ES2588483T3 - Audio decoder comprising a background noise estimator - Google Patents

Abstract

An audio decoder for decoding a data stream to reconstruct an audio signal therefrom, in which the data stream comprises at least one active phase (86) followed by an inactive phase 5 (88), where the Audio decoder comprises: a background noise estimator (90) configured to continuously update a parametric background noise estimate from the data stream (104) during the active phase (86); a decoder (92) configured to reconstruct the audio signal from the data stream during the active phase; a parametric random generator (94); a background noise generator (96) configured to synthesize the audio signal during the inactive phase (88) by controlling the parametric random generator (94) during the inactive phase (88) depending on the estimation of parametric background noise; where the decoder (92) is configured to, when reconstructing the audio signal from the data stream, form an excitation signal encoded by transformed in the data stream, according to linear prediction coefficients also encoded in the sequence of data; and where the background noise estimator (90) is configured to update the parametric background noise estimate using the excitation signal.

Description

5

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

60

65

DESCRIPTION

Audio decoder comprising a background noise estimator

The present invention concerns an audio codec that supports noise synthesis during inactive phases.

It is known in the art the possibility of reducing the transmission bandwidth by taking advantage of inactive voice nodes or other noise sources. Such schemes generally use some form of detection to distinguish between inactive (silent) and active (non-silent) phases. During inactive phases, a lower bit rate is achieved by stopping the transmission of the ordinary data sequence by encoding precisely the recorded signal, and instead sending only silence insertion description (SID) updates. SID updates can be transmitted at regular intervals or when changes in background noise characteristics are detected. The SID frames on the decoding side can then be used to generate a background noise with characteristics similar to the background noise during the active phases so that the transmission brake of the ordinary data sequence encoding the recorded signal does not leads to an unpleasant transition from the active phase to the inactive phase on the receiver side.

However, there is still a need to reduce the transmission rate further. A growing number of consumers of the amount of bits transmitted (bitrate), such as a growing number of mobile phones, and a growing number of applications that make more or less intensive use in terms of the amount of bits transmitted, require a permanent reduction of the number of transmitted bits consumed.

On the other hand, the synthesized noise should closely emulate the actual noise so that the synthesis is transparent to users.

Application WO 02/101722 A1 describes, among other things, a system for generating colored comfort noise. In particular, comfort noise is generated by spectrally coloring or shaping a white noise signal. Filter information is estimated based on pauses or intervals during a vocal signal.

Accordingly, it is an objective of the present invention to provide an audio codec scheme that supports noise generation during inactive phases, which allows reducing the amount of transmitted transmission bits while maintaining the attainable noise generation quality.

This objective is achieved by the subject of independent claims.

The basic idea of the present invention is that valuable amount of transmitted bits can be saved by maintaining the quality of noise generation within the inactive phases, if an estimation of parametric background noise is continuously updated during an active phase so that the generation Noise can be initiated immediately after the input of the inactive phase that follows the active phase. For example, continuous updating can be performed on the decoding side, and there is no need to provide a preliminary decoding side, with a coded representation of the background noise during a warm-up phase immediately following the detection of the inactive phase whose provision consumes valuable amount of transmitted bits, since the decoding side has the estimation of parametric background noise continuously updated during the active phase and therefore, is prepared at all times to immediately enter the active phase with a generation of appropriate noise. Similarly, a heating phase can be avoided as well if the estimation of parametric background noise is made on the coding side. Instead of continuing in a preliminary way with providing the decoding side with a conventionally coded representation of the background noise by detecting the input of the inactive phase to learn the background noise and consequently informing the decoding side after the learning phase, the The encoder can provide the decoder with the necessary estimation of parametric background noise immediately upon detecting the input of the inactive phase by resorting to the estimation of parametric ambient noise continuously updated during the active phase stopped, thereby avoiding the additional preliminary prosecution that consumes amount of bits transmitted from the very laborious background noise coding.

According to specific embodiments of the present invention, a more realistic noise generation is achieved with a moderate factor that lowers the performance in terms of, for example, number of transmitted bits and computational complexity. In particular, according to these embodiments, the spectral domain is used to parameterize the background noise thereby producing a background noise synthesis that is more realistic and therefore leads to a more transparent switching from active to inactive phase. Likewise, it has been found that parameterizing the background noise in the spectral domain allows separation of noise from the useful signal and therefore, parameterizing the background noise in the spectral domain has an advantage when combined with the continuous update mentioned before the estimation of parametric background noise during active phases since a better separation between noise and useful signal can be achieved in the spectral domain so that no additional transition from one domain to another is necessary when both advantageous aspects of the present application are combined.

5

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

60

65

Other advantageous details of embodiments of the present invention are the subject of the dependent claims.

Preferred embodiments of the present application are described below with respect to the Figures among which:

Figure 1 shows a block diagram of an audio encoder, in accordance with a comparison embodiment;

Figure 2 shows a possible implementation of the coding engine 14;

Figure 3 shows a block diagram of an audio decoder according to an embodiment;

Figure 4 shows a possible implementation of the decoding engine of Figure 3 according to one embodiment;

Figure 5 shows a block diagram of an audio encoder according to another more detailed comparison embodiment;

Figure 6 shows a block diagram of a decoder that could be used in connection with the encoder of Figure 5;

Figure 7 shows a block diagram of an audio decoder according to another more detailed embodiment;

Figure 8 shows a block diagram of a spectral bandwidth extension part of an audio encoder;

Figure 9 shows an implementation of the CNG spectral bandwidth extension encoder of Figure

8;

Figure 10 shows a block diagram of an audio decoder according to an exemplary embodiment using spectral bandwidth extension;

Figure 11 shows a block diagram of a possible more detailed description of an embodiment for an audio decoder using spectral bandwidth replication;

Figure 12 shows a block diagram of an audio encoder using spectral bandwidth extension; Y

Figure 13 shows a block diagram of another embodiment of an audio decoder.

Figure 1 illustrates an audio encoder 100 according to a comparison embodiment. The audio encoder of Figure 1 comprises a background noise estimator 12, an encoding engine 14, a detector 16, an audio signal input 18 and a data stream output 20. Supplier 12, the motor encoding 14 and the detector 16 have an input connected to the audio signal input 18, respectively. The outputs of the estimator 12 and the coding motor 14 are respectively connected to the data sequence output 20 through a switch 22. The switch 22, the estimator 12 and the coding motor 14 have a control input connected to a detector output 16, respectively.

The background noise estimator 12 is configured to continuously update a parametric background noise estimate during an active phase 24 based on an input audio signal that enters the audio encoder 10 at input 18. Although the Figure 1 suggests that the background noise estimator 12 may establish continuous updating of the parametric background noise estimate based on the audio signal as input at input 18, this is not necessarily the case. Alternatively or additionally, the background noise estimator 12 may obtain a version of the audio signal of the encoding engine 14 as illustrated by the dashed line 26. In that case, the background noise estimator 12, alternatively or additionally , will be connected to the input 18 indirectly via the connection line 26 and the coding motor 14, respectively. In particular, there are different possibilities for the background noise estimator 12 to continually update the background noise estimate and some of those possibilities are described below.

The encoding engine 14 is configured to encode the input audio signal that arrives at the input 18 in a sequence of data during the active phase 24. The active phase will cover any time when useful information is contained within the signal of Audio such as voice or other useful sound from a noise source. On the other hand, sounds with a characteristic almost invariable in time such as a spectrum with temporary invariance such as that caused by rain or traffic at the bottom of a speaker, will be classified as background noise and always

5

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

60

If this background noise is present, the respective time period will be classified as an inactive phase 28. The detector 16 is responsible for detecting the input of an inactive phase 28 following the active phase 24 based on the audio signal of input at input 18. In other words, the detector 16 distinguishes between two phases, namely the active phase and the inactive phase where the detector 16 decides as to which phase is currently present. The detector 16 informs the encoding engine 14 about the currently present phase and as already mentioned, the encoding engine 14 performs the encoding of the input audio signal in the data stream during the active phases 24. The detector 16 controls the switch 22 accordingly so that the sequence of data delivered by the encoding engine 14 is delivered at the output 20. During the inactive phases, the encoding engine 14 may stop encoding the input audio signal. At least, the data stream delivered at output 20 is no longer fed by any data stream possibly delivered by the encoding engine 14. In addition to that, the encoding engine 14 can only perform a minimum processing to support the estimator 12 with some variable status updates. This action will greatly reduce computing power. For example, switch 22 is set such that the output of the estimator 12 is connected to the output 20 instead of the output of the coding motor. In this way valuable amount of transmitted bits is reduced to transmit the series of bits in the time delivered at the output 20.

