EP0957471A2 - Measuring process for loudness quality assessment of audio signals - Google Patents

Abstract

The method involves using filters, time blurring (low-pass filtering), level and frequency adjustment. The audio test signal is compared with a reference signal. Both signals or signal pairs are prefiltered and then separated by a filter bank (3) into frequency ranges. The filter bank characteristics and a subsequent time blurring of the filter output signals are used to produce an perceptual representation of the test signal for evaluation. By comparing the audible representations of the test signal and the reference signal after nonlinear conversion, an estimate of the hearing impression is made.

Description

The invention relates to a measuring method for aurally correct Quality assessment of audio signals according to the generic term of claim 1.

Schroeder, M.R.; Atal, B. S.; Hall, J.L: Optimizing digital speech coders by exploiting masking properties of the human ear. J. Acoust. Soc. Am., Vol. 66 (1979), No. 6, December, Seiten 1647 - 1652.

Beerends, J.G.; Stemerdink, J.A.: A Perceptual Audio Quality Measure Based on a Psychoacoustic Sound Representation. J. AES, Vol. 40 (1992), No. 12, December, Seiten 963 - 978.

Brandenburg, K.H.; Sparer, Th.: NMR and Masking Flag: Evaluation of Quality Using Perceptual Criteria.

Proceedings of the AES 11th International Conference, Portland, Oregon, USA, 1992, Seiten 169-179.

Measurement methods for the aurally correct quality evaluation of audio signals are generally known. The basic structure of such a measuring method consists in the mapping of the input signals to a time-frequency representation that is appropriate for the ear, a comparison of this representation and the calculation of individual value values to estimate the perceptible disturbances. Please refer to the following publications:

Schroeder, MR; Atal, BS; Hall, JL: Optimizing digital speech coders by exploiting masking properties of the human ear. J. Acoust. Soc. Am., Vol. 66 (1979), No. 6, December, pages 1647-1652.

Beerends, JG; Stemerdink, JA: A Perceptual Audio Quality Measure Based on a Psychoacoustic Sound Representation. J. AES, Vol. 40 (1992), No. December 12, pages 963-978.

Brandenburg, KH; Sparer, Th .: NMR and Masking Flag: Evaluation of Quality Using Perceptual Criteria.

Proceedings of the AES 11th International Conference, Portland, Oregon, USA, 1992, pages 169-179.

As is apparent from these publications, the used for the evaluation of coded audio signals However, models use FFT algorithms and therefore require one Conversion from the linear given by the FFT Frequency division to a hearing appropriate Frequency division. This is the temporal resolution suboptimal. In addition, the folding takes place with Smear function after rectification or Amount formation.

The invention is therefore based on the object objective measurement method for correct hearing Quality evaluation of audio signals using new ones fast algorithms for the calculation of linear phase filters to create, the duration of the audible interference taking into account the time change of the Envelopes calculated at the individual filter outputs and an ear-matched filter bank is to be used, whereby an optimal temporal resolution can be achieved should and with significant saving of computing time compared to other filter banks.

The achievement of the object according to the invention is characteristic of claims 5 and 7 characterized.

Further solutions or refinements of the invention are in characterized the claims 2 to 23.

A major advantage of the method according to the invention is to get a more accurate hearing model since audible disturbances taking into account the temporal Change of the envelopes at the individual filter outputs be calculated.

In addition, an ear-adapted filter bank is used, whereby an optimal temporal resolution is achieved, and the temporal behavior of the filter (impulse response etc.) directly with the level dependency of the Transfer functions correspond. The Phase information in the filter channels is retained. How already carried out takes place in the previously known Solutions the folding with smudge function only after Rectification or amount formation. A signal dependency the filter characteristics is achieved in that the Filter outputs before rectification / amount formation with a level-dependent smear function in the frequency domain be folded.

Because a new fast algorithm for recursive Calculation using linear-phase filters results a significant saving in computing time simple design and filters that are easier to vary than the conventional recursive filters used so far are.

