EP0820212B1 - Acoustic signal processing based on loudness control - Google Patents

Description

The invention relates to a method for the loudness-controlled processing of acoustic signals in sound processing equipment and to an apparatus for carrying out the method according to the preambles of the independent claims. The invention is particularly suitable for use in hearing aids for the hearing impaired; incoming acoustic signals are processed in such a way that the loudness subjectively perceived by the hearing impaired again always corresponds to the loudness perceived by normal hearing persons.

The idea of the loudness-controlled processing of acoustic signals has been known for some time and has been described by various authors in the specialist literature, for example also by N. Dillier et al. in the Journal of Rehabilitation Research and Development, Vol. 1, 1993, pp. 100-103 , The method is based on the fact that normal hearing and hearing impaired known test signals for assessing the subjectively perceived loudness are presented. As test signals, harmonic sine signals or narrow band noise are used. The subjectively perceived loudness depends on the signal power and on the frequency of a sine signal, or on the frequency of the dominant signal components of a complex signal. The subjective loudness data are on a normalized scale with value range [0, 1]. By comparing the information of a hearing impaired person with those of a reference group of normal hearing impaired, hearing loss-specific, loudness-dependent correction data can be determined. In a suitable signal processing method, these correction data are then used to prepare the hearing impaired the acoustic signals of his environment in the targeted manner. In the aforementioned article, in comprehensibility tests with a group of 13 hearing impaired, remarkable improvements in intelligibility were demonstrated.

Despite the audiological effect, the loudness-controlled processing in the previously known form can not be used in practice. Namely, as described in the article mentioned, the processing takes place by Fourier transformation of short signal segments, modification of the short-term spectra and back transformation of the modified short-term spectra into the time domain. Due to the segment-by-segment processing, the processed signal has a delay of almost 20 ms. In intelligibility tests, this delay does not matter. However, in practice, when the hearing impaired person also speaks and perceives his own voice as being delayed, it is totally unacceptable. In the method described in the article mentioned the duration of the individual segments is 12.8 ms, and this value can not be significantly undershot, because to obtain a useful short-term spectrum, a minimum segment duration of this magnitude is essential.

As an alternative to segmental processing, an attempt has sometimes been made to divide the acoustic signal into subband signals and to process the individual subband signals with separate gain values (see, eg, US Pat US 5 500 902 and EP 0 661 905 ). From practical tests it is known that improvements can be achieved when splitting up to three subband signals. A breakdown into more However, subband signals again lead to worse results. One reason for this may be the discontinuities of the transfer function arising at the subband boundaries. If one compares the frequency resolution of the short-term spectra of the segment-by-segment processing to the division of the signal into three subband signals, it is clear in any case that the potential of segment-wise processing can not be exploited with the alternative approach. And even if dividing into more subband signals could lead to improved results, the problem of significantly increasing signal delay would again arise.

Another aspect of the goodness of the sound-controlled signal processing has to do with the loudness model used in the processing. In contrast to the simple test signals, the signal power of speech, music and noise is distributed over a wide frequency interval in a time-dependent and complex manner. With a loudness model, a loudness value is assigned to these complex signals in a time-dependent manner, which in the ideal case coincides exactly with the loudness perceived by normal hearing persons. The value determined with the loudness model is used for the time-dependent control of the signal processing. The loudness model described in the article mentioned takes into account not only the total energy of a signal segment but also the center of gravity frequency of its short-term spectrum. To calculate the center of gravity frequency is based on fundamentals of E. Zwicker, who in his textbook "Psychoacoustics", Springer Verlag, Berlin, 1982, on pages 51 to 53 are summarized. From the spectral lines of the short-term spectrum, the energies E (z) of the individual frequency groups are formed in a first step and then, in analogy to the calculation of the center of gravity in mechanics, to a center of gravity frequency

c = Σz * E (z) / ΣE (z) (1)

billed on the bark scale marked z. If one wanted to realize this loudness model by dividing the signal into subband signals, a total of 21 subband signals of different bandwidth corresponding to the known frequency group widths would have to be formed for processing a bandwidth of 7700 Hz. In addition to the already mentioned, strongly increasing signal delay, this procedure would also require extraordinarily large computational resources. However, with the technologies currently available for integrated circuits, as well as for the approach with segmental processing, the implementation into a hearing aid with today's usual geometric dimensions and power consumption is excluded.

It is an object of the present invention to specify a method for the loudness-controlled processing of acoustic signals in sound processing devices, which can be used in particular in hearing aids. The loudness subjectively perceived by the hearing aid user should always correspond to the loudness perceived by normal hearing persons. In particular, the signal delay should be so small that a hearing aid user is not irritated by the delayed perception of his own voice when speaking. It should also reduce the computational resources compared to known methods for the loudness-controlled processing of acoustic signals. Furthermore, an apparatus for carrying out the method according to the invention is to be created.

The object is achieved by the method and the device according to the independent patent claims.