The background noise estimator 12 is configured to continuously update a parametric background noise estimate during the active phase 24 based on the input audio signal 18 as already mentioned above, and because of this, the estimator 12 can insert into the data sequence 30 delivered at output 20, the estimation of parametric background noise as continuously updated during active phase 24, immediately following the transition from active phase 24 to inactive phase 28, that is, immediately after the entry into the inactive phase 28. For example, the background noise estimator 12 may insert a silence insertion descriptor box 32 in the data sequence 30 immediately following the completion of the active phase 24 and immediately following the instant of time 34 in which the detector 16 detected the input of the inactive phase 28. In other words, there is no time between detection of the input of the fa 28 is deactivated in the detectors and the insertion of the SID 32 necessary due to the continuous updating of the background noise estimator of the parametric background noise estimation during the active phase 24.

Thus, by synthesizing the description above, the audio encoder 10 of Figure 1 can function as follows. Imagine, for the purpose of illustration, that an active phase 24 is currently present. In this case, the encoding engine 14 currently encodes the input audio signal at input 18, forming data sequence 20. Switch 22 connects the coding motor output 14 to output 20. The coding engine 14 can use parametric coding and / / transform coding to encode the input audio signal 18 forming the data stream. In particular, the encoding engine 14 can encode the input audio signal in frame units with each frame encoding one of the consecutive time intervals - which partially overlap each other between sf - of the input audio signal. The coding engine 14 may additionally have the ability to switch between different coding modes between consecutive frames of the data stream. For example, some frames may be encoded using predictive coding such as CELP coding, and some other frames may be encoded using transform coding such as TCX or AaC coding. Reference is made, for example, to USAC and its coding modes as described in ISO / IEC CD 23003-3 dated September 24, 2010.

The background noise estimator 12 continuously updates the estimation of parametric background noise during the active phase 24. Accordingly, the background noise estimator 12 may be configured to distinguish between a noise component and a useful signal component within the Input audio signal to determine the estimation of parametric background noise merely from the noise component. As described further below, the background noise estimator 12 can perform this update in a spectral domain such as a spectral domain also used for transform coding within the encoding engine 14. However, other alternatives are also available, Just like time domain. If it is in the spectral domain, it may be a lapped transform domain such as an MDCT (Modified Discrete Cosine Transform) domain, or a filter bank domain such as a complex value filter bank domain such as a QMF domain (Quadrature Mirror Filters).

Likewise, the background noise estimator 12 can perform the update based on an excitation or residual signal obtained as an intermediate result within the coding engine 14, for example, during predictive coding and / or by transforming more than the audio signal. as it enters the entry 18 or an encoded with loss of data (lossy) forming the sequence of data. By doing so, a large amount of the useful signal component within the audio signal has already been removed so that the detection of the noise component is easier for the background noise estimator 12.

During the active phase 24, the detector 16 is also continuously operating to detect an input of the inactive phase 28. The detector 16 can be implemented as a voice / sound activity detector (VAD / SAD) or some other mechanism that Decide if there is a useful signal component currently present within the audio signal or not. A basic criterion for the detector 16 to decide whether or not to continue an active phase 24, could

5

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

60

65

be to verify if a power filtered by low pass of the audio signal, falls below a certain threshold, assuming that it enters an inactive phase as soon as the threshold is exceeded.

Regardless of the exact way in which the detector 16 detects the input of the inactive phase 28 following the active phase 24, the detector 16 immediately informs the other entities 12, 14 and 22, of the phase input inactive 28. Due to the continuous updating of the background noise estimator of the parametric background noise estimation during the active phase 24, it is immediately possible to prevent the data stream 30 delivered at the output 20 from being still fed from the motor coding 14. Instead, the background noise estimator 12 inserts, immediately after being informed of the input of the inactive phase 28, inserts the information on the last update of the parametric background noise estimate in the data stream 30 , in the form of SID frame 32. That is, SID frame 32 could immediately follow the last frame of the encoding engine encoding the frame of the audio signal concerning The time interval within which the detector 16 detected the inactive phase input.

Normally, the background noise does not change very often. In most cases, background noise tends to be somewhat invariant over time. Therefore, after the background noise estimator 12 inserted the SID frame 32 immediately after the detector 16 detecting the start of the inactive phase 28, any data stream transmission can be interrupted so that in this phase of interruption 34, the data stream 30 does not consume the amount of transmitted bits or merely a minimum of the quantity of transmitted bits required for some transmission purpose. To maintain a minimum amount of transmitted bits, the background noise estimator 12 can intermittently repeat the output of SID 32.

However, despite the tendency of the background noise not to change over time, it may still happen that the background noise changes. For example, imagine a mobile phone user leaving the car so that background noise changes from engine noise to traffic noise outside the car during the user's call. To track such changes in the background noise, the background noise estimator 12 may be configured to continuously scan the background noise during the inactive phase 28. Whenever the background noise estimator 12 determines that the parametric background noise estimate changes in a magnitude that exceeds some threshold, the background estimator 12 can insert an updated version of the parameterized background noise estimate into the data stream 20 via another SID 38, after which another interrupt phase 40 can follow , for example, another active phase 42 begins as detected by the detector 16, and so on. Naturally, the SID tables that reveal the currently updated parametric background noise estimate can, additionally or alternatively, be interleaved into the inactive phases in an intermediate manner dependent on the changes in the parametric background noise estimate.

Obviously, the data stream 44 delivered by the encoding engine 14 and indicated in Figure 1 through the use of shading, consumes more transmitted transmission bits than the data stream fragments 32 and 38 to be transmitted during the phases inactive 28 and consequently the savings in the amount of bits transmitted are considerable. Also, since the background noise estimator 12 can immediately begin with further feeding the data stream 30, it is not necessary to continue preliminary transmitting the data stream 44 of the encoding engine 14 beyond the detection time point. of inactive phase 34, thereby reducing even more the amount of transmitted bits consumed in total.

As will be explained in more detail below, the encoding engine 14 can be configured to, when encoding the input audio signal, predictively encode the input audio signal in linear prediction coefficients and an excitation signal with transform coding. excitation signal and encode the linear prediction coefficients in the sequence of data 30 and 44, respectively. A possible implementation is shown in Figure 2. According to Figure 2, the encoding motor 14 comprises a transformer 50, a noise domain corrector (shaper) of the frequency domain 52 and a quantizer 54 which are connected in series in the order in which they are mentioned between an input of audio signal 56 and a data sequence output 58 of the encoding engine 14. In addition, the encoding engine 14 of Figure 2 comprises a linear prediction analysis module 60 which is configured to determine linear prediction coefficients a starting from the audio signal 56 by respective window of analysis of portions of the audio signal and applying an autocorrelation to the sold portions, or determining an autocorrelation on the basis of the transformed ones in the transformed domain of the audio signal of input as output via transformer 50 using its power spectrum and applying a reverse DFT to it in order to determine autocorrelation, perform subsequent LPC estimate based on autocorrelation such as using an algorithm of (Wiener—) Levinson — Durbi.

Based on the linear prediction coefficients determined by the linear prediction analysis module 60, the data stream delivered at output 58 is fed with respective information about the LPCs, and the frequency domain noise corrector is controlled by so that spectrally correct the spectrogram of the audio signal in accordance with a transfer function corresponding to the transfer function of a linear prediction analysis filter determined by the linear prediction coefficients delivered by module 60. A quantification of the LPCs to transmit them in the data sequence, in the LSP / LSF domain using interpolation so as to reduce the transmission rate compared to the analysis rate in the

5

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

60

parser 60. In addition, the conversion of LPC to spectral weighting performed in the FDNS may involve applying an ODFT on the LPCs and applying the resulting weighting values to the transformer spectra as a divisor.

The quantifier 54 then quantifies the transformation coefficients of the spectrally formed (flattened) spectrogram. For example, the transformer 50 uses a lapped transform such as an MDCT to transfer the audio signal from the time domain to the spectral domain, thereby obtaining consecutive transforms corresponding to overlaid sold portions of the audio signal, which they are then spectrally formed by the frequency domain noise corrector 52 by weighing these transformed according to the transfer function of the LP analysis filter.

The corrected spectrogram can be interpreted as an excitation signal and as such is illustrated by the dashed arrow 62, the background noise estimator 12 may be configured to update the parametric noise estimate using this excitation signal. Alternatively, as indicated by the dashed arrow 62, the background noise estimator 12 can use the lapped transform representation as output by means of the transformer 50 as the basis for the update directly, that is, without the noise correction of the domain of the frequency using the noise corrector 52.

