RU2351024C2 - Method and device for noise reduction - Google Patents

Abstract

FIELD: physics, acoustics.

SUBSTANCE: invention relates to the method of decoding signals, which are coded using a hybrid coder. Essence of the invention lies in the method of reducing interference (S-OUT) in the decoded signal, which is formed from the first decoded signal constituent (S_CELP) and the second decoded signal constituent (S_TDAC), containing the following stages: obtaining the first enveloping characteristics (ENV_CELP) of energy and the second enveloping characteristics (EN_TDAC) of energy of the first decoded signal constituent (S_CELP) and the second decoded signal constituent (S_TDAC), forming the characteristic number (R) depending on the comparison of the first and second enveloping characteristics (ENV_CELP, ENV_TDAC) of energy; output of the amplification coefficient (G) depending on the characteristic number (R); in the preferred manner, multiplication of second decoded signal constituent with the amplification coefficient.

EFFECT: providing a reduction of interference noises anticipating an echo and the following echo for signals, coding using a hybrid coder without requiring additional information.

13 cl, 4 dwg

Description

The invention relates to a method for decoding a signal that is encoded by a hybrid encoder. The invention also relates to a decoding device made in an appropriate manner.

Various methods are known for encoding audio signals, which are particularly effective. So, for example, for qualitatively good coding of speech signals that are of good quality, and at the same time at low bit rates, the encoded data stream turned out to be favorable, in particular, the so-called CELP technologies (linear prediction with code excitation). The CELP method works in the time domain and is based on an excitation model for a variable filter. In this case, the speech signal is represented by both filter parameters and parameters that describe the excitation signal.

In most cases, a corresponding decoder is also considered in connection with the encoder, which can again decrypt or decode the encoded data. Corresponding communication devices contain a so-called codec in order to be able to transmit and receive data, which is required for communication.

For encoding music or speech signals, which must be of very high quality, in particular at high bit rates of the encoded data stream, so-called perceptual codecs (codec = encoder / decoder) are created. These perceptual codecs are based on the reduction of information in the frequency range and use masking effects characteristic of the human hearing organs, that is, for example, certain frequencies or changes that cannot be perceived by the person are also not represented. Thus, the complexity of the encoder or codec is reduced. Since these encoders most often work with the conversion of a temporary signal into a frequency range, and the conversion is carried out, for example, by means of a modified discrete cosine transform (MDCT), they are also often called transform encoders or codecs. This definition is applied in the following description.

Recently, so-called scalable codecs are being used more and more. Scalable codecs are those codecs that primarily provide excellent audio quality at a relatively high bit rate of the encoded data stream. Thus, relatively long packets are formed, which should be transmitted periodically.

A packet is a plurality of data that is included in one time slot and jointly transmitted in this packet. In packets, important data is often transmitted first, while less important data is transmitted after them. In the case of such long packets, however, it is possible to reduce these packets due to the fact that part of the data is deleted, in particular, the part of the packet transmitted by the time of the last is cut off. Naturally, this leads to some deterioration in quality.

Due to the above-mentioned properties, scalable codecs provide the opportunity to work with CELP codecs at low bit rates, and at converting codecs at higher bit rates. This has led to the development of hybrid CELP / transform codecs that generate the main signal with good quality through the CELP method and, in addition, an additional signal through the transform codec method, with which the main signal is improved. This results in the desired high quality.

The disadvantage of using these converting codecs is that the so-called “pre-emptive echo” effect is manifested. This is a noise interference that is evenly distributed over the entire length of the block of the transforming encoder. A block is understood as a collection of data that is coded together. For conversion codecs, the typical block length is 40 ms. The noise noise of the pre-emptive echo effect occurs due to the quantization error of the transmitted spectral components. With a uniform signal level, the level of this interference noise is everywhere below the level of the useful signal. However, if there is a useful signal with a zero level followed by an unexpected high level, this interference noise is clearly audible before a high level. A well-known example is given in the literature of such a change in the characteristics of a signal when a castanet is clicked.

Various methods have already been applied to reduce this effect. But they all work with the transfer of additional information, which, in turn, complicates the encoder circuitry or forces the encoder to work with temporarily increased bit rates.

Based on this prior art, it is an object of the present invention to provide simple means to reduce interference noise for signals encoded by a hybrid codec, which does not require additional information.

This problem is solved by sets of features of the independent claims. Preferred embodiments are provided in the dependent claims.

To reduce interference noise in the decoded signal, which is formed from the first signal created, for example, by the CELP decoder, and the second signal, created, for example, by the conversion decoder, the following steps are performed:

From both decoded signal components, the corresponding envelope energy characteristic is determined accordingly. The envelope characteristic of energy is understood, in particular, as the characteristic of the dependence of the signal energy on time.

