EP1382034B1 - Method for determining intensity parameters of background noise in speech pauses of voice signals - Google Patents

Abstract

A method for determining intensity characteristics of background noise during speech pauses of speech signals includes determining a proportion of speech pauses in the undisturbed source speech signal so as to define a frequency threshold. The disturbed speech signal is divided into short successive signal elements, an intensity value is determined for each of the signal elements, and a cumulative relative frequency distribution is formed from the determined intensity values of the signal elements. The cumulative relative frequency distribution is used to determine an intensity threshold value which corresponds to the defined frequency threshold. At least one intensity characteristic of the background noise during the speech pauses is determined using a region of the cumulative relative frequency distribution below the intensity threshold value.

Description

The invention relates to a method for the evaluation of background noise in Speech pauses of recorded or transmitted speech signals as defined in claim 1.

The perceived speech quality, e.g. in telephone connections or radio broadcasts, is mainly of language-simultaneous disorders, so of disorders during speech activity. But also noise in the speech pauses go in the quality judgment, especially in high-quality speech reproduction.

The intensity of background noise in the speech pauses can be considered complementary Characteristic value can be used to determine the voice quality (voice quality).

Speech quality determinations of speech signals are usually auditory ("subjective") investigations carried out with test subjects.

The goal of instrumental ("objective") speech quality assessment procedures is it, on the other hand, from properties of the speech signal to be evaluated by means of suitable Calculating characteristic values that determine the speech quality of the speech signal describe without having to resort to judgments from test subjects.

A safe determination of quality provide instrumental procedures that are based on a Comparison of undisturbed reference speech signal (source speech signal) and the disturbed one Speech signal based on the end of the transmission chain. There are many such Processes that are mostly used in so-called sample connection systems. there At the source, the undisturbed source speech signal is fed in and after transmission recorded again.

State of the art and disadvantages of known methods

Known methods for determining the intensity of background noise go mostly from the disturbed signal itself and use a fixed intensity threshold for Distinction between active speech and speech pauses (Fig. 1). This threshold is in the simplest case set constant in the process, but can also be based on the waveform be adapted (e.g., fixed distance to the signal peak). The aim is a safe distinction between language and language break. Manages the distinction For example, the sought after intensity characteristics of the background noise can be determined from the be determined as a voice break detected signal sections. This will be in the In general, the signal sections detected as a speech break again into shorter Segments (typically 8 ... 40ms) are divided and for these the intensity calculations (e.g., rms or loudness). From the results can then Intensity characteristics are determined.

The methods provide at low noise levels in speech pauses and at the same time high intensity of speech (large speech-to-noise ratio) safe readings, since the distinction between language and language break can be made safely (Fig.1).

With increasing noise intensities in speech pauses (decreasing speech-noise ratio) There are increasing uncertainties in the distinction between language and language Speech pauses. Here it is difficult to set the threshold so that on the one hand no sound sections with higher intensities than speech are detected (threshold too low) and on the other hand no linguistic sections of lesser intensity than language break be rated (threshold too high) (Fig. 2).

The intensity of the noise in the speech pauses reaches the intensity of the active Language or even exceeds this, no intensity threshold is to be found, the one Distinction between language and linguistic break allows.

Solutions to the problems described are possible if e.g. different spectral characteristics of speech and background noise are present. Here can by suitable pre-filtering of the signal or by a spectral analysis and Evaluation of selected frequency bands a higher ratio of language to Background noise can be achieved in the considered frequency ranges, so that again a safe distinction between active language and language break possible is.

Other solutions use certain parameters that are determined by speech coding and use them to distinguish between language and sections Background noise. The goal is to derive from the parameters whether the signal segment considered typical properties of speech (e.g., voiced parts) having. An example of this is "Voice Activity Detector" (ETSI Recommendation GSM 06.92, Valboune, 1989).

These methods work and become more robust at low speech-to-noise ratios primarily for suppressing the transmission of speech pauses e.g. in mobile used. However, the procedures show uncertainties when the background noise even language contains or is language-similar. Such sections are then called Language classified, although by a listener as a disturbing background noise be felt.

"A perceptual speech-quality measure based on a psychacoustic sound representation" (Beerends, J. G.; Stemerdink, J. A., J. Audio Eng. Soc. 42(1994)3, S. 115-123)

"Auditory distortion measure for speech coding" (Wang, S; Sekey, A.; Gersho, A.: IEEE Proc. lnt. Conf. acoust., speech and signalprocessing (1991), S.493-496).