More details related to possible implementations of the elements shown in Figures 1 to 2 can be established from the subsequent more detailed description and it is noted that all these details are individually transferable to the elements of Figures 1 and 2.

However, before this more detailed description, reference is made to Figure 3, which shows that, additionally or alternatively, the updating of parametric background noise estimation can be done on the decoder side.

The audio decoder 80 of Figure 3 is configured to decode a sequence of data that enters the input 82 of the decoder 80 so as to reconstruct from it an audio signal to be delivered to an output 84 of the decoder 80. The The data sequence comprises at least one active phase 86 followed by an inactive phase 88. Internally, the audio decoder 80 comprises a background noise estimator 90, a decoding engine 92, a parametric random generator 94 and a noise generator in the background 96. The decoding engine 92 is connected between input 82 and output 84 and similarly, the provider's serial connection 90, the background noise generator 96 and the parametric random generator 94 are connected between the input 82 and output 84. Decoder 92 is configured to reconstruct the audio signal from the data stream during the active phase, so that the audio signal 98 as delivered at output 84 it comprises noise and sound useful in an appropriate quality. The background noise estimator 90 is configured to continuously update a parametric background noise estimate from the data stream during the active phase. To this end, the background noise estimator 90 may not be connected to the input 82 directly but via the decoding engine 92 as illustrated by the dashed line 100 so as to obtain from the decoding engine 92 some reconstructed version of the audio signal In principle, the background noise estimator 90 may be configured to operate in a manner very similar to the background noise estimator 12, in addition to the fact that the background noise estimator 90 only has access to the rebuildable version of the signal of audio, that is, that includes the loss caused by quantification on the coding side.

The parametric random generator 94 may comprise one or more random number generators or pseudogenerators, the sequence of values delivered by which it can conform to a statistical distribution that can be parameterically set via the background noise generator 96.

The background noise generator 96 is configured to synthesize the audio signal 98 during the inactive phase 88 by controlling the parametric random generator 94 during the inactive phase 88 depending on the estimation of the parametric background noise as obtained from the background noise estimator 90. Although both entities, 96 and 94, are shown connected in series, the serial connection should not be construed as limiting. Generators 96 and 94 may be interconnected. In fact, generator 94 could be interpreted as being part of generator 96.

Thus, the operation mode of the audio decoder 80 of Figure 3 can be as follows. During an active phase 86, input 82 is continuously provided with a portion of data sequence 102 which is to be processed by the decoding engine 92 during active phase 86. The data sequence 104 entering input 82 then slows down the transmission of the data sequence portion 102 dedicated to the decoding engine 92 at some point in time 106. That is, there is no other frame of the data sequence portion at that time point 106 for decoding by the engine. 92. The signaling of the input of the inactive phase 88 may be the disturbance of the transmission of the data sequence portion 102, or it may be signaled by some information 108 arranged immediately at the start of the inactive phase 88.

5

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

60

65

In any case, the input of the inactive phase 88 occurs very suddenly, but this is not a problem since the background noise estimator 90 has continuously updated the estimation of parametric background noise during the active phase 86 on the basis of the portion of data sequence 102. Because of this, the background noise estimator 90 can provide the background noise generator 96 with the newest version of the parametric background noise estimate as soon as it starts at 106, the inactive phase 88 Consequently, from the moment 106 onwards, the decoding engine 92 stops delivering audio signal reconstruction since the decoding engine 92 is no longer fed with a portion of data sequence 102, but the parametric random generator 94 it is controlled by the background noise generator 96 in accordance with a parametric background noise estimate such that it can deliver an emulation of the background noise at output 84 immediately. Next to the instant of time 106 so as to continuously follow the reconstructed audio signal as delivered by the decoding engine 92 until the instant of time 106. Cross-fade can be used to transit from the last reconstructed frame of the active phase. as delivered by the engine 92 to the background noise as determined by the recently updated version of the parametric background noise estimate.

Since the background noise estimator 90 is configured to continually update the estimation of parametric background noise from the data stream 104 during the active phase 86, it can be configured to distinguish between a noise component and a component of Useful signal within the version of the audio signal as reconstructed from the data sequence 104 in the active phase 86 and to determine the estimation of parametric background noise merely from the noise component instead of the signal component Useful. The way in which the background noise estimator 90 make this distinction / separation corresponds to the manner outlined above with respect to the background noise estimator 12. For example, the internal residual excitation or signal reconstructed from the sequence can be used of data 104 within the decoding engine 92.

Similar to Figure 2, Figure 4 shows a possible implementation for the decoding engine 92. According to Figure 4, the decoding engine 92 comprises an input 110 for receiving the data sequence portion 102 and an output 112 for deliver the reconstructed signal within the active phase 86. Connected in series with each other, the decoding engine 92 comprises a quantizer 114, a frequency domain corrector 116 and a reverse transformer 118, which are connected between the input 110 and exit 112 in the order in which they are mentioned. The data sequence portion 102 that arrives at the input 110 comprises an encoded version of the excitation signal transform, that is, levels of transform coefficients that represent it, which are fed to the input of the quantifier 114, as well as also information on linear prediction coefficients, whose information is fed to the noise corrector of the frequency domain 116. The quantifier 114 decrypts the spectral representation of the excitation signal and sends it to the noise corrector of the frequency domain 116 which , in turn, spectrally forms the spectrogram of the excitation signal (together with the flat quantification noise) according to a transfer function corresponding to a linear prediction synthesis filter, thereby forming quantization noise. In principle, the FDNS 116 of Figure 4 acts similar to the FDNS of Figure 2: LPCs are extracted from the data stream and then subjected to spectral weight LPC conversion, for example, by applying an ODFT on the extracted LPCs , then applying the resulting spectral weights to the incoming quantified spectra from the quantifier 114 as multipliers. Transformer 118 then transfers the reconstruction of audio signal thus obtained from the spectral domain to the time domain and between the reconstructed audio signal thus obtained at output 112. The reverse transformer 118 can use a lapped transform such as an IMDCT. As illustrated by the dashed arrow 120, the spectrogram of the excitation signal can be used by the background noise estimator 90 for the updating of parametric background noise. Alternatively, but not according to the invention, the spectrogram of the audio signal itself can be used as indicated by dashed arrow 122.

With respect to Figure 2 and Figure 4, it should be noted that the implementations of the encoding / decoding engines described with respect thereto should not be construed as restrictive. Alternative examples are also feasible. Also, the encoding / decoding engines may be a multimodal codec type where the parts of Figures 2 and 4 merely assume responsibility for encoding / decoding frames having a specific frame coding mode associated therewith, while other frames are subjected to other parts of the encoding / decoding engines not shown in Figures 2 and 4. Such another mode of frame coding could also be a predictive coding mode that uses linear prediction coding for example, but with domain coding of time instead of using transform coding.

Figure 5 shows a more detailed example of the encoder of Figure 1. In particular, the background noise estimator 12 is shown in more detail in Figure 5.

According to Figure 5, the background noise estimator 12 comprises a transformer 140, an FDNS 142, an LP analysis module 144, a noise estimator 146, a parameter estimator 148, a stationarity meter 150, and a quantifier 152. Some of the components just mentioned may be totally or partially in the encoding motor 14. For example, the transformer 140 and the transformer 50 of Figure 2

5

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

60

65

they can be the same, the analysis modules of LP 60 and 144 can be the same, the FDNSs 52 and 142 can be the same and / or the quantifiers 54 and 152 can be implemented in a module.

Figure 5 also shows a time-series bit packer (bitstream packager) 154 which assumes a passive responsibility for the operation of switch 22 in Figure 1. In particular, the VAD as the detector 16 of the encoder of Figure 5 is called in an exemplary form, it simply decides on what path it should be taken, either the path of the audio coding 14 or the path of the background noise estimator 12. To be more precise, the coding engine 14 and the background noise estimator 12 are both connected in parallel between the input 18 and the packer 154, where inside the background noise estimator 12, the transformer 140, the FDNS 142, the analysis module of LP 144, noise estimator 146, parameter estimator 148, and quantifier 152, are connected in series between input 18 and packer 154 (in the order in which they are mentioned), while LP analysis module 144 is connected between entry 18 and an LPC input of the FDNS module 142 and an additional input of the quantizer 153, respectively, and a stationarity meter 150 is additionally connected between the analysis module of LP 144 and a control input of the quantizer 152. The series packer of bits in time 154 simply perform the packaging if you receive an input from any of the entities connected to your inputs.

