DD141356A1 - METHOD FOR THE ACOUSTIC GUETE ASSESSMENT OF MUSICAL INSTRUMENTS - Google Patents

Abstract

The invention relates to a method for objective quality assessment of musical instruments, which besides control and Ensuring the acoustic quality also important conclusions to further improve the sound quality permits. The goal of Invention is a meaningful and economical Procedure for comprehensive objective assessment of sound quality of musical instruments. The invention is thereby the task is to develop a method which the essential acoustic parameters in their entirety and interaction considered. According to the invention, this object is achieved solved that a five-stage Evolutionssystera, as in Fig.1 illustrated, designed with the following basic characteristics becomes: - Pypamid structure with largely horizontal and vertical breakdown structure - Central access to each subsystem. - Frequency selection and derivation of intensity and time characteristics in a first processing stage -Reduction of the side parameter in a second processing stage - Reduction of the frequency parameter in a third processing stage - Reduction of the game parameters in a fourth processing stage - Reduction of elementary codes in a fifth processing stage. -

Description

210 352

a) Title of the invention,.

Method for acoustic quality assessment of musical instruments

b) Field of application of the invention

The invention relates to a method for the acoustic quality assessment of musical instruments, which allows the derivation of meaningful Maßzahleη for describing the sound quality of musical instruments The method thus forms the basis for an objective quality assessment of musical instruments (modern GQA and quality assurance) and obtained due to the stormy development of microprocessor technology θ with the trend towards a complex, hierarchically structured information system (microprocessor hierarchy) is particularly topical «

1 O 352

After mastering a complex program of basic and applied research, this opens up completely new technical perspectives for ongoing quality control in production right through to automated production and manufacturing control.

Lifting control and assurance of the acoustic quality also result in important conclusions for the further improvement of the sound quality. The acoustic examinations make it possible to selectively focus the target direction of physics, construction and technology on the acoustically unsatisfactory bottlenecks. Such research is particularly important from an economic point of view (material saving, material substitution). Application · of the method in the context of an effective analysis-synthesis research concept finally results in essential impulses for electronic sound generation (electronic keyboard instruments, synthesizers, computer music) · As a partial result, the possibility of automatic music transcription opens up here (creation of the score based on the music signal) ,

c) Characteristic of the known technical solutions

The previously known methods and methods for acoustic quality assessment of musical instruments always relate only isolated aspects in the field of acoustic quality assessment. According to a strictly analytical approach, the individual acoustic phenomena are examined separately. As a rule, this is purely empirical, with additional knowledge from physics, objective signal analysis and psychophysics leading to a certain narrowing of the search space for possible effective solutions. "

All known methods and methods are particularly liable to the lack of ,, that they have no clear target criterion

-3- 210 352

so that there is not even an effective solution to the basic problem of acoustic quality assessment of musical instruments. In particular, the following deficiencies are highlighted:

1 · The feature selection methods found essentially by intuitive selection lead to an "objective" signal description, the nui? includes a certain part of the information relevant for the acoustic quality assessment, wherein moreover an information representation results, which has little direct relation to the subjective quality parameters. The resulting gap between the results of the empirically intuitive feature search on the one hand and the results of subjective evaluation tests (auditive assessment) on the other hand is attempted to be bridged with static methods (correlation analysis, etc.). However, this way of finding correlations between objective signal description and subjective sound perception has not yet led to satisfactory results.

2. Due to the detached analytical approach, partial procedures are often overrated. The focus is not on the optimization and re-activation of the overall system, taking into account the existing complex exchange relationships, but only on increasing the informative value of the subsystem, often under all available means. This one-sided approach leads JS * B * to an unreasonably high effort to measure certain physically relevant parameters; this would be much more effectively possible within the framework of an optimally coordinated complex system. The isolated analytical approach has led to a widespread dispersion in the search for an appropriate information reduction *

3 · This is in particular due to an often instrument-specific oriented view of the Binze!

210 352

aggravated. With that, however, the focus is on the uniform and consistent functional aspects of a complex measuring system. Cybernetic methods and strategies are not essential here. The many individual methods and individual methods are therefore also from this point of view as less effective.

4. The known methods and methods relate in the majority to the derivation of the quasi-stationary signal characteristics (tuning, level, duration of sound, spectrum). The dynamic signal characteristics (transients, frequency transients, amplitude, frequency and phase modulations, noise pulses, etc.) are hardly considered. Extensive studies of physiology and psychophysics, music synthesis, automatic speech recognition, etc., but it is known that just play the dynamic signal characteristics a serious role * It must therefore be noted that important features for the acoustic quality assessment is not used. This can be concluded on a total of only low significance of the known systems and methods.