Existing in the original signal and only in theirs spectral distribution changed signal components are from additive or by non-linearities Disorders separated, the separation by evaluating the Orthogonality relationship between the temporal courses of the envelopes at corresponding filter outputs of the signal to be evaluated and the original signal. The separation of these interference components corresponds better to that actual auditory impression.

• Aus jedem einlaufendem Impuls wird durch rekursive komplexe Multiplikation eine ungedämpfte Sinusschwingung mit der gewünschten Filtermittenfrequenz erzeugt.
• Die zu einem Eingangsimpuls gehörende Sinusschwingung wird durch Subtraktion des um dem Kehrwert der gewünschten Filterbandbreite entsprechende Zeit verzögerten und mit dem der Verzögerung entsprechenden Phasenwinkel multiplizierten Eingangsimpuls wieder abgebrochen.
• Durch Faltung im Frequenzbereich wird durch gewichtete Summation von je n Filterausgängen gleicher Bandbreite und um jeweils eine Periode versetzter Mittenfrequenz aus dem nach Schritt 2 resultierendem sin(x)/x- förmigen Dämpfungsverlauf ein der Fouriertransformierten eines cos∧ (n-1) förmigen Zeitfensters entsprechender Dämpfungsverlauf erzeugt. Hierdurch kann der Dämpfungsverlauf in der Umgebung der Filtermittenfrequenzen geformt und eine ausreichend hohe Sperrdämpfung ermöglicht werden.
• Der Dämpfungsverlauf in größerer Entfernung von der Filtermittenfrequenz kann durch eine weitere Faltung im Frequenzbereich bestimmt werden (Übergang zwischen Durchlaßbereich und Sperrbereich).

The filter bank algorithm is implemented in the following way:

• An undamped sine wave with the desired filter center frequency is generated from each incoming pulse by recursive complex multiplication.
• The sine wave associated with an input pulse is interrupted again by subtracting the time corresponding to the reciprocal of the desired filter bandwidth and multiplied by the phase angle corresponding to the delay.
• By convolution in the frequency domain, weighted summation of n filter outputs each with the same bandwidth and center frequency offset by one period results from the result of step 2 sin (x) / x-shaped Damping curve of one of the Fourier transforms cos ∧ (n-1) shaped Time window corresponding damping curve generated. As a result, the attenuation curve can be shaped in the vicinity of the filter center frequencies and a sufficiently high blocking attenuation can be made possible.
• The attenuation curve at a greater distance from the filter center frequency can be determined by a further convolution in the frequency range (transition between the pass band and the stop band).

Further advantages, features and possible applications of the present invention result from the following Description in conjunction with those in the drawing illustrated embodiments.

The invention is described below with reference to the drawing illustrated embodiments described in more detail. In the description, in the claims, the Summary and in the drawing are those in the list of reference numerals given below Terms and associated reference numerals used.

Fig. 1

eine Struktur des Meßverfahrens und

Fig. 2

eine Filterstruktur.

In the drawings:

Fig. 1

a structure of the measurement method and

Fig. 2

a filter structure.

The present measuring method evaluates the disturbances of a Audio signal by comparison with an undisturbed Reference signal. After filtering with the Transfer functions of the outer and middle ear are the Input signals through a hearing-adapted filter bank in converted a time-tonality representation. It will be the Amount squares of the filter output signals are calculated (Rectification) and it will fold the Filter outputs carried out with a smear function. The folding can be in contrast to the previously known Proceedings take place before or after the rectification. Level differences between test and reference signal as well linear distortions in the test signal are compensated for and evaluated separately. Then a frequency-dependent offset added to the intrinsic noise of the Model hearing and it becomes a temporal Output signals are smeared. A part this smearing of time can already be directly after rectification to save computing time. After the temporal smear (low-pass filtering) is then undersampling of the signals permitted. Through a Comparison between the resulting audible time-frequency patterns of test and reference signal can Series of output variables are calculated, the one Provide an estimate of the perceptible disturbances.