In the case of transmitting zero frames, that is, during the interruption phase of the inactive phase, the detector 16 informs the background noise estimator 12, in particular the quantifier 152, to stop processing and not send anything to the packer. bit string at time 154.

According to Figure 5, the detector 16 can operate in the time and / or the transform / spectral domain so as to detect active / inactive phases.

The operation mode of the encoder of Figure 5 is as follows. As will be clarified, the encoder of Figure 5 can improve the quality of comfort noise such as stationary noise in general, such as car noise, talk noise with many participants, some musical instruments, and in particular those that have many harmonics such as raindrops.

In particular, the encoder of Figure 5 is for controlling a random generator on the decoding side so as to excite transform coefficients such that the noise detected from the coding side is emulated. Accordingly, before discussing the functionality of the encoder of Figure 5 additionally, a brief reference is made to Figure 6 which shows a possible decoder which could emulate comfort noise on the decoding side according to the instruction of the encoder of Figure 5 More generally, Figure 6 shows a possible implementation of a decoder that fits the encoder of Figure 1.

In particular, the decoder of Figure 6 comprises a decoding engine 160 as to decode the data stream portion 44 during the active phases and a comfort noise generating part 162 to generate the comfort noise based on the information 32 and 38 provided in the data sequence concerning the inactive phases 28. The comfort noise generating part 162 comprises a parametric random generator 164, an FDNS 166 and a reverse transformer (or synthesizer) 168. Modules 164 and 168 are connected in series with each other so that at the output of the synthesizer 168 the comfort noise results, which fills the space between the reconstructed audio signal as delivered by the decoding engine 160 during the idle phases 28, as discussed with respect to to Figure 1. FDNS 166 processors and reverse transformer 168 may be part of decoding motor 160. In particular, they may be equal to FDNS 116 and 118 of Figure 4, for example.

The mode of operation and functionality of the individual modules of Figures 5 and 6 will be clearer from the following discussion.

In particular, transformer 140 spectrally decomposes the input signal into a spectrogram such as using a lapped transform. A noise estimator 146 is configured to determine noise parameters from it. Concurrently, the voice or sound activity detector 16 evaluates the traits established from the input signal so as to detect if a transition takes place from an active phase to an inactive phase or vice versa. These features used by the detector 16 may be in the form of a transient component / start-up detector, hue measurement, and residual LPC measurement. The transient component / start detector can be used to detect attack (sudden increase in energy) or active voice start in a clean environment or clean signal of noise; Tone measurement can be used to distinguish useful background noise such as a siren, telephone ringing or music; Residual LPC can be used to obtain an indication of the presence of voice in the signal. Based on these features, the detector 16 can give approximately information about whether the current frame can be classified, for example, as voice, silence, music or noise.

While noise estimator 146 may be responsible for distinguishing noise from within the spectrogram of the signal component useful therein as proposed [R. Martin, Noise Power Spectral Density Estimation Based Noise Power Spectral Density Estimation Based on Optimal Smoothing and Statistics

5

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

60

on Optimal Smoothing and Minimum Statistics), 2001], parameter estimator 148 may be responsible for statistically analyzing noise components and determining parameters for each spectral component, for example, based on the noise component.

The noise estimator 146 may be configured, for example, to search for local millimeters in the spectrogram and the parameter estimator 148 may be configured to determine the noise statistics in these portions assuming that the specimens of the spectrogram are primarily a noise attribute of background more than foreground sound.

As an intermediate note, it is emphasized that it is also possible to estimate using the noise estimator without the FDNS 142 since the mmf sf occur in the uncorrected spectrum. Most of the description in Figures 5 will remain the same.

The parameter quantifier 152 in turn can be configured to parameterize the parameters estimated by the parameter estimator 148. For example, the parameters can describe a mean amplitude and a first order moment, or higher order, of a distribution of values spectral within the spectrogram of the input signal while referring to the noise component. To save the amount of bits transmitted, the parameters can be sent to the sequence of data for insertion into the same within SID frames at a lower spectral resolution than the spectral resolution provided by the transformer 140.

The stationarity meter 150 may be configured to establish a stationarity measurement for the noise signal. Parameter estimator 148 in turn can use the stationarity measurement to decide whether or not a parameter update should be initiated by sending another SID table such as table 38 in Figure 1 or to influence the way in which they are estimated The parameters

Module 152 quantifies the parameters calculated by parameter estimator 148 and the analysis of LP 144 and sends the signals to the decoding side. In particular, before quantifying, the spectral components can be grouped into groups. Such grouping can be selected according to psychoacoustic aspects such as shaping the Bark scale or something similar. The detector 16 informs the quantifier 152 if the quantification needs to be done or not. In case the quantification is not necessary, zero tables must be followed.

When the description is transferred to a particular switching scenario from an active phase to an inactive phase, then the modules in Figure 5 act as follows.

During an active phase, the encoding engine 14 continues to encode the audio signal via the packet in series of bits in time. The coding can be done as pictures. Each frame of the data stream can represent a portion / time slot of the audio signal. The audio encoder 14 may be configured to encode all frames that use LPC encoding. Audio encoder 14 may be configured to encode some frames as described with respect to Figure 2, called TCX frame coding mode, for example. Remnants can be encoded using code-excited linear prediction (CELP) linear coding such as ACELP (algebraic code-excited linear prediction) coding mode, for example. That is, the portion 44 of the data stream may comprise a continuous update of LPC coefficients using some LPC transmission rate that may be equal to or greater than the frame rate.

In parallel, the noise estimator 146 inspects the LPC flattened spectra (filtered by LPC analysis) so as to identify the mmmins within the TCX spectrogram represented by the sequence of these spectra. Of course, these millimos can vary with time t, this is kmin (t). In any case, the mmimos can form traces in the output of the spectrogram by means of FDNS 142 and therefore, for each consecutive spectrum and at the instant you, the mmimo can be associable with the mmimos in preceding and following spectrum, respectively.

The parameter estimator then establishes parameters for estimating background noise thereafter such as, for example, a central tendency (mean, median or similar value) m / or dispersion (standard deviation, variance or similar) d for different components or spectral bands. The derivation may involve statistical analysis of the consecutive spectral coefficients of the spectrogram spectra in the mm, thereby producing m and d for each mm in kmin. Interpolation along the spectral dimension between the aforementioned mm of spectrum can be performed in order to obtain m and d for other predetermined spectral bands or components. The spectral resolution for the derivation and / or interpolation of the central tendency (average average) and the derivation of the dispersion (standard deviation, variance or similar) may differ.

The parameters just mentioned are continuously updated for each spectrum output by FDNS 142, for example.

5

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

60

65

As soon as the detector 16 detects the entry of an inactive phase, the detector 16 can inform the motor 14 accordingly so that no more active frames are sent to the packer 154. However, the quantifier 152 delivers the statistical noise parameters just mentioned in a first SID box within the inactive phase, instead. The first SID table may or may not include an update of the LPCs. If an LPC update is present, it can be conducted within the sequence of data in SID table 32 in the format used in portion 44, that is, during the active phase, such as using quantification in the LSF domain / LSP, or in a different way, such as using spectral weights corresponding to the LPC analysis or to the transfer function of the LPC synthesis filter such as those that would have been applied by the FDNS 142 within the framework of the coding engine 14 when proceeding With an active phase.

During the inactive phase, the noise estimator 146, the parameter estimator 148 and the stationarity meter 150 continue to cooperate in order to keep the decoding side updated on changes in the background noise. In particular, the meter 150 verifies the spectral weighting defined by the LPCs, so as to identify changes and inform the estimator 148 when an SID frame must be sent to the decoder. For example, the meter 150 could activate the estimator accordingly as long as the above-mentioned stationarity measure indicates a degree of fluctuation in the LPCs that exceeds a certain magnitude. Additionally or alternatively, the estimator could be triggered to send the updated parameters regularly. Among these SID 40 update boxes nothing is sent in the data streams, that is, "zero frames".