Although there are already known attempts to use SDVs for the rationalization of acoustic quality assessment sub-tasks, due to the lack of suitable methods for solving the problem of acoustic quality assessment, the actual possibilities of EDP in this field have not yet been developed. This is further reinforced when one considers that in the near future, the use of microprocessors for day-to-day production control through to automated production and manufacturing control will be crucial. "The absence of suitable algorithms for objective and complex quality assessment is a significant bottleneck for the introduction of modern rationalization methods in the musical instrument industry.

-> - 210 352

Thus, no method has yet become known concerning the solution of the problem of objective acoustic quality assessment of musical instruments in its entirety. This represents a major obstacle to the end-to-end rationalization of the musical instrument industry and touches on the crucial issue of quality improvement as well as a leg of related problems.

d) Object of the invention

The object of the invention is to provide a meaningful and economical method for the comprehensive objective assessment of the sound quality of musical instruments. ..

e) Presentation of the essence of the invention

The invention has for its object to provide a method for - "acoustic quality assessment of musical instruments, which takes into account the essential acoustic parameters in their entirety and interaction. Overall, the process is characterized by high informative value and performance, is characterized by a high, processing economy and ease of computing and is therefore efficient and of relatively low cost.

According to the invention, this object is achieved in that a five-stage evolution system is designed with the following basic characteristics;

I. Pyramidal structure with largely horizontal and vertical structure breakdown.

II. Central access to each subsystem (Active element).

III. Frequency selection and derivation of intensity and time parameters in a first processing stage.

IV. Reduction of the time parameter in a second processing stage.

V. Reduction of the frequency parameter in a third processing stage.

VI. Reduction of the game parameters in a fourth processing stage.

VII. Reduction of elementary numbers in a fifth stage of processing.

to I, pyramidal structure

The pyramidal structure is characterized firstly by the fact that there is an extensive breakdown of the structure into subsystems lying in a horizontal plane. This is identical to the assumption of a hierarchically organized Informationsverarbeiturigssy stems:

1. Breakdown of the overall system into a total of η subsystems or processing stages

2. Control of the function of a subsystem V "of the respectively upstream subsystem V, (absolute hierarchy)

3 · realization of complex interactions (hierarchy in the generalized sense),

Secondly, the pyramid structure is characterized by a differentiated structure breakdown of the individual subsystems. * This is identical to the assumption of parallel processing:

1. Breakdown of each processing stage V in the relatively autonomous parent system

2. Realization of complex interactions between the elementary systems

Third, finally, the pyrainoid structure is characterized by a targeted compression of the information from the bottom to the top *. This is also expressed by the fact that the number of processing steps is higher

-7- 210 352

Elementary systems usually decreases (convergence principle predominates). The time window of the processing increases towards the top.

The conception of a continuous pyramidal structure is based on the insight that biological evolution has arrived at extremely powerful and problematic measuring systems in just this way. So far, the technical realization of this general principle has failed because of blurred theoretical ideas about the function and structure of such systems as well as the still largely lacking technical possibilities of realization. However, with the recent theoretical information processing and the rapidly developing microprocessor technology with a trend towards a complex hierarchical information system, today a significant step has been taken towards the fusion between the theoretical solution concept and the technical realization possibilities.

to II. The active element

The active element has the following basic components:

1. Central information processing unit for the realization of specific information processing processes

2. Read-only memory for fixing the specific processing procedures of the element

3. temporary storage or operative storage for accumulating intermediate data

4 · Input and output systems.

5. Control unit for coordinating the subfunctions

The formulation of the active element is based on the insight that corresponding elementary structures are realized on the neuronal level in biological systems. The technical analogue to the active element is the calculating machine. The totality of active elements forms

210 352

the highly complex pyramidal structure. The technical concept that most accommodates these ideas at present is the microprocessor hierarchy.

