EP0642290A2 - Mobile communication apparatus with speech processing device - Google Patents

Abstract

The invention relates to a mobile radio unit having at least two microphones (M1, ...,MN). These are used for supplying microphone signals (x1, ..., xN), consisting of voice and noise signal components (s1, ..., sN, n1, ..., nN) to microphone signal branches which are connected to the inputs of an adding device (5) used for forming an aggregate signal (x). It is proposed to provide in the microphone signal branches means (3) for weighting the microphone signals (x1, ..., xN) with weighting factors (c1, ..., cN), to be calculated in a simple manner, in order to increase the signal/noise ratio (SNR) of the aggregate signal (x). The voice processing device is preferably integrated into a hands-free talking facility of the mobile radio unit. <IMAGE>

Description

The invention relates to a mobile radio device with a speech processing device with at least two microphones, which are used to deliver microphone signals consisting of speech and interference signal components to microphone signal branches, which are coupled to the inputs of an adding device used to form a sum signal.

From "Proceedings International Conference on Acoustics, Speech, and Signal Processing (ICASSP), pp. 2578-2581, New York, April 1988, IEEE" a microphone arrangement is known from four microphones located in the corners of a room with a square floor plan, the Microphone signals are further processed so that the influence of interference signals that are superimposed on speech signals is reduced. For this purpose, the microphone signals are first shifted against each other in time in order to compensate for time differences between a speaker and the individual microphones. The microphone signals with thus in-phase speech signal components are superimposed by an adding device to form a sum signal, so that the uncorrelated interference signal components of the microphone signals are weakened during the superimposition. The attenuation is not optimal if there is an inhomogeneous interference signal field. In this case, there are different levels of interference signals at the locations where the microphones are arranged. After attenuation, the superimposed microphone signals are fed to an adaptive filter (Wiener filter) by means of a correction factor serving to form the mean value. This is set by evaluating the in-phase microphone signals and further suppresses the interference signals.

The object of the invention is to improve the suppression of the interference signal component of the sum signal present at the output of the adding device.

• zum Empfang der Mikrophonsignale,
• zum Abschätzen der Störsignalanteile,
• zum Abschätzen der Sprachsignalanteile jeweils durch Bildung der Differenz von einem der Mikrophonsignale und dem geschätzten Störsignalanteil für dieses Mikrophonsignal,
• zur Auswahl eines der Mikrophonsignale als Referenzsignal bestehend aus einem Referenzstörsignalanteil und einem Referenzsprachsignalanteil,
• zur Bildung von Sprachsignalverhältnissen durch Division der geschätzten Sprachsignalanteile durch den geschätzten Referenzsprachsignalanteil,
• zur Bildung von Störsignalverhältnissen durch Division der Leistungen der geschätzten Störsignalanteile durch die Leistung des geschätzten Referenzstörsignalanteils und
• zur Bestimmung der Gewichtsfaktoren durch Division der Sprachsignalverhältnisse jeweils durch das zugehörige Störsignalverhältnis

vorgesehen ist.The object is achieved in that means for delaying the microphone signals and for weighting the microphone signals with weighting factors are provided in the microphone signal branches and
that an evaluation circuit

• for receiving the microphone signals,
• for estimating the interference signal components,
• for estimating the speech signal components in each case by forming the difference between one of the microphone signals and the estimated interference signal component for this microphone signal,
• to select one of the microphone signals as a reference signal consisting of a reference interference signal component and a reference speech signal component,
• to form speech signal ratios by dividing the estimated speech signal components by the estimated reference speech signal components,
• to form interference signal ratios by dividing the powers of the estimated interference signal components by the power of the estimated reference interference signal components and
• to determine the weighting factors by dividing the speech signal ratios by the associated interference signal ratio

is provided.