On the decoder side, during the active phase, the decoding engine 160 assumes responsibility for reconstructing the audio signal. As soon as the inactive phase begins, the random generator of adaptive parameters 164 uses the unquantified random generator parameters sent during the inactive phase within the data stream from the parameter quantizer 150, to generate random spectral components, thereby forming a random spectrogram which is spectrally formed inside the spectral energy processor 166 with the synthesizer 168 then performing a retransformation from the spectral domain to the time domain. For spectral formation within FDNS 166, either the most recent LPC coefficients from the most recent active tables can be used, or the spectral weighting to be applied by FDNS 166 can be derived therefrom by extrapolation, or the SID 32 box itself can drive the information. By this measure, at the beginning of the inactive phase, the FDNS 166 continues to spectrally weigh the incoming spectrum according to a transfer function of an LPC synthesis filter, with the LPS defining the LPC synthesis filter that is derived of the portion of active data 44 or of the SID table 32. However, with the beginning of the inactive phase, the spectrum to be corrected by the FDNS 166 is the randomly generated spectrum instead of a coded transformation as in the case of the mode of TCX frame coding. Likewise, the spectral correction applied in 166 is merely updated discontinuously through the use of SID frames 38. Interpolation or weakening may be performed to gradually switch from one spectral correction definition to the next during the interruption phases 36.

As shown in Figure 6, the adaptive parametric random generator 164 can additionally, optionally, use the quantized transform coefficients as contained within the most recent portions of the last active phase in the data sequence, namely, within the data sequence portion 44 immediately before the input of the inactive phase. For example, the meaning may then be that a smooth transition is made from the spectrogram within the active phase to the random spectrogram within the inactive phase.

With brief reference again to Figures 1 and 3, it follows from Figures 5 and 6 (and subsequently explained Figure 7) that the estimation of parametric background noise as generated within the encoder and / or decoder, may comprise information statistics on a distribution of temporarily consecutive spectral values for different spectral portions such as Bark bands or different spectral components. For each such spectral portion, for example, the statistical information may contain a measure of dispersion. The measure of dispersion, therefore, will be defined in the spectral information in a spectrally resolved manner, namely, sampled in / for the spectral portions. The spectral resolution, that is, the number of measures for dispersion and central tendency spreads along the spectral axis, may differ between, for example, measure of dispersion and the optionally present mean or measure of central tendency. The statistical information is contained within the SID tables. It can refer to a corrected spectrum such as the filtered spectrum of LPC analysis (that is, flattened LPC) such as a corrected MDCT spectrum which allows synthesis by synthesizing a random spectrum according to the statistical spectrum and correcting it from according to a transfer function of the LPC synthesis filter. In that case, the spectral correction information may be present within the SID frames, although it may not be used in the first SID frame 32, for example. However, as will be shown later, this statistical information may alternatively refer to an uncorrected spectrum. Also, instead of using a representation of the real value spectrum such as an MDCT, a complex spectrum filter bank spectrum such as QMF spectrum of the audio signal can be used. For example, the QMF spectrum of the audio signal can be used in an uncorrected form and can be described statistically by means of statistical information in which case there is no spectral correction other than that contained within the statistical information itself.

5

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

60

65

Similar to the relationship between Figure 3 with respect to Figure 1, Figure 7 shows a possible implementation of the decoder of Figure 3. As shown by using the same reference signs as in Figure 5, the The decoder of Figure 7 may comprise a noise estimator 146, a parameter estimator 148 and a stationarity meter 150, which function similarly to the same elements as in Figure 5, with the noise estimator 146 of the Figure 7 which, however, operates on the transmitted and unquantified spectrogram such as 120 or 122 of Figure 4. The parameter estimator 146 then operates as discussed in Figure 5. The same applies with respect to the stationarity meter 148, the which operates on the energy and spectral values or LPC data revealing the temporal development of the spectrum of the LpC analysis filter (or the LPC synthesis filter) as transmitted and quantified vfa / d This is the sequence of data during the active phase.

While elements 146, 148 and 150 act as the background noise estimator 90 of Figure 3, the decoder of Figure 7 also comprises an adaptive parametric random generator 164 and an FDNS 166 as well as an inverse transformer 168 and they are connected in series with each other as in Figure 6, so as to deliver the comfort noise at the output of the synthesizer 168. Modules 164, 166, and 168 act as the background noise generator 96 of Figure 3 with the module 164 assuming responsibility for the functionality of the parametric random generator 94. The adaptive parametric random generator 94 or 164 between spectral components generated randomly from the spectrogram according to the parameters determined by the parameter estimator 148 which in turn is triggered using the measurement of stationarity delivered by the stationarity meter 150. The processor 166 then spectrally corrects the spectrogram asf generated with the inverse transformer 168, then making the transition from the spectral domain to the time domain. Note that when during the inactive phase 88 the decoder is receiving information 108, the background noise estimator 90 is updating the noise estimates followed by some means of interpolation. Otherwise, if zero frames are received, it simply does the processing such as interpolation and / or fading.

Synthesizing Figures 5 to 7, these show that it is technically possible to apply a randomized controlled generator 164 to excite the TCX coefficients, which may be real values such as in MDCT or complex values as in FFT. It may also be advantageous to apply random generator 164 on groups of coefficients usually achieved through filter banks.

The random generator 164 is preferably controlled such that it models the type of noise as faithfully as possible. This could be done if soft noise is known in advance. Some applications may allow it. In many realistic applications where a subject can find different types of noise, an adaptive method is required as shown in Figures 5 to 7. Therefore, a random generator of adaptive parameters 164 is used, which can be briefly defined as g = f (x), where x = (xi, X2, ...) is a set of random generator parameters provided by parameter estimators 146 and 150, respectively.

To make the random parameter generator adaptive, the random generator parameter estimator 146 adequately controls the random generator. Slip compensation can be included to compensate for cases where the data is considered statistically insufficient. This is done to generate a statistically matched noise model based on the past tables and will always update the estimated parameters. An example is when it is assumed that random generator 164 generates a Gaussian noise. In this case, for example, only the average and variance parameters can be needed and a slip can be calculated and applied to those parameters. A more advanced method can handle any type of noise or distribution and the parameters are not necessarily the moments of a distribution.

For non-stationary noise, it is necessary to have a measure of stationarity and then a less adaptive parametric random generator can be used. The stationarity measurement determined by the meter 148 can be derived from the spectral shape of the input signal using various methods such as, for example, the Itakura distance measurement, the Kullback-Leibler distance measurement, etc.

To handle the discontinuous nature of noise updates sent through SID frames as illustrated by 38 in Figure 1, additional information such as energy and spectral noise form is usually sent. This information is useful for generating noise in the decoder having a smooth transition even during a period of discontinuity within the inactive phase. Finally, various smoothing or filtering techniques can be applied to help improve the quality of the comfort noise emulator.

As noted above, Figures 5 and 6 on the one hand, and Figure 7 on the other, belong to different scenarios. In a scenario corresponding to Figures 5 and 6, the estimation of parametric background noise is made in the encoder based on the processed input signal and then the parameters are transmitted to the decoder. Figure 7 corresponds to the other scenario where the decoder can deal with the estimation of parametric background noise based on the past frames received within the active phase. The use of a voice / signal activity detector or noise estimator may be beneficial to help extract noise components even during active voice, for example.

5

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

60

Among the scenarios shown in Figures 5 to 7, the scenario of Figure 7 may be preferred since this scenario results in a smaller amount of transmitted bits being transmitted. The scenario of Figures 5 and 6, however, has the advantage of having more accurate noise estimation available.

All of the above embodiments could be combined with bandwidth extension techniques such as spectral band replication (SBR), although in general bandwidth extension can be used.

To illustrate this, see Figure 8. Figure 8 shows modules by which the encoders of Figures 1 and 5 could be extended to perform parametric coding in relation to a higher frequency portion of the input signal. In particular, according to Figure 8, an input signal of the time domain is spectrally decomposed by an analysis filter bank 200 such as a QMF analysis filter bank as shown in Figure 8. The r The techniques above in Figures 1 and 5 are then only applied to a lower frequency portion of the spectral decomposition generated by the filter bank 200. To carry information on the higher frequency portion next to the decoder, it is also used parametric coding. To this end, a regular spectral band replication encoder 202 is configured to parameterize the highest frequency portion during active phases and to feed information about it in the form of spectral band replication information within the data sequence next to decoding A switch 204 may be provided between the output of the filter bank QMF 200 and the input of the spectral band replication encoder 202 to connect the output of the filter bank 200 with an input of a spectral band replication encoder 206 connected in parallel with encoder 202 so as to assume responsibility for bandwidth extension during inactive phases. That is, switch 204 can be controlled like switch 22 of Figure 1. As will be described in more detail below, the spectral band replication encoder module 206 may be configured to operate similarly to the band replication encoder. spectral 202: both can be configured to parameterize the spectral envelope of the input audio signal within the highest frequency portion, that is, the highest remaining frequency portion not subject to central coding by the encoding motor, by example. However, the spectral band replication encoder module 206 may use a minimum time / frequency resolution at which the spectral envelope is parameterized and conducted within the data stream, while the spectral band replication encoder 202 may be configured to adapt the time / frequency resolution to the input audio signal depending on the occurrences of transients within the audio signal.

