Comfort noise addition for modeling background noise at low bit-rates

Comfort noise addition for modeling background noise at low bit-rates

Description

The present invention relates to audio signal processing, and, in particular, to noisy speech coding and comfort noise addition to audio signals.

Comfort noise generators are usually used in discontinuous transmission (DTX) of audio signals, in particular of audio signals containing speech. In such a mode the audio signal is first classified in active and inactive frames by a voice activity detector (VAD). An example of a VAD can be found in [1]. Based on the VAD result, only the active speech frames are coded and transmitted at the nominal bit-rate. During long pauses, where only the background noise is present, the bit-rate is lowered or zeroed and the background noise is coded episodically and parametrically. The average bitrate is then significantly reduced. The noise is generated during the inactive frames at the decoder side by a comfort noise generator (CNG). For example the speech coders AMR-WB [2] and ITU G.718 [1] have the possibility to be run both in DTX mode.

The coding of speech and especially of noisy speech at low bit-rates is prone to artefacts. Speech coders are usually based on a speech production model which doesn’t hold anymore in presence of background noise. In that case, the coding efficiently drops and the quality of decoded audio signal decreases. Moreover certain characteristics of speech coding may be especially perturbing when handling noisy speech. Indeed at low rates, the coarse quantization of coding parameters produces some fluctuation over time, fluctuations perceptually annoying when coding speech over stationary background noise.

Noise reduction is a well-known technique for enhancing the intelligibility of speech and improving the communication in the presence of background

CA 02895391 2016-11-08 noise. It was also adopted in speech coding. For example the coder G.718 uses noise reduction for deducing some coding parameters like the speech pitch. It has also the possibility to code the enhanced signal instead of the original signal. The speech is then more predominant compared to the noise level in the decoded signal. However, it usually sounds more degraded or less natural, as noise reduction might distort the speech components and cause audible musical noise artifacts in addition to the coding artifacts.

The object of the present invention is to provide improved concepts for audio signal processing.

In one aspect the invention provides a decoder being configured for processing an encoded audio bitstream, wherein the decoder comprises:

a bitstream decoder configured to derive a decoded audio signal from the bitstream, wherein the decoded audio signal comprises at least one decoded frame;

a noise estimation device configured to produce a noise estimation signal containing an estimation of the level and/or the spectral shape of a noise in the decoded audio signal;

a comfort noise generating device configured to derive a comfort noise signal from the noise estimation signal; and a combiner configured to combine the decoded frame of the decoded audio signal and the comfort noise signal in order to obtain an audio output signal.

CA 02895391 2016-11-08

The bitstream decoder may be a device or a computer program capable of decoding an audio bitstream, which is a digital data stream containing audio information. The decoding process results in a digital decoded audio signal, which may be fed to an A/D converter to produce an analogous audio signal, which then may be fed to a loudspeaker, in order to produce an audible signal.

The decoded audio signal is divided into so called frames, wherein each of these frames contains audio information referring to a certain time interval. Such frames may be classified into active frames and inactive frames, wherein an active frame is a frame, which contains wanted components of the audio information, such as speech or music, whereas an inactive frame is a frame, which does not contain any wanted components of the audio information. Inactive frames usually occur during pauses, where no wanted components, such as music or speech, are present. Therefore, inactive frames usually contain solely background noise.

In discontinuous transmission (DTX) of audio signal only the active frames of the decoded audio signal are obtained by decoding the bitstream as during inactive frames the encoder does not transmit the audio signal within the bitstream.

In non- discontinuous transmission (non-DTX) of audio signal the active frames as well as the inactive frames are obtained by decoding the bitstream.

Frames which are obtained by decoding the bitstream by the bitstream decoder are referred to as decoded frames.

The noise estimation device is configured to produce a noise estimation signal containing an estimation of the level and/or the spectral shape of a noise in the decoded audio signal. Further, the comfort noise generating device is configured to derive a comfort noise signal from the noise

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 estimation signal. The noise estimation signal may be a signal, which contains information regarding the characteristics of the noise contained in the decoded audio signal in a parametric form. The comfort noise signal is an artificial audio signal, which corresponds to the noise contained in the decoded audio signal. These features allow the comfort noise to sound like the actual background noise without requiring any side information regarding the background noise in the bitstream.

The combiner is configured to combine the decoded frame of the decoded audio signal and the comfort noise signal in order to obtain an audio output signal. As a result the audio output signal comprises decoded frames, which comprise artificial noise. The artificial noise in the decoded frames allows masking artifacts in the audio output signal especially when the bitstream is transmitted at low bit-rates. It smooths the usually observed fluctuations and in the meantime masks the predominant coding artifacts.

In contrast to prior art, the present invention applies the principle of adding artificial comfort noise to decoded frames. The inventive concept may be applied in both DTX and non-DTX modes.

The invention provides a method for enhancing the quality of noisy speech coded and transmitted at low bit-rates. At low bit-rates, the coding of noisy speech, i.e. speech recorded with background noise, is usually not as efficient as the coding of clean speech. The decoded synthesis is usually prone to artifacts. The two different kinds of sources, the noise and the speech, can’t be efficiently coded by a coding scheme relying on a singlesource model. The present invention provides a concept for modeling and synthesizing the background noise at the decoder side and requires very small or no side-information. This is achieved by estimating the level and spectral shape of the background noise at the decoder side, and by generating artificially a comfort noise. The generated noise is combined with the decoded audio signal and allows masking coding artifacts.