In the method according to the invention, the processing of the acoustic signal without Fourier transformation takes place, ie completely in the time domain, and also without division into subband signals. The peculiarity of the inventive method is that a characteristic of the loudness control ψ is calculated in an iterative manner and used to control a time-dependent correction filter. By the term "iterative calculation mode" is meant that a new value is calculated for the control variable ψ at each sampling time, using values which have the quantities necessary for their calculation in the respectively preceding sampling time. In contrast to the known segment-wise method, the loudness-specific control variable is thus determined not only as an average of successive signal segments but as a continuous time function. The short signal delay, typically measured at 2 ms, represents the observation time required for reliable estimation beyond the respective validity time point, and thus, unlike the segment-wise method, is not merely the result of an adverse property of the chosen implementation. The iterative method of calculation takes place in the method according to the invention by means of particularly efficient and at the same time original method steps.

The time-dependent correction filter is controlled by assigning new values to parameters of the correction filter at each sampling time by interpolation with the aid of the control variable ψ. In contrast to the segment-by-segment method, in which the hearing-impaired correction data are stored as gain values for the individual spectral lines of a short-term spectrum, in the method according to the invention for well-defined values of the control variable ψ coefficient sets for prototype filters are determined in advance and stored. The transfer functions of these prototype filters run along the corresponding gain values, which are determined in the segment-by-segment procedure for the individual spectral lines of a short-term spectrum. To characterize the prototype filters, sets of coefficients are used in the process according to the invention which are known to be suitable for interpolation, ie that the transfer function determined by interpolated coefficients is expected to run between the transfer functions determined by the coefficient sets on which the interpolation is based.

With the method according to the invention, completely new approaches are taken. There are in the mentioned article by N. Dillier et al. achieved good intelligibility results described. In addition, however, the inventive method reduces the signal delay to about 2 ms and at the same time achieves a drastic reduction of the computational resources. Therefore, it is possible to implement the erfndungsgemässe method in a hearing aid today common design.

The invention further relates to a device for carrying out the method according to the invention. This device contains a stage for the iterative calculation of the loudness characteristic control variable ψ and thus a time-dependent controlled correction filter stage, which prepares incoming acoustic signals target setting. The already mentioned drastic reduction of the required processing resources has different causes. On the one hand, the iterative calculation method eliminates the segment-by-segment buffering of the input and output signals. Then, saving the coefficient sets for the prototype filters also results in a substantial saving compared to storing the gain values for the individual spectral lines of the short-term spectra.

Fig. 1

ein Blockdiagramm der lautheitsgesteuerten Verarbeitung im Überblick,

Fig. 2

ein Blockdiagramm zur Ermittlung der für die Lautheit charakteristischen Steuergrösse,

Fig. 3

ein Signalflussdiagramm eines rekursiven Digitalfilters,

Fig. 4

ein Signalflussdiagramm einer einfachen Schätzwertberechnungseinheit,

Fig. 5

ein Signalflussdiagramm einer Schätzwertberechnungseinheit für die Signalleistung,

Fig. 6 und 7

Schemas zur Gewinnung von Tabellenadressen,

Fig. 8

ein Signalflussdiagramm einer Schätzwertberechnungseinheit für die Schwerpunktsfrequenz,

Fig. 9

ein Signalflussdiagramm eines nichtlinearen Glättungsfilters,

Fig. 10

ein Diagramm für den Zusammenhang interner Grössen des nichtlinearen Glättungsfilters,

Fig. 11

ein Schema für eine zweidimensionale Interpolation,

Fig. 12

ein Blockdiagramm der Interpolation von Parametern des Korrekturfilters,

Fig. 13

ein Schema zur Gewinnung von Tabellenadressen und Proportionalgrössen für Interpolationen,

Fig. 14

ein Blockdiagramm des zeitabhängigen Korrekturfilters,

Fig. 15

ein Signalflussdiagramm eines Kreuzgliedfilters zur Realisierung von Nullstellen,

Fig. 16

ein Signalflussdiagramm eines Kreuzgliedfilters zur Realisierung von Polstellen,

Fig. 17 und 18

Schemas für zweistufige lineare Interpolationen und

Fig. 19 und 20

Schemas zur Gewinnung von Tabellenadressen und Proportionalgrössen für Interpolationen.

In the following the invention with reference to the drawings and a detailed embodiment will be explained in more detail. Show it:

Fig. 1

a block diagram of the loudness-controlled processing at a glance,

Fig. 2

a block diagram for determining the characteristic of the loudness control,

Fig. 3

a signal flow diagram of a recursive digital filter,

Fig. 4

a signal flow diagram of a simple estimated value calculation unit,

Fig. 5

a signal flow diagram of an estimated value calculation unit for the signal power,

6 and 7

Schemas for obtaining table addresses,

Fig. 8

a signal flow diagram of an estimated value calculation unit for the center of gravity frequency,

Fig. 9

a signal flow diagram of a nonlinear smoothing filter,

Fig. 10

a diagram for the relationship between internal magnitudes of the non-linear smoothing filter,

Fig. 11

a scheme for a two-dimensional interpolation,

Fig. 12

a block diagram of the interpolation of parameters of the correction filter,

Fig. 13

a scheme for obtaining table addresses and proportional quantities for interpolations,

Fig. 14

a block diagram of the time-dependent correction filter,

Fig. 15

a signal flow diagram of a cross-member filter for the realization of zeros,

Fig. 16

a signal flow diagram of a cross-member filter for the realization of poles,

FIGS. 17 and 18

Schemas for two-stage linear interpolations and

FIGS. 19 and 20

Schemas for obtaining table addresses and proportional quantities for interpolations.