Figure 9 shows a possible implementation of the bandwidth extension coding module 206. A time / frequency grid fixator 208, an energy calculator 210 and an energy encoder 212 are connected in series with each other between an input and a output of the coding module 206. The time / frequency grid fixator 208 may be configured to set the time / frequency resolution at which the envelope of the highest frequency portion is determined. For example, a minimum resolution of time / frequency allowed is continuously used by the encoding module 206. The energy calculator 210 can then determine the energy of the highest frequency portion of the spectrogram delivered by the filter bank 2'00 within of the highest frequency portion in time / frequency tiles corresponding to the time / frequency resolution, and the energy encoder 212 may use entropy coding, for example, to insert the energies calculated by the calculator 210 into the sequence of data 40 (see Figure 1) during inactive phases such as within SID boxes, such as SID table 38.

It should be noted that the bandwidth extension information generated in accordance with Figures 8 and 9 can also be used in connection with using a decoder according to any of the embodiments described above, such as those in Figures 3 , 4 and 7.

Thus, Figures 8 and 9 clarify that the generation of comfort noise as explained with respect to Figures 1 to 7, can also be used in connection with spectral band replication. For example, the audio encoders and decoders described above may operate in different modes of operation, among which some comprise spectral band replication and some do not. Super broadband modes of operation, for example, could involve spectral band replication. In any case, the examples of Figures 1 to 7 showing examples for generating comfort noise can be combined with bandwidth extension techniques in the manner described with respect to Figures 8 and 9. The replication coding module Spectral band 206 which is responsible for bandwidth extension during inactive phases may be configured to operate over a very low resolution of time and frequency. Compared to regular spectral band replication processing, encoder 206 can operate in a different frequency resolution which results in an additional frequency band table with very low frequency resolution along with IIR smoothing filters in the decoder for all bands. of comfort noise generation scale factor that interpolates the energy scale factors applied to the envelope adjuster during the inactive phases. As mentioned earlier, the time / frequency grid can be set to correspond to a temporary resolution as low as possible.

5

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

60

65

That is, the bandwidth extension encoding can be performed differently in QMF or spectral domain depending on the silence or active phase that is present. In the active phase, that is, during active frames, regular SBR encoding is carried out by means of encoder 202, resulting in a normal SBR data sequence that accompanies data sequences 44 102, respectively. In the inactive phases or during frames classified as SID frames, only information about the spectral envelope, represented as energy scale factors, can be extracted by applying a time / frequency grid that exhibits a very low frequency resolution, and for example, the lowest possible resolution of time. The resulting scale factors can be efficiently encoded by encoder 212 and written to the data stream. In zero frames or during interruption phases 36, lateral information cannot be written to the data sequence by means of the spectral band replication coding module 206 and therefore no energy calculation can be carried out by means of the calculator 210 .

In accordance with Figure 8, Figure 10 shows a possible extension of the exemplary embodiments of the decoder of Figures 3 and 7 to bandwidth extension coding techniques. To be more precise, Figure 10 shows a possible example of realization of an audio decoder according to the present invention. A decoder per core 92 is connected in parallel with a comfort noise generator, the comfort noise generator being indicated with the reference sign 220 and comprising, for example, the noise generation module 162 or the modules 90, 94 and 96 of Figure 3. A switch 222 is shown as distributing the frames within the data stream 104 and 30 respectively, over the decoder per core 92 or the comfort noise generator 220 depending on the type of frame, namely, if the frame concerns or belongs to an active phase, or concerns or belongs to an inactive phase such as SID frames or zero frames that concern interrupt phases. The outputs of the core decoder 92 and the comfort noise generator 220 are connected to an input of a bandwidth extension decoder 224, the output of which reveals the reconstructed audio signal.

Figure 11 shows a more detailed embodiment of a possible implementation of a bandwidth extension decoder 224.

As shown in Figure 11, the bandwidth extension decoder 224 according to the embodiment of Figure 11 comprises an input 226 for receiving the time domain reconstruction of the low frequency portion of the complete audio signal To be rebuilt. It is the input 226 that connects the bandwidth extension decoder 224 with the outputs of the core decoder 92 and the comfort noise generator 220 so that the time domain input at the input 226 can be rebuilt the portion of low frequency of an audio signal comprising both noise and useful component, or the comfort noise generated to bridge the time between active phases.

As in accordance with the embodiment of Figure 11, the bandwidth extension decoder 224 is constructed to perform a spectral bandwidth replication, the decoder is called SBR decoder in the following. With respect to Figures 8 to 10, however, it is emphasized that these examples are not restricted to spectral bandwidth replication. Rather, an alternative, more general, way of bandwidth extension can also be used, in relation to these examples.

In addition, the decoder SBR 224 of Figure 11 comprises an output of the time domain 228 to deliver the reconstructed audio signal, that is, either in active phases or in inactive phases. Between the input 226 and the output 228 of the decoder SBR 224, they are connected in series in the order in which they are mentioned, a spectral decomposer 230 which can be, as shown in Figure 11, a bank of analysis filters such as a bank of QMF analysis filters, an HF generator 232, an envelope adjuster 234, and a converter of the spectral domain at time 236 which can be, as shown in Figure 11, made as a filter bank of synthesis such as a bank of QMF synthesis filters.

Modules 230 to 236 operate as follows. The spectral decomposer 230 spectrally decomposes the input signal of the time domain so as to obtain a reconstructed low frequency portion. The HF generator 232 generates a high frequency replica portion based on reconstructed low frequency portion and envelope adjuster 234 spectrally forms or corrects the high frequency replica using a representation of a spectral envelope of the high frequency portion as is carried by means of the SBR data sequence and provided by modules not yet discussed but shown in Figure 11 above the envelope adjuster 234. Thus, the envelope adjuster 234 adjusts the envelope of the high frequency replica portion according to the time / frequency grid representation of the transmitted high frequency envelope, and forwards the high frequency portion thus obtained to the spectral to temporal domain converter 236 for a complete frequency spectrum conversion, that is, high frequency portion formed spectrally along with the reconstructed low frequency portion, at the signal of the time domain rec built at exit 228.

As already mentioned above with respect to Figures 8 to 10, the high frequency portion spectral envelope can be carried within the sequence of data in the form of energy scale factors and the decoder SBR 224 comprises an input 238 for receive this information about the spectral envelope of portions of

5

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

60

65

high frequency. As shown in Figure 11, in the case of active phases, that is, active frames present in the data sequence during active phases, the inputs 238 can be connected directly to the spectral envelope input of the envelope adjuster 234 via a respective switch 240. However, the decoder SBR 224 further comprises a scale factor combiner 242, a scale factor data storage 244, an interpolation filter unit 246, such as a IIR filter unit, and a gain adjuster 248. Modules 242, 244, 246 and 248 are connected in series between sf between 238 and the spectral envelope input of envelope adjuster 234 with switch 240 which are connected between gain adjuster 248 and envelope adjuster 234 and an additional switch 250 that is connected between the scale factor data storage 244 and the filter unit 246. The switch 250 is configured to co nectar this storage of scale factor data 244 with the input of the filter unit 246, or a restorer of scale factor data 252. In the case of SID frames during inactive phases - and optionally in cases of active frames for the which a very gross representation of the high frequency portion spectral envelope is acceptable - switches 250 and 240 connect the sequence of modules 242 to 248 between input 238 and envelope adjuster 234. The scale factor combiner 242 adapts the frequency resolution in which the spectral envelope of high frequency portions has been transmitted via the data sequence to the resolution, which expects to receive the envelope adjuster 234 and a scale factor data storage 244 stores the spectral envelope until a next update. The filter unit 246 filters the spectral envelope in temporal and / or spectral dimension and the gain adjuster 248 adapts the gain of the spectral envelope of the high frequency portion. To that end, the gain adjuster can combine the envelope data as obtained by the unit 246 with the actual envelope as can be derived from the output of the QMF filter bank. The scale factor data restorer 252 reproduces the scale factor data representing the spectral envelope within interrupt phases or zero frames as stored by the scale factor storage 244.

