Noise reduction method and apparatus

(54) (56)

INVENTOR INVENTOR

TERMINAL DESIGNATION (57)

Telefonaktiebolaget LM Ericsson, 126 25 Fisseha Mekuria, Lund SE, Hans Svensson,

CALLED PUBLICATIONS:

Ericsson Mobile Communications AB Device and method for noise reduction

Stockholm SE Staffanstorp SE

EP Al 0 459 364 (G01L 3/02), EP Al 0 645 756 (G01L 7/08), US A 5 012 519 (704/226)

SUMMARY:

Noise reduction devices and noise treatment methods for noise reduction in audio signals are indicated. The noise level of an input signal at an input terminal (23) is measured and the noise / signal ratio determined. A reduced voice activity detector (15) is used to determine whether the input signal comprises speech or not. If the measured noise level exceeds a threshold, a switch (18) connects the input signal to noise reduction means (13). However, if the measured noise level does not exceed the threshold level, ie. if no noise reduction is required, the switch (18) disables the noise reduction means (13) and the input signal is forwarded unchanged. Energy is saved by turning off the energy supply of the organ (13) for noise reduction when needed. 10

<img file="SE515674C2_D0001.tif" />

The numbers in brackets indicate international identification code, INID code. Letters in clamps indicate international document code.

515 674

TECHNICAL FIELD OF THE INVENTION

The present invention relates to noise reduction and, in particular, to noise reduction devices, communication devices and systems including such noise reduction devices, and to brush management methods for noise reduction in audible signals.

DESCRIPTION OF THE PRIOR ART

A voice communication communication system normally includes a microphone for recording an acoustic signal which is assumed to contain speech intended for communication. In fact, however, not only the speech but also the noise present in the environment of the speaking person will be picked up by the microphone. Typical noise environments are car environments, shopping centers and streets with lively traffic. It should be noted that people often use mobile communication devices, such as cell-based telephones, in this kind of noise-affecting environments and therefore the implementation of an effective noise reduction method is essential for these types of devices.

A known form of digital processing method for detecting and separating noise from real-time speech is disclosed in US-A-5,012,519. The noise in an input signal is suppressed by dividing the input signal into spectral channels and decreasing the gain in each channel having a low signal to noise ratio.

US-A-5,533,133 discloses a method of noise suppression where noise is suppressed during breaks and silent periods during a call and voice signals are allowed to pass freely. A voice activity detector is used to determine whether the signal contains speech.

515 674

Although the known methods of noise reduction described above work quite satisfactorily, they have a number of disadvantages.

The implementation of the method of detection and separation of noise disclosed in US-A-5,012,519 is complicated and expensive. A lot of (expensive) memory and energy is required to perform the necessary fast Fourier transforms, FFT (Fast Fourier Transforms), and additional calculations. Memory is expensive especially in highly integrated equipment where the available chip space is limited. Furthermore, energy is always a scarce resource, especially in small battery-powered handheld devices such as mobile communication devices (eg cell-based phones).

The implementation of the noise suppression method disclosed in US-A-5,533,133 is also complicated and expensive. Using a digital signal processor, DSP, to implement the voice activity detector requires a lot of (expensive) memory and a lot of energy to perform all the necessary calculations. Memory is expensive, especially in highly integrated equipment, where the available chip space is limited. Furthermore, energy is always a scarce resource, especially in small battery-powered handheld devices such as mobile communication devices (eg cell-based phones).

It is an object of the present invention to provide noise reduction apparatus, communication devices and systems including such noise reduction apparatus and noise processing methods for noise reduction in audio signals that overcome or mitigate the aforementioned problems.

515 674

SUMMARY OF THE INVENTION

According to one aspect of the present invention, a noise reduction apparatus is provided which comprises an input terminal for receiving an input signal, a first switching means connected to the input terminal, a noise measuring means connected to the input terminal and arranged to measure the noise level of the input signal, a first comparison means receiving a measured noise level, which first comparison means is arranged to compare the measured noise level with a first predetermined noise level and generate a first control signal accordingly, the first comparison means being connected to the first switching means and the first switching means being controlled by the first control signal, a noise reducing means connected to the first the coupling means and arranged to make noise reduction of the input signal and to provide a noise reduced output signal; when it is activated, and skipped or switched off when it is deactivated, and the first control signal is arranged to control the first switching means such that the first switching means deactivates the noise reducing means when the measured noise level is lower than the first predetermined noise level.

According to a further aspect of the present invention, a mobile communication device, such as a cellular telephone, is provided comprising a microphone, a radio receiver / transmitter and an air interface where the signal from the microphone is connected to a noise reduction apparatus according to the above apparatus, the output of the noise reduction apparatus is transmitted by the receiver / transmitter and air interface.