FIG. 1 shows the use of the inventive method and the method itself in a schematic overview. An acoustic signal is converted by a microphone 1 into an electrical signal, which is digitized by a signal converter 2 and then freed in a high-pass filter 3 of any offset and lowest-frequency interference signal components.

The essential steps of the method according to the invention consist in the processing of an output signal x of the high-pass filter 3. In a processing stage 4, the iterative calculation of the control variable ψ takes place. In a subsequent interpolation stage 5, the parameters of a time-dependent correction filter 7 are thus determined and transferred to this. A delay stage 6 provides with respect to the filtering with the correction filter 7 for the synchronization of the signal x with the filter parameter values derived therefrom by causing a corresponding signal delay, for example by 2 ms. At a sampling rate of 16 kHz, the delay stage 6 is advantageously designed as a cyclic buffer with 32 memory locations.

The filtered with the correction filter 7 signal y passes to a signal converter 8 and is converted there into an analog electrical signal. In an analog amplifier stage 9, it is amplified with a hearing impaired but constant gain value g _e , and then fed to an electro-acoustic signal converter 10. The value of g _e is determined in processing the coefficient sets for the prototype filters, such that the 16-bit number format used in the apparatus for performing the method is used as optimally as possible, with limitation of the processed signals due to the assumption in the apparatus However, saturation arithmetic only exceptionally effective.

As already mentioned, the loudness of complex signals can be determined on the basis of the total energy of short signal segments and the center of gravity frequency of their short-term spectra. The loudness depends approximately quadratically on the signal energy expressed on a logarithmic scale. As will be shown, the loudness model can be implemented in the method according to the invention with a two-dimensional linear interpolation. This interpolation gives more accurate results when the control quantity

ψ = (√L '- √L _min ) / (√L _max - √L _min ) (2)

Consequently, it depends approximately linearly on the logarithmic signal energy. In this case, L 'represents the loudness limited to the value range [L _min , L _max ], and L _min and L _max are sensibly selected minimum and maximum values of loudness, thus defining the working range of the method within which the correction filter due to smallest changes the loudness is constantly updated. Based on the formula (2), ψ is a control value normalized to the value range [0, 1], and for loudness values outside the value range [L _min , L _max ], the correction filter for ψ = 0 or that for ψ = 1 is used.

The block diagram in Fig. 2 shows in more detail how the control variable ψ is obtained from the input signal x. Compared with the known segment-wise method, in the inventive iterative signal processing method, a momentary signal power q takes the place of the signal energy of a short signal segment and an instantaneous center of gravity c takes the place of the center frequency of its short-term spectrum. These quantities are determined in processing stages 11-15. After a processing stage 13, corresponding output signal values c _r and q _r due to the iterative type of calculation still have an undesired scattering, which is eliminated in subsequent smoothing filters 14 and 15. The smoothed signals c and q are applied in a processing stage 16 to the already mentioned two-dimensional interpolation, wherein the successive output signal values ψr also have a still undesirable scattering, which is eliminated with a subsequent smoothing filter 17.

An essential aspect of the method according to the invention lies in the iterative calculation mode of the logarithmic signal power q as well as the center of gravity frequency c expressed on a bar scale, ie the conversion of the formula (1) into an iterative calculation scheme. Instead of forming the Frequency-group-specific energies E (z) in the inventive method, a frequency-selective weighting of the input signal x is made with a filter, which is hereinafter referred to as a frequency group filter. The frequency group filter is in Fig. 2 shown as processing stage 11, and its output signal is denoted by φ. Its frequency-dependent transfer function

H _FG (f) = √Δf _G (f) / √Δf _G (f _N ) (3)

results from the frequency group width function Δf _G (f). The denominator in formula (3) effects a normalization, where f _N denotes the Nyquist frequency, in the exemplary embodiment thus 8 kHz. The normalization serves the optimal use of the 16-bit wide fixed-point format specified in the exemplary embodiment. In the exemplary embodiment, the transfer function H _FG (f) is approximated by a second-order recursive filter 11. The structure of the frequency group filter 11 is for the sake of completeness in FIG Fig. 3 illustrated.

Instead of the weighting of the frequency group energies E (z) with the frequency group indices z in the numerator of formula (1), a frequency-selective weighting of the signal φ is carried out in the method according to the invention with a filter which is referred to below as a bark filter. The bark filter is in Fig. 2 shown as processing stage 12, and its output signal is denoted by θ its transfer function

H _B (f) = √z (f) / √z (f _N ) (4)

results from the pitch function z (f). The denominator in formula (4) in turn causes a normalization for the purpose of optimal use of the given number format. In the embodiment, the transfer function H _B (f) is also approximated by a second order recursive digital filter 12, which in turn maps the in Fig. 3 having shown structure.