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527

Furthermore, the concept can be combined with a noise reduction scheme applied at the encoder side. Noise reduction enhances the signal-to-noise ratio (SNR) level, and improves the performance of the subsequent audio coding. The missing amount of noise in the decoded audio signal is then compensated by the comfort noise at the decoder side. However, it usually sounds more degraded or less natural, as noise reduction might distort the audio components and cause audible musical noise artifacts in addition to the coding artifacts. One aspect of the present invention is to mask such unpleasant distortions by adding a comfort noise at the decoder side. When using a noise reduction scheme, the addition of comfort noise does not deteriorate the SNR. Moreover, the comfort noise conceals a great part of the annoying musical noise typical to noise reduction techniques.

In a preferred embodiment of the invention the decoded frame is an active frame. This feature extends the principle of comfort noise addition to decoded active frames.

In a preferred embodiment of the invention the decoded frame is an active frame. This feature extends the principle of comfort noise addition to decoded inactive frames.

In a preferred embodiment of the invention the noise estimating device comprises a spectral analysis device configured to create an analysis signal containing the level and the spectral shape of the noise in the decoded audio signal and a noise estimation producing device configured to produce the noise estimation signal based on the analysis signal.

In a preferred embodiment of the invention the comfort noise generating device comprises a noise generator configured to create a frequency domain comfort noise signal based on the noise estimation signal and a spectral

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 synthesizer configured to create the comfort noise signal based on the frequency domain comfort noise signal.

In a preferred embodiment of the invention the decoder comprises a switch device configured to switch the decoder alternatively to a first mode of operation or to a second mode of operation, wherein in the first mode of operation the comfort noise signal is fed to the combiner, whereas the comfort noise signal is not fed to the combiner in the second mode of operation. These features allow to cease the use of the artificial comfort noise in situations, where it is not needed.

In a preferred embodiment of the invention the decoder comprises a control device configured to control the switch device automatically, wherein the control device comprises a noise detector configured to control the switch device depending on a signal-to-noise ratio of the decoded audio signal, wherein under low-signal-to-noise-ratio-conditions the decoder is switched to the first mode of operation and under high-signal-to-noise-ratio-conditions to the second mode of operation. By these features the comfort noise may be triggered in noisy speech scenarios only, i.e., not in clean speech or clean music situations. For the purpose of discriminating between low-signal-tonoise-ratio-conditions and high-signal-to-noise-ratio-conditions a threshold for the signal-to-noise ratio may be defined and used.

In a preferred embodiment of the invention the control device comprises a side information receiver configured to receive side information contained in the bitstream, which corresponds to the signal-to-noise ratio of the decoded audio signal, and configured to create a noise detection signal, wherein the noise detector controls the switch device depending on the noise detection signal. These features allow controlling the switch device based on a signal analysis done by an external device producing and/or processing the received bitstream. The external device especially may be an encoder producing the bitstream.

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527

In a preferred embodiment of the invention the side information corresponding to the signal-to-noise ratio of the decoded audio signal consists of at least one dedicated bit in the bitstream. A dedicated bit in general is a bit, which contains, alone or together with other dedicated bits, defined information. Here, the dedicated bit may indicate, if the signal-tonoise ratio is above or below a predefined threshold.

In a preferred embodiment of the invention the control device comprises a wanted signal energy estimator configured to determine an energy of a wanted signal of the decoded audio signal, a noise energy estimator configured to determine an energy of a noise of the decoded audio signal and a signal-to-noise ratio estimator configured to determine the signal-tonoise ratio of the decoded audio signal based on the energy of wanted signal and based on the energy of the noise, wherein the switch device is switched depending on the signal-to-noise ratio determined by the control device. In this case no side information in the bitstream is necessary. As the energy of the wanted signal usually exceeds the energy of the noise of the decoded signal, the total energy of the decoded audio signal, including the energy of the wanted signal as well as the energy of the noise, gives a rough estimation of the energy of the wanted signal of the decoded audio signal. For this reason, the signal-to-noise ratio may be calculated in an approximation by dividing the total energy of the decoded audio signal by the energy of the noise of the decoded signal.

In a preferred embodiment of the invention the bitstream contains active frames and inactive frames, wherein the control device is configured to determine the energy of the wanted signal of the decoded audio signal during the active frames and to determine the energy of the noise of the decoded audio signal during inactive frames. By this, a high accuracy in estimating the signal-to-noise ratio may be achieved in an easy way.

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527

In a preferred embodiment of the invention the bitstream contains active frames and inactive frames, wherein the decoder comprises a side information receiver configured to discriminate between the active frames and the inactive frames based on side information in the bitstream indicating whether the present frame is active or inactive. By this feature active frames or in active frames respectively may be identified without calculating effort.

In a preferred embodiment of the invention the side information indicating whether the present frame is active or inactive consists of at least one dedicated bit in the bitstream.

In a preferred embodiment of the invention the control device is configured to determine the energy of the wanted signal of the decoded audio signal based on the analysis signal. In this case the analysis signal, which usually has to be computed for the purpose of noise estimation, may be reused, so that the complexity may be reduced.

In a preferred embodiment of the invention the control device is configured to determine the energy of the noise of the decoded audio signal based on the noise estimation signal. In such an embodiment the noise estimation signal, which typically has to be computed for the purpose of comfort noise generating, may be reused, so that the complexity may be further reduced.

In a preferred embodiment of the invention the comfort noise generating device is configured to create the comfort noise signal based on a target comfort noise level signal. The level of added comfort noise should be limited to preserve intelligibility and quality. This may be achieved by scaling the comfort noise using a target noise signal which indicates a pre-determined target noise level.