Instrumental speech quality measurement methods are mostly based on the principle of signal comparison of undisturbed reference speech signal and disturbed signal to be evaluated. Examples are the publications:

"A perceptual speech-quality measure based on a psychacoustic sound representation" (Beerends, JG; Stemerdink, JA, J. Audio Eng. Soc. 42 (1994) 3, pp. 115-123)

"Auditory distortion measure for speech coding" (Wang, S; Sekey, A., Gersho, A .: IEEE Proc. Int. Conf. Acoust., Speech and signal processing (1991), pp. 493-496).

The currently valid ITU-T standard P.861 also describes such a procedure: "Objective quality measurement of telephone-band speech codecs" (ITU-T Rec. P.861, Geneva 1996).

Such measuring methods are used in so-called sample connection systems, in to which a known reference speech signal (source speech signal) is fed at the source, over z. B. transmitted a telephone connection and recorded on the sink. To the recording of the speech signal are used to evaluate the speech quality of the possibly disturbed signal whose characteristics with those of undisturbed Source speech signal compared.

Is the undisturbed for determining the background noise in speech pauses Source speech signal available, then this can be used to determine the transition times from speech to speech pause or from speech pause to speech. For this purpose, e.g. a method with threshold determination - as described above - on the source speech signal applied. The method provides reliable distinctions between speech and linguistic break, since the speech-to-noise ratio is undisturbed Source speech signal is sufficiently high (Fig. 3a). The times of the threshold passage, i. Start or end of voice activity, can now access the disturbed voice signal be transmitted (Fig. 3b).

Such a method can easily be modified if a constant time difference (eg delay due to signal transmission) occurs between the source speech signal and the disturbed signal. The condition is, however, that this time difference can be determined in advance safely and then used to correct the times end or beginning of voice activity. This is usually possible with time-invariant systems, since they have a constant delay (FIG. 3c).
In principle, such a method also works if the time offset between the two signals is not constant for the entire signal length but runs variably. These time-invariant systems include, in particular, packet-based transmission systems in which significant fluctuations in the system delay can occur due to different packet delays and appropriate management in the receiver. In order to prevent losses due to late arrival of parcels, some language breaks in the recipient are extended and later shortened again. A transmission of the times of the beginning or end of the voice activity is only possible with knowledge of the current delay at these points. The adaptive determination of the time offset is computationally intensive and often succeeds only inadequately with reduced speech-noise ratios. If the adaptive determination of the time offset is not successful, the beginning and end of speech pauses can not be determined exactly or not at all. As a result, no or only an uncertain determination of the intensity characteristics of pause noise is possible.

Examples of the determination of background noise according to the The prior art are from US6044342A, US5598466A, WO0052683A and US4811404A.

task

• ein geringes Verhältnis von Sprache zu Hintergrundgeräusch vorliegt,
• das Hintergrundgeräusch Sprache beinhaltet oder selbst sprachähnlich ist,
• der Zeitversatz zwischen ungestörtem Quellsprachsignal und gestörtem Sprachsignal nicht konstant über die gesamte Signallänge ist.

As described, the determination of background noise in speech pauses is difficult or partially impossible, even if the undisturbed source speech signal is known, especially if

• there is a low ratio of speech to background noise,
• the background noise includes speech or is itself speech-like,
• the time offset between undisturbed source speech signal and disturbed speech signal is not constant over the entire signal length.

It should be presented a procedure, with which also under the conditions mentioned a safe and fast determination of background noise intensity characteristics is ensured in speech pauses. Condition is that both source speech signal as well as disturbed speech signal are fully recorded available.

solution principle

The known methods assume the time from the beginning and end of a Language break as accurately as possible to determine. The result is then the signal from the Pause sections available for further evaluation. Separated from these Pause sections of the signal, the intensity characteristics are determined.

With the present method, intensity characteristics of background noise in speech pauses of speech signals are determined without the exact times from the beginning and end of a break period. Also is one Separation of the speech pause signal is not required for the evaluation.

Basis for the method described here for the determination of intensity characteristics of Background noise in speech pauses of speech signals is the cumulative frequency distribution the intensity values of the signal segments into which the speech signal previously divided. These short-term signal intensities refer to signal segments with a duration of e.g. 8ms or 16ms. The frequency distribution indicates how high the Proportion of short-term intensities below a defined threshold is.