Thus, on the decoder side, the following processing can be carried out. In active frames or during active phases, regular spectral band replication processing can be applied. During these periods, the scale factors from the data stream, which are typically available for a larger number of scale factor bands compared to comfort noise generation processing, are converted to the frequency generation resolution of comfort noise by the scale factor combiner 242. The scale factor combiner combines the scale factors for the highest frequency resolution to result in a number of CNG compliant scale factors taking advantage of common frequency band edges of The different frequency band tables. The resulting scale factor values at the output of the scale factor combination unit 242 are stored for reuse in zero frames and subsequent reproduction by the recuperator 252 and are subsequently used to update the filter unit 246 for the cNg mode of operation. . In the SID tables, a modified SBR data sequence reader is applied which extracts the scale factor information from the data sequence. The remaining configuration of the SBR processing is initialized with predefined values, the time / frequency grid is initialized to the same time / frequency resolution used in the encoder. Extracted scale factors are fed to the filter unit 246 where, for example, a smoothing filter IIR interpolates the progression of energy for a low resolution scale factor band over time. In the case of zero frames, there is no payload to read from the series of bits in time and the SBR configuration that includes the time / frequency grid is the same as that used in the SID boxes. In zero frames, the smoothing filters of the filter unit 246 are fed with a scale factor value delivered from the scale factor combination unit 242 that have been stored in the last frame containing information of valid scale factors. In the case that the current frame is classified as an inactive frame or SID frame, the comfort noise is generated in the TCX domain and is transformed back to the time domain. Subsequently, the time domain signal containing the comfort noise is fed to the analysis filter bank QMF 230 of module SBR 224. In the QMF domain, the bandwidth extension of the comfort noise is performed by means of Copy-up transposition within the HF 232 generator and finally the spectral envelope of the artificially created high frequency part is adjusted by applying information of energy scale factors in envelope adjuster 234. These energy scale factors they are obtained by the output of the filter unit 246 and are scaled by the gain adjustment unit 248 before application in the envelope adjuster 234. In this gain adjustment unit 248 a gain value is calculated to adjust the scale factors and applied to compensate for large differences in energy at the edge between the low frequency portion and the high frequency content of l to signal.

The embodiments described above are used in common in Figures 12 and 13. Figure 12 shows an audio encoder according to a comparison embodiment, and Figure 13 shows an embodiment of an audio decoder. The details revealed with respect to these figures apply equally to the previously mentioned elements individually.

The audio encoder of Figure 12 comprises a bank of QMF 200 analysis filters to spectrally decompose an input audio signal. A detector 270 and a noise estimator 262 are connected to an output of the QMF analysis filter bank 200. The noise estimator 262 assumes responsibility for the functionality of the background noise estimator 12. During the active phases, the QMF spectra coming from

5

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

60

65

QMF analysis filter bank are processor by a parallel connection of an estimator of spectral band replication parameters 260 followed by some SBR 264 encoder on the one hand, and a concatenation of a QMF 272 synthesis filter bank followed by an encoder by core 14 on the other side. Both parallel paths are connected to a respective output of the serial bit packer at time 266. In the case of outgoing SID frames, the SID frame encoder 274 receives the data from the noise estimator 262 and delivers the SID frames to the packer Serial bit time 266.

The spectral bandwidth extension data delivered by estimator 260 describes the spectral envelope of the high frequency portion of the spectrogram or spectrum delivered by the QMF 200 analysis filter bank, which are then encoded, such as by coding by Entropy, by the SBR encoder 264. The data sequence multiplexer 266 inserts the spectral bandwidth extension data of active phases into the data sequence delivered at an output 268 of the multiplexer 266.

Detector 270 detects whether an active or inactive phase is currently active. Based on this detection, an active frame, an SID frame or a zero frame, this inactive frame, is currently to be delivered. In other words, module 270 decides whether an active phase or an inactive phase is active and if the inactive phase is active, whether or not an SID box is to be delivered. The decisions are indicated in Figure 12 using I for zero frames, A for active frames, and S for SID frames. The tables corresponding to time intervals of the input signal where the active phase is present are also sent to the concatenation of the QMF 272 filter bank and the core 14 encoder. The QMF 272 filter bank has a lower frequency resolution or operates at a lower number of QMF subbands when compared to the QMF 200 analysis filter bank so as to achieve a reduction in the sampling rate by means of the number of subbands transfer the active frame portions of the input signal to the time domain again. In particular, the QMF 272 synthesis filter bank is applied to the lower frequency portions or lower frequency subbands of the QMF analysis filter bank spectrogram within the active frames. The coder per core 14 thus receives a reduced sampling rate version of the input signal, which thus converts merely a lower frequency portion of the original input signal entered into the QMF 200 analysis filter bank. The remaining portion of the Higher frequency is parameterically coded by modules 260 and 264.

The SID tables (or, to be more precise, the information to be carried by them) are sent to the SID 274 encoder which assumes responsibility for the functionalities of module 152 of Figure 5, for example. The only difference: module 262 operates on the spectrum of the input signal directly - without correction by LPC. Also, as the QMF analysis filtering is used, the operation of module 262 is independent of the frame mode chosen by the coder per core or the spectral bandwidth extension option that is applied or not. The functionalities of module 148 and 150 of Figure 5 can be implemented within module 274.

The multiplexer 266 multiplexes the respective encoded information forming the data sequence at the output 268.

The audio decoder of Figure 13 can operate on a sequence of data such as that delivered by the encoder of Figure 12. That is, a module 280 is configured to receive the sequence of data and to classify the frames within the sequence of data in active boxes, SID boxes and zero boxes, that is, a lack of frame in the data stream, for example. The active frames are sent to a concatenation of a decoder per core 92, a bank of QMF analysis filters 282 and an extension module of spectral bandwidth 284. Optionally, a noise estimator 286 is connected to the output of the bank of QMF analysis filters. The noise estimator 286 may operate in a similar manner, and may assume responsibility for the functionalities of the background noise estimator 90 of Figure 3, for example, with the exception that the noise estimator operates on the non-spectrum corrected instead of over excitation spectra. The concatenation of modules 92, 282 and 284 is connected to an input of the QMF 288 synthesis filter bank. The SID frames are sent to a SID 290 frame decoder which assumes responsibility for the functionality of the background noise generator 96 of Figure 3, for example. A comfort noise generation parameter updater 292 is fed by the information from the decoder 290 and the noise estimator 286 with this updater 292 that governs the random generator 294 which assumes responsibility for the functionality of the parametric random generators of Figure 3. As inactive or zero frames are missing, it does not have to be sent anywhere, but they trigger another random generation cycle of the random generator 294. The output of the random generator 294 is connected to the QMF 288 synthesis filter bank, the output of which reveals the audio signal reconstructed in silence and active phases in the time domain.

Thus, during active phases, the decoder per core 92 reconstructs the low frequency portion of the audio signal including both noise and useful signal components. The QMF analysis filter bank 282 spectrally decomposes the reconstructed signal and the spectral bandwidth extension module 284 uses the spectral bandwidth extension information within the sequence of data and active frames, respectively, to add the portion high frequency The noise estimator 286, if present, performs noise estimation based on a spectrum portion as reconstructed by the decoder per core, that is, the low frequency portion. In the inactive phases, the SID tables carry information that describes

5

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

60

parametrically the background noise estimate derived by the noise estimate 262 on the encoder side. Parameter updater 292 can mainly use the encoder information to update its estimation of parametric background noise, using the information provided by the noise estimator 286 primarily as an information support position in case of transmission loss relative to SID frames. . The QMF 288 synthesis filter bank converts the spectrally decomposed signal as output by means of the spectral band replication module 284 in active phases and the signal spectrum generated from comfort noise in the time domain. Thus, Figures 12 and 13 make it clear that a QMF filter bank framework can be used as a basis for comfort noise generation based on QMF. The QMF framework provides a convenient way to reduce the sampling rate of the input signal to the sampling rate of the coder per core, to increase the sampling rate of the output signal of the coder per core of the decoder per core 92 on the decoder side using the QMF 288 synthesis filter bank. At the same time, the QMF framework can also be used in combination with bandwidth extension to extract and process the high frequency components of the signal which they are postponed by the core encoder and core decoder modules 14 and 92. Consequently, the QMF filter bank can offer a common framework for various signal processing tools. According to Figures 12 and 13, the generation of comfort noise is successfully included within this framework.

In particular, according to Figures 12 and 13, it can be seen that it is possible to generate comfort noise on the decoder side after the QMF analysis, but before the QMF synthesis by applying a random generator 294 to excite the real and imaginary parts of each QMF coefficient of the QMF 288 synthesis filter bank, for example. The amplitude of the random sequences, for example, are computed individually in each QMF band such that the spectrum of the generated comfort noise resembles the spectrum of the actual input background noise signal. This can be achieved in each QMF band using a noise estimator after the QMF analysis on the coding side. These parameters can then be transmitted through the SID frames to update the amplitude of the random sequences applied in each QMF band on the decoder side.