According to a further aspect of the present invention, an accessory equipment is provided, e.g. a hands-free equipment, for a mobile communication device such as a

515 674 cell-based telephone, comprising a microphone in which the signal from the microphone is connected to a noise reduction apparatus according to the above apparatus.

According to a further aspect of the present invention there is provided a communication system comprising a microphone and a noise reduction apparatus according to the above apparatus.

Preferably, the noise measuring means comprises a detector of reduced voice activity which is not based on the pitch but on the autocorrelation and energy content of the input signal. This has the advantage that a small buffer memory is needed and that a small amount of energy is consumed when the necessary calculations are performed.

According to a further aspect of the present invention, a noise processing method for reducing noise in audio signals is provided comprising the steps of measuring the noise level of an input signal, determining whether the measured noise level is lower than a first predetermined noise level, and output the input signal as the output signal of the measured noise level. first predetermined noise level and, otherwise, perform noise reduction of the input signal and output the thus processed input signal as the output signal.

According to a further aspect of the present invention there is provided a noise processing method for reducing noise in audio signals comprising the steps of measuring the noise level of an input signal, determining whether the measured noise level is lower than a first predetermined noise level or higher than a second predetermined noise level, and providing an output signal. if the measured noise level is lower than the first predetermined noise level, perform noise reduction of the input signal and output the thus processed input signal as the output signal if the measured noise level is higher than or equal to the first predetermined noise level but lower than the second predetermined noise level

515 674 and, otherwise, low pass filter the input signal and output the thus filtered input signal as the output signal.

The noise reduction apparatus, the mobile communication device, the accessory equipment for a mobile communication device and the noise processing method according to the present invention solve the problems of the prior art. Compared to US-A-5,012,519, no FFT need be performed to divide the input signal into spectral channels. Therefore, a simple implementation is obtained which requires less memory capacity and is therefore cheaper to implement and which consumes less energy. Further, in comparison with US-A-5,533,133, a reduced voice activity detector may be selected instead of a conventional voice activity detector. Therefore, a simple implementation is achieved which requires less memory capacity and thus is cheaper to implement and which consumes less energy.

Furthermore, the present invention has the advantage that noise reduction is performed only when needed and, thereby, energy is saved when there is no need for noise reduction. The need for noise reduction performance is set to occur when the noise level is above a predetermined noise threshold level. The advantage of energy saving is even more pronounced when the noise reduction is switched off and replaced with a simple low pass filter, if the noise level is high, ie. above a second predetermined threshold level.

It can be noted that energy consumption is always an important aspect, especially in small hand-held battery-powered equipment where energy is a scarce resource.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 is a block diagram illustrating a first embodiment of the present invention;

515 674 FIG. 2 is a block diagram illustrating a second embodiment of the present invention;

Fig. 3 is a block diagram illustrating a conventional voice activity detector;

Fig. 4 is a block diagram illustrating a detector for reduced voice activity;

Fig. 5 is a flowchart illustrating a first mode of operation of the present invention;

FIG. 6 is a flow chart illustrating a second mode of operation of the present invention.

DETAILED DESCRIPTION OF EMBODIMENTS

Fig. 1 is a block diagram illustrating a first embodiment of the present invention. A microphone 11 is connected to an analog / digital converter 12. The output of the analog / digital converter 12 is connected to an input terminal 23. The input terminal 23 is connected to a first input of a switch 18, to an input of a reduced voice activity detector, RVAD, And to a first input of a noise-to-signal ratio means, NSR, 16. The output of RVAD 15 is connected to a second input of NSR 16. The output of NSR is connected to an input of a comparison / control device, DECISION 17. The output of DECISION 17 is connected to a control input of switch 18. The switch 18 is so connected that, depending on the output signal of DECISION 17, the input terminal 23 is either connected to a first select terminal (choice A) connected to an input of a noise reduction means, NR, 13, or to a second select terminal (choice B) connected to the output of NR 13. The output of NR 13 is connected to a speech encoder SPE (SPeech Encoder), 14. As shown in FIG. 1 RVAD 15, NSR 16, DECISION 17, switch 18, NR 13 and SPE 14 are implemented through a digital signal processor, DSP, 19. RVAD 15 and NSR 16 form noise measuring means for measuring the noise level of an input signal at input terminal 23. NR 13 further forms

515 674 means for reducing the noise level of the input signal at the input terminal 23.