In a preferred embodiment of the invention the target comfort noise level signal is adjusted depending on a bit-rate of the bitstream. Typically, the

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 decoded audio signal exhibits a higher signal-to-noise ratio than the original input signal, especially at low bit-rates where the coding artifacts are the most severe. This attenuation of the noise level in speech coding is coming from the source model paradigm which expects to have speech as input. Otherwise, the source model coding is not entirely appropriate and won’t be able to reproduce the whole energy of non-speech components. Hence, the target comfort noise level signal may be adjusted depending on the bit-rate to roughly compensate for the noise attenuation inherently introduced by coding process.

In a preferred embodiment of the invention the target comfort noise level signal is adjusted depending on a noise attenuation level caused by a noise reduction method applied to the bitstream. By this features the noise attenuation caused by a noise reduction module in an encoder may be compensated.

In a preferred embodiment of the invention an energy of the frequency domain comfort noise signal of the random noise w(fc) is adjusted depending on the target comfort noise level signal, which indicates a target comfort noise level 0_tar, for each frequency k as E_w(fc) = maxf^,· - 1) ; 0}, wherein E_n(k) refers to an estimate of the energy of the noise of the decoded audio signal at frequency k, as delivered by the noise estimation producing device. By these features intelligibility and quality of the output signal may be enhanced.

In a preferred embodiment of the invention the decoder comprises a further bitstream decoder, wherein the bitstream decoder and the further bitstream decoder are of different types, wherein the decoder comprises a switch configured to feed either the decoded signal from the bitstream decoder or the decoded signal from the further bitstream decoder to the noise estimation device and to the combiner. As the comfort noise addition is done when

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 using the bitstream decoder as well as when using the further bitstream decoder, transition artefacts when switching between the bitstream decoder and the further bitstream decoder may be minimized. For example, the bitstream decoder may be an algebraic code excited linear prediction (ACELP) bitstream decoder, whereas the further bitstream decoder may be a transform-based core (TCX) bitstream decoder.

The invention further provides an audio signal processing encoder being configured for producing an audio bitstream, wherein the encoder comprises:

a bitstream encoder configured to produce an encoded audio signal corresponding to an audio input signal and to derive the bitstream from the encoded audio signal;

an signal analyzer having a signal-to-noise ratio estimator configured to determine the signal-to-noise ratio of the audio input signal based on an energy of a wanted signal of the audio signal determined by a wanted signal energy estimator and based on an energy of a noise of the audio input signal determined by noise energy estimator;

a noise reduction device configured to produce an noise reduced audio signal; and a switch device configured to feed, depending on the determined signal-tonoise ratio of the audio input signal, either the audio input signal or the noise reduced audio signal to the bitstream encoder for the purpose of encoding the respective signal, wherein the bitstream encoder is configured to transmit a side information, which indicates whether the audio input signal or noise reduced audio signal is encoded, within in the bitstream.

The bitstream encoder may be a device or a computer program capable of encoding an audio signal, which is a digital data signal containing audio

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 information. The encoding process results in a digital bitstream, which may be transmitted over a digital data link to a decoder at a remote location.

The audio input signal is directly coded by the bitstream encoder. The bitstream encoder can be a speech encoder or a low-delay scheme switching between a speech coder ACELP and a transform-based audio coder TCX. The bitstream encoder is responsible for coding the audio input signal and generating the bitstream needed for decoding the audio signal. In parallel, the input signal is analyzed by any module called signal analyzer. In a preferred embodiment the signal analysis is the same as the one used in

G.718. It consists of a spectral analysis device followed by the noise estimation producing device. The spectrums of both the original signal and the estimated noise are input in the noise reduction module. The noise reduction attenuates the background noise level in the frequency domain. The amount of reduction is given by the target attenuation level. The enhanced time-domain signal (noise reduced audio signal) is generated after spectral synthesis. The signal is used for deducing some features, like the pitch stability which is then exploited by the VAD for discriminating between active and inactive frames. The result of the classification can be further used by the encoder module. In the preferred embodiment, a specific coding mode is used to handle inactive frames. This way, the decoder can deduce the VAD flag from the bit-stream without requiring a dedicated bit.

To avoid unnecessary distortions in noiseless situations (clean speech or clean music), noise reduction is applied only in case of noisy speech and is bypassed otherwise. The discrimination between noisy and noiseless signals is achieved by estimating the long-term energy of both the noise and the desired signal (speech or music). The long-term energy is computed by a first-order auto-regressive filtering of either the input frame energy (during active frames) or using the output of the noise estimation module (during inactive frames). In this way an estimate of the signal-to-noise ratio can be computed, which is defined as the ratio of the long-term energy of the speech

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 or music over the long-term energy of the noise. If the signal-to-noise ratio is below a predetermined threshold, the frame is considered as noisy speech otherwise it is classified as clean speech. As the bitstream encoder is configured to transmit within in the bitstream side information, which indicates whether the audio input signal or noise reduced audio signal is encoded, the decoder may adjust the target comfort noise level signal automatically to the mode of operation of the encoder.

In the preferred embodiment of the invention during active frames, only the long-term speech/music energy estimate is updated. During inactive frames, only the noise energy estimate is updated.

The invention further provides a system comprising an audio signal processing decoder and an audio signal processing encoder, wherein the decoder is designed according to the claimed invention and/or the encoder is designed according to the claimed invention.

In another aspect the invention provides a method of decoding an audio bitstream, wherein the method comprises:

deriving a decoded audio signal from the bitstream, wherein the decoded audio signal comprises at least one decoded frame;

producing a noise estimation signal containing an estimation of the level and/or the spectral shape of a noise in the decoded audio signal;

deriving a comfort noise signal from the noise estimation signal; and combining the decoded frame of the decoded audio signal and the comfort noise signal in order to obtain an audio output signal.