First of all, that in FIG. 1 is intended as an exemplary embodiment structure shown or the structure of the measuring method be explained. The test signals 1a, 1b for the left and right channel and the reference signals 1c, 1d, for the left and right channels are used for pre-filtering given on pre-filter 2. After the pre-filtering, the actual filtering in the filter bank 3 the spectral smear 4 and the calculation of the Amount squares 5. The boxes marked 6 in the Figures represent the temporal smear symbolically. This is followed by level and frequency response equalization 7, whereby output parameters 11 can also be supplied. After the level and Frequency adjustment 7, the addition of Own noise 8 and then the temporal smear 9.

The calculation of output parameters 11 takes place in the structure shown in the symbolically represented Block 10. The level and frequency response adjustment 7 can also between step or operation 9 and 10.

First, the calculation of the excitation pattern using the hearing-adapted filter bank 3 described.

The filter bank 3 consists of an arbitrarily selectable number of filter pairs for test and reference signal 1a, b or 1d, c (values between 30 and 200 are sensible). The filters can be distributed evenly on largely any pitch scales. A suitable pitch scale is e.g. B. The following approximation proposed by Schroeder: e.g. / Bark = 7 arsinh f / Hz 650

The filters are linear phase and are defined by impulse responses of the following form:

The value n determines the blocking attenuation of the filter and should > = 2.

The output values of filter bank 3 are spectrally smeared with 31 dB / Bark on the lower flank and between -24 and -6 dB / Bark on the upper flank to take account of the simultaneous masking, which means that crosstalk is generated between the filter outputs. The upper edge is calculated depending on the level: s = min -6 dB Bark , -24 dB Bark +0.2 Bark -1 · L / dB

The level L is calculated independently for each filter output from the magnitude square 5 of the corresponding output value that is low-pass filtered with a time constant of 10 ms. This smearing is carried out independently for the filters which represent the real part of the signal (Eq. 2) and the filters which represent the imaginary part (Eq. 3) of the signal. As an alternative, the level can also be calculated without a low-pass filter and, instead, the factor determining the crosstalk, which results from delogarithming the slope (Eq. 4), can be low-pass filtered. Since this convolution operation is quasi linear and therefore preserves the relation between the resulting frequency response and the resulting impulse response, it can be understood as part of the filter bank 3.

Since the filter bank provides 3 pairs of output signals with phases shifted by 90 °, the rectification can be carried out by forming the squares 5 of the filter outputs: E ( f c , t ) = A re 2nd ( f c , t ) + A in the 2nd ( f c , t )

The filter output signal is smeared in two stages. In the first stage, the signals are over a cos 2nd -shaped Time window averaged, which primarily models the pre-masking. The second masking is then modeled, which will be described in more detail later. The cos 2nd -shaped The time window has a length of 400 samples at a sampling rate of 48 kHz. The distance between the maximum of the time window and its 3 dB point is therefore approximately 100 samples or 2 ms, which corresponds to a period of time often assumed for the pre-masking.

Level differences and linear distortions (frequency responses of the test object) between test and reference signal 1a, b or 1c, d can be compensated and thus by the evaluation other types of faults are separated.

For level adjustment, the instantaneous amount squad rates at the filter outputs are smoothed in time by first-order low-pass filters. The time constants used are selected depending on the center frequency of the respective filter:

A correction factor corr _{total is} calculated from the filter output values P _test and P _ref smoothed in this way: corr total = Σ P test · P Ref Σ P test 2nd

If this correction factor is greater than one, it will Reference signal 1a; b divided by the correction factor, otherwise the test signal 1c; d with the Correction factor multiplied.

For each filter channel, correction factors from the orthogonality relationship between the time envelopes of the filter outputs of test and reference signals 1a, b; 1c, d calculated: ratio f , t = -∞ 0 e t τ · X test · X Ref German ∞ 0 e t τ · X Ref · X Ref German

The time constants are calculated according to Eq. 6 determined. If ratio _{f, t is} greater than one, the correction factor for the test signal is set to ratio _{f, t} ^-1 and the correction factor for the reference signal is set to one. In the opposite case, the correction factor for the reference signal is set to ratio _{f, t} and the correction factor for the test signal is set to one.