With the conception of the active element as the elemental component of the highly complex processing system, the following advantages are initially linked;

1 · Storage of intermediate values possible

2. The data can be recalled at any time for comparisons, secondary evaluations, remote data transmission, etc.

3 · Output of documentable intermediate results · possible

4. The decisive consequence is the step-by-step optimization of the overall system in the sense of a specifically adaptive evolutionary strategy.

to III. The first processing stage:

The conception of this first processing stage is based on the following findings on basic principles of the peripheral hearing system:

1st sequence selection (local principle)

2. Evaluation of intensity and time variables in the auditory nerve

3. Temporal integrations and temporal differentiations.

According to the invention, the following measuring system is proposed:

1. filter bank with.-Suitable choice of corner frequencies Cz.B · frequency group width, filter overlapping; octave wide)

2. Downstream a method 1 for the formation of amplitude measurements of the filter channel time function (F-function), with the time-limited intervals in the order of 30 ms are evaluated.

3. Furthermore, a method 2 for forming envelope measurements of the F-function (level gradient) with high time resolution ( & 1 ms) continues.

-9- 210 352

4 · Also followed by a method 3 for the low-inertia formation of time characteristics of the F-function.

The conception of these subsystems is not new in itself:

1- * generally known

2 - »z. Establishment of Kupfmüller 1964 (patent

Scripture DB-PS.1 163 567;

Method and apparatus of Fuchs 1971

(Patent DD-PS 84 747)

3 -> Procedure of Blutner 1978

(Patent in preparation) 4-τ ¹ method and arrangement of Steinbuch 1962 (DE-PS 1 122 274)

However, hay is the specific combination of these methods with the aim of economically and comprehensively recording all the signal parameters that are essential for acoustic quality assessment. It is an optimally coordinated measuring system whose subsystems fulfill very specific functions in the context of the overall system:

/ !../Selective acquisition of the quasi-static spectral signal properties

(continuous spectral time changes with a temporal resolution ^ 30 ms):

Spectrum, in particular in the relatively stationary signal region (core region).

slow in-and-out processes slow compensation processes slow fluctuations of spectral signal parameters

/2./Selective detection of the rapid changes of the signal envelope:

Transient responses below 50 ms Amplitude fluctuations above 16 Hz

/ 3 · / '&) Selective acquisition of the quasistatic properties of the fine time structure:

 Mood ·

Partialton shift slow pitch and phase changes

210 352

b) Selective detection of the dynamic properties of the fine time structure: Frequency transients (portamento) phase transients * frequency and phase fluctuations Noise pulses

This measurement system shows an extensive agreement with the biological processing system. "Thus, the fundamental target criterion of the acoustic quality evaluation is already taken into account in the first processing stage: Highly specialized measuring devices realize a first data reduction (filtering out the information relevant for the acoustic quality assessment). The data processing is carried out in such a convenient way that at the higher processing levels, the further data reduction in the direction of less meaningful quality features is possible. This prevents the emergence of a gap between objective signal description and the subjective perception of sound from the outset «

The measuring system was designed on the basis of uniform and consistent functional aspects. The adaptation to the processing principles proven in the course of biological evolution thus ensures the optimization of the overall system according to the specific acoustic signal properties. The economics of the overall system is made possible in this case essentially by the high selectivity and signal matching of the subsystems and their optimal tuning. "The uniform and consistent functional aspects of the solution according to the invention therefore do not exclude an adaptation to the concrete specifics of the musical instruments, but make them virtually possible.

210 352

In addition to the quasi-static signal characteristics (/ 1 /, / 3α /), the solution according to the invention also includes the dynamic signal properties (/ 2 /, / 3 b /). This takes account of a large number of investigations which signal the seriousness of these properties. It must therefore be stated that the concept according to the invention comprehensively takes into account the essential characteristics for acoustic quality assessment. This is the

high significance of the overall system from the conception of the first stage of processing secured.

to IV. The second stage of processing:

As output variables of the selective subsystems of the 1 × processing stage, there are certain functions for each frequency channel which are called Α functions. According to the invention, the reduction of the time parameter takes place at the second processing stage in such a way that characteristic characteristic values are extracted within a specific time interval (time window). There are two principal possibilities:

1. Formation of Amplitude Measurements 2 "Formation of Timing Measurements

For this purpose, the already known from the measurement technique methods offer:

1.1 maximum value formation

1 "2 Arithmetic mean 1.3 RMS value

1.4- Otherwise method

2 * 1 t-Werfc to specified Α-threshold (quantile)

2.2 t value to characteristic points of the waveform (maximum, inflection point, etc.)

