KR20070030816A - Audio encoding - Google Patents

Abstract

Hybrid sinusoidal / pulse excitation encoders have recently been proposed to construct scalable audio encoders. The base layer consisting of the data supplied by the sinusoidal encoder retains the main characteristics of the input signal, achieving medium-high quality audio at very low bit rates. Quality can be further improved by adding an excitation signal layer associated with decreasing decimation that further models the finer aspects of the original signal. The present invention provides a method of mixing different excitation signal layers such that the overall concept of scale capability is achieved without compromising the quality of the encoded signal. Mixing is controlled through quality parameters that add importance to the previous layer when constructing a new high layer.

Description

Audio Encoding {AUDIO ENCODING}

The present invention relates to the encoding and decoding of wideband signals, in particular audio signals. The present invention relates to both encoders and decoders, and to audio streams encoded according to the present invention and to data storage media on which such audio streams are stored.

When transmitting a wideband signal, for example an audio signal such as voice, compression or encoding techniques are used to reduce the bit rate of the signal. Reducing the bit rate is equivalent to reducing the bandwidth required for transmission.

1 shows a schematic diagram of a known parameter encoder, in particular a sinusoidal encoder, used in the present invention and described in WO 01/69593. In such an encoder, the input audio signal x (t) is generally divided into several (possibly overlapped) time segments or frames, each having a duration of 20 ms. Each segment is decomposed into transient, sinusoidal and noise components, and the parameters representing these signal components are generated with C _T , C _S and C _N , respectively. Although components are not relevant to the present invention, other components of the input audio signal, such as harmonic complex numbers, can be derived.

The first stage of the encoder comprises a transient encoder 11 comprising a transient detector (TD) 110, a transient analyzer (TA) 111, and a transient synthesizer (TS) 112. Detector 110 estimates the presence of transient signal components and positions. This information is supplied to the transient analyzer 111. Once the location of the transient signal component is determined, the transient analyzer 111 tries to extract the transient signal component or the most significant portion of this component. The transient analyzer preferably determines the content under the shape function by matching the shape function to the signal segment starting at the estimated starting position and using, for example, (small) number of sinusoidal components. This information is included in the transient code C _T.

The transient code C _T is supplied to the transient synthesizer 112. The synthesized transient signal component is subtracted from the input signal x (t) in subtractor 16, resulting in signal x _A. Gain control mechanism GC 12 generates x _B from x _A. Used. The signal x _B analyzed by the sinusoidal analyzer (SA) 130 is supplied to the sinusoidal encoder 13, which determines the sinusoidal component, that is, the crystal component. The final result of the sinusoidal encoding is a sinusoidal code C _S , and a detailed example illustrating the conventional generation of an exemplary sinusoidal code C _S is provided in International Patent Publication WO 00 / 79519A1.

From the sinusoidal code C _S generated by the sinusoidal encoder, the sinusoidal signal component is reconstructed by the sinusoidal synthesizer (SS) 131. This signal is subtracted from subtractor 17 from input x _B of sinusoidal encoder 13, resulting in a residual signal x _C without (large) transient signal components and (major) crystal sinusoidal components.

The remaining signal x _C is considered to mainly contain noise, and noise analyzer 14 generates a noise code C _N representing this noise as described in WO 01 / 89086A1.

2A and 2B generally show the form of an encoder NA and a corresponding decoder ND suitable for use as the noise analyzer 14 of FIG. 1. The first audio signal r ₁ , corresponding to the residual signal x _C of FIG. 1, comprises a first linear prediction (SE) stage which in particular flattens the signal and produces a prediction coefficient Ps of a given order. Is attracted to the encoder. More specifically, the Laguerre filter, Leuven, Belgium, November 15, 2002, a bulletin of the IEEE Benelux Workshop (MPCA-2002) on Model-Based Processing and Coding of First Audio, EGP Schuijers, AWJ Oomen , AC den Brinker and AJGerrits, "Advances in parametric coding for high-quality audio", pp.73-79. The residual signal r ₂ is input to a time envelope estimator TE which produces a set of parameters Pt and possibly a temporarily flattened residual signal r ₃ . The parameter Pt may be a gain set representing a temporal envelope. Alternatively, the parameter may be a parameter derived from linear prediction in the frequency domain, such as a line spectral pair (LSP) or a spectral frequency (LSF), representing a standardized time envelope that is augmented with a gain parameter per frame. .