With the signals θ and φ, it is possible in the method according to the invention to calculate the instantaneous centroid frequency according to formula (1) in an iterative manner. For this purpose, the quotient of its signal powers is calculated in a processing stage 13.

gilt, lässt sich die Geschwindigkeit steuern, mit der das Ausgangssignal v der sich verändernden Eingangssignalleistung folgt.For the iterative calculation of signal powers, in the method according to the invention a simple first-order estimated value calculation unit is used for the time-exponentially weighted expected value of the squared input signal. Such an estimated value calculation unit is for the general case, with input signal u and output signal v, in Fig. 4 shown. In this signal flow diagram, a new output signal value v results from the fact that the output signal value of the preceding sampling instant is multiplied by the constant (1-ε) and the square of the new input signal value u multiplied by the constant factor ε is added to this product. With the adaptation constant ε, for which $$\mathrm{0}\mathrm{<}\mathrm{\epsilon }\mathrm{<}\mathrm{1}$$

the speed at which the output signal v follows the changing input signal power can be controlled.

The simple estimate calculation unit of Fig. 4 has the disadvantages that a doubly wide number format is required for processing the squared input signal, and that the logarithm of the output signal v is additionally required for the subsequent calculations. Both Aspects in the inventive method in a simple way, as in Fig. 5 by embedding the simple estimate computing unit of Fig. 4 solved in a digital control loop.

The operation of the signal flow diagram in Fig. 5 is based on the fact that the variable v is controlled to a fixed setpoint. For this purpose, for each newly calculated signal value v, the incremental logarithmic increase or decrease in the signal power is determined, which corresponds to the deviation of the value v from the predetermined desired value. The desired logarithmic signal power p results in the sequence by simply accumulating the successive incremental change values. For the proper functioning of the control loop, it is necessary for each input signal value x to be scaled with a scaling factor which corresponds in each case to the estimated value p, and for the quantity v itself to be adjusted multiplicatively even before the renewed updating with an adjustment value corresponding to the change in power. The determination of both the incremental change as well as the scaling and Justierwerte takes place in the inventive method at each sampling time for values of the sizes v and p whose accuracy is limited by truncation to 6 or 7 decimal places. This makes it possible to use tables efficiently in which the 64 or 128 predefined, suitable values are stored. Finally, to address the tables then only need, as in the 6 and 7 shown to extract the relevant bit fields from the quantities v and p. In Fig. 5 is the table with the incremental logarithmic power changes denoted by Δp To save otherwise separately performed multiplications, the table contains S in Fig. 5 furthermore modified scaling values which were obtained from the original scaling values by multiplication with the root from the constant ε. For the same purpose, the calibration values in the table labeled A are already multiplied by the constant (1 - ε). Further is sufficient for storing the quantities v and p as well as all table values in Fig. 5 the usual 16-bit fixed-point format.

As already mentioned, in the method according to the invention the iterative calculation of the center of gravity frequency is based on the calculation of the quotient of the signal powers of the signals θ and φ, for example in the processing stage 13. The calculation of the signal powers is based on the in Fig. 5 mapped signal flow diagram returned. This results in the calculation of the center of gravity frequency in Fig. 8 illustrated signal flow diagram. The lower part of the diagram is identical to the Fig. 5 , It is used to calculate the power of the signal φ. The upper part is used to calculate the power of the signal θ. In this calculation, the scaling and adjustment values are taken from the lower part of the circuit, which compares the signal flow diagram in the upper part Fig. 5 simplified. With this arrangement, the optimal use of the number format is also guaranteed for the calculation of the power of the signal θ, and the sought center frequency results, as mentioned, by quotient of the two signal powers.

nur unwesentlich von 1 verschiedene Werte an, und anstelle der Division durch (1 + δ) kann der Quotient

        Q ≈ Z·(1 - δ)     (7)

durch Multiplikation des Zählers Z mit (1 - δ) angenähert werden.As in Fig. 8 1, the calculation of a quotient Q = Z / N formed from a counter Z and a denominator N is carried out on the basis of the signal power values already tracked with an adjustment value from the table A. This has the advantage that the otherwise required, unfavorable division can be significantly simplified. The denominator decreases in a number format standardized to the specified nominal value $$\mathrm{N}\mathrm{=}\mathrm{1}\mathrm{+}\mathrm{\delta \; with}\left|\mathrm{\delta }\right|\mathrm{<}\mathrm{1}$$

only slightly different from 1 different values, and instead of dividing by (1 + δ), the quotient

Q ≈ Z · (1 - δ) (7)

by multiplying the counter Z by (1 - δ).

As already stated above, the loudness can be determined from the signal power p and the center of gravity frequency c. The direct solution would be to place the signal flow diagrams in the Fig. 5 and 8th and their output signals after passing through smoothing filters of the interpolation stage 16 (see Fig. 2 ). However, the method according to the invention involves a further substantial simplification due to the fact that the frequency group filter 11 performs only a frequency-selective weighting of the input signal x. This makes it possible to modify the entries in the original interpolation tables in such a way that the same value results for the control variable ψ when, instead of the logarithmic signal power p of the input signal x, the logarithmic signal power q of the signal φ is used together with the modified tables. This eliminates the separate calculation of the signal power p, and the processing stage 13 in the inventive method Fig. 2 includes only the in Fig. 8 illustrated signal flow diagram.