2.3 Otherwise. method

-Ia dependence on the choice of the time window (width and shape of the time window) will continue to result in a whole series of modifications. "In principle, here are two cases:

210 352

1. The time window is chosen so that the entire duration of a single sound is detected sufficiently.

2, fixed time window with defined measuring interval length (about 30 ms). The beginning of the time window, that is to say the starting time of the measuring interval, is thereby optimally determined in accordance with the signal amplitude (clock synchronization with the signal discontinuities),

Case 1 is the simplest elementary case for the processing of Binze! klängeη. Case 2 takes into account the principles of timing and clock synchronization used at the higher level of biological evolution in close adaptation to nature's acoustic signal structure.

In case 1, the so-called frequency domain characteristic diagrams (FK diagrams) result directly from the mentioned measuring methods 1.1. 2.3, in the case 2 FK diagrams for the successive time segments t. This results in an FK-mountain containing the discrete time parameter t , To reduce this parameter, the method of classification (histogram formation) known from statistics is proposed. In particular, this has already become known relatively complex hardware sealings for data preparation in the context of music-acoustic sound research. In the context of the overall system designed here, the use of classification according to the invention pursues the goal of determining the essential parameters of the entire F-K mountains in a simple and rational manner. Thus, on the basis of the F-K mountains, even in the case of 2 F-K diagrams, these diagrams differ essentially from the representations according to case 1 in that the parameters of the entire time shape were taken into account comprehensively.

The partial algorithms of the described 2 processing stage are already known in detail. What is new is the specific combination of these elements in the context of a coordinated overall system for the acoustic quality evaluation of musical instruments.

2fO3S2

Analogous to the first processing stage, the basic target criterion of the acoustic quality evaluation is therefore also taken into account in the second processing stage. The frequency-differently structured FK diagrams (a few broad filter areas: coarse-structure features, several relatively narrow filter areas: fine-structure features) represent the immediate precursor for the derivation of meaningful quality features (see V.), which directly affect the subjective "elementary sound attributes". , Respectively. Thus, even at the 2nd processing stage, every ground is removed from the emergence of a gap between objective measuring results and subjective sound characteristics.

The approach, which is based on uniform and consistent functional points of view, reflects the extensive structuring of the biological perception system (local principle) in the specific arrangement of the resulting frequency domain characteristics in the LC diagram. This is the first time that a very simple and clear form has been found: finding information in a consistent way . The FK diagrams capture both the quasi-stationary signal characteristics and the dynamic signal characteristics in accordance with the upstream measurement systems. Differences in the signal properties relevant to the acoustic quality evaluation are thus expressed in a different configuration of the FK diagrams. This also reflects the peculiarities of a particular instrument in a given typical configuration.

The extensive formalization also allows a very efficient output for EDP of incurred intermediate values by means of graphical representations in the form of x / y diagrams on parallel printer, plotter, display. This presentation is largely adapted to the possibilities of visual invariant formation and thus very user-friendly. friendly. Here are crucial possibilities for a gradual system optimization, especially with regard to the downstream processing stages. That largely

2 1 ü 352

The empirical procedure is thereby decisively simplified and made more effective in the process of adaptation of the Bvolution system to the instrument specifics via the already optimized T _s il ^s 7 ^s ^ ^eme .

to V. The third stage of processing:

The task of the third processing stage according to the invention is to derive characteristic features for the elementary sound attributes. The entire configuration of the respective F-K diagram is taken into account. This

The aim is to extract the structural characteristics of the F-K diagrams that are essential for the acoustic quality evaluation.

The erfindiINGSgemäße solution of the problem is characterized in that for filtering the relevant features a learning system is used, which builds and stores the characteristic shape configurations during the learning phase and thus largely suppressed in the Kannphase a pattern comparison, the irrelevant properties.

The relevant characteristics for the quality evaluation can be calculated according to the general theory of object recognition on the basis of a specific F-K diagram no. T as follows:

U C

= Wi. fc. "" Xi. +. ... + Wi ₁ JJ K- "Xj ^ r. + * · · + W. -p, _c

XU L GC U TJI /

Where:

t = 1 ... T: number of the F-K diagram ki. = 1 ... K.: Number of the feature for the

UU

F-K diagram no "t"

f _fc = 1 ... F _fc : Number of the filter channel corresponding to the

F-K diagram no. "T"

210 352

: K-th feature for the sound quality entsprech ^u end by the FK-diagram ITr. "t ^lf sensing

th elementary sound attributes. The totality of these features forms the M-phrase.

xt: ΐ \ - dimensional feature vector of the

c υ

F-K Diagramins No. "t" (Column Rector) wf. : Reference coefficient vector (row vector)

iTfor all t: third-level reference knowledge.