In the parametric noise decoder ND, a synthesized white noise sequence (in WNG) is generated, resulting in a signal r ₃ ′ having an envelope temporally and spectrally flat. The temporal envelope generator (TEG) adds a temporal envelope based on the received and quantized parameter (Pt '), thereby generating r ₂ ', and the spectral envelope generator (SEG, time varying filter) receives the received and quantized parameters ( Adding a spectral envelope based on P _s ') results in a noise signal r ₁ '.

In the multiplexer 15, an audio stream AS comprising codes C _T , C _S and C _N is constructed.

The sinusoidal encoder 13 and the noise analyzer 14 are used for all or most segments and are the largest part of the bit rate budget.

It is well known that parametric audio coders can provide good quality at relatively low bit rates, for example 20 kbit / s. However, at higher bit rates, the quality is increased because the function of increasing the bit rate is rather low. Therefore, excessive bit rate is necessary to obtain good or transparent quality. Therefore, it is difficult to obtain transparency using parameter encoding, for example, at a bit rate comparable to the bit rate of the waveform coder. This means that it is difficult to construct a parametric audio coder that is excellent in transparency quality without excessive use of bit cost.

The reason for the basic difficulty in parameter encoding reaching transparency is in a limited subject. Parametric encoders are very effective for encoding tone components (sine waves) and noise components (noise encoders). In real audio, however, many signal components are in the gray region: they are not properly modeled by noise or cannot be modeled as (a small number) sinusoids. Therefore, although very advantageous from the point of view of the bit rate for intermediate quality levels, the parametric audio encoder is a bottleneck in achieving a good resolution or reaching a transparent quality level.

At the same time, conventional audio coders (subbands and conversions) are generally good at transparent encoding quality at certain bit rates, such as about 80-130 kbit / s for stereo signals sampled at 44.1 kHz. The combination of transform and parameter coders (so-called hybrid coders) has been proposed, for example, as described in European patent application 02077032.7, filed May 24, 2002. The spectral-time interval of the audio signal that would otherwise be sub-band coded is optionally coded with a noise parameter during an attempt to reduce the bit rate while maintaining audio quality.

Alternatively, the transform or sub-band encoder can be cascaded into a parametric encoder of the type shown in FIG. 1. However, the predicted encoding gain for such an arrangement where the parametric encoder is earlier than the transform or sub-band encoder is minimized. This makes the perceptually most important area of the audio signal captured by the sinusoidal encoder, thereby reducing the possibility for encoding gain in the transform / sub-band encoder.

Audio coders using spectral flattening and residual signal modeling with a small number of bits per sample are described in the AES 17th International Conference Bulletin, A. Harma and U.K. Laine's "Warped low-delay CELP for wide-band audio coding"; Florence, Italy, September 2-5, 1999, High-Quality Audio Coding, pages 207-215; 1990, NJ, IEEE Piscataway, Atlanta GA, Bulletin of the 1990 International Conference on Acoustic Speech Signal Processing (ICASSP90), "High quality audio coding using multi-pulse LPC" by S. Singhal, pages 1101-1104; 1991, NJ, IEEE Piscataway, Atlanta OA, Bulletin of the 1991 International Conference on Acoustic Speech Signal Processing (ICASSP91), X.Lin, "High quality audio coding using analysis-by synthesis technique," pages 3617-3620. In many papers, it can be seen that this encoding method is superior in transparency quality at a bit rate corresponding to 2 bits / sample for mono signals (88.2 kbit / s for 44.1 kHz audio). In this respect, the performance of the sub-band or the transform coder is not exceeded.

The possibility of scaling the bit stream appears to be very attractive in applications that often have to offer the possibility that audio material is accessed at different signal qualities or bit rates, as in the case of music distribution. Bit stream scalability allows the content provider to store only one version of the encoded material. Another interesting application may be the use of the first (base) layer of the encoded signal to provide audio "thumbnails", where subsequent access to the full version of the file does not require retransmission of the base layer material. Do not. RPE-based coders for generating layered bit streams are described in 1997, Transactions for IEEE Voice and Audio Processing, Volume 5 (4), 367-371, by S. Zhang and G. Lockhart, "Embedded RPE based. on multistage coding ".

