, and the resonance characteristics of the musical instrument are added to the pitch-converted signal. Pitch conversion methods include, for example, the conventional pitch shift technique1.
By reading every other piece of data from 9, it is possible to obtain a pitch signal having a pitch one octave higher than the collected musical instrument sound. Filter 33 filters the residual signal with double the pinch.
, the residual signal is output with constant formant characteristics that are not related to changes in pitch.

As described above, according to the present invention, the pitch is changed by performing pitch conversion on the residual signal, which is only the pitch signal from which the resonance characteristics of the musical instrument have been removed, and the pitch is changed to the residual signal from which the pitch is converted. Since a formant characteristic representing the characteristic is added, a sound close to the musical instrument sound when an actual musical instrument is played is generated.

As already explained, the cepstral method is a method suitable for removing the line spectrum structure and extracting the envelope of the spectrum. The pstrum is defined as the Fourier-transformed logarithm of the power spectrum, which is the square of the absolute value of the spectrum obtained by Fourier-transforming the waveform data.

A signal mainly corresponding to the spectrum envelope characteristic appears in the low quefrency cheek region of the quefrency axis, and a signal corresponding to the pitch component appears in the high quefrency region. Therefore, by applying a low-pass lifter to this cepstrum, a signal corresponding to the spectrum envelope characteristic is separated and extracted, and by inverse Fourier transform of this signal, a logarithmic power spectrum envelope characteristic is obtained.

Up to now, the formant characteristics of the filter have remained unchanged, but if the pitch changes significantly, the resonance efficiency of the instrument body may change depending on the fundamental frequency supplied from the sound source. In such a case, the formant characteristics of the second filter to which the residual signal is supplied can be changed according to the pitch by dividing the pitch into several intervals and finding a plurality of formant characteristics. Good too.

As described above, according to the present invention, a filter and an inverse filter having formant characteristics representing the resonance characteristics of the instrument are constructed from the musical instrument sound produced by an actual musical performance, and a residual signal from which the formant characteristic is removed from the musical instrument sound is constructed. Extract and memorize. The stored residual signal was subjected to pitch conversion according to the pitch, and then passed through a filter having characteristics representing the resonance characteristics of the musical instrument.

"Effects of the Invention" This invention provides a residual signal from which resonance characteristics of a musical instrument have been removed;
In other words, the pitch signal is subjected to pinch transformation according to the pitch, and the pitch-converted residual signal is passed through a filter having constant formant characteristics that is independent of pitch fluctuations. Therefore, it is now possible to generate instrument sounds that do not change the characteristics of the instrument as a resonator due to changes in pitch, and it is now possible to reproduce instrument sounds that provide a natural hearing sensation. became.

Moreover, although the explanation has been given so far in which musical instrument sounds are collected and outputted after changing the pitch, the present invention can be applied not only to musical instrument sounds.

[Brief explanation of the drawing]

Fig. 1 is a configuration diagram showing an embodiment of the present invention, Fig. 2 is a diagram showing an example of the configuration of a filter having formant characteristics of collected musical instrument sounds, and Fig. 3 is a diagram showing a configuration example of a filter having formant characteristics of collected musical instrument sounds. Figure 4 is a diagram showing an example of the configuration of a conventional sampling synthesizer, and Figure 5 shows the resonance characteristics when outputting an instrument sound sampled by a conventional sampling synthesizer with different pitches. FIG. 4 is a diagram for explaining that the pitch also changes as the pitch changes. 1): Microphone, 12: Control device, 13: Amplifier,
14: Sample hold circuit, 15: A/D converter, 1
6: waveform memory, 17: keyboard, 18: bass register, 19: D/A converter, 21: amplifier, 22: speaker, 31: memory, 32: linear predictive analysis section, 33: filter, 34: inverse filter , 35: Delay element, 36: Coefficient unit, 37.38: Adder, 39: Memory. Patent Applicant: Korg Co., Ltd. Agent: Takumi Kusano
1 k Thurs 20 L--111---------------1'' Thurs 3 Kokushu 4 Yen 5 Illustration Procedures Amendment Letter Released December 5, 1985 1. Display of the incident Patent application No. 61-2367282,
Title of the invention Musical instrument sound reproduction method 3, person making the amendment Relationship to the case Patent applicant Co., Ltd. KORG 5, subject of the amendment Drawing 6 of the specification, contents of the amendment (1) Figure 3 is corrected to the attached drawing.

Claims (1)

[Claims] (1) Take in a desired sound signal, find the characteristic parameters that determine its timbre, and use those characteristic parameters to construct an inverse filter that has the inverse characteristics of the sound and a filter that has the same formant characteristics as the sound. Then, the residual signal of the above-mentioned sound using the above-mentioned inverse filter is extracted and stored, and this residual signal is pitch-converted and then D/A-converted through the filter having the above-mentioned formant characteristics to reproduce it as a sound. Configured instrument sound playback method.