CA 02895391 2016-11-08

The invention further provides a method of audio signal encoding for producing an audio bitstream, wherein the method comprises:

determining the signal-to-noise ratio of an audio input signal based on a determined energy of a wanted signal of the audio input signal and a determined energy of a noise of the audio input signal;

producing an noise reduced audio signal;

producing an encoded audio signal corresponding to the audio input signal, wherein, depending on the determined signal-to-noise ratio of the audio input signal, either the audio input signal or the noise reduced audio signal is encoded;

deriving the bitstream from the encoded audio signal; and transmitting a side information, which indicates whether the audio input signal or the noise reduced audio signal is encoded, within the bitstream.

The invention further provides a bitstream produced according to the method above. The claimed bitstream contains side information, which indicates whether the audio input signal or the noise reduced audio signal is encoded.

A further aspect the invention provides a computer program for performing, when running on a computer or a processor, the inventive methods.

Preferred embodiments of the invention are subsequently discussed with respect to the accompanying drawings, in which:

Fig. 1 illustrates a first embodiment of a decoder according to the invention;

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527

Fig. 2 illustrates a second embodiment of a decoder according to the invention;

Fig. 3 illustrates an encoder according to prior art;

Fig. 4 illustrates a first embodiment of an encoder according to the invention;

Fig. 5 illustrates a second embodiment of an encoder according to the invention; and

Fig. 6 illustrates an embodiment of a frame format of the bitstream according to the invention.

Fig. 1 illustrates a first embodiment of a decoder 1 according to the invention. The decoder 1 is configured for processing an encoded audio bitstream BS, wherein the decoder 1 comprises:

a bitstream decoder 2 configured to derive a decoded audio signal DS from the bitstream BS, wherein the decoded audio signal DS comprises at least one decoded frame;

a noise estimation device 3 configured to produce a noise estimation signal NE containing an estimation of the level and/or the spectral shape of a noise N in the decoded audio signal DS;

a comfort noise generating device 4 configured to derive a comfort noise audio signal CN from the noise estimation signal NE; and a combiner 5 configured to combine the decoded frame of the decoded audio signal DS and the comfort noise signal CN in order to obtain an audio output signal OS.

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527

The bitstream decoder 2 may be a device or a computer program capable of decoding an audio bitstream BS, which is a digital data stream containing audio information. The decoding process results in a digital decoded audio signal DS, which may be fed to an A/D converter to produce an analogous audio signal, which then may be fed to a loudspeaker, in order to produce an audible signal.

The decoded audio signal DS comprises so called frames, wherein each of these frames contains audio information referring to a certain time. Such frames may be classified into active frames and inactive frames, wherein an active frame is a frame, which contains wanted components WS of the audio information, also referred to as wanted signal WS, such as speech or music, whereas an inactive frame is a frame, which does not contain any wanted components of the audio information. Inactive frames usually occur during pauses, where no wanted components, such as music or speech, are present. Therefore, inactive frames usually contain solely background noise N.

The noise estimation device 3 is configured to produce a noise estimation signal NE containing an estimation of the level and/or the spectral shape of a noise in the decoded audio signal DS. Further, the comfort noise generating device 4 is configured to derive a comfort noise audio signal CN from the noise estimation signal NE. The noise estimation signal NE may be a signal, which contains information regarding the characteristics of the noise N contained in the decoded audio signal DS in a parametric form. The comfort noise signal CN is an artificial audio signal, which corresponds to the noise N contained in the decoded audio signal DS. These features allow the comfort noise CN to sound like the actual background noise N without requiring any side information in the bitstream BS regarding the background noise N.

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527

The combiner 5 is configured to combine the decoded frame of the decoded audio signal DS and the comfort noise signal CN in order to obtain an audio output signal OS. As a result the audio output signal OS comprises decoded frames, which comprise artificial noise CN. The artificial noise CN in the decoded frames allows masking artifacts in the audio output signal OS especially when the bitstream BS is transmitted at low bit-rates.

In contrast to prior art, the present invention applies the principle of adding artificial comfort noise CN to decoded active or non-active frames. The inventive concept may be applied in both DTX and non-DTX modes.

The invention provides a method for enhancing the quality of noisy speech coded and transmitted at low bit-rates. At low bit-rates, the coding of noisy speech, i.e. speech recorded with background noise N, is usually not as efficient as the coding of clean speech WS. The decoded synthesis is usually prone to artifacts. The two different kinds of sources, the noise N and the speech WS, can’t be efficiently coded by a coding scheme relying on a single-source model. The present invention provides a concept for modeling and synthesizing the background noise N at the decoder side and requires very small or no side-information. This is achieved by estimating the level and spectral shape of the background noise N at the decoder side, and by generating artificially a comfort noise CN. The generated noise CN is combined with the decoded audio signal DS and allows masking coding artifacts during decoded frames.

Furthermore, the concept can be combined with a noise reduction scheme applied at the encoder side. Noise reduction enhances the signal-to-noise ratio (SNR) level, and improves the performance of the subsequent audio coding. The missing amount of noise N in the decoded audio signal DS is then compensated by the comfort noise CN at the decoder side. However, it usually sounds more degraded or less natural, as noise reduction might distort the audio components and cause audible musical noise artifacts in

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 addition to the coding artifacts. One aspect of the present invention is to mask such unpleasant distortions by adding a comfort noise CN at the decoder side. When using a noise reduction scheme, the addition of comfort noise does not deteriorate the SNR. Moreover, the comfort noise conceals a great part of the annoying musical noise typical to noise reduction techniques.

In a preferred embodiment of the invention the decoded frame is an active frame. This feature extends the principle of comfort noise addition to decoded active frames.

In a preferred embodiment of the invention the decoded frame is an active frame. This feature extends the principle of comfort noise addition to decoded inactive frames.