The inventors have recognized that known techniques for generating hierarchical bit streams are limited in quality due to loss of scale capability. It is an object of the present invention to mitigate the loss of quality when generating a hierarchical bit stream.

The invention thus relates to a method of encoding a digital audio signal, for each time segment of the signal, the following steps:

Encoding the audio signal to provide a code representative of the audio signal;

Subtracting the signal corresponding to the code from the audio signal to obtain a first residual signal;

Spectral planarizing the first residual signal to obtain a spectral planarized residual signal r and a spectral planarization parameter;

Calculating a first excitation signal from the spectral planarized residual signal using a pulse train encoder;

Determining the quality of the first excitation signal as a similarity with the residual signal flattened in a spectral manner;

Subtracting a portion of the first excitation signal from the spectral planarized residual signal to obtain a second residual signal whose portion depends on the determined quality of the first excitation signal;

Calculating a second excitation signal from the second residual signal using a pulse train encoder;

Creating an audio stream,

      A first excitation signal,

      A second excitation signal, and

      A parameter representing the quality of the first excitation signal

Including generating an audio stream.

The invention also relates to an audio encoder adapted to encode each time segment of a digital audio signal using the method, wherein the encoder comprises:

An encoder for encoding a digital audio signal and providing a code representing said signal;

A subtractor for subtracting the signal corresponding to the code from the audio signal to obtain a first residual signal;

A spectral flattening unit for spectral planarizing the first residual signal to obtain a spectral flattened residual signal and a spectral flattening parameter;

A pulse train encoder for calculating a first excitation signal for the spectralized planarized residual signal;

Means for determining the quality of the first excitation signal as a similarity with the residual signal flattened in a spectral manner;

A subtractor for subtracting the portion of the first excitation signal from the spectral planarized residual signal to obtain a second residual signal whose portion depends on the determined quality of the first excitation signal;

A pulse train encoder for calculating a second excitation signal for the second residual signal;

A bit stream generator for generating audio streams,

      A first excitation signal,

      A second excitation signal, and

      A parameter representing the quality of the first excitation signal

It includes a bit stream generator.

Moreover, the present invention relates to a method for decoding a received audio stream, such as an encoded audio stream using the method or encoder, wherein the audio stream is for each of a plurality of segments of an audio signal,

A first excitation signal,

A second excitation signal,

A parameter indicative of the quality of the first excitation signal,

The method,

Combining the first and second excitation signals according to a quality parameter to obtain a combined excitation signal;

Synthesizing the first residual signal from the combined excitation signal using a linear predictive synthesis filter.

Accordingly, the present invention relates to an audio player for receiving and decoding an audio stream, wherein the audio stream is for each of a plurality of segments of the audio signal,

A first excitation signal,

A second excitation signal, and

A parameter indicative of the quality of the first excitation signal,

The audio player,

Means for combining the first and second excitation signals according to a quality parameter to obtain a combined excitation signal;

Means for synthesizing the first residual signal from the combined excitation signal using linear prediction.

Finally, the present invention relates to each of a plurality of segments of an audio signal,

A first excitation signal resulting from pulse train encoding of a spectral planarized residual signal, the residual signal resulting from subtracting the encoded audio signal from the audio signal;

A second excitation signal resulting from the pulse train encoding of the second residual signal, the signal being generated by subtracting a portion of the first excitation signal from the spectral planarized residual signal, wherein the portion is A second excitation signal, dependent on the determined quality;

A parameter indicative of the determined quality of the first excitation signal;

An audio stream is included and a storage medium for storing such an audio stream.

Embodiments of the present invention will now be described by way of example with reference to the accompanying drawings.

1 shows a conventional parametric encoder.

2A and 2B show a conventional parametric noise encoder (NA) and a corresponding noise decoder (ND), respectively.

3 shows an outline of an encoder;

4 shows an overview of a first decoder compatible with the encoder of FIG.

