EP3014611A1 - Improved frequency band extension in an audio signal decoder - Google Patents

Abstract

The invention relates to a method for extending the frequency band of an audio signal during a decoding or improvement process comprising a step of decoding or extracting, in a first so-called low frequency band, an excitation signal and coefficients of a linear prediction filter. The method comprises the following steps: - obtaining a signal (U_HB2(k), E403)) extended in at least a second frequency band higher than the first frequency band from an oversampled excitation signal extended in at least a second frequency band (UHB1(k), E401);- scaling (E406) the extended signal by means of a gain defined by subframe on the basis of an energy ratio of a frame and of a subframe; - filtering (E404) said scaled extended signal with a linear prediction filter of which the coefficients are derived from the coefficients of the low frequency band filter. The invention also relates to a frequency band extension device implementing the described method and a decoder comprising such a device.

Description

 Improved frequency band extension in an audio-frequency decoder

The present invention relates to the field of coding / decoding and audio-frequency signal processing (such as speech, music or other signals) for their transmission or storage.

 More particularly, the invention relates to a method and a device for extending the frequency band in a decoder or a processor performing an audio-frequency signal improvement.

 Many techniques exist to compress (with loss) an audiofrequency signal such as speech or music.

 Conventional methods of coding for conversational applications are generally classified in waveform coding (MIC for "pulse modulation and coding", ADPCM for "Pulse Modulation and Adaptive Differential Coding, transform coding, etc.), parametric coding (LPC for "Linear Predictive Coding" in English, sinus coding ...) and parametric hybrid coding with a quantification of the parameters by "analysis by synthesis" whose coding CELP (for "Code Excited Linear Prediction" in English) is the best known example.

 For non-conversational applications, state of the art audio signal coding (mono) consists of perceptual encoding by transform or subband, with parametric coding of high frequencies by tape replication.

 A review of conventional speech and audio coding methods can be found in W. B. Kleijn and K. K. Paliwal (Eds.), Speech Coding and Synthesis, Elsevier, 1995;

M. Bosi, R. E. Goldberg, Introduction to Digital Audio Coding and Standards, Springer 2002; J. Benesty, M. Sondhi M., Y. Huang (Eds.), Handbook of Speech Processing, Springer 2008.

Of particular interest here is the 3GPP AMR-WB ("Adaptive Multi-Rate Wideband") codec and decoder, which operates at an input / output frequency of 16 kHz and in which the signal is divided into two sub-bands, the low band (0-6.4 kHz) which is sampled at 12.8 kHz and coded by CELP model and the high band (6.4-7 kHz) which is parametrically reconstructed by "band extension" ( or BWE for "Bandwidth Extension" with or without additional information depending on the mode of the current frame. It can be noted here that the limitation of the coded band of the AMR-WB codec at 7 kHz is essentially related to the fact that the transmission frequency response of the broadband terminals has been approximated at the time of standardization (ETSI / 3GPP then ITU-T T) according to the frequency mask defined in the ITU-T P.341 standard and more precisely by using a so-called "P341" filter defined in the ITU-T G.191 standard which cuts frequencies above 7 kHz (this filter respects the mask defined in P.341). However, in theory, it is well known that a signal sampled at 16 kHz may have a defined audio band of 0 to 8000 Hz; the AMR-WB codec thus introduces a limitation of the high band in comparison with the theoretical bandwidth of 8 kHz.

 The 3GPP AMR-WB speech codec was standardized in 2001 mainly for circuit-mode (CS) telephony applications on GSM (2G) and UMTS (3G). This same codec was also standardized in 2003 in ITU-T as Recommendation G.722.2 "Wideband coding speech at around 16kbit / s using Adaptive Multi-Rate Wideband (AMR-WB)".

 It includes nine speeds, called modes, from 6.6 to 23.85 kbit / s, and includes continuous transmission mechanisms (DTX for "Discontinuous Transmission") with voice activity detection (VAD for "Voice Activity Detection") and noise generation of comfort (CNG for "Comfort Noise Generation") from frames of silence description (SID for "Silence Insertion Descriptor"), as well as mechanisms of correction of lost frames (FEC for "Frame Erasure Concealment", sometimes called PLC for "Packet Loss Concealment").

 The details of the AMR-WB coding and decoding algorithm are not repeated here, a detailed description of this codec is found in the 3GPP specifications (TS 26.190, 26.191, 26.192, 26.193, 26.194, 26.204) and ITU-TG .722.2 (and the corresponding Appendices and Appendix) and in the article by B. Bessette et al. entitled "The adaptive multirate broadband speech coded (AMR-WB)", IEEE Transactions on Speech and Audio Processing, vol. 10, no. 8, 2002, pp. 620-636 and associated 3GPP and ITU-T standard source codes.