In operation, speech is recorded by the microphone 11 which emits an analog audio signal. In fact, however, not only speech is recorded by the microphone 11 but also noise that is in the vicinity of the speaking person. Typical noise environments are car environments, shopping centers and heavily congested streets. It should be noted that people often use mobile communication devices, such as cell-based telephones in this kind of noise environments and therefore the need for implementation of an effective noise reduction method is essential for these types of devices. The analog audio signal is converted into a digital signal by means of the analog / digital converter 12. The digital signal will act as an input signal on the input terminal 23. It is to be understood that the analogue / digital converter 12 is shown schematically and includes conventional pre-filtering means (e.g. low pass filter) as well as sampling and holding means. The sequence of digital values ​​appearing on the input terminal as an input signal can be designated z (n) where n denotes the nth sample.

The noise level of the input signal at the input terminal 23 is measured, as described below, by RVAD 15 and NSR 16. The comparison / controller DECISION 17 compares the measured noise level with a first predetermined noise level, NL1. If the noise level of the input signal is lower than the first predetermined noise level, the switch 18 is controlled so that the means for reducing the noise level of the input signal, NR, 13 is deactivated. In this case, the input signal at the input terminal 23 is connected directly by the switch 18 to the input of SPE 14 and, accordingly, NR 13. This corresponds to choice B at the selector 18. The fact that NR is deactivated means that its energy supply can cease to save energy. If NR is implemented as a software routine of a DSP 19, as indicated in Figure 1, this means that this software routine is not operational. About the noise level

515 674 of the input signal is higher than or equal to the first predetermined noise level, NL1, the switch 18 is controlled so that the noise reduction means of the input signal, NR, 13 is activated. In this case, the input signal at the input terminal 23 is processed by the NR and the thus treated noise-reduced signal forms the output signal of the noise reduction apparatus. This corresponds to choice A at the switch

18th In the embodiment shown in Fig. 1, this output signal is given to SPE 14, where it is further processed.

The following describes the operation of the noise measuring means for measuring the noise level of an input signal at the input terminal 23. First, the reduced voice activity detector, RVAD, 15 is used to determine whether or not there is a voice signal at the input terminal 23.

It should be noted that a conventional voice activity detector, VAD, can be used instead of a reduced voice activity detector. Such a conventional VAD is known, for example, from the ETSI standard for GSM (No. 06.32), whose method and implementation are hereby incorporated by reference.

Fig. 3 is a block diagram illustrating a conventional voice activity detector. A buffer memory, BUFFERT, 31 has an input for receiving a sampled signal in (n). The output of buffer memory 31 is connected to the input of a pitch detector, TONIGHT, 32 and to the input of a combined detector for autocorrelation function, ACF, and energy detector, ENERGY, 33. The output of the pitch detector 32 and the output of the autocorrelation function detector as well as the energy detector 33 are connected to the inputs of a decision unit, DECISION UNIT, 34. The output of the decision unit 34 is designated 0_WHAT. In operation, a sequence of sampled signals (sampled at times n) is stored in buffer memory 31. A control means (not shown) outputs the stored signals to the pitch detector 32, which determines the fundamental pitch of the delivered signals. At the same time

515 674, the stored signals are output to the autocorrelation function detector and the energy detector 33 which determines the autocorrelation function and the energy content of the output signals. The autocorrelation function detector ACF and the energy detector ENERGY are shown in a common box 33 in Fig. 3 because the square of the output signal needs to be determined by both the autocorrelation function detector and the energy detector. The decision unit 34 uses the output of the pitch detector 32 and the detector for autocorrelation function and the energy detector 33 to provide an output signal 0_WHAT. The output signal O_WHAT either assumes a first state (e.g., binary 1) or a second state (e.g., binary 0), depending on whether the decision unit 34 has determined that numbers exist or not.

The decision unit 34 operates according to the basic idea that when numbers are present, the pitch detector can determine a fundamental pitch and, at the same time, the energy content of the output signal is comparatively large. On the other hand, if there is no number, the pitch detector has difficulty determining the basic pitch, the autocorrelation is close to zero and the energy content is relatively low.

One problem with the conventional VAD is that it requires a large buffer memory and a lot of computational energy to determine the pitch of the signal. Assuming that a signal has sampled at 8000 samples / second and 13 bits of resolution, and that the pitch detector needs one second to determine the pitch, the buffer memory needs to store 104,000 bits. Further, the pitch detector consumes a great deal of energy when performing the required calculations. Therefore, it is advantageous to use a detector for reduced voice activity, RVAD, which can do without the pitch information.