Ideally, note that the noise estimate 262 applied on the encoder side should be able to operate both during inactive periods (that is, only noise) and active (which typically contain voice noises) so that comfort noise parameters can be updated immediately at the end of each active period. In addition, noise estimation could also be used on the decoder side. Since only noise frames are discarded in a coding / decoding system based on DTX, noise estimation on the decoder side can favorably operate on noisy voice contents. The advantage of performing noise estimation on the decoder side, in addition to the encoder side, is that the spectral form of comfort noise can be updated even when the packet transmission from the encoder to the decoder fails for the (the ) first SID table (s) following a period of activity.

The noise estimation should be able to follow, precisely and quickly, variations of the spectral content of the background noise and ideally it should be able to be carried out during both frames, active and inactive, as expressed above. One way to achieve these objectives is to track the millimeters taken in each band by the power spectrum using a sliding sale of finite length, as proposed in [R. Martin, Noise Power Spectral Density Estimation Based on Optimal Smoothing and Minimum Statistics), 2001]. The idea behind this is that the power of a loud voice spectrum often decays to the power of the background noise, for example, between words or syllables. Tracking the minimum of the power spectrum, therefore, provides an estimate of the noise floor in each band, even during voice activity. However, these noise floors are generally underestimated. Likewise, they do not allow to capture rapid fluctuations of the spectral powers, especially sudden increases in energy.

However, the computed noise floor as described above in each band provides very useful lateral information for applying a second stage of noise estimation. In fact, we can expect that the power of a noisy spectrum is close to the estimated noise floor during inactivity, while the spectral power will be well above the noise floor during activity. Noise floors computed separately in each band can then be used as gross activity detectors for each band. Based on this knowledge, the background noise power can be easily estimated as a recursively smoothed version of the power spectrum as follows:

image 1

where ^ uiV2 (mJ

l (mJ

denotes the spectral power density of the input signal in frame m and band ^

refers to the estimation of noise power, and J

It is a forgetting factor (necessarily between 0 and

1) which controls the magnitude of smoothing for each band and each frame separately. Using the noise floor information to reflect the state of activity, it should take a small value during inactive periods (that is, when the power spectrum is close to the noise floor), while a high value must be chosen to apply more

5

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

60

l (mJ

smoothing (ideally keeping J'1 '' ■ "‘ ’" S constant) during active frames. To achieve this, a weak decision can be made by calculating the forgetting factors as follows:

= i _ e Wr'tWf)

where 0flF is the noise floor power and a is a control parameter. A higher value for a results in larger forgetting factors and therefore more overall smoothing.

Thus, a concept of Comfort Noise Generation (CNG) has been described where artificial noise is produced on the decoder side in a transform domain. The above embodiments can be applied in combination with virtually any type of spectrum-temporal analysis tool (that is, a transformation or a bank of filters) that breaks down a time domain signal into multiple spectral bands.

Thus, the above, inter alias, embodiments described a CNG based on TCX where a comfort noise generator employs random pulses to model the residual.

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 stage or a characteristic of a method stage. Similarly, the aspects described in the context of a method step also represent a description of a corresponding block or component or feature of a corresponding apparatus. Some or all of the method steps can be executed by (or using) a hardware device, such as a microprocessor, a programmable computer or an electronic circuit. In some embodiments, some or more of the most important method steps can be performed by such an apparatus.

Depending on certain implementation requirements, embodiments of the invention can be implemented in hardware or software. The implementation can be performed using a digital storage medium, for example, a floppy disk, a DVD, a CD, a Blu-Ray, a read-only memory (ROM), a PROM, an EPROM, an EEPROM or a memory FLASH, having electronically readable control signals stored therein, which cooperate (or are able to cooperate) with a programmable computer system such that the respective method is executed. Therefore, the digital storage medium can be computer readable.

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

Generally, embodiments of the present invention can be implemented as a computer program product with a program code, the operating program code being used to execute one of the methods when the computer program product is run on a computer. The program code can be stored, for example, on a carrier readable by a machine.

Other embodiments comprise the computer program for executing one of the methods described herein, stored in a carrier readable by a machine.

In other words, an embodiment of the inventive method is, therefore, a computer program with a program code to execute one of the methods described herein, when the computer program is run on a computer.

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

A further embodiment of the inventive method is, therefore, a data stream or a sequence of signals representing the computer program for executing one of the methods described herein. The data stream or the signal sequence can be configured, for example, 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 for or adapted to execute one of the methods described herein.

A further embodiment comprises a computer that has one of the methods described here installed in the computer program.

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

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

The embodiments described above are purely illustrative for the principles of the present invention. It is understood that the possible modifications and variations of the provisions and details described herein will be apparent to those skilled in the art. Therefore, it is the intention that the invention is limited only by the scope of the following patent claims and not by the specific details presented by the description and explanation of the embodiments herein.

Claims (10)

5 10 fifteen twenty 25 30 35 40 Four. Five fifty 55 60 65 1. An audio decoder for decoding a data stream to reconstruct an audio signal therefrom, in which the data stream comprises at least one active phase (86) followed by an inactive phase (88), where The audio decoder comprises: a background noise estimator (90) configured to continuously update a parametric background noise estimate from the data stream (104) during the active phase (86); a decoder (92) configured to reconstruct the audio signal from the data stream during the active phase; a parametric random generator (94); a background noise generator (96) configured to synthesize the audio signal during the inactive phase (88) by controlling the parametric random generator (94) during the inactive phase (88) depending on the estimation of the parametric background noise; where the decoder (92) is configured to, when reconstructing the audio signal from the data stream, form an excitation signal encoded by transformed in the data stream, according to linear prediction coefficients also encoded in the sequence of data; Y where the background noise estimator (90) is configured to update the parametric background noise estimate using the excitation signal. 2. An audio decoder according to claim 1, wherein the background noise estimator is configured to, when updating the parametric background noise estimate, identify local millimeters in the excitation signal and perform a statistical analysis of the excitation signal in local mmimos to derive the estimation of parametric background noise. 3. An audio decoder according to any one of the preceding claims, wherein the decoder is configured to, when reconstructing the audio signal, use predictive decoding and / or transform to reconstruct a lower frequency portion of the signal. of audio from the data stream, and to synthesize a higher frequency portion of the audio signal. 4. An audio decoder according to claim 3, wherein the decoder is configured to synthesize the highest frequency portion of the audio signal from a spectral envelope of the highest frequency portion of the audio signal input, parameterically encoded in the data stream, or to synthesize the highest frequency portion of the audio signal by blind bandwidth extension based on the lowest frequency portion. 5. An audio decoder according to claim 4, wherein the decoder is configured to interrupt the predictive decoding and / or by transformed into inactive phases and perform the synthesis of the highest frequency portion of the audio signal forming spectrally a replica of the lowest frequency portion of the audio signal according to the spectral envelope in the active phase, and spectrally forming a replica of the audio signal synthesized according to the spectral envelope in the inactive phase. 6. An audio decoder according to claim 4 or 5, wherein the decoder comprises a reverse filter bank for spectrally composing the input audio signal from a set of subbands of the lowest frequency portion , and a set of subbands of the highest frequency portion. 7. An audio decoder according to any one of claims 1 to 6, wherein the audio decoder is configured to detect an inactive phase input whenever the data sequence is interrupted, and / or whenever the sequence of data indicates the input of the data stream. An audio decoder according to any one of claims 1 to 7, wherein the background noise generator (96) is configured to synthesize the audio signal during the inactive phase (88) by controlling the parametric random generator ( 94) during the inactive phase (88) depending on the parametric background noise as it is continuously updated by the background noise estimator only in the absence of any information on the estimation of the parametric background noise in the data stream immediately after of a transition from an active phase to an inactive phase. 9. An audio decoding method for decoding a data stream to reconstruct an audio signal therefrom, in which the data stream comprises at least one active phase (86) followed by an inactive phase (88) , where the method comprises: continuously update an estimation of parametric background noise from the data stream (104) during the active phase (86); reconstruct the audio signal from the data stream during the active phase; synthesize the audio signal during the inactive phase (88) by controlling a parametric random generator (94) during the inactive phase (88) depending on the estimation of parametric background noise; where the reconstruction of the audio signal from the data sequence comprises forming an excitation signal encoded by transformed in the data sequence, according to linear prediction coefficients also encoded in the data sequence; Y where continuous updating of the parametric background noise estimation is carried out using the excitation signal. 10. A computer program having a program code configured to implement, when executed on a computer, a method according to claim 9.