In a preferred embodiment of the invention the noise estimating device 3 comprises a spectral analysis device 6 configured to create an analysis signal AS containing the level and the spectral shape of the noise in the decoded audio signal DS and a noise estimation producing device 7 configured to produce the noise estimation signal NE based on the analysis signal AS.

In a preferred embodiment of the invention the comfort noise generating device comprises 4 a noise generator 8 configured to create a frequency domain comfort noise signal FD based on the noise estimation signal NE and a spectral synthesizer 9 configured to create the comfort noise CN signal based on the frequency domain comfort noise signal FD.

In a preferred embodiment of the invention the decoder 1 comprises a switch device 10 configured to switch the decoder 1 alternatively to a first mode of operation or to a second mode of operation, wherein in the first mode of operation the comfort noise signal CN is fed to the combiner, whereas the

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 comfort noise signal CN is not fed to the combiner 5 in the second mode of operation. These features allow to cease the use of the artificial comfort noise CN in situations, where it is not needed.

In a preferred embodiment of the invention the decoder 1 comprises a control device 11 configured to control the switch device 10 automatically, wherein the control device 10 comprises a noise detector 12 configured to control the switch device 10 depending on a signal-to-noise ratio of the decoded audio signal DS, wherein under low-signal-to-noise-ratio-conditions the decoder is switched to the first mode of operation and under high-signal-to-noise-ratioconditions to the second mode of operation. By these features the use of comfort noise CN may be triggered in noisy speech scenarios only, i.e., not in clean speech or clean music situations. For the purpose of discriminating between low-signal-to-noise-ratio-conditions and high-signal-to-noise-ratioconditions a threshold for the signal-to-noise ratio may be defined and used.

In a preferred embodiment of the invention the control device 11 comprises a side information receiver 13 configured to receive side information contained in the bitstream BS, which corresponds to the signal-to-noise ratio of the decoded audio signal DS, and configured to create a noise detection signal ND, wherein the noise detector 12 switches the switch device 11 depending on the noise detection signal ND. These features allow to control the switch device 10 based on a signal analysis done by an external device producing and/or processing the received bitstream BS. The external device especially may be an encoder producing the bitstream BS.

In a preferred embodiment of the invention the side information corresponding to the signal-to-noise ratio of the decoded audio signal DS consists of at least one dedicated bit in the bitstream BS. A dedicated bit in general is a bit, which contains, alone or together with other dedicated bits, defined information. Here, the dedicated bit may indicate, if the signal-tonoise ratio is above or below a predefined threshold.

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527

In a preferred embodiment of the invention the comfort noise generating device 4 is configured to create the comfort noise signal CN based on a target comfort noise level signal TNL. The level of added comfort noise CN should be limited to preserve intelligibility and quality. This may be achieved by scaling the comfort noise CN using a target noise signal TNL which indicates a pre-determined target noise level.

In a preferred embodiment of the invention the target comfort noise level signal TNL is adjusted depending on a bit-rate of the bitstream BS. Typically, the decoded audio signal DS exhibits a higher signal-to-noise ratio than the original input signal, especially at low bit-rates where the coding artifacts are the most severe. This attenuation of the noise level in speech coding is coming from the source model paradigm which expects to have speech as input. Otherwise, the source model coding is not entirely appropriate and won’t be able to reproduce the whole energy of no-speech components. Hence, the target comfort noise level signal TNL may be adjusted depending on the bit-rate to roughly compensate for the noise attenuation inherently introduced by coding process.

In a preferred embodiment of the invention the target comfort noise level signal TNL is adjusted depending on a noise attenuation level caused by a noise reduction method applied to the bitstream BS. By this features the noise attenuation caused by a noise reduction module in an encoder may be compensated.

In a preferred embodiment of the invention an energy of the frequency domain comfort noise signal FD of the random noise w(k) is adjusted depending on the target comfort noise level signal TNL, which indicates a target comfort noise level g_tar, for each frequency k as E_w(fc) = max{(^_tar -

1) £_n(k) ; O), wherein F_n(fc) refers to an estimate of the energy of the noise N

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 of the decoded audio signal DS at frequency k, as delivered by the noise estimation producing device 7. By these features intelligibility and quality of the output signal OS may be enhanced.

Fig. 2 illustrates a second embodiment of a decoder 1 according to the invention. The second embodiment of the decoder 1 is based on the decoder 1 of the first embodiment. In the following only the differences to the first embodiment discussed and explained.

In a preferred embodiment of the invention the control device comprises a wanted signal energy estimator 14 configured to determine an energy of a wanted signal WS of the decoded audio signal DS, a noise energy estimator 15 configured to determine an energy of a noise N of the decoded audio signal DS and a signal-to-noise ratio estimator 16 configured to determine the signal-to-noise ratio of the decoded audio signal DS based on the energy of wanted signal WS and based on the energy of the noise N, wherein the switch device 10 is switched depending on the signal-to-noise ratio determined by the control device 11. In this case no side information in the bitstream regarding the signal-to-noise ratio is necessary. Therefore, the side information receiver 13 of the first embodiment is not necessary as well.

In a preferred embodiment of the invention the bitstream BS contains active frames and inactive frames, wherein the control device 11 is configured to determine the energy of the wanted signal WS of the decoded audio signal DS during the active frames and to determine the energy of the noise N of the decoded audio signal DS during inactive frames. By this, a high accuracy in estimating the signal-to-noise ratio may be achieved in an easy way.

In a preferred embodiment of the invention the bitstream BS contains active frames and inactive frames, wherein the decoder 1 comprises a side information receiver 17 configured to discriminate between the active frames and the inactive frames based on side information in the bitstream indicating

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 whether the present frame is active or inactive. By this feature active frames or in active frames respectively may be identified without calculating effort.