For the calculation of the frequency distribution, the speech signal to be analyzed in subdivided into short consecutive signal segments and from each signal segment of the Intensity value (e.g., Loudness or RMS).

Fig. 4 shows a typical waveform for speech signals with stationary background noise (Voice-noise distance approx. 10dB). The cumulative frequency distribution is using the example of short-term loudness (loudness calculated according to ISO532). 2000 segments of 16ms length were evaluated. It can be seen that none of the Segments has a value less than 30 sone (P = 0%) and no segment achieved a higher loudness than 80 sone, since the value P = 100% is already reached here. The steep increase in the function at approx. 30 sone points to a low fluctuation of the Signal intensity in large areas (nearly 70%) of the signal. As signal was here uses a speech signal with additive white noise.

Such a distribution function should now be used to determine intensity characteristics of background noise in the speech pauses. For this it is necessary to know the proportion of speech pauses in the overall signal. This proportion can be determined from the undisturbed source speech signal (FIG. 3a). Total length of speech pauses = (t 1 - t0) + (t3 - t2) Total length of the signal section = (t4 - t0) Language break share = Total length of the speech pauses Total length of the signal section

It is assumed that the ratio of active language to language pauses While the transmission remains largely constant, this value can also be applied to the disturbed signal to be transmitted.

If the proportion of speech pauses in the entire speech signal is known and becomes this proportion defined as the frequency threshold, it can be determined from the frequency distribution of the short-term intensities determines the intensity threshold corresponding to the frequency threshold become.

In Fig. 4 is entered as an example, a proportion of voice pauses of 58%. This frequency threshold P _z = 0.58 corresponds to an intensity threshold of N = 34.5 sone, which means that of 58% of the signal segments the intensity value (loudness) of 34.5 sone is not exceeded.

The area below the intensity threshold shows the frequency distribution for Intensity values of signal segments in the speech pauses and can for the determination of intensity characteristics of the background noise used in the speech pauses become.

It is assumed that no speech pause segment has a higher intensity value has as a speech segment, so that the intensity threshold value as the maximum value for the Background noise in speech pauses can be viewed.

Determination of the arithmetic mean of intensities

The cumulative distribution function can also be used to derive the arithmetic mean of all segments whose intensities are below a previously established frequency threshold. For this purpose, first a differentiation of the cumulative distribution function P (x) into a distribution density function p (x) is to be carried out.
The arithmetic mean of all evaluated intensities X of the total signal is calculated as known from the integral of the distribution density function p (x):

A limitation of the integration at a certain value x _G makes it possible to determine the arithmetic mean over all values X which are below this limit value. However, the result should be weighted with the frequency P (x _G ) . This frequency corresponds to the integral over p (x) up to the value x _G.

The intensity threshold x _G can be derived from the distribution function P (x) . In the example according to FIG. 4, the frequency threshold value P (x _G ) is the proportion of speech pauses in the overall signal P _z = 0.58 to which the intensity threshold value x _G = 34.5 sone is assigned. The arithmetic mean of all segments with an intensity less than x _G is calculated according to Eq. 2, where x _G = 34.5 sone . The frequency of 58% here corresponds to the weighting value P (x _G = 34.5) = 0.58 . Graphically, this procedure is shown in FIG. 5.

If it is assumed again that the intensities of segments in speech pauses, do not exceed the intensities of speech segments or the background noise has only weak temporal fluctuations, the calculated arithmetic Mean should be considered as the mean of the intensity in speech pauses.

Simplified method for determining the arithmetic mean

A simplified method for determining the mean over all X is based on the assumption that the relative frequency distribution of the intensity values of the signal segments in the range P (x) = 0 to the frequency threshold of speech pauses P _{z is represented} by a weighted normal distribution G (x, μ σ ² ) can be approximated. The value for the distribution function G (x, μ, σ ² ) for x → ∞ is 1. As is known, the value x , where G (x, μ, σ ² ) = 0.5, corresponds to the arithmetic mean over all individual values X.

If an approximation of the relative frequency distribution P (x) is obtained in the range from P (x) = 0 to P _z with a weighted normal distribution κ P _z G (x, μ, σ ² ), then the arithmetic mean over X corresponds to the weighted normal distribution the value x for the following applies: G (x, μ σ ² ) = 0.5 κ P _z . By assuming that κP _z G (x, μ, σ ² ) closely approximates the distribution P (x) in the range from P (x) = 0 to P _z and κ ≥ 1, the arithmetic mean sought is equal to x _A , for which P (x _A ) = 0.5 κ P _z .