From a comparison of both envelopes of the energy characteristics, a certain characteristic number is formed, for example, a ratio.

This characteristic number is used to output the gain.

This method is particularly advantageous when energy, for example, in the encoding method resulting in the first decoded signal component, is recognized in a more reliable manner. Then, it is by means of a characteristic number or gain that a deviation can be detected.

In particular, the second decoded signal component may be multiplied by a gain. Thus, the aforementioned deviation can be corrected.

All signals can be divided into time intervals, and in particular, time intervals that are used for the first decoded signal component may be shorter than time intervals for the second signal component.

Due to this, due to the higher temporal resolution, energy deviations in the second signal component can be better corrected.

The first signal component can be generated by a CELP decoder, which decodes the signal encoded according to the CELP algorithm, and the second signal component, by a conversion decoder, which decodes the signal encoded with the conversion. This transform encoded signal may, in particular, also comprise a first signal component decoded according to the CELP algorithm, which, after decoding, was encoded with transform, summed with the transform encoded signal transmitted from the transmitter (i.e. already in the frequency domain) and then decoded in the conversion decoder as a contribution to the second signal component.

Alternatively, the summation of the transmitted signal encoded according to the CELP algorithm and the transmitted signal encoded with the conversion can also be performed in the time domain.

The gain may, in particular, be equal to a characteristic number. Then, when forming a suitable ratio, a corresponding attenuation of the second decoded signal component can be obtained if it presumably contains anticipatory echo noise.

In particular, the first decoder may be a CELP-based decoder, and / or the second decoder may be a transform decoder. This provides a particularly effective noise reduction while at the same time high quality decoded signal.

The change in the received full signal on the side of the decoder can be carried out, in particular, if there are predetermined criteria.

In particular, it is envisaged that the change in the received full signal on the side of the decoder is carried out only if the change in signal level exceeds a certain threshold value. This provides a particularly effective reduction of the pre-emptive echo, since the effect of the pre-emptive echo, as mentioned above, occurs mainly when the level changes, since then the noise of the pre-emptive echo lies above the signal level. On the other hand, such a selective change does not unnecessarily result in a quality improvement by the second encoder.

According to another aspect of the invention, a method is created in which, based on the described method, a decoded signal or its first and second decoded signal components are processed separately in frequency ranges. This has the following advantage. When decoding for many frequency bands, the nominal energy for these frequency bands is known, in particular, from the energy of individual first decoded signal components separated by frequency ranges, for example, CELP signals. By means of the second decoded signal component, an additional signal component can now be provided, which, however, can deviate substantially in energy. It is problematic, first of all, if the energy of the second decoded signal component is significantly too high, for example, due to the effects of pre-emptive echo. The method provides for each individual processed frequency band the energy (or level) of the second signal component is limited depending on the energy of the first signal component. This method is all the more effective the more frequency bands are processed in a separately described manner.

Other advantages of the invention are presented by way of example embodiments.

The drawings show the following.

Figure 1 is a representation of the essential components on the encoder side and on the decoder side to explain an example implementation of the encoding / decoding process.

Figure 2 - schematic representation of a communication device for transmitting an encoded signal between communication devices in a communication network.

Figure 3 is a decoding device or noise suppression device for explaining the reduction of the pre-emptive echo by means of gain adaptation, which is based on the CELP signal.

4 is another embodiment of matching levels or reducing pre-emptive echoes.

Figure 1 shows a schematic diagram of the encoding and decoding process according to one embodiment. On the C side of the encoder, the analog signal S, which is to be transmitted to the receiver, is pre-processed by the pre-processing unit PP or prepared for encoding, for example, by digitization. In addition, the signal is decomposed into time segments or frames in the division block F. The signal thus prepared is fed to the COD coding block. The coding unit COD comprises a hybrid encoder that includes a first encoder representing a CELP encoder COD1 and a second encoder representing a transform encoder COD2. The CELP encoder COD1 includes a plurality of CELP encoders COD1_A, COD1_B, COD1_C, which operate in different frequency ranges. Due to this division into different frequency ranges, particularly accurate coding can be ensured. In addition, this division into different frequency ranges very well supports the principle of a scalable codec, since depending on the desired scaling, only one frequency range, several or all frequency ranges can be transmitted. The CELP encoder COD1 provides the main component S_G to the encoded full signal S_GES. The conversion encoder COD2 provides an additional component S_Z to the encoded full signal S_GES. The encoded full signal S_GES is transmitted by the communication device KS on the encoder side C to the communication device KD on the decoder side D. Here, if necessary, processing (for example, dividing the encoded full signal into S_G and S_Z components) of the data or the received encoded full signal S_GES is carried out in the processing unit PROC, and then the processed data or the processed signal is supplied to the decoding unit DEC for subsequent decoding (see also figure 3 and 4). A noise reduction unit NR, which is presented in more detail in FIG. 3, is connected to the decoding unit.