The signal-to-noise ratio corresponds to the ratio of the powers of the speech and interference signal components of the sum signal. The influence of an inhomogeneity of the interference signal field is minimized. Microphone signals with small interfering signal components are compared to the microphone signals large interfering signal components amplified. Due to the correlated nature of the speech signals and the uncorrelated nature of the interference signals, this leads to the sum signal present at the output of the adding device having a reduced interference signal component or an increased signal / noise ratio, as a result of which the speech signal of the sum signal is better understood.

The computation of the weight factors, which does not require much computation, leads to an increased signal / noise ratio and improved speech intelligibility. Because of the efficient calculation of the weighting factors, a calculation that is often required in speech processing is possible in real time, so that there is no annoying delay during a conversation conducted via the speech processing device.

In a further embodiment of the invention, an adaptation of the weighting factors to changes in the interference signal components over time is provided.

In the case of transient, i.e. time-dependent interference signal statistics deteriorate with constant weight factors, the interference signal suppression with the change in the signal statistics. An adjustment of the weight factors prevents this. The weighting factors are kept constant in periods in which a satisfactory stationarity of the signal statistics of the interference signals is assumed. The length of these time segments depends on the nature of the respective interference signal field.

Another embodiment of the invention is characterized in that in each microphone signal branch a transformation device for spectral transformation of the assigned microphone signal, it is provided that the evaluation circuit is provided to form weight factors for each section of the spectral range of the microphone signals and that in each microphone signal branch a reverse transformation device is arranged after a means for weighting the spectral section sections.

The interference signal components of the microphone signals generally have no spectra with spectral values of the same size. For this reason, it makes sense to determine the weighting factors of the microphone signals and the weighting not in the time domain but in the spectral range, for which purpose a transformation of the microphone signals - for example with a Fourier transformation - is required. The spectral range is divided into sections with at least one spectral value. For each spectral range section, the optimal weight factors are determined with which the corresponding spectral values of the microphone signals are weighted. An improved reduction of the interference signal components of the microphone signals is achieved and speech intelligibility is further increased.

Exemplary embodiments are explained in more detail below with reference to the drawings.

• Fig. 1 eine Sprachverarbeitungseinrichtung mit einer Anordnung zur Reduzierung von Störsignalen,
• Fig. 2 eine Ausgestaltung der Sprachverarbeitungseinrichtung durch eine Verarbeitung im Spektralbereich,
• Fig. 3 ein Schaltungselement der in Fig. 2 dargestellten Sprachverarbeitungseinrichtung und
• Fig. 4 ein Mobilfunkgerät, in das die Sprachverarbeitungseinrichtung integriert ist.

Show it:

• 1 shows a speech processing device with an arrangement for reducing interference signals,
• 2 shows an embodiment of the speech processing device by processing in the spectral range,
• Fig. 3 shows a circuit element of the speech processing device shown in Fig. 2 and
• 4 shows a mobile radio device in which the speech processing device is integrated.

(i=1, ..., N) um, die zur Weiterverarbeitung von Analog-Digital-Umsetzern 1 digitalisiert werden. x_i steht für das vom Mikrophon M_i erzeugte Mikrophonsignal, s_i für den darin enthaltenen Sprachsignalanteil und n_i für den entsprechenden Störsignalanteil jeweils im i-ten Mikrophonsignalzweig. Für die digitalisierten Signale sollen im folgenden dieselben Bezeichnungen wie für die entsprechenden analogen Signale gelten. Die Störsignale sind normalerweise Rauschsignale, die beim Einsatz in Fahrzeugen beispielsweise durch Motor- oder Fahrtwindgeräusche verursacht werden. Die Ausgänge der Analog-Digital-Umsetzer 1 sind mit N Eingängen einer Vorverarbeitungseinheit 2 verbunden. Diese enthält für jeden Mikrophonsignalzweig jeweils ein Verzögerungsglied T₁, ..., T_N, wodurch Laufzeitunterschiede von Sprachsignalen einer Sprachsignalquelle zu den Mikrophonen M₁, ..., M_N ausgeglichen werden. Die Verzögerungsglieder T₁, ..., T_N werden adaptiv an diese Laufzeitunterschiede angepaßt. Die Ausgänge der Vorverarbeitungseinheit 2 sind mit steuerbaren Multiplizierern 3 verbunden, die für eine Gewichtung mit Gewichtsfaktoren c_i (i=1, ..., N) in den Mikrophonsignalzweigen sorgen. Die Gewichtsfaktoren c₁, ..., c_N werden durch eine Auswerteeinheit 4 eingestellt, die diese durch Auswertung der Mikrophonsignale x₁, ..., x_N nach einem noch zu erläuternden Schema ermittelt. Kann eine näherungsweise zeitliche Stationarität der statistischen Eigenschaften der Störsignalanteile n_i vorausgesetzt werden, reicht eine einmalige Berechnung der Gewichtsfaktoren aus. Die Ausgänge der Multiplizierer 3, die gleichzeitig die Ausgänge der Mikrophonsignalzweige darstellen, sind mit N Eingängen einer Addiervorrichtung 5 verbunden. Diese erzeugt aus den Ausgangssignalen der Multiplizierer 3 ein Summensignal $\text{x=s+n}$