For the application case of speech with additive background noise considered here, values for κ = 1 ... 1.3 show good approximation results. FIG. 6 shows an example of the approximation by weighted normal distributions. In this case, a value κ = 1. 1 selected. The graph shows speech as background noise and has a speech pause rate of 58%. The strong temporal fluctuation of the language background can be clearly seen as a flatter slope in the range N = 0 ... 40 sone. The arithmetic mean derived from the normal distribution function with P (x _A ) = 0.5 κ P _z = 0.32 is 20 sone.

The advantage of this simplified method is the lower computational intensity, since it is possible to dispense with the computation of the distribution density and its integration. It is also not necessary to exactly determine the normal distribution function κP _z G (x, μ, σ ² ), it is sufficient to determine κ. Since P _{z is} known, the mean value over all X <x _{G is determined} as value x _A , where P (x _A ) = 0.5 κP _z . The arithmetic mean over all X to x _G thus corresponds to the intensity value which corresponds to a frequency value of 0.5 * κ * proportion of the speech pauses in the total signal, ie the intensity which is not exceeded by a proportion of segments of 0.5 * κ * proportion of the speech pauses ,

Determination of further statistical characteristics

Other statistical intensity parameters can also be determined with this method become. In Fig. 7 is demonstrated by the example of FIG. 4, as from the function of Intensity value can be determined by only 20% of the speech pause segments is exceeded (20% percentile loudness).

In the example given, the intensity value is sought, which is undercut by 80% of the segments in speech pauses, ie the abscissa value is searched for, which applies to the ordinate value P = 0.58 * 0.8 = 0.46 . The value is only slightly less than the maximum value due to the example given little fluctuating noise.

Exercise example for the determination of the arithmetic mean value from the Verteitungsdichtefunkfion

The presented here embodiment of the method for determining the intensity of Background noise determines the arithmetic mean of all the loudnesses of the Segments that are below a certain frequency threshold. This frequency threshold corresponds to the proportion of speech pauses in the signal and the calculated one Arithmetic mean is considered as average loudness in speech pauses. This will be In this embodiment, the distribution density function is used.

Prerequisite is that both signals, i. the undisturbed source speech signal and the Disturbed signal to be evaluated, fully recorded.

First, the proportion of speech pauses P _z in this signal is determined by means of a suitable threshold on the basis of the source speech signal.

The second step is the calculation of the desired intensity values for successive short signal segments of the speech signal to be evaluated. In this embodiment, the loudnesses are calculated according to ISO532 in successive signal portions of 16ms length. The distribution function is approximated by a series of individual values (discrete relative frequency distribution). These individual values are designated by successive indices m. The series of individual values is limited at a maximum value M (eg: P ₀ ... P ₂₀₀ ). In the evaluation of each individual value P _m - whose index exceeds the determined intensity X of the evaluated signal segment - increased by the counter 1. After evaluation of the entire signal, all individual values are divided by the number of evaluated signal segments. Each individual value P _m then contains the relative frequency of the signal segments which have a loudness smaller than the value of the index.

On the basis of the previously determined proportion of speech pauses P _z , that frequency value P _{s is} determined which has the lowest absolute difference to P _z . The index S of this individual value P _s indicates the corresponding loudness, ie the loudness, which is not exceeded by a proportion P _{s of} all segments. In order to determine the arithmetic mean of the loudnesses of all segments whose loudnesses are below the given frequency threshold P _s , the conversion of the discrete frequency distribution P ₀ ... P _M , into a discrete frequency density (stripe frequency) p ₀ .. _M-1 . For this purpose, the differences of two consecutive individual values are formed and stored as the value sequence p ₀ ... P _N-1 : p m = p m + 1 - p m for all m = 0 ... M-1

The value p _m then contains the relative frequency of the segments whose loudness is between m and m + 1 . The sought arithmetic mean value corresponds to the weighted sum over the strip frequency P _m to m = S , ie the loudness, which is not exceeded by a proportion P _{s of} all segments:

The correction value ½ corresponds to half the distance between two consecutive indices. The value p _m contains the relative frequency of segments whose loudnesses are between m and m + 1 . The expected value of all loudnesses recorded here is, assuming an equal distribution of the loudnesses of m .... m + 1, therefore m + 0.5.