In the preferred embodiment of the invention the side information receiver 17 may be configured to control and a switch 17a, which alternatively feeds an output signal OW of the wanted signal energy estimator 14 or an output signal ON of the noise energy estimator 15 to the signal-to-noise ratio estimator 16, wherein the output signal OW of a wanted signal energy estimator 14 is fed to the to the signal-to-noise ratio estimator 16 during active frames and wherein the output signal ON of the noise energy estimate of 15 is fed to the to the signal-to-noise ratio estimator 16 during inactive frames. By these features the signal-to-noise ratio may be calculated in an easy and accurate manner.

In a preferred embodiment of the invention the control device 11 is configured to determine the energy of the wanted signal of the decoded audio signal based on the analysis signal AS. In this case the analysis signal AS, which usually has to be computed for the purpose of noise estimation, may be reused, so that the complexity may be reduced.

In a preferred embodiment of the invention the control device 11 is configured to determine the energy of the noise N of the decoded audio signal DS based on the noise estimation signal NE. In such an embodiment the noise estimation signal NE, which typically has to be computed for the purpose of comfort noise generating, may be reused, so that the complexity may be further reduced.

In a preferred embodiment of the invention the decoder 1 comprises a further bitstream decoder (not shown in the figures), wherein the bitstream decoder 2 and the further bitstream decoder are of different types, wherein the decoder 1 comprises a switch (not shown in the figures) configured to feed either the decoded signal DS from the bitstream decoder 2 or the decoded

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 signal from the further bitstream decoder to the noise estimation device 3 and to the combiner 5. As the comfort noise addition is done when using the bitstream decoder 2 as well as when using the further bitstream decoder, transition artefacts when switching between the bitstream decoder 2 and the further bitstream decoder may be minimized. For example, the bitstream decoder 2 may be an algebraic code excited linear prediction (ACELP) bitstream decoder, whereas the further bitstream decoder may be a transform-based core (TCX) bitstream decoder.

The decoder 1 of the invention is described in figures 1 and 2, where the comfort noise addition is done blindly in the frequency domain. To have a comfort noise CN which looks like the actual background noise N, a noise estimation device 3 is used at the decoder 1 to determine the level and spectral shape of the background noise N, without requiring any sideinformation.

The comfort noise generating device 4 is triggered in noisy speech scenarios only, i.e., not in clean speech or clean music situations. The discrimination can be based on the detection performed in the encoder. In this case, the decision should be transmitted using a dedicated bit. In a preferred embodiment, in contrast, a noise estimation producing device 7 is applied which is similar to the noise estimation device used in the encoder. It consists in estimating the long-term signal-to noise ratio by separately adapting longterm estimates of either the energy of the noise N or the energy of the wanted signal WS, such as speech and/or music, depending on the VAD decision. The latter may be deduced directly from the index of the ACELP and TCX modes. Indeed, TCX and ACELP can be run in a specific mode called TCX-ΝΑ and ACELP-NA, respectively, when the signal is non-active speech/music frames, i.e., frames with background noise only. All other modes of ACELP and TCX refer to active frames. Hence the presence of a dedicated VAD bit in the bit-stream can be avoided.

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527

The level of added comfort noise should be limited to preserve intelligibility and quality. The comfort noise is hence scaled to reach a pre-determined target noise level. If g_tar denotes the target noise amplification level after comfort noise addition, the energy E_w of the random noise w(k) is adjusted for each frequency k as

E_w(k) = max{(£_tar - 1) Ê_n(k) ; θ), where Ê_n(fc) refers to an estimate of the noise energy present in the decoded audio output at frequency k, as delivered by the noise estimation module.

Typically, the decoded audio signal DS exhibits a higher signal-to-noise ratio than the original input signal, especially at low bit-rates where the coding artifacts are the most severe. This attenuation of the noise level in speech coding is coming from the source model paradigm which expects to have speech as input. Otherwise, the source model coding is not entirely appropriate and won’t be able to reproduce the whole energy of no-speech components. Hence, for the first aspect of the invention using the encoder depicted in figure 3, the target comfort noise level i/_tar is adjusted depending on the bit-rate to roughly compensate for the noise attenuation inherently introduced by coding process.

For the second aspect of the invention using the encoder depicted in figures 4 and 5, the target comfort noise level g_tar should, in addition, account for the noise attenuation caused by the noise reduction module in the encoder.

Furthermore, the comfort noise addition as described herein allows to smooth the transition artefact between one coding type (e.g.) to another one (e.g. TCX) by adding uniformly a comfort noise over all frames.

Fig. 3 illustrates an encoder according to prior art which can be used in combination with the decoders depicted in figures 1 and 2.

CA 02895391 2016-11-08

The input signal IS is directly coded by the bitstream encoder 20. The bitstream encoder 20 can be a speech coder or a low-delay scheme switching between a speech coder ACELP and a transform-based audio coder TCX. The bitstream encoder 20 comprises a signal encoder 21 for coding the signal IS and a bit stream producer 22 for generating the bitstream BS needed for producing the decoded signal DS at the decoder 1. In parallel, the input signal IS is analyzed by the module called signal analyzer 23, which comprises a noise estimation device 24. In the preferred embodiment the noise estimation device 24 is the same as the one used in G.718. It consists of a spectral analysis device 25 followed by a noise estimation producing device 26. The spectrum SI of the original signal IS and the spectrum Nl of the estimated noise are input in the noise reduction module 27. The noise reduction module 27 is attenuates the background noise level in the enhanced frequency domain signal FS. The amount of reduction is given by the target attenuation level signal TAS. The enhanced time-domain signal (noise reduced audio signal) is TS is generated after spectral synthesis done by the spectral synthesis device 28. The signal TS is used for deducing some features, like the pitch stability which is then exploited by the signal activity detector 29 for discriminating between active and inactive frames. The result of the classification can be further used by the encoder module 18. In a preferred embodiment, a specific coding mode is used to handle inactive frames. This way, the decoder 1 can deduce the signal activity flag (VAD flag) from the bit-stream without requiring a dedicated bit.