, das einem adaptivem Filter 6 - beispielsweise ein als Wiener-Filter ausgeführtes FIR-Filter - zugeführt wird. Das Filter 6 wird mit Hilfe der Auswerteeinheit 4 durch Auswertung der Mikrophonsignale z.B. wie im eingangs zitierten Stand der Technik eingestellt.The speech processing device shown in FIG. 1, which is integrated, for example, into hands-free devices of vehicles, contains N microphones M _i (i = 1, ..., N). These convert acoustic signals, which are composed of speech and interference signal components, into electrical microphone signals ${\text{x}}_{\text{i}}{\text{= s}}_{\text{i}}{\text{+ n}}_{\text{i}}$

(i = 1, ..., N), which are digitized for further processing by analog-digital converters 1. x _i stands for the microphone signal generated by the microphone M _i , s _i for the speech signal component contained therein and n _i for the corresponding interference signal component in each case in the i-th microphone signal branch. The following designations apply to the digitized signals as for the corresponding analog signals. The interference signals are normally noise signals that are caused, for example, by engine or wind noise when used in vehicles. The outputs of the analog-digital converter 1 are connected to N inputs of a preprocessing unit 2. This contains for each microphone signal branch a delay element T₁, ..., T _N , whereby differences in transit time of speech signals from a speech signal source to the microphones M₁, ..., M _{N are} compensated. The delay elements T₁, ..., T _N are adaptively adapted to these time differences. The outputs of the preprocessing unit 2 are connected to controllable multipliers 3, which ensure weighting with weighting factors c _i (i = 1, ..., N) in the microphone signal branches. The weight factors c₁, ..., c _N are set by an evaluation unit 4, which determines this by evaluating the microphone signals x₁, ..., x _N according to a scheme to be explained. If an approximate temporal steadiness of the statistical properties of the interference signal components n _{i can be} assumed, a one-off calculation of the weighting factors is sufficient. The outputs of the multipliers 3, which simultaneously represent the outputs of the microphone signal branches, have N inputs an adding device 5 connected. This generates a sum signal from the output signals of the multipliers 3 $\text{x = s + n}$

, which is fed to an adaptive filter 6 - for example a FIR filter designed as a Wiener filter. The filter 6 is set with the aid of the evaluation unit 4 by evaluating the microphone signals, for example as in the prior art cited at the beginning.

In the following, the scheme with which the evaluation unit 4 determines the weighting factors c _{i will} be explained. Sampled values of the microphone signals x _{i are} read into a buffer memory arranged in the evaluation unit 4. Estimates for the amplitudes or the interference signal components n _{i are} obtained by evaluating the sample values of the microphone signals x _i stored in the buffer memory from the periods in which there are no or negligibly small speech signal components s _i . Such speech pauses can be detected due to the striking signal curve or spectrum of speech signals compared to interference signals. By subtracting the determined estimates of the amplitudes of the interference signals n _i from estimates of the amplitudes of microphone signals x _i (with speech signal components s _i ) which are outside the speech pauses and which are likewise determined from sample values stored in the buffer memory, the estimates of the amplitudes of the speech signal components s _{i are} obtained by Determination of difference determined.