As described in the application, the method provides a discrete frequency distribution with a resolution of 1 sone, since the index m is an integer and the loudness values are assigned directly to the corresponding indexes. In order to achieve other higher or lower resolutions, the loudness value has to be multiplied with corresponding factors before calculating the relative frequency distribution.

To demonstrate the measurement reliability of the presented method are shown in Table 1 Measured values for various signals and background noise are listed. It was Speech signals of 32 s length and different percentage of speech pauses (35%, 58% and 91%) each mixed with different sounds. As noises became first white noise with different voice-noise intervals used. Furthermore was also continuously spoken language as well as two sounds from real acoustic environments (street and office).

Prior to calculating the frequency distribution, a multiplication of all the launess values by a factor of 2 is performed in order to increase the resolution of the representation when using integer indexes. This then corresponds to a loudness grading at integer indexes of 0.5 sone. With a limitation of the frequency distribution function at P ₂₀₀ , loudnesses from 0 ... 100 sone can be mapped in steps of 0.5 sone . It should be noted, however, that this factor must be applied as a divisor to all results for correction. In the exemplary embodiment selected here, this means that the calculated arithmetic mean value is to be divided by 2.

Explanatory notes to Table 1: The voice-to-noise ratio is for information purposes only; The basis for this is the distance between the average level of activity in speech activity and the mean effective level of background noise. The mean loudness value (target value) was determined in a reference measurement, in which the speech pauses were marked manually and evaluated in segments of 16 ms. The calculated standard deviations refer to the thus measured reference loudnesses and give information about the magnitude of the occurring fluctuations. The measured values in column 5 were determined using the method described in this exemplary embodiment. noise SNR mean loudness (sone) target value Standard deviation of segment loudness mean loudness (sone) measured with described methods Deviation (measurement error) abs. / rel. Pausenanteil of the speech signal 91% White noise 6 dB 41.4 1:55 42.0 0.6 / 1.4% White noise 10 dB 32.3 1.22 32.6 0.3 / 0.9% White noise 16 dB 22.2 0.87 22.3 0.1 / 0.4% language 6 dB 21.3 11.7 20.6 -0.7 / -3.3% language 10 dB 16.5 9.16 16.2 -0.3 / -1.8% language 16 dB 11.2 6.21 11.3 0.1 / 0.9% road noise 10 dB 26.0 3.22 26.2 0.2 / 0.8% office noise 10 dB 26.3 2.78 26.6 0.3 / 1.1% Pausenanteil of the speech signal 58% White noise 6 dB 41.3 1:55 44.8 3.5 / 8.5% White noise 10 dB 32.3 1.22 34.2 1.9 / 6.0% White noise 16 dB 22.1 0.87 22.6 0.5 / 2.2% language 6 dB 20.7 11.7 19.0 -1.7 / -8.2% language 10 dB 16.0 9.16 15.4 -0.6 / -3.8% language 16 dB 10.7 6.21 10.8 0.1 / 0.9% road noise 10 dB 26.1 3.22 27.0 0.9 / 3.4% office noise 10 dB 26.3 2.78 27.3 1.0 / 3.8% Pausenanteil of the speech signal 35% White noise 6 dB 41.3 1:55 46.1 4.8 / 11.6% White noise 10 dB 32.3 1.22 35.6 3.3 / 10.2% White noise 16 dB 22.1 0.87 23.3 1.2 / 5.4% language 6 dB 20.0 11:22 17.6 -2.4./ -12% language 10 dB 15.6 8.7 15.0 -0.6 / -3.8% language 16 dB 10.9 5.93 11.8 0.9 / 8.3% road noise 10 dB 26.1 3.22 27.3 1.2 / 4.6% office noise 10 dB 26.3 1.78 27.9 1.6 / 6.1%

First, it should be noted that the measurement reliability increases with increasing pause portion in the signal to be evaluated. An increase in the measurement reliability can also be observed with decreasing noise intensity and a lesser temporal fluctuation of the background noise. Based on a typical proportion of speech pauses in a telephone communication of P _z > 50% , the measured values achieved with the presented method are satisfactory, even in the case of stronger fluctuations in background noise (eg speech).

Embodiment for the determination of the arithmetic mean with simplified procedure

This particular embodiment shows an application of the described simplified method for the determination of the arithmetic mean using a weighted normal distribution.