As already mentioned, the successive signal values of the output signals of the processing stages 13 and 16 have an undesired scattering, which is eliminated with smoothing filters 14, 15 and 17. The use of classical, linear low-pass filter would be obvious, but is completely unacceptable because of the associated delay times in the inventive method. In its place, therefore, becomes a nonlinear Smoothing filter according to Fig. 9 used, which, in addition to minimal delay time yet characterized by significantly reduced computational effort. A new output value c is obtained by adding a correction quantity D to the output value of the preceding sampling instant. The correction quantity D is determined from the difference d, which results from the new input signal value c _r and the previous output signal value. The quantity d is first multiplied by a constant factor α> 1. In the smoothing filters 14, 15 and 17, the value of α is set to, for example, 2 and 3, respectively, and the result of the multiplication is limited to the value range [-1, 1] by a saturation arithmetic. The product w is then squared and limited to a value β, and the correction quantity D is obtained by multiplying the value thus calculated by the quantity w.

The effect of the nonlinear smoothing filter, whose signal flow diagram in Fig. 9 is shown is from the Fig. 10 understandable, which shows the connection of the internal quantities d and D. First, it should be noted that these smoothing filters make use of the normality of the signals to be filtered, so that their range of values includes the interval [0, 1]. Thus, the difference d assumes values from the interval [-1, 1]. In the Fig. 10 shown mapping curve D (d) is composed of five different curve parts 27.1-27.5. For absolute values of the difference d, for example for -0.2 <d <0.2, the correction quantity D in the third power depends on the difference d; this corresponds to a first curve part 27.1. The slight scattering of successive signal values with values from the value range [-0.1, 0.1] are thus effectively suppressed. For absolute values of the difference d, for example for 0.2 ≦ | d | <0.5, the mapping curve D (d) merges into linear parts; this corresponds to a second and third curve part 27.2 and 27.3. With significant changes in the input signal, these parts make sure that the output signal follows with only minimal delay. Finally, fourth and fifth parts 27.4 and 27.5 of the imaging curve, where a limit is set to a constant value, guarantees a smooth transition even in the case of extremely unsteady changes in the input signal d.

With the filtered center of gravity frequency c and the filtered signal power q, the calculation of the control variable ψ takes place in the processing stage 16. As already mentioned, this process takes place by means of a two-dimensional interpolation, which in Fig. 11 is shown in a detailed scheme. The schema comprises three tables. The table designated by ψ _{0 contains} the interpolation point values for fixed values of the input quantities c and q. The other two tables, denoted by ∂ψ / ∂c and ∂ψ / ∂q, contain the gradient values of the function ψ (c, q) in the direction of the c and q coordinates, which match the interpolation points. The value of the control variable ψ for arbitrary input signal values c and q thus results approximately by

ψ _r = ψ ₀ (c _i , q _k ) + (c - c _i ) · (∂ψ / ∂c) | _{ci, qk} + (q - q _k) · (∂ψ / ∂q) | _{ci, qk} (8)

where c _i and q _{k represent} the nearest to c and q interpolation point coordinates, which are at the same time not greater than c or q itself. Due to the input variables c and q normalized to the value range [0, 1], the values c _i and q _k as well as (c - c _i ) and (q - q _k ) can be simply masked out in the method according to the invention Fig. 11 determine bit fields from the quantities c and q. For addressing the table values, finally, the according to Fig. 11 summarized values c _i and q _k used.

Another aspect of the method according to the invention relates to the use of optimal table values in two-dimensional interpolation. The Values of the function ψ (c, q) at the corners of a rectangle defined by successive vertex coordinates are shown schematically as ψ (c _i , q _k ), ψ (c _{i + 1} , q _k ), ψ (c _i , q _{k + 1} ) and ψ (c _{i + 1} , q _{k + 1} ). Then in the process according to the invention the table values

ψ ₀ (c _i , q _k ) = ψ (c _i , q _k ) + [ψ (c _{i + 1} , q _k ) + ψ (c _i , q _{k + 1} ) - ψ (c _{i + 1} , q _{k + 1} ) - ψ (c _i , q _k )] / 4 (9)

(∂ψ / ∂c) | _{ci, qk} = {[ψ (c _{i + 1} , q _{k + 1} ) - ψ (c _i , q _{k + 1} )] + [ψ (c _{i + 1} , q _k ) - ψ (c _i , q _k )] / 2 (10)

and

(∂ψ / ∂q) | _{ci, qk} = {[ψ (c _{i + 1} , q _{k + 1} ) - ψ (c _{i + 1} , q _k )] + [ψ (c _i , q _{k + 1} ) - ψ (c _i , q _k )] / 2 (11)

used. Thus, the unavoidable interpolation errors are more evenly distributed than with the approximate table values ψ (c _i , q _k ), [ψ (c _{i + 1} , q _k ) - ψ (c _i , q _k )] and [ψ (c _i , q _{k + 1} ) - ψ (c _i , q _k )]. As has already been mentioned, the successive signal values ψ _{r have} an undesirable scattering, which is combined with the smoothing filter 17 (cf. Fig. 2 ) is eliminated. The output signal of the smoothing filter 17 is the control variable ψ, which in the interpolation stage 5 (see. Fig. 1 ) is used to determine filter parameters of the correction filter 7.