5 shows an overview of a second decoder compatible with the encoder of FIG.

6 is a schematic diagram of an encoder according to the present invention;

7 is a schematic diagram of a decoder according to the present invention;

1-5 and the corresponding description reflect the description in the closed European patent application number 03104472.0 (Applicant's internal control number PHNL031414EPP) filed December 1, 2003.

1 shows a sinusoidal encoder 1 of the type described in WO 01/69593 and used in the preferred embodiment of the invention. The operation of such conventional encoders and corresponding decoders is well described and the description is provided only where relevant to the present invention.

The audio encoder 1 receives a digital audio signal x (t) sampled at a specific sampling frequency. The encoder 1 then separates the sampled input signal into three components: the transient signal component, the sustained decision component, and the sustained stochastic component. The audio encoder 1 comprises a transient encoder 11, a sinusoidal encoder 13, and a noise encoder 14.

The transient encoder 11 includes a transient detector (TD) 110, a transient analyzer (TA) 111, and a transient synthesizer (TS) 112. First, the signal x (t) is input to the transient detector 110. This detector 110 estimates whether there is a transient signal component and its location. This information is supplied to the transient analyzer 111. Once the location of the transient signal component is determined, the transient analyzer 111 attempts to extract the major portion of the transient signal component. The transient analyzer preferably matches the shape function with the signal segment starting at the estimated starting position and determines the content below the shape function by using, for example, a (small) number of sinusoidal components. This information is contained in the transient code (C _T), more detailed information on generating the transient code (C _T) is provided in WO 01/69593.

The transient code C _T is provided to the transient analyzer 112. The synthesized transient signal component is subtracted from input signal x (t) in subtractor 16, resulting in signal x _A. Gain control mechanism (GC) 12 is used to generate x _B from x _A.

The signal x _B is provided to a sinusoidal encoder 13 which is analyzed by a sinusoidal analyzer (SA) 130 that determines the (deterministic) sinusoidal component. Therefore, the presence of the transient analyzer is desirable, but not required, and it will be appreciated that the present invention can be implemented as such an analyzer. Alternatively, as mentioned above, the invention can also be implemented with, for example, a harmonic complex analyzer. Briefly, the sinusoidal encoder encodes the input signal x _B as a track of sinusoidal components linked from one frame segment to the next.

The encoder shown in FIG. 3 is, in 1986, IEEE Trans. Acoust. Speech, Signal Process, 34, P. Kroon, E.F. It is supplemented by a pulse train encoder of the type described in Deprettere and R. J. Sluijter's "Regular Pulse Excitation-A novel approach to effective and efficient multipulse coding of speech". Nevertheless, although the embodiment is described in a Regular Pulse Excitation (RPE) encoder, the Multi-Pulse Excitation (MPE) technology described in US Pat. No. 4,932,061, or April 21-24, 1997, Munich, Germany, ICASSP-97 K. Jarvinen, J. Vainio, P. Kapanen, T. Honkanen, P. Haavisto, R. Salami, C. Laflamme, JP in the newsletter, Vol. 2, pages 771-774. It will be appreciated that the same can be implemented with the ACELP encoder described in Adoul's "GSM enhanced full rate speech codec", each of which includes a first LP based spectral smoothing stage.

In the encoder shown in FIG. 3, the overall bit rate cost determined according to the quality required from the encoder is the bit rate B usable by the parametric encoder and the RPE encoding cost from which the RPE decimation factor D can be derived. Divided.

In FIG. 3, the input audio signal x is first processed in a block (TSA: Transient and Sinusoidal Analysis) corresponding to blocks 11 and 13 of the parametric encoder of FIG. 1. Thus, this block generates associated parameters for transients and noise as described in FIG. Given a bit rate B, the block rate control (BRC) preferably limits the number of sinusoids, preferably such that the overall bit rate for sinusoids and transients is at most equal to B, which is generally set at about 20 kbit / s. To preserve the transition;

The waveform is generated by block TSA and corresponds to blocks 112 and 131 of FIG. 1 using transient and sinusoidal parameters C _T and C _S modified by block BRC. Sinusoidal Synthesiser). This signal is subtracted from the input signal x, resulting in a signal r ₁ corresponding to the residual signal x _C in FIG. In general, the signal r ₁ does not include substantial sinusoidal and transient components.