The simplified method dispenses with the calculation of the strip frequency and derives an estimated value for the arithmetic mean of the loudnesses of all segments whose loudnesses are below the predetermined frequency threshold P _z , directly from the relative frequency distribution P _m . As described, only the value κ needs to be set for the estimation.

in this embodiment, κ = 1.1 is defined. The estimated value then corresponds to the loudness value, which is not exceeded by a proportion of 0.5 * 1.1 * P _{z of} all evaluated segments. In the exemplary embodiment, this estimated value of the arithmetic mean corresponds to the loudnesses, the index m of the frequency value, which has the smallest absolute difference to 0.55 P _z . Table 2 lists the measurements obtained by this simplified procedure. Again, to increase the resolution to 0.5 sone, all loudness values were multiplied by a factor of 2 before calculating the frequency distribution, and the results were corrected accordingly. noise SNR mean loudness (sone) target value Standard deviation of segment loudness Medium loudness (sone) measured with simplified procedure Deviation (measurement error) abs. / rel. Pausenanteil of the speech signal 91% White noise 6 dB 41.4 1:55 41.5 0.1 / 0.2% White noise 10 dB 32.3 1.22 32.5 0.2 / 0.6% White noise 16 dB 22.2 0.87 22.5 0.3 / 1.3% language 6 dB 21.3 11.7 20.5 -0.8 / -3.8% language 10 dB 16.5 9.76 16.5 0.0 / 0.0% language 16 dB 11.2 6.21 11.0 -0.2 / 1.8% road noise 10 dB 26.0 3.22 26.0 0.0 / 0.0% office noise 10 dB 26.3 2.78 26.5 0.2 / 0.6% Pausenanteil of the speech signal 58% White noise 6 dB 41.3 1:55 41.50 0.2 / 0.5% White noise 10 dB 32.3 1.22 32.5 0.2 / 0.6% White noise 16 dB 22.1 0.87 22.5 0.4 / 1.8% language 6 dB 20.7 11.7 20.0 -0.7 / -3.4% language 10 dB 16.0 9.16 16.0 0.0 / 0.0% language 16 dB 10.7 6.21 11.0 0.3 / 2.8% road noise 10 dB 26.1 3.22 26.0 -0.1 / -0.4% office noise 10 dB 26.3 2.78 26.5 0.2 / 0.8% Pausenanteil of the speech signal 35% White noise 6 dB 41.3 1:55 41.0 -0.3 / 0.7% White noise 10 dB 32.3 1.22 32.5 0.2 / 0.6% White noise 16 dB 22.1 0.87 22.5 0.4 / 1.8% language 6 dB 20.0 11:12 19.0 -1.0 / -5% language 10 dB 15.6 8.7 15.5 -0.1 / -0.6% language 16 dB 10.9 5.93 11.5 0.6 / 5.5% road noise 10 dB 26.1 3.22 25.5 -0.6 / -1.4% office noise 10 dB 26.3 2.78 26.5 0.2 / 0.8%

The simplified method not only saves computation time, but in the evaluated examples provides measured values with a significantly higher accuracy compared to the values from Table 1. Since the index m is used directly as an estimate, the accuracy of the estimation is based on the resolution of the relative discrete frequency distribution (here: 0.5 sone ) limited.

With the described simplified measuring method, good measured values are achieved even with noises with greater fluctuation. At the selected voice-noise intervals of 6dB, it can no longer be assumed that all loudnesses in speech pauses have less loudness than speech segments. Even so, the readings were hardly distorted. The described simplified method is also suitable for signals with a lower percentage of pauses.

Embodiment for the determination of percentile loudnesses from the relative Häufigkeitsverteitung

The percentile loudness of all segments which are below a certain frequency threshold P _z can be obtained by multiplying this relative frequency P _z by a value 1- percentile value (eg 10% percentile loudness: P _Z10% = 0.9 * P _z ). The integer index m of the frequency value P _m , which has the least absolute difference to P _S10% , provides the searched percentile loudness value.