 The principle of band extension in the AMR-WB codec is rather rudimentary. Indeed, the high band (6.4-7 kHz) is generated by formatting a white noise through a temporal envelope (applied in the form of gains per subframe) and frequency (by the application of a linear prediction synthesis filter or LPC for "Linear Predictive Coding"). This band extension technique is illustrated in Figure 1.

A white noise, u _HB1 (n), n = 0, · · ·, 79, is generated at 16 kHz per 5 ms subframe per linear congruent generator (block 100). This noise u _HB1 (n) is shaped in time by applying gains per subframe; this operation is broken down into two processing steps (blocks 102, 106 or 109):

· A first factor is calculated (block 101) to set the white noise u _HB1 (n) (block

102) at a level similar to that of excitation, u (n), "= (), · · -, 63, decoded at

12.8 kHz in the low band:

 63

 2

Σ (l) It can be noted here that the normalization of the energies is done by comparing blocks of different size (64 for u (n) and 80 for u _HBl (n)), without compensation of the differences of sampling frequencies (12.8 or 16 kHz) .

 • The excitation in the high band is then obtained (block 106 or 109) in the form:

u _HB (n) = g _HB u _HB2 (n)

where the gain g _HB is obtained differently depending on the flow rate. If the rate of the current frame is <23.85 kbit / s, the gain g _HB is estimated to be "blind" (ie without additional information); in this case, the block 103 filters the low-band decoded signal by a high-pass filter having a cut-off frequency at 400 Hz to obtain a signal s _hp (n), "= (), · -, - 63 - high-pass filter eliminates the influence of the very low frequencies that can bias the estimate made in block 104 - then the "tilt" (signal of spectral slope) noted e _tiU of the signal s _hp (n) is calculated by standard autocorrelation (block 104):

 63 e = -

ctilt 63 and finally we calculate g _HB in the form:

SHB = ^W SPSSP + ( ¹ - ^W SP) SBG

where g _sp = l - e _tilt is the gain applied in active speech frames (SP for speech), g _BG = l .25g _sp is the gain applied in inactive speech frames associated with background noise (BG for Background) and w _sp is a weighting function that depends on Voice Activity Detection (VAD). It is understood that the estimation of the tilt (e _tilt ) makes it possible to adapt the level of the high band according to the spectral nature of the signal; this estimate is particularly important when the spectral slope of the decoded signal CELP is such that the average energy decreases when the frequency increases (case of a voiced signal where e _tiU is close to 1, so g _sp = l - e _tilt is thus reduced). Note also that the factor g _HB in the decoding AMR-WB is bounded to take values in the interval [0.1, 1.0].

At 23.85 kbit / s, correction information is transmitted by the encoder AMR-WB and decoded (blocks 107, 108) in order to refine the estimated gain per subframe (4 bits every 5ms, ie 0.8 kbit / s) . The artificial excitation u _HB (n) is then filtered (block 111) by a transfer function function LPC (block 111) synthesis filter 1 / A _HB (z) and operating at the sampling frequency of 16 kHz. The realization of this filter depends on the rate of the current frame:

• At 6.6 kbit / s, the filter \ l A _HB (z) is obtained by weighting by a factor ^ = 0.9 an LPC filter of order 20, I l À ^ext (z) which "extrapolates" the LPC filter from order 16, 1 / A (z), decoded in the low band (at 12.8 kHz) - details of the extrapolation in the domain of the ISF parameters (for "Imittance Spectral Frequency") are described in the standard G. 722.2 in section 6.3.2.1; in that case,

\ IA _HB (z) = \ IA ^ext (zl Y)

· At rates> 6.6 kbit / s, the filter 1 / A _HB (z) is of order 16 and simply corresponds to:

 where ^ = 0.6. Note that in this case the filter II A (zlf) is used at 16 kHz, which results in a spread (by homothety) of the frequency response of this filter from [6.4 kHz] to [0.8 kHz] ].

The result, s _HB (n), is finally processed by a band-pass filter (block 112) of type FIR ("Finite

Impulse Response "), to keep only the band 6 - 7 kHz, at 23.85 kbit / s, a low-pass filter also FIR type (block 113) is added to the treatment to further attenuate frequencies above 7 kHz. The synthesis at high frequencies (HF) is finally added (block 130) to the low frequency synthesis (BF) obtained with the blocks 120 to 123 and resampled at 16 kHz (block 123). theoretically extends from 6.4 to 7 kHz in the AMR-WB coding, the HF synthesis is rather in the band 6-7 kHz before addition with the synthesis BF.

 Several disadvantages can be identified with the AMR-WB coding extension technique:

• The signal in the high band is white noise formatted (by temporal gains per subframe, filtering by 1 / A _HB (z) and bandpass filtering), which is not a good general model signal in the band 6.4-7 kHz. For example, there are very harmonic music signals for which the 6.4-7 kHz band contains sinusoidal components (or tones) and no noise (or little noise), for these signals the band extension of the AMR-WB coding degrades. strongly the quality.