The application of this method in the third processing stage is based on the insight that here another basic principle of biological information processing is considered: the formation of conditional links and subsequent summations in analogy to the Steinbuch learning matrix.

From this, the significant conclusion is drawn that an optimal reduction of the FK diagrams can be achieved by implementing corresponding methods. This is ultimately justified by the fact that the entire configuration of the FK contour is taken into account and thus in principle no circumcision of the relevant information takes place. The insignificant characteristics of the FK contour, which strongly tolerate the different realizations of the training sample, are filtered out by the learning system. The essential characteristics, on the other hand, only tolerate very little within the corresponding learning sample and are therefore fully captured by the learning system in the can phase. This means that the resulting M-phrase fully reflects the static and dynamic signal characteristics essential for acoustic quality assessment. The peculiarities of a particular instrument, which are decisive for the acoustic sound quality, are also completely reflected in the M-phrase:

The individual components of the M ~ set correspond directly to the characteristics of the elementary subjective sound attributes

210 352

(e.g., dark-light, dull-sharp, fully-empty, soft-rough, broad-thin, round-hollow, corporeal, dense, shiny, gloomy, vowelike). This also prevents the third stage of the emergence of a gap between objective measurement result and subjective sound perception.

The resulting features are thus also directly related to the subjective sound quality. This ensures the economics of the overall system: the subsequent processing stages can be designed very simply based on this "compactness" criterion while ensuring high performance. .

The consistent application of the principle of the learning matrix in the 3-stage processing allows a uniform and thus feasible technical implementation of

- EDVA or microprocessor (Liniarkifikator as proven

and computer-friendly variant).

- Adaptive threshold elements

to VI. The fourth processing stage According to the invention, the object of the fourth processing stage is to derive elementary quality parameters (Q-phrase) from the low-level M-set, which are combined in a subsequent fifth processing stage to form global acoustic quality indicators (G-phrase).

In the solution of this problem, it is first assumed that the acoustic quality characteristics of a particular instrument are found in the entirety of all the sound structures to be physically generated by the instrument, whereby the totality of all possible individual sounds is initially limited. In an expedient representation, one can represent the M-phrase in a M-dimensional parametric representation space. The parameters of this space correspond to the relatively independent characteristics for the elementary sound attributes. Each single sound can thus be assigned a point in the representation space.

-.17 -

The sum of the individual sounds for a particular instrument forms a point cloud. The specific structure of this point cloud contains the information about the acoustic sound quality of the instrument, but in a form that is less suitable for the evaluation. For this reason we enter a second parameter sat 25, the set of game parameters:

5 ₁ = pitch

5 ₂ = Anspielstärke So = type of excitation S * = register

Sn . - Transitions / Glissandi.

We. thus converts to the Mx - ^ dimensional parameter _ra . The totality of the individual sounds of a particular instrument is represented in this space as a 5-dimensional hypersurface.

The specific shape of the 5-dimensional hypersurface thus contains the essential information about the acoustic quality properties of the instrument. The advantage of this presentation globally lies in the fact that by introducing the game parameters into the parameter space, the physical and technological-constructional peculiarities of the musical instruments can be adequately taken into account. In "adaptation of the evolutionary system to the given environmental conditions," a substantial reduction in effort is possible compared to a general consideration of the object recognition problem.

If one assumes that in the general parametric Hevorenuatiortsraum (according to Meyer -Bppler) in principle an infinite variety of possible sound structures can be represented, it should be noted that due to the physical laws and technological-constructive and Mafcerialeigenheiten the classical musical instruments a whole decisive reduction of the quantity

- ¹⁸ - Al P 352

possible sound structures takes place. However, this means that in the parameter space only a limited number of characteristic hypersurfaces can occur, which just reflect these instrument-specific peculiarities.

The consideration of these specifics in the fourth processing stage as well as the observance of the specific peculiarities of the hearing system, especially in the upstream stages, are fundamental characteristics of the proposed system for acoustic quality assessment. This is the basis for a truly economical and efficient system in the sense of the Svelution strategy. In solving the task of quality assessment can be considered in particular from the fact that the d-characteristics are directly related to the sound quality of the instrument · This is due to the fact that d-Lierkmale the

Qn

correspond to elementary subjective sound attribute. In the estimation of quality, the subjective system of perception is unimaginably based on these quantities, as extensive studies on the method of subjective scaling show. However, this means that there is a simple monotone relationship between quality value and d-value.