Fig. 4 illustrates a first embodiment of an encoder 18 according to the invention. The encoder 18 depicted in figure 4 is based on the encoder 18 shown in figure 3.

The encoder 18 shown in figure 4 is configured for producing an audio bitstream BS, wherein the encoder 18 comprises:

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 a bitstream encoder 20 configured to produce an encoded audio signal ES corresponding to an audio input signal IS and to derive the bitstream BS from the encoded audio signal ES;

an signal analyzer 19 having a signal-to-noise ratio estimator 33 configured to determine the signal-to-noise ratio of the audio input signal IS based on an energy of a wanted signal WS of the audio input signal IS determined by a wanted signal energy estimator 31 and based on an energy of a noise N of the audio input signal IS determined by noise energy estimator 32;

a noise reduction device 27, 28 configured to produce a noise reduced audio signal TS; and a switch device 35 configured to feed, depending on the determined signalto-noise ratio of the audio input signal IS, either the audio input signal IS or the noise reduced audio signal TS to the bitstream encoder 20 for the purpose of encoding the respective signal IS, TS, wherein the bitstream encoder 20 is configured to transmit a side information within in the bitstream, which indicates whether the audio input signal IS or the noise reduced audio signal TS is encoded.

The bitstream encoder 20 may be a device or a computer program capable of encoding an audio signal, which is a digital data signal containing audio information. The encoding process results in a digital bitstream, which may be transmitted over a digital data link to a decoder at a remote location.

The encoder part of one embodiment of the invention is given in figure 4. The main difference compared to figure 3 is coming from the fact that this time it encodes the output of the noise reduction, i.e., the enhanced signal TS. To avoid unnecessary distortions in noiseless situations (clean speech or clean music), noise reduction is applied only in case of noisy speech and is bypassed otherwise. The discrimination between noisy and noiseless signals

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 is achieved by estimating the long-term energy of the wanted signal WS (speech or music) by the wanted signal energy estimator 31 and by estimating the long-term energy of the noise N by the noise energy estimator 32. For this purpose the wanted signal energy estimator 31 receives the spectrum SI signal for the input signal IS as provided by the spectral analysis device 25. Further, the noise energy estimator receives the noise estimation signal Nl for the input signal IS as provided by the noise estimation producing device 26. During active frames, only the long-term speech/music energy estimate WE is updated. During inactive frames, only the noise energy estimate NE is updated. The long-term energy is computed by a first-order auto-regressive filtering of either the input frame energy (during active frames) or using the output of the noise estimation module (during inactive frames). In this way a signal-to-noise ratio signal RS can be computed by the signal-to-noise ratio estimator 33, which contains the ratio of the long-term energy of the speech or music WS over the long-term energy of the noise N. The signal-to-noise ratio signal RS is fed to a noise detector 34 which determines whether the present frame contains a noisy audio signal or a clean audio signal If the signal-to-noise ratio signal RS is below a predetermined threshold, the frame is considered as noisy speech otherwise it is classified as clean speech.

The result of the classification is outputted as a noise flag signal NF, which is used to control the switch 35. Furthermore, the noise takes signal NF is fed to the bitstream encoder 20. The bitstream encoder 20 is configured to produce and to transmit a side information based on the noise flag signal NF within in the bitstream, which indicates whether the audio input signal IS or the noise reduced audio signal TS is encoded. By decoding this flag a decoder may adjust the target noise level automatically without the necessity of classifying the decoded signal DS as being a noisy or as being clean.

Fig. 5 illustrates a second embodiment of an encoder 18 according to the invention. The encoder 18 depicted in figure 5 is based on the encoder a

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 team shown in figure 4. In the following additional features be explained. In figure 4 the signal analyzer 30 comprises a signal activity detector 36 which receives the spectrum signal SI for the input signal IS and the noise estimation signal Nl. The signal activity detector 36 is configured to discriminate between active frames and inactive frames based on these two signals. The signal activity detector produces a signal activity signal SA which on one hand is transmitted to the bitstream encoder 20 for the purpose of adapting the bitstream BS to the signal activity and on the other hand is used to switch a switch 37 which is configured to alternatively fed the wanted signal energy signal WE or the noise energy signal EN two the signal-tonoise ratio estimator 33.

Fig. 6 illustrates an embodiment of a frame format FF of the bitstream BS according to the invention. The frame according to the frame format FF comprises a signal vector SV having a plurality of bits which are located on the positions from 0 to n. At the position n+1 a bit being an activity flag AF indicating whether the frame is in active frame and inactive frame is located. Furthermore, the position n+2 a bit being a noise flag NF indicating whether the frame contains a noisy signals or a team signal is foreseen. At the position n+3 and bit being padding bit PB is arranged.

In a preferred embodiment of the invention the side information indicating whether the present frame is active or inactive consists of at least one dedicated bit in the bitstream.

As a summary it may be said that in one aspect of the invention, the original signal is encoded and at decoder 1 it is decoded before being added to an artificially generated comfort noise CN. The comfort noise generating device 4 requires no or very small amount of side-information. In a first embodiment, the comfort noise generating device 4 requires no side-information and all the processing is done blindly. In the preferred embodiment, the comfort noise generating device 4 needs to recover the VAD information (active and

CA 02895391 2016-11-08 inactive frame classification result) from the bit-stream BS, which can be already present in the bit-stream and used for other purposes. In a third embodiment, the comfort noise generating device 4 requires from the encoder 18 a noisy speech flag discriminating between clean and noisy speech. One can also imagine any kinds of information parametrically coded which can help to drive the comfort noise generating device 4.