σ_s und σ_n sind die Standardabweichungen des Sprachsignalanteils s und des Störsignalanteils n des Summensignals x. Weiterhin sind durch

${\text{s}}_{\text{i}}{\text{= a}}_{\text{i}}\text{s₁, i=1,...,N}$

Sprachsignalverhältnisse a_i durch das Verhältnis der geschätzten Amplituden der Sprachsignalanteile s_i zu der geschätzten Amplitude des als Referenzsprachsignalanteil dienenden Sprachsignalanteils s₁ bestimmt, wenn x₁ als Referenzmikrophonsignal zugrunde gelegt wird. n₁ dient damit als Referenzstörsignal. Als Referenzgrößen sind ohne Einschränkung auch alle anderen Mikrophonsignale bzw. Sprach- und Störsignalanteile mit einem Index i≠1 festsetzbar. Unter der Voraussetzung, daß die Störsignalanteile n_i unkorreliert und mittelwertfrei sind, gilt:

${\text{E{n}}_{\text{i}}{\text{n}}_{\text{j}}\text{} = 0 für alle i≠j}$

und

${\text{E{n}}_{\text{i}}{\text{²} = σ}}_{\text{ni}}{\text{² = b}}_{\text{i}}{\text{² σ}}_{\text{n1}}\text{²}$

mit E{} als Erwartungswertoperator und σ_n1² als Referenzstörleistung. Damit sind sind Störsignalverhältnisse b_i² durch das Verhältnis der geschätzten Leistungen σ_ni² der Störsignalanteile zu der geschätzten Leistung σ_n1² des Referenzstörsignalanteils definiert.The weighting factors c 1, ..., c _N are to be dimensioned such that the so-called signal-to-noise ratio (SNR) of the sum signal x at the output of the adding device 5 is maximized. The SNR results from the ratio of the power (variance) of the speech signal component to the power (variance) of the interference signal component of the sum signal x.

σ _s and σ _n are the standard deviations of the speech signal component s and the interference signal component n of the sum signal x. Furthermore are through

${\text{s}}_{\text{i}}{\text{= a}}_{\text{i}}\text{s₁, i = 1, ..., N}$

Speech signal ratios a _i determined by the ratio of the estimated amplitudes of the speech signal components s _i to the estimated amplitude of the speech signal component serving as the reference speech signal component s 1 if x 1 is used as the reference microphone signal. n₁ thus serves as a reference interference signal. All other microphone signals or speech and interference signal components with an index i ≠ 1 can also be set as reference variables without restriction. Provided that the interference signal components n _{i are} uncorrelated and free of mean values, the following applies:

${\text{E {n}}_{\text{i}}{\text{n}}_{\text{j}}\text{} = 0 for all i ≠ j}$

and

${\text{E {n}}_{\text{i}}{\text{²} = <figure-callout id="2" label="σ ni" filenames="imgf0001.png,imgf0002.png" state="{{state}}">σ </figure-callout>}}_{\text{ni}}{\text{² = <figure-callout id="2" label="b i" filenames="imgf0001.png,imgf0002.png" state="{{state}}">b </figure-callout>}}_{\text{i}}{\text{² <figure-callout id="2" label="σ n1" filenames="imgf0001.png,imgf0002.png" state="{{state}}">σ </figure-callout>}}_{\text{n1}}\text{²}$

with E {} as the expected value operator and σ _n1 ² as the reference interference power. Interference signal ratios b _i ² are thus defined by the ratio of the estimated powers σ _ni ² of the interference signal components to the estimated power σ _n1 ² of the reference interference signal component.

beschrieben wird. Damit ergibt sich als Formel für das SNR des Summensignals x:

Die Maximierung dieses Ausdrucks bezüglich der Gewichtsfaktoren c_i ergibt:

Dieses Ergebnis erhält man beispielsweise über die Bildung der partiellen Ableitungen des obigen Ausdrucks für das SNR. Man erhält eine sehr einfache Formel zur Berechnung der Gewichtsfaktoren c_i.It is further assumed that the speech and interference signal components are not correlated with one another and are free of mean values, which is due to the expression

${\text{It}}_{\text{i}}{\text{n}}_{\text{j}}\text{} = 0 for all i, j}$

is described. The formula for the SNR of the sum signal x is:

Maximizing this expression with respect to the weighting factors c _i gives:

This result can be obtained, for example, by forming the partial derivatives of the above expression for the SNR. A very simple formula for calculating the weighting factors c _{i is obtained} .

The speech processing device described by FIGS. 2 and 3 represents an embodiment of the speech processing device shown in FIG. 1. The N output signals of the preprocessing unit 2, which represent the samples of the microphone signals x 1,..., _N , are converted into spectral transformation devices 7 in the Spectral range transformed, for example by fast Fourier transform (FFT). The spectral range is divided into M sections that contain at least one spectral value. The spectral values are given to N multiplication devices 8, each section of the spectral range weighted or multiplied by a weight factor c _{i, j} calculated separately for each spectral range section. i is the index of the microphone signal branch. j represents the spectral or frequency index of the respective spectral range section. In FIG. 3 one of the multiplication devices 8 is shown in its basic structure, which multiplies the spectral range sections of the respective microphone signal branch by the weighting factors c _{i, j} . The spectral range contains M spectral range sections, so that M multipliers are required for each microphone signal branch. The weighting factors c _{i, j} are set by an evaluation unit 9. They are determined analogously to the calculation of the weighting factors c _i in the description of FIG. 1 by maximizing the signal / noise ratio (SNR) in the respective spectral range sections. The estimated values of the amplitudes of the speech and interference signal components s _i , n _i in the time domain are to be replaced by corresponding estimated values in the frequency domain. The spectral values weighted in this way are fed back transformation devices 10, which transform the weighted spectra of the respective microphone signal branches back into the time domain. The signals obtained in this way are added up as in FIG. 1 by the adding device 5 and fed to the adaptive filter 6. This is set by an evaluation unit 11 which, analogous to the evaluation unit 4 in FIG. 1, evaluates the microphone signals x _{i present} at the outputs of the analog-digital converter 1.

With the aid of a speech processing device designed in this way, the signal-to-noise ratio (SNR) of the sum signal x can be further increased and speech intelligibility can be improved, since it is taken into account that the power of the interference signal components in the spectral range is not uniformly distributed over all spectral values.

In the case of time-variant interference signal statistics, ie that the standard deviations σ _ni are not approximately independent of time, the weighting factors c _i and c _{i, j are} constantly recalculated and set. This depends on the nature of the respective interference signal field. For example, the interference signal statistics of a vehicle change considerably when accelerating from a standing position, since noise is now generated, for example, by the headwind.

In Fig. 4, a mobile device 12 is shown, in which a voice processing device 13 is integrated, which are supplied via an arrangement of three microphones M₁, M₂ and M₃ microphone signals. The structure of the speech processing device 13 can be found either in FIG. 1 or in FIGS. 2 and 3 with the associated descriptions. Output signals of the speech processing device 13 are fed to a function block 14, which combines the other functional units of the mobile radio device 12 and to which a loudspeaker 15 and an antenna 16 are coupled. The microphones M 1, M 2 and M 3, the speech processing device 13 and the loudspeaker 15 act with the help of the function block 14 as parts of a hands-free device of the mobile radio device 12.