In other words, the mapping of the instrument shafts in the parameter space leads to a specific structure of the object set. This structure is characterized by the fact that objects of close quality have only a small distance in the parameter space * They form a cluster. On the other hand, objects spaced apart in quality have a large distance in the parameter space. Thus, according to the compactness hypothesis, the parameter space is characterized by a typical identity structure. Bine such a structure is the prerequisite for the formation of measures of similarity for the effective solution of the valuation problem «

-19- 210 352

The merits of such a similarity assessment, as specified in the example embodiment, are:

1. Very vivid representation of the sound quality instrument features in the S-M chart. On the basis of the specific mode of representation selected, direct conclusions can be drawn from this with regard to the realization of subjectively essential elementary qualities as a function of the reflection parameters given by the design.

This is the basic requirement of using the quality assessment system beyond objective quality estimation and quality control. In fact, to decisively improve quality characteristics, it is not only important to know whether an instrument is good or bad. The decisive factor is rather the statement about which discrete properties are good and which are bad. From the S-M-Diagramni, the comparison of the instrument to be optimized with the Heister instruments reveals these critical areas to which the further efforts of physics, design and technology must be focused. The set of S-M diagrams represents, for the first time in a consistent way, a synthesis between subjective sound characteristics and the structurally determined possibilities of influencing them via the motion parameters.

2. Simple and computer-friendly as well as uniform algorithms while ensuring a high informative value of the entire system under the specific usage conditions.

3 · Ensuring a computer-friendly, step-by-step optimization procedure in line with the general evolutionary concept for improving the adaptation of the system to the specific characteristics of the groups of instruments to be examined.

4 ·. The resulting analogue quantities of the Q-set almost completely reflect the static and dynamic signal characteristics that determine sound quality. Information loss through discrete classroom

352

education accounts. This transition will take effect only in the process of forming hyper-classes at the following stage of processing.

on VII · The fifth stage of processing:

According to the invention, the object of the fifth processing stage is to map the analog Q set obtained on the input side into the global quality characteristic numbers (G set) 'via a hierarchical process of the super character formation.

The definition of the structure of this information cascade as well as the adjustment of the weighting coefficients ("point evaluation") represent detailed tasks in the optimization according to the concept of the transformation.

1. Musical-performance requirements of a certain style epoch

2. the sound concept of a regional time

3 · on. technological specifications and material requirements (cost function!)

4, to largely fashion-conditioned customer requirements.

Corresponding optimization strategies are very computer-friendly.

The whole concept of the system is thus rounded off by the fact that at the fifth stage of processing, all those factors which fluctuate greatly over time can be considered, which ultimately determine the disposal of a particular instrument. Dn's system of Acoustic Quality Assessment of musical instruments also allows a flexible and simple economic system adaptation in this respect.

10 3

f) embodiment

The invention will be explained in more detail below with reference to an example "in the accompanying drawings:

Pig.1: the block diagram of the whole procedure

Fig. 2: a demonstration of the selection of design criteria

3ί a demonstration of the selection of similarity criteria

Fig. 4? a demonstration of the hierarchy of supercharacters.

According to the invention, the acoustic signals are converted via a microphone 1 into electrical vibrations and then placed on a filter bank 2, for example, about 1? In the relevant listening area, it is possible to realize a frequency-group-side filter, but also a coarser frequency distribution, such as, for example, allows octave filtering. The resulting filter channel time functions are applied to a system and to known, but specifically assembled, processing units 3, 8, 11, 13 for selective acquisition of intensity and time characteristics.

The system 3 detects the temporally slowly changing spectral configurations and, as is known and customary, can be implemented via the following series of function stages: rectification, filtering, sampling, analog-to-digital conversion. As a result of this processing procedure, as a function of the frequency channel, the frequency-domain characteristics diagrams for the successive Zsitsegmente of the sample result. In order to reduce the time parameter, the method of level classification 4 known per se is used according to the invention. On the basis of the characteristic value-time diagrams for the individual frequency ranges, the density function ρ (χ) of the level is determined such that the time duration, the signal in a certain level class χ. spends, is measured.

210 352

To describe the density function, the following secondary parameters are derived. The mean output level of the corresponding filter channel can be interpreted as the center of gravity § of the area inscribed by ρ (χ.) (5). The determination of the maximum of the density function P _max Cx ^) (6) leads to the description of the decay, in particular in the case of ramped decay behavior. Deviations from the ramp can be effectively described by specifying Pearson's skewness and other secondary parameters (7).