In another aspect of the invention, noise reduction is first applied to the original signal IS and an enhanced signal TS is conveyed to the bitstream encoder 20, coded, and transmitted. At the end of the decoding, an artificially-generated comfort noise CN is then added to the decoded (enhanced) signal DS. The target attenuation level used for noise reduction at the encoder is a static value shared with the CNG module at the decoder. Hence, the target attenuation level does not need to be explicitly transmitted.

Although some aspects have been described in the context of an apparatus, it is clear that these aspects also represent a description of the corresponding method, where a block or device corresponds to a method step or a feature of a method step. Analogously, aspects described in the context of a method step also represent a description of a corresponding block or item or feature of a corresponding apparatus. Some or all of the method steps may be executed by (or using) a hardware apparatus, like for example, a microprocessor, a programmable computer or an electronic circuit. In some embodiments, some one or more of the most important method steps may be executed by such an apparatus.

Depending on certain implementation requirements, embodiments of the invention can be implemented in hardware or in software. The implementation can be performed using a non-transitory storage medium such as a digital storage medium, for example a floppy disc, a DVD, a Blu-Ray™, a CD, a ROM, a PROM, and EPROM, an EEPROM or a FLASH memory, having electronically readable control signals stored thereon, which

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 cooperate (or are capable of cooperating) with a programmable computer system such that the respective method is performed. Therefore, the digital storage medium may be computer readable.

Some embodiments according to the invention comprise a data carrier having electronically readable control signals, which are capable of cooperating with a programmable computer system, such that one of the methods described herein is performed.

Generally, embodiments of the present invention can be implemented as a computer program product with a program code, the program code being operative for performing one of the methods when the computer program product runs on a computer. The program code may, for example, be stored on a machine readable carrier.

Other embodiments comprise the computer program for performing one of the methods described herein, stored on a machine readable carrier.

In other words, an embodiment of the inventive method is, therefore, a computer program having a program code for performing one of the methods described herein, when the computer program runs on a computer.

A further embodiment of the inventive method is, therefore, a data carrier (or a digital storage medium, or a computer-readable medium) comprising, recorded thereon, the computer program for performing one of the methods described herein. The data carrier, the digital storage medium or the recorded medium are typically tangible and/or non-transitionary.

A further embodiment of the invention method is, therefore, a data stream or a sequence of signals representing the computer program for performing one of the methods described herein. The data stream or the sequence of signals

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 may, for example, be configured to be transferred via a data communication connection, for example, via the internet.

A further embodiment comprises a processing means, for example, a computer or a programmable logic device, configured to, or adapted to, perform one of the methods described herein.

A further embodiment comprises a computer having installed thereon the computer program for performing one of the methods described herein.

A further embodiment according to the invention comprises an apparatus or a system configured to transfer (for example, electronically or optically) a computer program for performing one of the methods described herein to a receiver. The receiver may, for example, be a computer, a mobile device, a memory device or the like. The apparatus or system may, for example, comprise a file server for transferring the computer program to the receiver.

In some embodiments, a programmable logic device (for example, a field programmable gate array) may be used to perform some or all of the functionalities of the methods described herein. In some embodiments, a field programmable gate array may cooperate with a microprocessor in order to perform one of the methods described herein. Generally, the methods are preferably performed by any hardware apparatus.

The above described embodiments are merely illustrative for the principles of the present invention. It is understood that modifications and variations of the arrangements and the details described herein will be apparent to others skilled in the art. It is the intent, therefore, to be limited only by the scope of the impending patent claims and not by the specific details presented by way of description and explanation of the embodiments herein.

Reference signs:

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 decoder bitstream decoder noise estimation device comfort noise generating device combiner spectral analysis device noise estimation producing device noise generator spectral synthesizer switch device control device noise detector side information receiver wanted signal energy estimator noise energy estimator signal-to-noise ratio estimator side information receiver

17a switch encoder signal analyzer bitstream encoder signal encoder bitstream producer signal analyzer noise estimation device spectral analysis device noise estimation producing device noise reduction module spectral synthesis device signal activity detector signal analyzer

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527 wanted signal energy estimator noise energy estimator signal-to-noise ratio estimator noise detector switch signal activity detector switch

BS encoded audio bitstream

DS decoded audio signal

NE noise estimation signal

N noise

CN comfort noise signal

OS audio output signal

AS analysis signal

FD frequency domain comfort noise signal

ND noise detection signal

TNL target comfort noise level

IS input signal

ES encoded signal

OW output signal of the wanted signal energy estimator

ON output signal of the noise energy estimator

SI spectrum signal for the input signal

Nl noise estimation signal for the input signal

TAS target attenuation signal

FS enhanced frequency domain signal

TS noise reduced audio signal

AD activity detector signal

WE wanted signal energy signal

EN noise energy signal

RS signal-to-noise ratio signal

NF noise flag

CA 02895391 2015-06-17

WO 2014/096280

PCT/EP2013/077527

SA

FF signal activity signal frame format

SV

AF

NF

PB signal vector activity flag noise flag signal padding bit

References:

[1] [2]

Recommendation ITU-T G.718: “Frame error robust narrow-band and wideband embedded variable bit-rate coding of speech and audio from 8-32 kbit/s

3GPP TS 26.190 “Adaptive Multi-Rate wideband speech transcoding,” 3GPP Technical Specification.