The interpolation stage 5 is in the block diagram of Fig. 12 shown in more detail. The control variable ψ arrives at a processing stage 18, from where it is masked out for the subsequent interpolations by the in Fig. 13 shown bit fields a table address ψ _a and a proportional size ψ _{f are} obtained. A processing stage 19 represents a 3-bit-wide counter whose count is denoted by j. In a processing stage 20, a gain value g of the correction filter 7 is determined, and in a processing stage 21, filter coefficients k _j ⁽ⁿ⁾ and k _j ^{(p) are} determined. The count value j and the interpolated filter parameters g, k _j ⁽ⁿ⁾ and k _j ^(p) are designated as a whole with m

The count value j and the interpolated filter parameters g, k _j ⁽ⁿ⁾ and k _j ^(p) reach the correction filter 7, which is shown in the block diagram of FIG Fig. 14 is shown in more detail. It comprises an amplifier stage 22, a cross-member filter 24 for the realization of zeros and a cross-member filter 26 for the realization of poles. For completeness, the structures of the cross-member filters 24 and 26 in the signal flow diagrams of FIGS. 15 or 16 reproduced in detail.

At each sampling instant, an interpolated gain value g reaches the amplifier stage 22 (cf. Fig. 14 ) and is multiplied by the input signal x _d delayed by, for example, 2 ms. The filter coefficients k _j ⁽ⁿ⁾ and k _j ^(p) reach processing stages 23 and 25, respectively, to which the counter value j is also passed. The processing stages 23 and 25 are merely switches which assign the interpolated filter coefficient values corresponding to the counter value j to the correct filter coefficients in the cross-member filters 24 and 26, respectively. The counter values 0 to 7 are assigned the filter coefficients with the indices 1 to 8 in ascending order.

The interpolation stages 20 and 21 (cf. Fig. 12 ) are in the Figures 17 or 18 detailed. As already mentioned, the hearing correction data determined from the individual loudness data in the method according to the invention are stored as filter parameters in a form suitable for interpolation. The gain is a logarithmic gain value

γ = γ ₀ (ψ _a ) + ψ _f Δ Δγ (ψ _a ) (12)

using the tables γ ₀ and Δγ, as in Fig. 17 shown, is interpolated from the input variables ψ _a and ψ _f . For the sake of completeness, it should be mentioned that in the present case of a one-dimensional interpolation, the table Δγ is omitted and the corresponding value can be recalculated each time by subtracting the read value γ _{0 from} the value tabulated below.

In order to determine the gain value g required in the amplifier stage 22, the value .gamma Fig. 19 shown bit fields, the address γ _a and the proportional value γ _f won. With their help becomes a gain value

g = exp (γ _a ) + γ _f · Δexp (γ _a ) (13)

from the tables designated by exp and Δexp which contain values of the exponential function, obtained in a further interpolation. The Fig. 17 thus represents a two-stage interpolation scheme, which makes use of the normality of the signal values and matched tables for the efficient determination of the required output value.

In the case of the filter coefficients, the hearing-impaired-specific values are stored in the form of the log area-ratio coefficients. In contrast to the amplification value, in each case only one coefficient of the two cross-link filters 24 and 26 is newly determined at each sampling instant. As already mentioned, the modulo 7 counter represented by the processing stage 19 controls the selection mechanism. In the two-stage interpolation scheme of Fig. 18 is therefore the three-bit value of the counter with the size ψ _a joined to the current table address. For each of the two cross-member filters 24 and 26, the log area ratio coefficient results

λ ^(ν) = λ ₀ ^(ν) (ψ _a , j) + ψ _f · Δλ ^(ν) (ψ _a , j) (14)

by interpolation with the tables λ ₀ ^(ν) and Δλ ^(ν) , where ν stands for one of the symbols n or p, which distinguish the cross-member filters 24 and 26 for the realization of zeros or poles.

The filter coefficients k _j ⁽ⁿ⁾ and k _j ^(p) required in the cross-member filters 24 and 26 are determined in a renewed interpolation, wherein from each of the log area ratio coefficients λ, in turn, by masking out the in Fig. 20 shown bit fields, an address value λ _a and a proportional variable λ _{f are} obtained. For the coefficients of the two cross-link filters 24 and 26, this process, as well as the subsequent interpolation itself, can take place one after the other, which in Fig. 18 is indicated with the multiplexer M and in particular has the consequence that the tanh and Δtanh designated tables of the tangent hyperbolic function must be stored only once. The filter coefficients

k ^(ν) = tahn (λ _a ^(ν) ) + λ _f ^(ν) · Δtanh (λ _a ^(ν) ) (15)

result from a further interpolation, whereby the normalization of the signal quantities and the corresponding tables are again used for the efficient implementation.

In summary, it can be said that in the method according to the invention for the loudness-controlled processing of acoustic signals in sound processing devices, an acoustic signal x to be processed is completely in the time domain is processed. In this case, starting from the signal x to be processed, a control variable ψ which is characteristic of the subjective loudness sensation is continuously calculated. The input signal x is processed with a time-dependent filter 7, the parameters of which are newly determined by means of the control variable ψ continuously by interpolation in user-specific correction data previously calculated and stored in tables and applied to the time-dependent filter 7. A device according to the invention for carrying out the method has a processing stage 4 for the iterative calculation of the control variable ψ and a time-dependent controlled correction filter stage 7.