From signal r ₁ , the spectral envelope is estimated and removed at block SE using a linear prediction filter based on a tapping-delay-line or Lager filter, for example as in FIG. 2A of the prior art. The prediction coefficient Ps of the selected filter is recorded in the bit stream AS for transmission to the decoder as part of the conventional type noise code C _N. The temporal envelope is then removed in block TE, which, for example, produces a line spectral pair (LSP) or line spectral frequency (LSF) coefficient with gain, again as described in Figure 2a of the prior art. In any case, the number resulting from the time-based planarization (Pt) is written in the bit stream (AS) for transmitting to the decoder as part of a prior art type of noise code (C _N). In general, the coefficients P _s and P _t require bit rate guarding of 4-5 kbit / s.

Since the pulse train coder uses the first spectral flattening stage, the RPE encoder is selectively applied to the spectralized flattened signal r ₂ generated by the block SE depending on whether the bit rate cost is assigned to the RPE encoder. Can be. In an alternative embodiment, as indicated by the dotted line, the RPE encoder is applied with a time flattened signal r ₃ in a spectral fashion generated by the block TE.

As is known from the literature mentioned in the background art, the RPE encoder performs a search in an analysis-synthesis manner on the residual signal r ₂ / r ₃ . Given the decimation factor (D), the RPE retrieval procedure uses an offset (value between 0 and D1, D1 depends on D), a ternary with an RPE pulse (eg, values -1, 0 and 1). Pulse amplitude), and gain parameter. This information is stored in the layer L ₀ included in the audio stream AS for transmission by the multiplexer MUX to the decoder when RPE encoding is used.

RPE encoders are operable at different bit rates and thus supply different quality levels. The bit rate is effectively tunable by the decimation factor (D) and the quantization grid, and by accurately setting these parameters, monotonically increasing quality is obtained at increasing bit rates, which is competitive with prior art encoders over a significant range of bit rates. do.

Experiments have shown that RPE encoders sometimes result in loss of brightness in the reconstructed signal when using high decimation factors (eg, D = 8). Adding some low level noise to the RPE sequence alleviates this problem. To determine the level of noise, the gain g is calculated based on the energy / power difference between the signal generated from the coded RPE sequence and the residual signal r ₂ / r ₃ , for example. This gain is also sent to the decoder as part of the layer (L ₀ ) information.

4 illustrates a decoder that is compatible with the encoder of FIG. 3. The demultiplexer (DeM) reads the incoming audio stream (AS '), and the sinusoidal, transient and noise codes {C _S , C _T and C _N (Ps, Pt)}, as in the prior art, are combined with each synthesizer (SiS, TrS and TEG / SEG). As in the prior art, the white noise generator (WNG) supplies an input signal to the time envelope generator (TEG). In an embodiment where the above information is available, the pulse train generator PTG generates a pulse train from layer L ₀ , which is output by the TEG at block Mx to provide an excitation signal r ₂ ′. It is mixed with the noise signal. When the noise code C _N (Ps, Pt) and the layer L ₀ are generated independently of the same residual signal r ₂ , the signal produced by the encoder is accurate to the synthesized excitation signal r ₂ ′. It will be appreciated from the encoder that the gain needs to be modified to provide an energy level. In this embodiment, in the mixer Mx, the signals generated by the blocks TEG and PTG are combined.

)를 생성하기 위해 종래의 과도 및 정현파 합성기에 의해 생성된 합성된 신호에 추가된다.The excitation signal r ₂ ′ is then supplied to a spectral envelope generator SEG that produces a synthesized noise signal r ₁ ′ according to the code Ps. This signal is the output signal (

Is added to the synthesized signal produced by the conventional transient and sinusoidal synthesizers to produce the < RTI ID = 0.0 >

In an alternative embodiment, the parameters generated by the pulse train generator (PTG) are used in combination with the noise code (Pt) to form a time envelope of the signal output by the WNG to generate a time formed noise signal (dotted line). Indicated by).