FIG. 2 shows a first communication device COM1 (for example, represented by components on the C side of the encoder in FIG. 1), which has a transmission and reception unit ANT1 (for example, corresponding to a communication device CC) for transmitting and / or receiving data, as well as computing a CPU1 unit configured to implement components on the C side of the encoder or implement the encoding method shown in FIG. 1 (processing on the C side of the encoding). Data is transmitted via the transmit and receive unit ANT1 via the CN communication network (which, for example, depending on the communication devices used, can be implemented as the Internet, a telephone network, or a mobile radio communication network). Reception is carried out by means of a second communication device COM2 (for example, represented by components on the right side of FIG. 1), which again contains a transmission and reception unit ANT2 (for example, corresponding to an HF communication device), as well as a computing unit CPU2 configured to implement components on the side D of the decoder or implementation of the decoding method shown in FIG. 1 (processing on the D side of the decoder). Examples of possible implementations of COM1 and COM2 communication devices in which this method may find application include IP telephones, voice gateways, or mobile telephones.

The following are links to FIG. 3, which shows a decoding unit DEC and a noise reduction unit NR with essential components for schematically presenting a pre-emptive suppression procedure. The signal S_COD, CELP encoded using the CELP algorithm (corresponding to the signal S_G) is decoded using the full-band CELP decoder DEC_GES, CELP. The decoded signal S_CELP is supplied, on the one hand, to the (first) energy envelope determining unit GE1 to determine the corresponding envelope ENV_CELP, and on the other hand, to the time domain spectral overlap compensation encoder (COD_TDAC). Time Domain Overlay Compensation (TDAC) coding is an example of transform coding.

The encoded signal S_COD, CELP, TDAC together with the transmitter-encoded signal S_COD, TDAC (corresponding to the signal S_Z) formed on the receiver side is supplied to the conversion decoder DEC_TDAC to generate the decoded signal S_TDAC. Also from this decoded signal S_TDAC in the second energy envelope determination unit GE2, the corresponding energy envelope ENV_TDAC is determined. In the block D determining the relationship on time segments, the ratio R of the energy envelopes to each other in the form of a characteristic number is determined. In the condition determination unit BFE, it is established whether the set minimum distance from 1 is related (1: both energy envelopes are the same), that is, whether the levels of both signals are equal or deviate from each other only by a given percentage.

The result is a gain or attenuation coefficient G, which in the case shown is equal to the ratio R (characteristic number) by which the transform-encoded signal component S_TDAC is multiplied in the multiplication block M to obtain the final signal S_OUT with reduced interference noise. More precisely, it can be assumed, for example, that the relation R is formed by R = ENV_CELP / ENV_TDAC, and it is established that this ratio cannot fall below a predetermined threshold value SW, then when the value falls below the threshold value SW, the signal component encoded with conversion S_TDAC is multiplied by the gain, for example, by G = R, which leads to a weakening of the signal component of S_TDAC. In addition, it is possible, in the case when there is no fall below the threshold value SW, to assign the gain G to the value “1”, so that when the signal component S_TDAC is multiplied, which can then be implemented in any case, the value S_TDAC remains unchanged.

Thus, in case of deviation of the energy of the signal component S_TDAC encoded with the conversion, the deviation being due to the aforementioned effect of the pre-emptive echo, the energy or level of this signal component is brought to the permissible value of the S_CELP signal decoded using the CELP algorithm, so that in the final S-OUT signal interference noise is reduced.

The following are links to FIG. 4, with which another embodiment for reducing the effect of proactive echo is explained.

It is possible that instead of just one CELP codec, there are several codecs separated by frequency ranges (CELP and others). The embodiment shown in FIG. 4 corresponds for the most part to the embodiment shown in FIG. 3, and with respect to it provides further development, namely that the method shown in FIG. 3 is not applied to complete signals (CELP and other) a decoder and a conversion decoder, and this method is applied separately in the frequency ranges. This means that first, the complete signal or individual signal components are divided into frequency ranges, the method of FIG. 3 then being applied to each frequency range to the individual signal components.