The correction factors are over several neighboring ones Filter channels, and temporally with the same time constants smoothed as indicated above.

A frequency-dependent offset for modeling the intrinsic noise of the hearing is added to the amount squares at all filter outputs. Another offset to take account of background noise can also be added (but is normally set to 0) E ( f c , t ) = E ( f , t ) +10 0.364 f c kHz -0.8

The current ones are used to model the subsequent masking Squares of amounts in each filter channel through a low-pass filter first order with a time constant of approx. 10 ms smeared in time. The time constant can also optionally depending on the center frequency of each Filters are calculated. In this case it is 50 ms for low frequencies and 8 ms for high ones Frequencies (like Eq. 6).

Before the second stage of temporal smearing just described, a simple approximation for the loudness is calculated by taking the squares of magnitude at the filter outputs high 0.3. This value E and the amount of its time derivative d E / dt are smoothed with the same time constants as already described. From the result of the temporal smoothing E _which is determined a measure for the envelope modulation in each channel:

The most important output parameter of the method, which is most correlated with subjective hearing test data, is the loudness of the interference when throttled by the useful signal. The input values for this are the amount squares in each filter channel E _ref and E _test ("excitation"), the envelope curve modulation, the intrinsic noise of the hearing ("basic excitation") E _HS and the constants E ₀ and α. The reduced noise level is reduced NL ( f c , t ) = 1 s test · E HS E 0 0.23 · 1+ Max( s test · E test - s ref · E ref , 0) E HS + s ref · E ref · Β 0.23 -1 calculated, where:

Eq. 11 has been designed here to provide the specific loudness of the disturbance when there is no masker and to provide approximately the ratio between disturbance and masker when the disturbance is very small in relation to the masker. The factor β determining the throttling is calculated according to the following equation: β = exp -α · E test - E ref E ref

The "throttled noise level" corresponds to the mean of this variable over time and filter channels. In order to determine linear distortions, the same calculation is carried out again without frequency response adjustment, the test and reference signals being interchanged in the equations given above. The resulting output parameter is referred to as "loudness of missing signal components". With the help of these two output variables, a good prediction of the subjectively perceived signal quality of a coded audio signal is possible. Alternatively, linear distortions can also be determined by using the reference signal as a test signal before the signal adjustment. Another output variable is the modulation difference, which results from normalizing the amount of the difference between the modulation of the test and reference signals to the modulation of the reference signal. When normalizing to the reference signal, an offset is added in order to limit the calculated values with very small modulation of the reference signal: Modulation difference = modtest - modref Offset + modref

The modulation difference is over time and filter bands averaged.

The modulation used on the input side results from Normalization of the time derivative of the instantaneous values their time-smoothed value.

2 shows a filter structure for recursive calculation a simple bandpass with a finite impulse response (FIR) shown.

The signal is processed separately according to the real part (upper path) and imaginary part (lower path). Since the input signal X is originally purely real, the lower path is initially missing. The input signal X is delayed by N samples (1) and after multiplication by a complex factor cos (N.ϕ) + j.sin (N.ϕ) subtracted from the original input signal (2). The resulting signal V is added to the output signal delayed by one sample (3). The result multiplied by another complex factor cos (ϕ) + j.sin (ϕ) gives the new output signal Y (4). The overlined identifiers for V and Y each mark the imaginary part.

The second complex multiplication sets the input signal periodically. The addition of the delayed and by the first complex multiplication of weighted input signal breaks the continuation of the input signal to N Samples again.

The entire filter, consisting of real and imaginary part output, has the amplitude frequency response A ( f ) = N · si N 2nd ϕ- 2 · π · f f A si 1 2nd ϕ- 2 · π · f f A ,

Where f _A denotes the sampling frequency.