FIG. 5 shows a second embodiment of a decoder corresponding to the embodiment of FIG. 3 in which the RPE block processes the residual signal r ₃ . Here, the signal generated by the white noise generator (WNG) and processed by the block (We) based on the gain (g) and C _N determined by the encoder, and the pulse generated by the pulse train generator (PTG). The train is added to construct an excitation signal r ₃ ′. Of course, if the layer L ₀ information is not available, the white noise is not affected by the block We and is provided to the temporal envelope generator block TEG as the excitation signal r ₃ ′.

Then, the temporal envelope coefficient Pt is imposed on the excitation signal r ₃ ′ by the block TEG to provide the synthesized signal r ₂ ′ that has been previously processed. As mentioned above, this is advantageous because, in general, pulse train excitation results in some loss in manageable brightness, with a properly weighted additional noise sequence. The weighting may include simple amplitude or spectral formation based on gain factors g and C _N , respectively.

As before, the signal is filtered by a linear predictive synthesis filter, for example in a block that adds a spectral envelope to the signal (SEG). The resulting signal is then added to the synthesized sinusoidal and transient signals as before.

4 or 5, it will be appreciated that if no PTG is used, the decoding scheme is similar to a conventional sinusoidal encoder using only a noise encoder. If PTG is used, an RPE sequence is added that improves the reconstructed signal, ie provides higher audio quality.

In the embodiment of Figure 5, it should be noted that the time envelope is merged in the signal r ₂ ', compared to a standard pulse encoder (RPE or MPE) where fixed gain is used for the complete frame. By using such a temporal envelope, better sound quality can be obtained because the flexibility in the gain profile is higher than the fixed gain per frame.

The hybrid method described above can operate at a variety of bit rates, and, for each bit rate, provides a quality comparable to that of prior art encoders. In the method, the base layer constituted by the data supplied by the parametric (sinusoidal) encoder comprises the main or basic characteristics of the input signal, in which the medium-high quality audio signal is obtained at a very low bit rate.

However, it is desirable for the generated bit stream to be scalable so that the layer can be extracted. It is assumed that there are aligned layers. Thus, it is desirable for the encoder to be able to structurally add information to obtain optimal quality for a given bit rate. Stratification of the bit stream generally means a reduction in quality (so-called loss of scale capability) induced by the requirements of the scalable bit stream. The present invention seeks to alleviate this problem. For this reason, the encoder, decoder and bit stream are adapted.

In the following description, a mixture of different excitation signal layers is performed at the decoder to give a description of the method according to the invention in which the entire concept of scale capability is realized without compromising the quality of the coded signal. Mixing is controlled through one or more parameters determined at the encoder and stored in the bit stream. These parameters reflect the importance of the previous layer when constructing a new higher layer.

6 shows a fully scalable combined parameter (sine wave) and waveform (pulse) encoder in accordance with the present invention. It is noted that the present invention may utilize any other encoder than the encoders described herein. The input signal is received by the parameter encoder, which is a sinusoidal SSC encoder 1 as in FIG. 1 in the illustrated embodiment. The residual signal r _SSC from the SSC encoder is preferably first flattened spectrally using LPC analysis, through which its dynamic range is reduced, which in turn reduces errors in the quantization step. The spectrally flattened residual signal r is then fed to a first waveform encoder which is an RPE-8 stage with decimation factor 8, which is the first from the spectrally flattened residual signal r. Generate an excitation signal (x ₈ ).

The new residual signal r ₈ is generated by combining the residual signal r with an already calculated excitation signal x ₈ . In particular, r ₈ is defined as the difference between the original residual signal r and the weighted excitation signal x ₈ in accordance with r ₈ = r-ρx ₈ .

The parameter ρ is optimized so that the combined layers achieve maximum quality.

Note that setting ρ equal to 0 means creating an independent layer on which information is not reusable. Setting ρ to 1 is a known technique for generating dependent layers in a scalable bit stream, but hinders the achievement of the best quality.

The residual signal r ₈ is here fed to a second waveform encoder which is an RPE-2 stage with decimation factor 2. The RPE-2 stage generates an excitation signal x ₂ .