The advantage provided by this is explained below. In the decoder for several frequency bands, the nominal energy for these frequency bands is known, namely from the energy of individual CELP signals divided into frequency ranges. The conversion decoder now produces an additional signal (additional signal component), which, however, can deviate substantially in energy. It is problematic, first of all, if the energy of the signal from the converting decoder is too high, for example, due to the effects of pre-emptive echo. This method now introduces, for each separately processed frequency band, the energy restriction of the transforming codec depending on the CELP energy. This method is all the more effective the more frequency bands are separately processed in this way.

This can be illustrated by the following example.

Let the complete signal consist of a tone with a frequency of 2000 Hz, which is completely formed by a component of the CELP codec. Additionally, based on the effects of pre-emptive echo, the converting codec generates another interfering signal with a frequency of 6000 Hz; let the energy of the interfering signal be 10% of the tone energy with a frequency of 2000 Hz. Let the criterion for limiting the component of the transforming codec be that it should be as equal to the CELP component as possible.

Case 1: bandwidth division is not performed (first embodiment). Then the interfering signal at a frequency of 6000 Hz is not suppressed, since it has only 10% of the tone energy with a frequency of 2000 Hz from the CELP codec.

Case 2: Frequency bands A: 0-4000 Hz and B: 4000-8000 Hz are processed separately (second embodiment). In this case, the interfering signal is completely suppressed, since in the upper frequency band the CELP component is equal to zero, and thus, the signal encoded with the conversion is limited to a value equal to zero.

FIG. 4 shows (respectively FIG. 3) a decoding unit DEC and a noise reduction unit NR with components essential for a schematic representation of a level matching procedure or for reducing pre-emptive echo. As for generating encoded signals or transmitting to the receiver, references will again be given in FIGS. 1 or 2.

The signal S_COD, CELP encoded by the CELP algorithm (corresponding to the signal component S_G) is decoded by the full-band CELP decoder, DEC_GES, CELP '. In this case, the full-band CELP decoder includes two decoding units, a first decoding unit, DEC_FB_A, for decoding the S_COD, CELP signal in the first frequency band A and a second decoding unit, DEC_FB_B, for decoding the S_COD, CELP signal in the second frequency band B. The first decoded signal S_CELP_A is supplied to the first energy envelope determination unit GE1_A to determine the corresponding envelope ENV_CELP_A, while the second decoded signal S_CELP_B is supplied to the second energy envelope determination unit GE1_B to determine the corresponding envelope ENV_CELP_B.

The transform-encoded S_COD, TDAC signal generated at the receiver side (corresponding to the S_Z signal) is supplied to the DEC_TDAC transform decoder to generate a decoded S_TDAC signal, which is again fed to the FBS frequency band splitter. This splitter divides the S_TDAC signal into two signals, namely, the S_TDAC_A signal for frequency band A and the S_TDAC_B signal for frequency band B. The division into frequency bands may, as an option, be carried out in the frequency domain, before being converted back to the time domain. There is no delay introduced by the frequency band separator operating in the time domain (high-pass filter, low-pass filter or band-pass filter). Also from these decoded frequency-dependent signals S_TDAC_A and S_TDAC_B in the third energy envelope determination unit GE2_A or in the fourth energy envelope determination unit GE2_B, the corresponding envelope ENV_TDAC_A or ENV_TDAC_B is determined.

In the first block BD_A for determining the gain for frequency band A using the energy envelopes ENV_CELP_A and ENV_TDAC_A, the gain G_A is determined (or the attenuation coefficient, since the gain is negative), and in the second block BD_B for determining the gain for the frequency band B using envelopes ENV_CELP_В and ENV_TDAC_В of energy the gain G_B (or attenuation coefficient) is determined. The determination of the corresponding gain can occur according to the definition in figure 3 (see components D, BFE). In this case, the corresponding ratio (characteristic number) R_A, R_B of energy envelopes for the corresponding frequency band A and B can be formed again, namely: R_A = ENV_CELP_A / ENV_TDAC_A or R_B = ENV_CELP_B / ENV_TDAC_B, and the threshold value SW_A or SW_B is set for the corresponding frequency band , when falling below which the corresponding gain G_A (for example, G_A = R_A) or G_B (for example, G_B = R_B) is formed, which should be applied to the signal S_TDAC_A or S_TDAC_B of the corresponding frequency band (in order to introduce the corresponding attenuation). If there is no fall below the threshold value, the corresponding gain G_A or G_B can be set to “1”, so when you multiply by it, the signal S_TDAC_A or S_TDAC_B, depending on the frequency band, remains unchanged.