The initially low blocking attenuation of these bandpasses can be increased by calculating K + 1 of such bandpasses with the same impulse response length N, but different values of ϕ in parallel, adapting their phase responses to one another by means of a further complex multiplication, and adding up their weighted output signals: A ( f ) = k = 0 K w k · A k ( f ) With ϕ k = 2 · π f M f A + k - K 2nd · 2π N (f _M : center frequency of the bandpass) and w k = 2π N · 2 - K · K k

The blocking attenuation of the resulting filter decreases with the (K + 1) th power of the distance between the signal frequency and the center frequency of the filter: The impulse response of the entire filter has the form a K ( n ) = sin K π N n · Cos 2 · π · f M f A · n | 0≤ n < N for the real part and a K ( n ) = sin K π N n · Sin 2 · π · f M f A · n | 0≤ n < N for the imaginary part. This corresponds to that in Eq. 2 and 3 described characteristics.

List of reference numbers

1a

Test signal, left channel

1b

Test signal, right channel

1c

Reference signal, left channel

1d

Reterence signal, right channel

2nd

Pre-filtering

3rd

Filter bank

4th

spectral smear

5

Calculation of the squares of amounts

6

temporal smear

7

Level and frequency response adjustment

8th

Addition of inherent noise

9

temporal smear

10th

Calculation of output parameters

11

Output parameters

Claims (23)