Ideally, the excitation signal x ₈ calculated at the RPE-8 encoder should be used in the decoder whenever it provides a fairly good approximation of the residual signal r, in which case the RPE-2 discards it and It is better to work directly on r than r ₈ . This evaluates the quality as a similarity or goodness-of-fit of x ₈ with respect to r, ie how well r is modeled by x ₈ , and thus processes it in terms of combining it with x _2. It is suggested that there be a mechanism to do this. In its simplest form, this mechanism consists of only simple gains. In the following, we also describe how the gain ρ, referred to as the mixing coefficient, can be used and calculated to evaluate and process x ₈ .

Finally, the parameter code (SSC code), the first excitation signal x ₈ , the second excitation signal x ₂ , the mixing coefficient ρ and preferably also the spectral smoothing parameter form an encoded audio stream AS. To be combined. In general, the bit stream then consists of three layers: a base parameter layer, a first refinement layer comprising a first excitation signal, and a second layer comprising a second excitation signal, the first layer The reusability of is expressed by the parameter ρ.

Spectral smoothing parameters need not be included in the audio bit stream. If such an audio stream is received at the audio player without the spectral flattening parameter, the decoder at the audio player can determine the spectral flattening parameter by backward adaptation.

7 shows a decoder according to the invention. The encoded audio stream AS is received and its components, i.e., the parameter code (SSC code), the first excitation signal x ₈ , the second excitation signal x ₂ , the blending coefficient ρ and the spectral smoothing parameter are It is identified and processed as follows.

The parameter code is supplied to a parameter decoder (SSC decoder) to decode sinusoidal and transient components. Here, a spectral shaping filter, such as an LPC synthesis filter, receives a first excitation signal x ₈ or a combined excitation signal x ₂ + ρx ₈ . Using the received spectral flattening parameters, LPC synthesis filter estimated SSC residual signal (r _'SSC) The SSC residual signal regeneration, and estimates the (r' _SSC) having an original formed spectrometry to form a decoded signal It is added to the decoded sinusoidal and transient components. In addition, some of the parametric noise may be inserted into the excitation signal, similar to the scheme used in FIGS. 4 and 5.

One of the possible criteria for determining the usefulness of x ₈ in the next RPE stage is the similarity with the input residual signal r. Therefore, it is common for the gain p to be somewhat related to the correlation of these two signals. With the goal of eliminating similarities between signals r and x ₈ (FIG. 4), the optimal value for ρ can be calculated as:

Where x ₈ is thus the signal identified in FIG. 6 and N is the window length for which p is optimized. The gain is preferably calculated based on frame-by-frame, ie N is the frame length. From Equation 1, it can be seen that the optimum gain is that standardized x across the ₈ x ₈ myeoksu and correlation of r relationship. Other gains with properties similar to those of Equation 1 may also be defined (eg, the representation of Equation 1 is optimized in terms of squared error criteria; other criteria may also be used).

If the model of r provided by x ₈ is complete (ie r = x ₈ ), then the blending coefficient is 1 and r ₈ is 0 because no further modeling is needed. On the other hand, if x ₈ is not a good model of r, the mixing coefficient takes less value and the second RPE acts primarily at r rather than r ₈ , ie the decimation 2 layer is provided by the decimation 8 layer. Use information limitedly.

The described technique can be applied to a full bandwidth signal or a specific frequency band. The quality parameter ρ represents the possibility for a complete filter to produce r ₈ implying several but not single parameters. The method provided herein is performed on a hierarchical bit stream comprising more than two excitation signals.

As mentioned above, the present invention relates to the encoding and decoding of wideband signals, in particular audio signals, to both encoders and decoders, to audio streams encoded according to the invention, and to data storage media on which such audio streams are stored. Used for

Claims (12)