Claims (5)

Mobile radio device with a speech processing device with at least two microphones (M₁, ..., M _N ), which are used to supply microphone signals consisting of speech and interference signal components (s₁, ..., s _N , n₁, ..., n _N ) ( x₁, ..., x _N ) are used on microphone signal branches which are coupled to the inputs of an adding device (5) serving to form a sum signal (x),
characterized by
that in the microphone signal branches means (T₁, ..., T _N ) for delaying the microphone signals (x₁, ..., x _N ) and means (3) for weighting the microphone signals (x₁, ..., x _N ) with weighting factors (c₁, ..., c _N ) are provided and that an evaluation circuit (4) - to receive the microphone signals (x₁, ..., x _N ), - to estimate the interference signal components (n₁, ..., n _N ), - to estimate the speech signal components (s 1, ..., s _N ) in each case by forming the difference between one of the microphone signals (x _i ) and the estimated interference signal component (n _i ) for this microphone signal (x _i ), - to select one of the microphone signals as a reference signal (x₁) consisting of a reference interference signal component (n₁) and a reference speech signal component (s₁), - to form speech signal ratios (a₁, ..., a _N ) by dividing the estimated speech signal components (s₁, ..., s _N ) by the estimated reference speech signal component (s₁), - For the formation of interference signal ratios (b₁², ..., b _N ²) by dividing the powers (σ _n1 ², ..., σ _nN ²) of the estimated interference signal components (n₁, ..., n _N ) by the Power (σ _n1 ²) of the estimated reference noise component (n₁) and - To determine the weighting factors (c₁, ..., c _N ) by dividing the speech signal ratios (a₁, ..., a _N ) each by the associated interference signal ratio (b _i ²) is provided. Mobile radio device according to claim 1,
characterized by
that the speech processing device is integrated into a hands-free device. Mobile radio device according to claim 1 or 2,
characterized by
that an adaptation of the weighting factors (c₁, ..., c _N ) to temporal changes in the interference signal components (n₁, ..., n _N ) is provided. Mobile radio device according to one of Claims 1, 2 or 3, characterized in that
that in each microphone signal branch a transformation device (7) for spectrally transforming the assigned microphone signal (x _i ) is provided,
that the evaluation circuit (9) for forming weight factors (c _{i, j} ) is provided for each section of the spectral range of the microphone signals (x₁, ..., x _N ) and
that in each microphone signal branch a means (8) for weighting the spectral range sections is followed by a reverse transformation device (10). Speech processing device with at least two microphones (M₁, ..., M _N ) which are used to supply microphone signals (x₁,... Consisting of speech and interference signal components (s₁, ..., s _N , n₁, ..., n _N ). .., x _N ) are used on microphone signal branches which are coupled to the inputs of an adding device (5) serving to form a sum signal (x),
characterized by
that in the microphone signal branches means (T₁, ..., T _N ) for delaying the microphone signals (x₁, ..., x _N ) and means (3) for weighting the microphone signals (x₁, ..., x _N ) with weighting factors (c₁, ..., c _N ) are provided and that an evaluation circuit (4) - to receive the microphone signals (x₁, ..., x _N ), - to estimate the interference signal components (n₁, ..., n _N ), - to estimate the speech signal components (s 1, ..., s _N ) in each case by forming the difference between one of the microphone signals (x _i ) and the estimated interference signal component (n _i ) for this microphone signal (x _i ), - to select one of the microphone signals as a reference signal (x₁) consisting of a reference interference signal component (n₁) and a reference speech signal component (s₁), - to form speech signal ratios (a₁, ..., a _N ) by dividing the estimated speech signal components (s₁, ..., s _N ) by the estimated reference speech signal component (s₁), - to form interference signal ratios (b₁², ..., b _N ²) by dividing the powers (σ _n1 ², ..., σ _nN ²) of the estimated interference signal components (n₁, ..., n _N ) by the power ( σ _n1 ²) of the estimated reference noise component (n₁) and - To determine the weighting factors (c₁, ..., c _N ) by dividing the speech signal ratios (a₁, ..., a _N ) each by the associated interference signal ratio (b _i ²) is provided.

Images