515 674

Fig. 4 is a block diagram showing a detector for reduced voice activity. A buffer memory, BUFFERT, 41 has an input for receiving a sampled signal in (n). The output of the buffer memory 41 is connected to the input of a detector for autocorrelation function, ACF, and an energy detector, ENERGY, 43. The output of the detector for autocorrelation function and the energy detector 43 is connected to a decision unit, DECISION UNIT, 44. Output signal_RVAD. In operation, a sequence of sampled signals (sampled at times n) is stored in buffer memory 41. A controller (not shown) outputs the stored signals to the autocorrelation function detector and the energy detector 43 which determines the autocorrelation function and energy content of the delivered signals. The decision unit 44 uses the output of the autocorrelation function detector and the energy detector 43 to provide an output signal 0._RVAD. The output signal 0_RVAD either assumes a first state (e.g., binary 1) or a second state (e.g., binary 0), depending on whether the decision unit 44 has determined that numbers exist or not.

The decision unit 44 operates according to the basic idea that when there are numbers, the autocorrelation is not close to zero and the energy content of the delivered signal is relatively high. On the other hand, when there is no number, the autocorrelation is close to zero and the energy content is relatively low.

It should be noted that the format of the buffer memory 41 of the RVAD in Figure 4 need not be as large as the buffer memory 31 of the VAD of Figure 3. The reason for this is that the buffer memory 41 of the RVAD need only store a sufficient number of samples to determine the autocorrelation function and the energy content of the signal emitted. However, the buffer memory 41 of the VAD also needs to store the large amount of samples required to allow determination of the pitch of the output signal. Furthermore, RVAD consumes much less energy than VAD due to

515 674 because the energy consumed by the pitch detector to calculate the pitch of the output signal is considerable.

Returning to Fig. 1, the output of RVAD 15 (or if a VAD is used, the output of VAD) includes the binary information about whether or not numbers are present in the input signal z (n) at input terminal 23. If RVAD 15 indicates that numbers are present, the input signal is designated z (n) as x (n) and if RVAD indicates that number does not exist, the input signal z (n) is designated as v (n).

The noise / signal ratio R is then calculated by NSR 16 according to the algorithm:

l.fv (n) r&gt; tl = 1 ^R “--- N ä · Σ η - 1 where N is a predetermined number of samples. It may be noted that x (n) is set equal to zero if no voice is present in sample n and v (n) is set equal to zero if voice exists at sample n. Noise / signal ratio R is passed to the comparison / control, DECISION, 17, where R is compared to a first predetermined noise level, NL1. Switch 18 is controlled in accordance with the above description depending on whether R is lower than NL1 or not.

Fig. 2 is a block diagram illustrating a second embodiment of the present invention. The second embodiment is similar to the first embodiment and corresponding elements have received the same reference numerals and numbers. The second embodiment differs from the first embodiment in that switch 18 is replaced by a switch 28 having an additional third select clamp. Furthermore, the

515 674 comparison / control means DECISION 17 has been replaced with a comparison / control means DECISION 27 which compares the measured noise level with the first predetermined noise level NL1 and a second predetermined noise level NL2. DECISION 27 generates a control signal that assumes a unique state depending on whether the measured noise level is lower than the first predetermined noise level NL1 or if the measured noise level is higher than or equal to the first predetermined noise level NL1 but lower than or equal to the second predetermined noise level NL2, or if the measured noise level is higher than the other predetermined noise level NL2. The output of DECISION 27 is connected to the control input of switch 28. The switch 28 is connected so that, depending on the state of the control signal from DECISION 27, the input terminal 23 is connected either to the first selection terminal (choice A) connected to the input of the noise reduction means NR 13 or to a second selection terminal (choice B) connected to the the output of NR 13 or to a third select terminal (choice C) connected to an input of a low pass filter LP 21. The second embodiment further differs from the first embodiment in that the input of the low-pass filter 21 is connected to the third select terminal of switch 28 and the output of the low-pass filter 21 is connected to the output of NR 13. As shown in Fig. 2, the low-pass filter is implemented by the digital signal processor. 19th It will be appreciated that switch 28 can be implemented by a first switch means corresponding to switch 18 in the first embodiment and a second switch means corresponding to a switch allowing the low-pass filter 21 to be connected between the input terminal 23 and the output of NR 13. LP 21 further forms a filtering means for filtering the input signal. at the input terminal.

In operation, speech is recorded by the microphone 11 and converted into a digital signal in the same way as discussed above in connection with the first embodiment.