A method of encoding a digital audio signal, wherein for each time segment of the signal, Encoding the audio signal to provide a code SSC representing the audio signal; Subtracting the code from the audio signal to obtain a first residual signal r _SSC ; Spectrally planarizing the first residual signal r _SSC to obtain a spectral planarized residual signal r and a spectral planarization parameter; Calculating a first excitation signal from the spectral planarized residual signal r using a pulse train encoder; Determining the quality of the first excitation signal x ₈ as a similarity with the residual signal r flattened in a spectral manner; - by subtracting the portion of the first excitation signal (x ₈₎ from the residual signal (r) flattening the spectral method, the second residual signal (r in which the part is dependent on the determined quality of the first excitation signal (x _{8) 8} ) a subtraction step; Calculating a second excitation signal (x ₂ ) from the second residual signal r ₈ using a pulse train encoder; Creating an audio stream, A first excitation signal (x ₈ ), A second excitation signal x ₂ , and A parameter p representing the quality of the first excitation signal x ₈ Generating an audio stream. And a digital audio signal. The method of claim 1, wherein the parameter code comprises sinusoidal and noise components of the audio signal. The method of claim 1, wherein the spectral flattening is performed using linear prediction encoding (LPC). The method of claim 1, wherein the first quality of an excitation signal (x ₈₎ is a digital audio signal that is based on a correlation between the first excitation signal (x ₈₎ and the smoothed spectrum scheme residual signal (r) How to encode. An audio encoder, adapted to encode a time segment of a digital audio signal, An encoder for encoding a digital audio signal and providing a code (SSC) representing the signal; A subtractor for subtracting the signal corresponding to the code from the audio signal to obtain a first residual signal r _SSC ; A spectral flattening unit for spectrally flattening the first residual signal r _SSC to obtain a spectral flattened residual signal r and a spectral flattening parameter; A pulse train encoder for calculating a first excitation signal for the residual signal r flattened in a spectral manner; Means for determining the quality of the first excitation signal x ₈ as a similarity with the residual signal r flattened in a spectral manner; - by subtracting the portion of the first excitation signal (x ₈₎ from the residual signal (r) flattening the spectral method, the second residual signal (r in which the part is dependent on the determined quality of the first excitation signal (x _{8) 8} ) a subtractor to obtain; A pulse train encoder for calculating a second excitation signal x ₂ for the second residual signal r ₈ ; A bit stream generator 15 for generating an audio stream AS, A first excitation signal (x ₈ ), A second excitation signal x ₂ , and A parameter p representing the quality of the first excitation signal x ₈ Bit stream generator 15, including Included, audio encoder. 6. The audio encoder of claim 5 wherein the parameter code comprises sinusoidal and noise components of the audio signal. 6. The audio encoder of claim 5 comprising a linear prediction encoder (LPC) adapted to perform spectral smoothing. The audio encoder according to claim 5, wherein the fragment (ρ) is based on a correlation between the first excitation signal (x ₈ ) and the residual signal (r) flattened in a spectral manner. A method of decoding a received audio stream (AS), wherein the audio stream is for each of a plurality of segments of an audio signal, A first excitation signal (x ₈ ), A second excitation signal (x ₂ ), A parameter p representing the quality of the first excitation signal x ₈ , the method comprising: Combining the first and second excitation signals x ₈ , x ₂ to obtain a combined excitation signal according to the quality parameter p; Synthesizing the first residual signal r ' _SSC from the combined excitation signal using linear prediction And decoding the received audio stream. An audio player for receiving and decoding an audio stream (AS), the audio stream for each of a plurality of segments of an audio signal, A first excitation signal (x ₈ ), A second excitation signal x ₂ , and A parameter indicative of the quality of the first excitation signal x ₈ , wherein the audio player comprises: Means for combining the first and second excitation signals x ₈ , x ₂ to obtain a combined excitation signal according to the quality parameter p; Means for synthesizing the first residual signal r ' _SSC from the combined excitation signal using linear prediction Including, audio player. As an audio stream (AS), for each of a plurality of segments of an audio signal, A first excitation signal x ₈ resulting from the pulse train encoding of the spectral planarized residual signal r, the residual signal r resulting from subtracting the encoded audio signal from the audio signal, A first excitation signal x ₈ ; A second excitation signal (x ₂ ) resulting from the pulse train encoding of the second residual signal, said signal being generated by subtracting a portion of the first excitation signal (x ₈ ) from the spectral planarized residual signal (r) Wherein the portion is the second excitation signal x ₂ , which depends on the determined quality of the first excitation signal x ₈ ; A parameter p representing the determined quality of the first excitation signal x ₈ An audio stream. A storage medium for storing the audio stream (AS) according to claim 11.

Images