Claims (29)

1. A method for loudness-controlled processing of acoustic signals in sound processing devices, wherein an acoustic signal (x) to be processed is processed entirely in the time domain such that based on the signal (x) to be processed, a control variable (ψ) characteristic of a subjective loudness perception of a person with ordinary hearing abilities is continuously computed, and that the input signal (x) is processed through a time-based filter (7), the parameters of which are continuously determined anew based on the control variable (ψ) through interpolation of user-specific correction data computed in advance and stored in tables, and the parameters are applied to the time-based filter (7).
2. The method according to claim 1, wherein the acoustic signal (x) is processed in an iterative manner without a division into sub-band Signals.
3. The method according to claim 1 or 2, wherein the control variable (ψ) is defined as the square root of the loudness standardized for a defined loudness interval.
4. The method according to one of the claims 1 through 3, wherein the control variable (ψ) is continuously determined through a two-dimensional interpolation, that is, on the basis of two iteratively computed variables, wherein a first iteratively computed variable (p) represents an estimated value for the current signal power expressed on a logarithmic scale, and a second iteratively computed variable (c) represents an estimated value for the center frequency of the current signal power distribution on a Bark scale.
5. The method according to claim 4, wherein the first iteratively computed variable (p) is determined by an iterative estimated value computing unit of the first order, which is embedded in a digital control loop, for a time based exponentially weighted estimated value for the squared input signal.
6. The method according to claim 4 or 5, wherein the second iteratively computed variable (c) is computed by dividing an iteratively determined dividend by an iteratively determined divisor, wherein the divisor is an estimated value for the current power of the signals (φ), which signal is weighted by a frequency group filter, and the dividend is an estimated value for the current power of the signal (θ), which, in addition, is weighted with a Bark filter, wherein the transfer function of the frequency group filter corresponds to the square root of a standardized frequency group width function, and the transfer function of the Bark filter corresponds to the square root of a standardized tonalness function.
7. The method according to claim 6, wherein the divisor and also the dividend are determined through an iterative estimated value computation unit of the first order embedded in a digital control loop for a time based exponentially weighted estimated value of the squared input signal, wherein the unit for determining the dividend receives the regulation signals from the unit for the divisor and applies them to its signals.
8. The method according to claim 6, wherein the division is computed based on the regulated estimated values and approximated by a multiplication with (1 - δ), wherein 1 is the target value and |δ|<< 1 holds.
9. A method according to one of the claims 5 - 8, wherein the scaling variables required for regulating the iterative estimated value computation unit and the incremental change values required for updating the logarithmic estimated value are read out from the tables (S, A, Δp) stored in advance.
10. A method according to claim 9, wherein the readout is performed from tables organized so that the table indices for finding the desired variables can be derived from the still unregulated estimated value (v) and the logarithmic estimated value (p).
11. A method according to one of the claims 1 through 3, wherein the control variable (ψ) is continuously determined through a two dimensional interpolation and thus based on two iteratively computed variables, of which a first iteratively computed variable (q) is an estimated value expressed on a logarithmic scale for the current power of a signal (φ) weighted with a frequency group filter, wherein the weighting is compensated by changing the entries in the original interpolation table, and a second iteratively computed variable (c) represents an estimated value for the center frequency of the current power distribution on a Bark scale.
12. A method according to one of the claims 4 through 11, wherein the control variable (ψ) and/or the first iteratively computed variable (p or q) and/or the second iteratively computed variable (c) are smoothed with a non linear filter and thus so that a new start value is generated through adding a correction value (D) to the preceding start value, wherein the correction value (D) is computed from the difference (d) between new input signal and preceding output signal and that the correction value (D) for small absolute values (|d|) of the difference (d) is a function of the third power of the difference (d), for medium absolute values (|d|) of the difference (d) it is a linear function of the difference (d), and for large absolute values (|d|) of the difference (d) it is constant.
13. A method according to one of the claims 4 through 12, wherein the interpolation of the control variable (ψ) is performed with tables organized, so that the table index for finding the interpolation point value and the incremental growth variables in both dimensions, and also the proportional variables by which the incremental growth variables are multiplied before being added to the interpolation point value can be generated from the iteratively computed variables (p or q; c) by simple masked reading of bit fields.
14. The method according to claim 13, wherein optimized values according to the following formulas are used in the tables for two-dimensional interpolation of the control variable (ψ): $${\mathrm{\psi }}_0\left({\mathrm{c}}_{\mathrm{i}},{\mathrm{q}}_{\mathrm{k}}\right)=\psi \left({\mathrm{c}}_{\mathrm{i}},{\mathrm{q}}_{\mathrm{k}}\right)+\left[\psi \left({\mathrm{c}}_{\mathrm{i}+1},{\mathrm{q}}_{\mathrm{k}}\right)+\psi \left({\mathrm{c}}_{\mathrm{i}},{\mathrm{q}}_{\mathrm{k}+1}\right)-\psi \left({\mathrm{c}}_{\mathrm{i}+1},{\mathrm{q}}_{\mathrm{k}+1}\right)-\psi \left({\mathrm{c}}_{\mathrm{i}},{\mathrm{q}}_{\mathrm{k}}\right)\right]/4$$