515 674

The noise level of the input signal at the input terminal 23 is measured, as described above, by RVAD 15 and NSR 16. DECISION 27 compares the measured noise level with the first predetermined noise level NL1 and with the second predetermined noise level NL2. If the noise level of the input signal is lower than a first predetermined noise level NL1, the switch 28 is controlled so that the means for reducing the noise level of the input signal, NR 13, is deactivated. In this case, the input signal at input terminal 23 by switch 28 is directly connected to the input of SPE 14 and is accordingly passed by NR 13. This corresponds to choice B at switch 28. The fact that NR is deactivated means that its energy can be switched off to save energy. If NR is implemented as a software routine of a DSP 19 as indicated in Figure 2, this means that the software routine does not run. If the noise level of the input signal is higher than or equal to the first predetermined noise level NL1 but lower than or equal to a second predetermined noise level NL2, the switch 28 is controlled so that the noise reduction means of the input signal NR 13 is activated. In this case, the input signal at the input terminal 23 is processed by the NR and the thus treated noise-reduced signal produces an output signal from the noise reduction apparatus. This corresponds to choice A of switch 28. If the noise level of the input signal is higher than the other predetermined noise level NL2, the switch 28 is controlled so that the low pass filter LP 21 is activated. In this case, the input signal at the input terminal 23 is processed by LP and the thus filtered signal forms the output of the noise reduction apparatus. At the same time, NR 13. is deactivated. This corresponds to choice C of switch 28. The fact that NR is deactivated again means that its energy can be switched off to save energy. The idea is that at high noise levels, it is assumed that a (simple) low pass filter will work at least as well as a (complex) noise reduction algorithm of NR. By turning off the NR power supply and instead using the less energy consuming low pass filter, the total

515 674 energy consumption in this case. It will be appreciated that LP 21 can be deactivated and, consequently, its energy can be turned off by means of switch 28 when the noise level of the input signal is lower than or equal to the other predetermined noise level NL2. In the embodiment shown in Figure 2, this output signal is fed to SPE 14 where it is processed further.

The noise measuring means for measuring the noise level of an input signal at the input terminal 23 operates in the same manner as discussed above in connection with the first embodiment. The only difference is that the noise / signal ratio R given to the comparison / control means DECISION 27 is compared by DECISION 27 with a first predetermined noise level NL1 and a second predetermined noise level NL2. Switch 28 is controlled in accordance with the description above depending on whether R is lower than NL1, higher than or equal to NL1 but lower than or equal to NL2, or higher than NL2.

Fig. 5 is a flow chart showing a first mode of operation of the present invention. This mode of operation corresponds to the mode of operation described in connection with the first embodiment above. First, the noise level of an input signal is measured. Then the measured noise level is compared with a predetermined noise level NL1. If the measured noise level is lower than the predetermined noise level, the input signal is output as the output signal. Otherwise, the noise of the input signal is reduced, e.g. with a suitable noise reduction algorithm, and the thus noise-reduced signal is output as the output signal.

Fig. 6 is a flow chart showing a second mode of operation of the present invention. This method corresponds to the mode of operation discussed in connection with the second embodiment above. First, the noise level of an input signal is measured. Then the measured noise level is compared with a first predetermined noise level NL1. If the measured noise level is lower than the predetermined noise level, the input signal is output as the output signal. About the measured noise level

515 674 is higher than or equal to the first predetermined noise level, the measured noise level is compared to a second predetermined noise level NL2. If the measured noise level is higher than the other predetermined noise level, the input signal is filtered by a low pass filter and the thus filtered signal is output as the output signal. Otherwise, the noise of the input signal is reduced, e.g. by a suitable noise reduction algorithm, and thus the noise-reduced signal is output as the output signal.

The noise reduction apparatus and the noise processing method are particularly suitable in communication systems (not shown) and mobile communication devices (not shown), such as cellular telephones. The communication system or mobile communication device comprises a microphone, a radio receiver / transmitter and an air interface. By coupling the signal from the microphone to the noise reduction apparatus and the output of the noise reduction apparatus to the receiver / transmitter, the noise processing method of the present invention is performed on the signal from the microphone before the signal is transmitted by the receiver / transmitter and the air interface.

If a hands-free equipment for mobile communication devices, such as cell-based telephones, is used, the noise reduction apparatus and the noise processing method of the present invention can preferably be implemented in the hands-free equipment. The output of the hands-free equipment to the mobile communication device will then have been processed according to the noise processing method of the present invention before entering the mobile communication device.

It will be appreciated by those skilled in the art to vary the characteristics of the described embodiments and use well-known equivalents without departing from the present invention. Although

515 674

<td> 16</td><td></td>

<td>the embodiments shown use signal processor for implementation</td><td>and digital of the present</td>

the invention, for example, is in no way limited to this mode of implementation. Instead, the entire structure or parts of it can be implemented through hardware.

515 674