In contrast, the system 8 detects the rapid spectral time changes and is composed according to an already known method as follows: half-wave rectification, envelope deformation, detection of envelope changes, sampling, analog-to-digital conversion. In order to reduce the time parameter, two ways are provided here. "Sinmally, a maximum value detection 9 supplies the hardness. Spectra representing the maximum gradient across the frequency channel. The configuration of these hardness spectra contains important features for the subjective tone hardness. On the other hand, roughness spectra can be derived via an average value detection IO for characterizing the mean amplitude fluctuations over the frequency channel. The configuration of these roughness spectra provides features for describing subjective sound roughness.

In order to select the quasistatic properties of the fine time structure, an instantaneous value detection 11 with a subsequent time averaging 12 is provided in a manner known per se. Here, the mean period between the occurrence of certain phase positions is measured. For the selection of the dynamic properties of the fine time structure, an instantaneous value detector with a downstream differentiating stage 13 is used. It is important to pay attention to the bandwidth of the upstream filter, so that the filter settling time is negligible compared to the signal changes to be measured. A downstream maximum detector

210 352

14 detects the speed; Frequenztransienteη for the individual frequency channels, a downstream time averaging the 'frequency fluctuations.

For the first time, the combination of the four measuring methods comprehensively covers the essential quasi-static and dynamic signal parameters in the formally uniform frequency range characteristic diagrams. Based on this, the subsequent processing stage 16 carries out the reduction of the frequency parameter via a selection of design criteria. In this case, according to the example shown in FIG. 2, a rough approximation of the v / vector takes place via adaptation of reference templates to the design configurations of the relevant frequency range characteristic diagrams. In the. Representation are plotted on the abscissa of the frequency ranges F and on the ordinate the reference template w or the frequency range characteristic K. When defining the reference template, the procedure is as follows:

1. Representation of typical classified samples (eg master instruments, series instruments of quality grades I-IIIj in picture KS 1 and KS 2) in the relevant frequency range characteristic diagram

2 "Visual assessment of the frequency domain characteristic contours for the individual instruments of each sample, in particular with regard to parameter dispersion and typical shape configurations.

3 "Accordingly, select a suitable Referenzschab-. lpne from a present repertoire of different Gestaltkonfigurationen

4. Optimal adaptation of the. Reference templates to the available training sample by definition of the template parameters (eg for the referential template in FIG. 2: f _u = 1 kHz, f _Q = 3 kHz)

- 24 - 210 352 '

Here it is to be noted that, by specifying the template, the respective w vector is uniquely fixed (learning process finished) and now, in the can phase, the current feature value of the individual implementations of each random sample can be calculated. In accordance with the target criterion, these resulting feature values are intended to separate the quality characteristics of the instruments as well as possible, that is to say instruments with widely differing characteristics. Quality should have very different d-values and instruments with roughly the same quality should differ only slightly in the d-phrase. Based on the resulting d-phrase, the subsequent processing stage 17 performs the seduction of the game parameters via a selection of similarity criteria. According to the typical identity structure of the maximum Mx5-dimensional parameter space, it is possible to proceed according to FIG. 3 as follows:

1. Splitting the 5-dimensional hypersurface into maximum K χ T χ Κ .. Game parameter sound characteristic diagrams (SM diagrams) where:

K: mean number of features for an F-K diagram Q?: 'Number of P-K diagrams

Ko "Number of game parameter combinations according to the design (eg Variation.S- - parameter (pitch) for 3rd values of S ₂ (mf, f, ff-int) for 2 values of S ^ (legato and staccato-int), p ^ and Β, - are kept constant, so that K 3 =

3x2x1 = 6

The abscissa of the SM diagram is plotted with the value of the spatial parameter S _y , and the ordinate d <value of the elementary timbre attribute

-25- 210 352

2nd learning phase

2.1. Representation of a classified sample (learning sample LP) with excellent quality characteristics (e.g., master instruments, Q-grade instruments) in the S-M diagram and determination of the ideal-approximate reference pattern for each S-M diagram through arithmetic averaging

2.2. Sending the standard deviation of the training sample from the mean reference signal for all values of the respective game parameter S _y

2.3. Optionally further optimization of the upstream shape detectors (reference templates) with the objective:

1. Referential patterns in the S-Ivl diagram are as smooth as possible (characteristic values should change constantly with constant variation of the game parameter; if possible, they should remain constant at all times

2 «The standard deviations belonging to the reference pattern should remain as constant as possible over S (smooth diffusion tube)