In Table 3, for the examples already given in Tables 1 and 2, the 10% percentile loudnesses are listed and compared with a manually determined reference value. noise SNR 10% percentile loudness (sone) target value Standard deviation of segment loudness 10% percentile loudness (sone) measured by frequency distribution Deviation (measuring instrument) abs. rel. Pausenanteil of the speech signal 91% White noise 6 dB 42.5 1:55 43.0 0.5 / 1.2% White noise 10 dB 33.0 1.22 34.0 1.0 / 3.0% White noise 16 dB 22.5 0.87 23.5 1.0 / 4.4% language 6 dB 37.0 11.7 34.5 -2.5 / -6.8% language 10 dB 28.5 9.16 27.5 -1.0 / -3.5% language 16 dB 19.0 6.21 19.5 0.5 / 2.6% road noise 10 dB 29.5 3.22 30.0 0.5 / 1.7% office noise 10 dB 29.0 2.78 29.5 0.5 / 1.7% Pausenanteil of the speech signal 58% White noise 6 dB 42.5 1:55 42.5 0.0 / 0.0% White noise 10 dB 33.0 1.22 33.5 0.5 / 1.5% White noise 16 dB 22.5 0.87 23.0 0.5 / 2.2% language 6 dB 36.0 11.7 29.0 -7.0 / -19% language 10 dB 28.5 9.16 24.5 -4.0 / -14% language 16 dB 19.0 6.21 18.0 -1.0 / -5.3% road noise 10 dB 30.0 3.22 29.0 -1.0 / -3.3% office noise 10 dB 29.0 2.78 28.5 -0.5 / -1.6% Pausenanteil of the speech signal 35% White noise 6 dB 42.5 1:55 42.5 0.0 / 0.0% White noise 10 dB 33.0 1.22 33.5 0.5 / 1.5% White noise 16 dB 22.5 0.87 23.5 1.0 / 2.2% language 6 dB 35.5 11:21 24.0 -11.5 / -33% language 10 dB 27.5 8.7 21.0 - 6.5 / -24% language 16 dB 19.0 5.93 17.5 -1.5 / -7.9% road noise 10 dB 29.5 3.22 28.0 -1.5 / -4.8% office noise 10 dB 29.0 1.78 28.5 -0.5 / -1.6%

The measurements show a good estimate of percentile loudness for background noise with low turnover, for language - especially at low Pause percentage - only insufficient accuracy achieved. Only at higher voice-noise intervals The results are usable until good.

Claims (4)

1. Method for the determination of intensity characteristic values of background noises in speech pauses of speech signals of which the undisturbed source speech signal and the disturbed speech signal are available in recorded form, wherein, from the undisturbed source speech signal, the proportion of speech pauses in the total signal is determined according to known methods and the disturbed speech signal is broken down into short consecutive signal elements and an intensity value is determined for each signal element, wherein
   the cumulative relative frequency distribution (1) is formed from the intensity values of the individual signal elements of the disturbed speech signal;
   the determined proportion of speech pauses in the source speech signal is defined as a frequency threshold and the frequency threshold is applied to the disturbed speech signal; the intensity threshold value (3) corresponding to the defined frequency threshold (2) is determined from the frequency distribution of the intensity values of the signal segments; all the signal segments with an intensity value lower than that of the intensity threshold value are assessed as belonging to the speech pauses;
   the distribution function for the intensity values of the signal segments in the region below the intensity threshold value represents the frequency distribution for the intensity values in the speech pauses (4), and this region of the distribution function can be used for the determination of intensity characteristic values of the background noises in the speech pauses.
2. Method according to claim 1, characterized in that the arithmetic mean value of the intensity values of the signal elements in the speech pauses is determined as the intensity characteristic value of the background noises in the speech pauses, and in that the arithmetic mean value is calculated in that the distribution density is derived from the frequency distribution and the arithmetic mean value of the intensity values in the speech pauses is determined by subsequent integration over the distribution density in the region below the intensity threshold value.
3. Method according to claim 1, characterized in that the arithmetic mean value of the intensity values of the signal elements in the speech pauses is determined as the intensity characteristic value of the background noises in the speech pauses, and in that the arithmetic mean value is determined from the frequency distribution in that the intensity distribution in the region below the intensity threshold value is approximated by a normal distribution weighted with a factor and, for calculation of the arithmetic mean value, the intensity threshold value is multiplied by 0.5 and the weighting factor.
4. Method according to claim 1, characterized in that percentile characteristic values can be determined as intensity characteristic values of the background noises in the speech pauses, in that the percentile characteristic values can be determined from the frequency distribution, in that the specified percentile value is subtracted from 100 percent, the difference is multiplied by the frequency threshold value and, for the resulting frequency value, the intensity value corresponding to said value is determined as a percentile characteristic value from the distribution function.

Images