Measuring method for aurally correct quality evaluation of audio signals with the help of filters, temporal smearing, level and frequency response adjustment, characterized in that that the audio signal to be evaluated is compared as a test signal (1a, b) with an original signal supplied as a reference signal (1c, d), that both signals or signal pairs (1a, b; 1c, d) are separated into the frequency range after a pre-filtering (2) by a filter bank (3), that the characteristics of the filter bank (3) and subsequent temporal smearing (9) of the filter output signals produce a hearing-appropriate representation of the audio signals to be evaluated as a test signal (1a, b) and that by comparing the hearing-appropriate representations of test signal (1a, b) and reference signal (1c, d) after non-linear transformations, an estimate of the auditory impression to be expected is provided. Method according to claim 1, characterized in that the filter bank (3) is adapted to the hearing and generates an undamped sine wave with the desired filter center frequency from each incoming signal by recursive complex multiplication, that the sine wave belonging to a test signal (1a, b) is interrupted again by subtracting the input test signal (1a, b) delayed by the reciprocal of the desired filter bandwidth by a corresponding time and multiplied by the phase angle corresponding to the delay. Method according to claim 1, characterized in that one of the Fourier transforms one by convolution in the frequency range from n filter outputs of the same bandwidth and center frequency offset by the inverse of the window length cos n (n-1) shaped A corresponding damping curve is generated in the time window. Method according to one of claims 2 or 3, characterized in that that the attenuation curve at a greater distance from the filter center frequency in the transition between the pass band and the stop band is determined by a further convolution in the frequency range. Method according to the preamble of patent claim 1, characterized in that that an undamped sine wave with the desired filter center frequency is generated from each incoming test signal (1a, b) by recursive complex multiplication, that the sine wave associated with an input test signal (1a, b) is delayed again by subtracting the time corresponding to the reciprocal of the desired filter bandwidth and multiplied by the phase angle corresponding to the delay, the input test signal (1a, b) is stopped again, that by convolution in the frequency range from n filter outputs of the same bandwidth and the reciprocal of the window length offset center frequencies one of the Fourier transforms one cos n (n-1) - shaped Time window corresponding damping curve is generated and that the attenuation curve at a greater distance from the filter center frequency is determined by a further convolution in the frequency range. Method according to one of the claims 1 to 5, characterized in that that the input test signals (1a, b) and the reference signals (1c, d) are each introduced as an input variable for a left and a right channel, that is to say in pairs. Method according to the preamble of patent claim 1, characterized in that that the test signals (1a, b) and the reference signals (1c, d) are first subjected to pre-filtering (2), then passed into a filter bank (3), that spectral smearing (4) then takes place, then the calculation of squares of amounts (5) takes place, whereupon a temporal smearing is carried out, that the output variables thus obtained are subjected to a level and frequency response adjustment (7) and that afterwards an addition of intrinsic noise (8) takes place, whereupon in turn there is a temporal smearing (9) and a calculation (10) of output parameters (11) or step (7) is carried out between steps (9) and (10). Method according to one of the claims 1 to 7, characterized in that that after filtering with transfer functions of the outer and middle ear, input signals are converted into a time-tonality representation by a hearing-adapted filter bank (3), that amount squares (5) of the filter output signals are then calculated and a convolution of the filter output signals is carried out with a blurring function (6). Method according to claim 8, characterized in that the folding takes place before or after the rectification. Method according to one of claims 1 to 9, characterized in that that level differences between the test and reference signals (1a, b or 1c, d) and linear distortions of the reference signal (1c, d) are compensated for and evaluated separately. Method according to one of claims 1 to 9, characterized in that that part of the time smearing occurs directly after rectification. Method according to one of claims 1 or 5, characterized in that that an ear-matched filter bank (3) is used which achieves a signal dependency of the filter characteristics in that the filter outputs are folded with a level-dependent smearing function in the frequency range before the rectification / amount formation. Method according to one of claims 1 to 12, characterized in that that signal components present in the reference signal (1c, d) and only changed in their spectral distribution are separated from additive or from interferences generated by non-linearities and that the separation of these interference components is carried out by evaluating the orthogonality relationship between the time profiles of the envelopes at corresponding filter outputs of the test signal (1a, b) to be evaluated and the reference signal (1c, d). Method according to one of the claims 1 to 13, characterized in that that the filter bank (3) consists of any selectable number of filter pairs for test and reference signal (1a, b or 1c, d) and that the filters are evenly distributed on largely any pitch scales. Method according to one of the claims 1 to 14, characterized in that that the output values of the filter bank (3) are spectrally smeared to take account of the simultaneous masking on the upper flank that the level (L) is calculated as a function of each filter output from the magnitude square (5) of the corresponding output value, which is low-pass filtered with a time constant, or is determined without a low-pass filter, and instead the smear factor is low-pass filtered and that the smearing is performed independently for the filters that represent the real part of the signal and the filters that represent the imaginary part of the signal. Method according to one of the claims 1 to 15, characterized in that that the temporal smearing of the filter output signals takes place in two stages, the signals in the first stage via one Cosine 2nd - shaped Time windows are determined and a second masking is modulated in the second stage. Method according to claim 16, characterized in that that the Cosine 2nd -shaped Time windows have a length between 1 and 16 ms. Method according to one of the claims 1 to 17, characterized in that that the instantaneous amount squares (5) at the filter outputs are smoothed in time by low-pass filters of the first order, that the time constants used are selected as a function of the center frequency of the respective filter and that a correction factor is calculated from the orthogonality relationship between spectral envelopes of the temporally smoothed filter outputs of the test and reference signals (1a, b; 1c, d). Method according to claim 18, characterized in that that the test signal is multiplied by the correction factor if the correction factor is <1 and the reference signal is divided by the correction factor if the correction factor is> 1. Method according to one of the claims 1 to 19, characterized in that that correction factors for each filter channel are calculated from the orthogonality relationship between the time envelopes of the filter outputs of the test and reference signals (1a, b; 1c, d). Method according to claim 1, characterized in that a modulation difference is determined from the (absolute) difference of the envelope curves of test and reference signal normalized to the modulation of the reference signal for each filter channel and filter band, which is suitable for estimating certain audible disturbances after temporal and spectral averaging. Method according to one of the claims 1 to 21, characterized in that that from input values in the form of the squares (5) in each filter channel, the envelope modulation, the intrinsic noise of the hearing and constants, a reduced noise level is determined and averaged over time and filter channels. Method according to one of claims 1 to 22, characterized in that that the input signal (X) is delayed by N samples and is subtracted from the original input signal after multiplication by a complex value factor, that the resulting signal (V) is added to the output signal delayed by one sample and that the result multiplied by another complex-valued factor gives the new output signal.

Images