In the first multiplication device M_A for frequency band A, the gain G_A is multiplied by the signal S_TDAC_A, and the gain G_B is multiplied by the signal S_TDAC_B. Then, the multiplied (ultimately attenuated) band-dependent signals are combined to obtain the final full-band signal S_OUT 'with suppressed interference noise.

It should be noted that although the decoded signal components S_CELP_A, S_CELP_B, S_TDAC_A, S_TDAC_B were only divided into two frequency bands A and B, the separation into three or more frequency bands was possible and may be preferred.

Claims (13)

1. A method of reducing interference noise (S_OUT) in a decoded signal, which is formed from a first decoded signal component (S_CELP) and a second decoded signal component (S_TDAC), comprising the following steps:
a) determining the first energy envelope of the characteristic (ENV_CELP) and the second energy envelope of the characteristic (ENV_TDAC) of the first decoded signal component (S_CELP) and the second decoded signal component (S_TDAC);
b) generating a characteristic number (R) depending on a comparison of the first and second envelopes of the energy characteristics (ENV_CELP, ENV_TDAC);
c) deriving a gain (G) depending on a characteristic number (R);
d) multiplying the second decoded signal component (S_TDAC) by the gain (G) if the characteristic number (R) falls below a predetermined threshold value. 2. The method according to claim 1, in which the decoded signal components (S_TDAC, S_CELP) are divided into time intervals, and steps (a) to (d) are performed at time intervals. 3. The method according to claim 2, in which the length of the time intervals for the first and second decoded signal components (S_TDAC, S_CELP) is different, and steps (a) to (d) are performed in time intervals for a shorter time interval. 4. The method according to claim 1, in which the first decoded signal component (S_CELP) is formed by decoding the first encoded component (S_COD, CELP) by the first decoder (DEC_GES, CELP), and the second decoded signal component (S_TDAC) is formed by decoding the second encoded component (S_COD, TDAC, S_COD, CELP, TDAC) by the second decoder (DEC_TDAC). 5. The method according to claim 4, in which the second encoded signal component (S_TDAC) contains a first encoded signal component (S_CELP). 6. The method according to claim 1, in which the characteristic number (R) is formed by the ratio of the first and second envelopes of the characteristics (ENV_CELP, ENV_TDAC) of the energy. 7. The method according to claim 1, in which the gain (G) is equal to the characteristic number (R). 8. The method according to claim 1, in which the first decoded signal (S_CELP) is formed by decoding the first signal (S_COD, CELP), which is formed by a plurality of first encoders (COD1_A, COD1JB, COD_C) that operate in different frequency ranges. 9. The method according to claim 4, in which the first decoder (DEC_GES_CELP) is formed by a CELP decoder. 10. The method according to claim 4, in which the second decoder (DEC_TDAC) is formed by a converting decoder. 11. The method according to any one of claims 4, 5, 9, 10, in which the first and second decoders (DEC_TDAC, DEC_CELP) include the same frequency range. 12. A method for reducing interference noise in a decoded signal relating to a frequency band that is formed from a corresponding first decoded signal component (S_CELP-A, S_CELP_B) and a corresponding second decoded signal component (S_TDAC_A, S_TDAC_B), for a corresponding partial frequency band of said frequency band containing the following steps:
a) determining the first envelope of the characteristic (ENV_CELP_A, ENV_CELP_B) of the energy corresponding to the first decoded signal component and the second envelope of the characteristic (ENV_CELP_A, ENV_TDAC_A, ENV_TDAC_B) of the energy corresponding to the second decoded signal component for the corresponding partial frequency band;
b) the formation of the corresponding characteristic number (R_A, R_B) depending on the comparison of the first and second envelopes of the energy characteristics for the corresponding partial frequency band;
c) outputting the corresponding gain factor (G_A, G_B) depending on the corresponding characteristic number for the corresponding partial frequency band; and
d) multiplying the second decoded signal component (S_TDAC_A, S_TDAC_B) by the corresponding gain (G_A, G_B) for the corresponding partial frequency band if the characteristic number (R_A, R_B) falls below a predetermined threshold value. 13. A device for reducing interference noise in a decoded signal for implementing the method according to any one of claims 1 to 12, containing
a) means for determining the first envelope of the energy characteristic (ENV_CELP) and the second energy envelope (ENV_TDAC) of the first decoded signal component (S_CELP) and the second decoded signal component (S_TDAC);
b) means for generating a characteristic number (R) depending on a comparison of the first and second envelopes of the energy characteristics (ENV_CELP, ENV_TDAC);
c) means for deriving a gain (G) depending on a characteristic number (R); and
d) means for multiplying the second decoded signal component (S_TDAC) by a gain (G) if the characteristic number (R) falls below a predetermined threshold value.

Images