    

    $${\left(\mathit{\delta \psi }/\mathit{\delta }\mathrm{c}\right)|}_{\mathrm{ci},\mathrm{qk}}=\left\{\left[\psi \left({\mathrm{c}}_{\mathrm{i}\mathrm{+}\mathrm{1}},{\mathrm{q}}_{\mathrm{k}\mathrm{+}\mathrm{1}}\right)-\psi \left({\mathrm{c}}_{\mathrm{i}},{\mathrm{q}}_{\mathrm{k}+1}\right)\right]+\left[\psi \left({\mathrm{c}}_{\mathrm{i}+1},{\mathrm{q}}_{\mathrm{k}}\right)-\psi \left({\mathrm{c}}_{\mathrm{i}},{\mathrm{q}}_{\mathrm{k}}\right)\right]\right\}/2$$

    

    $${\left(\mathit{\delta \psi }/\mathit{\delta }\mathrm{q}\right)|}_{\mathrm{ci},\mathrm{qk}}=\left\{\left[\psi \left({\mathrm{c}}_{\mathrm{i}\mathrm{+}\mathrm{1}},{\mathrm{q}}_{\mathrm{k}\mathrm{+}\mathrm{1}}\right)-\psi \left({\mathrm{c}}_{\mathrm{i}+1},{\mathrm{q}}_{\mathrm{k}}\right)\right]+\left[\psi \left({\mathrm{c}}_{\mathrm{i}},{\mathrm{q}}_{\mathrm{k}+1}\right)-\psi \left({\mathrm{c}}_{\mathrm{i}},{\mathrm{q}}_{\mathrm{k}}\right)\right]\right\}/2$$

    
15. The method according to one of claims 1 through 14, wherein the values stored in tables for interpolating the user specific correction data are stored as amplification values in the logarithmic portion and as filter coefficients in the log-area-ratio-portion.
16. The method according to claim 15, wherein the interpolation of the user-specific correction data is performed with tables organized so that the table index for finding the interpolation point value and the table index for finding the proportional variable, by which the difference between the subsequent interpolation point value and instant interpolation point value itself is multiplied before being added to the support value, are generated from the control variable (ψ) by simple masked reading of bit fields.
17. The method according to claim 15 or 16, wherein the amplification value is determined from the interpolated logarithmic amplification value and the filter coefficients are interpolated from the interpolated log-area-ratio-coefficient through interpolation with stored tables of the exponential function and the hyperbolic tangent function and tables of the incremental variables values of these functions.
18. A method according to claim 17, wherein the interpolation is performed with tables organized, so that the table indices for finding the interpolation point values and the incremental growth variables and the proportional variables, by which the incremental growth variables are multiplied before addition to the support point values, are generated by simple masked reading of bit fields of the interpolated augmentation value and the interpolated log-area-ratio-coefficients.
19. A method according to one of claims 15 through 18, wherein the amplification value in each scanning interval and from the filter coefficients in each scanning interval respectively only the coefficients of a pole- / null-pair are determined, anew, wherein a invariably continuous sequence applies for Updating the filter coefficients.
20. The method according, to one of claims 1 to 19, wherein the input signal for the time-based filter is delayed, so that the filter coefficients and amplification values continuously determined anew through the computation of the variable (ψ) are applied to the signal in a timely manner, on which the computation is based.
21. An apparatus for implementing the method according to claim 1, comprising a processing stage 4 for iterative computation of the control variable (ψ), and thus a correction filter stage (7) controlled time based.
22. The apparatus according to claim 21, comprising two dimensional interpolation stage (16) for determining the control variable (ψ) from a signal power (q) and a center frequency (c).
23. The apparatus according to claim 21 or 22, comprising a frequency group filter (11) and a Bark filter (12) for determining the filtered signals (φ, θ) from an input signal (x).
24. The apparatus according to claim 23, wherein the frequency group filter and the Bark filter are configured as recursive filters.
25. The apparatus according to one of claims 21 through 24, comprising an estimated value computation unit (13) for computing the signal power (q) and the center frequency (c) from the filtered input signals (φ, θ).
26. The apparatus according to one of claims 21 through 25, comprising a smoothing filter (14, 15, 17) for eliminating an undesirable scatter of sequential signal values (c_r, q_r, ψ_r).
27. The apparatus according to one of claims 21 through 26, comprising a serial circuit with an amplifier stage (22), a cross member filter stage (24) for implementing nulls, and a cross member filter stage (26) for implementing poles.
28. The apparatus according to one of claims 21 through 27, comprising two-stage interpolation stages for determining the amplification value (g) and the coefficients (k_j ⁽ⁿ⁾ and k_j ^(p) of the correction filter (7) from the control variable (ψ).
29. The apparatus according to one of claims 21 through 28, comprising a signal delay unit (6) for synchronizing the input signal (x) with respect to processing through the correction filter (7) whose filter parameters are derived from the input signal (x).

Images