2.4. Definitive determination of the ideal (reference pattern) by:

1st average value

2. Function of standard deviations for each S-M chart

3 · Can phase

3.1. Representation of the instrument I to be evaluated with still unknown quality properties in the S-M diagram

3.2. Determination of the distance function-ά d (S ν for all

LL V J

Values of the respective game parameter

210 352

3.3. Determination of the mean distance of the instrument to be evaluated from the reference pattern and the fluctuation of the distance function over either the entire range of the respective S parameter or a suitably fixed subrange (for example in pitch parameter: bass midrange treble)

3.4-. Forwarding of the thus resulting elementary quality indicators to the subsequent processing stage 18th

For the considered in Fig. 3 S-M diagram arise as four elementary quality indicators. These indices relate to:

1 · & bsoluter <6d _u - value (eg Sc harfe at tri but) in the lower S - range 1 (eg pitch range)

2. Absolute äd value in upper S _v range 2

3. Balance of A ^ in the lower 3 _γ range

4. Balance νοηώ in the upper S range 2

In any case: the larger the index, the worse the quality compared to the reference sample (ideal).

These resulting elementary codes of the Q sentence are further reduced via a hierarchy of super character formation 18. FIG. 4 demonstrates this for the simplest case of a cascade of exposed pulses with downstream threshold elements. In the figure, analogous output-side G features are initially formed over 5 summations, these G features are further combined over two subsequent summation stages, and finally the resulting features are summed once more. The resulting final value is given to three threshold value switches with successively staggered threshold values S 1 -S 3, thus forming the three discrete features I-III. ,

Claims (6)

lrfihdungsanspruch A method for acoustic quality assessment of musical instruments for deriving meaningful metrics for the acoustic sound quality, characterized in that a five-level svolution system with extensive horizontal and vertical structural breakdown is designed for central access to each subsystem, which as a hierarchically organized information system, a targeted compression of the information from bottom to top in the direction of less global quality indicators. 2. Verfallren after item 1, characterized in that at a first processing stage, a frequency selection with subsequent evaluation of intensity and time ^ - metrics in a well-coordinated measurement system for selectively detecting the quasi-static spectral signals eigenstates, the rapid changes of the signal envelope, the quasi-static Properties of the fine Time structure and the dynamic properties of the fine time structure is made, which are screened through a first Datenredukbion the most essential evaluated by the hearing instrument static and dynamic signals Eigen th properties. 3 Method according to points 1 and 2, characterized in that, at a subsequent second processing stage, a reduction of the time parameter takes place in such a way that characteristic within a certain time interval. 210-352 amplitude and Zeitmeßwerte be extracted, in particular a maximum value formation, time averaging and level classification is performed and thus arise in a continuous manner frequency domain characteristic diagrams, which reflect the specific acoustic characteristics of a musical instrument in certain typical configurations of the determined secondary characteristic values on the filter channel , 4. Method according to points 1 to 3 », characterized in that on a further third processing stage, a reduction of the frequency parameter is performed, wherein a learning system is used for the extraction of the essential design features of the sequence area characteristic diagrams, which during the learning phase, the characteristic shape configurations builds and stores and thus largely suppresses the irrelevant properties in the can phase via a pattern comparison and which, based on the subjective processing system, uses a learning matrix to filter out the elementary subjective tonal attributes that are directly related to the subjective sound quality, resulting in further economic data reduction of the successor systems while retaining the essential static and dynamic signal properties. Method according to points 1 to 4, characterized in that on a further fourth processing stage an information representation takes place in the game parameter sound characteristic diagram, where on the abscissa the respectively examined game parameter is plotted and on the ordinate the elementary sound color attribute measured on the third processing stage and based thereon on the basis of the typical identity structure of this information presentation, whereby instruments which are closely related in terms of quality have only a small distance in the parameter space and differ in terms of quality Instruments show a large distance, through a formation of the mean distance between ideal (reference pattern) and instrument with still unknown quality characteristics further reduced characteristics for describing the mang quality; be derived. 6. Method according to points 1 to 5 »characterized in that, at a final fifth processing stage, the global quality indicators are adapted via a hierarchical process of super-character formation, which in the simplest case can be realized via a cascade of weighted summations with downstream threshold elements musical-performance requirements of a certain style epoch, the sound-sounding concept of a period of time, technological specifications and material requirements and fashion-related customer requirements are derived. Kierzü ...: i - .. Se? Ten drawings