• The 7 kHz low-pass filter (block 113) introduces an offset of nearly 1 ms between the low and high bands, which can potentially degrade the quality of some signals by slightly desynchronizing the two bands at 23.85 kbit / s - this Desynchronization can also be a problem when switching from 23.85 kbit / s to other modes.

 • The estimate of gains per subframe (block 101, 103 to 105) is not optimal. In part, it is based on an equalization of the "absolute" energy per sub-frame (block 101) between signals at different frequencies: the artificial excitation at 16 kHz

(white noise) and a signal at 12.8 kHz (ACELP decoded excitation). It can be noted in particular that this approach implicitly induces an attenuation of the high band excitation (by a ratio 12.8 / 16 = 0.8); in fact, it will also be noted that no deemphasis is performed on the high band in the AMR-WB coding, which implicitly induces a relative amplification close to 0.6 (which corresponds to the value of the frequency response s / (l - 0.68z ^_1) to 6400 Hz).

In fact, the factors of 1 / 0.8 and 0.6 compensate each other approximately.

 • On speech, the 3GPP AMR-WB coding characterization tests documented in the 3GPP TR 26.976 report showed that the 23.85 kbit / sa mode is not as good as 23.05 kbit / s, its quality is in fact similar to that of the 15.85 kbit / s mode. This shows in particular that the level of artificial RF signal must be controlled very carefully, because the quality is degraded to 23.85 kbit / s while the 4 bits per frame are supposed to better approach the energy of the original high frequencies.

 "The limitation of the 7 kHz coded band results from the application of a strict model of the emission response of acoustic terminals (filter P.341 in ITU-T G.191). However, for a sampling frequency of 16 kHz, frequencies in the band 7-8 kHz remain important, especially for music signals, to ensure a good level of quality.

The AMR-WB decoding algorithm has been improved in part with the development of the ITU-T G.718 scalable codec which was standardized in 2008.

 ITU-T G.718 includes an interoperable mode, for which core coding is compatible with 12.65 kbit / s G.722.2 (AMR-WB) coding; in addition, the G.718 decoder has the particularity of being able to decode a bit stream AMR-WB / G.722.2 at all possible bit rates of the AMR-WB codec (from 6.6 to 23.85 kbit / s).

 The G.718 interoperable decoder in low delay mode (G.718-LD) is illustrated in FIG. 2. Below are the improvements made to the bit-stream decoding functionality AMR- WB in the G.718 decoder, with references to Figure 1 when necessary:

• The band extension (described in clause 7.13.1 of Recommendation G.718, block 206) is identical to that of the AMR-WB decoder, except that the 6-7 kHz band-pass filter and the filter Synthesis 1 / A _HB (z) (Blocks 111 and 112) are in reverse order. In addition, at 23.85 kbit / s the 4 bits transmitted by AMR-WB encoder subframes are not used in the interoperable G.718 decoder; the synthesis of high frequencies (HF) at 23.85 kbit / s is therefore identical to 23.05 kbit / s which avoids the known problem of quality of AMR-WB decoding at 23.85 kbit / s. A fortiori, the low-pass filter at 7 kHz (block 113) is not used, and the specific decoding mode 23.85 kbit / s is omitted (blocks 107 to 109).

 • A post-processing of the 16 kHz synthesis (see clause 7.14 of G.718) is implemented in G.718 by "noise gâte" in block 208 (to "improve" the quality of the silences by reducing the level ), high-pass filtering (block 209), low-frequency post-filter (so-called "bass posfilter") in block 210 attenuating interharmonic noise at low frequencies, and conversion to 16-bit integers with saturation control (with control of gain or AGC) in block 211.

However, the band extension in the AMR-WB and / or G.718 codecs (interoperable mode) is still limited in several respects.

 • In particular, the synthesis of high frequencies by shaped white noise (by a temporal approach of source-filter type LPC) is a very limited model of the signal in the frequency band higher than 6.4 kHz.

 • Only the 6.4-7 kHz band is artificially re-synthesized, whereas in practice a wider band (up to 8 kHz) is theoretically possible at the sampling frequency of 16 kHz, which can potentially improve signal quality, if not pre-processed by a P.341 (50-7000 Hz) filter as defined in the ITU-T Software Tool Library (G.191 standard).

There is therefore a need to improve the band extension in an AMR-WB codec or an interoperable version of this codec or more generally to improve the band extension of an audio signal.

 The present invention improves the situation.

 To this end, the invention proposes a method of extending the frequency band of an audiofrequency signal during a decoding or improvement process comprising a decoding or extraction step, in a so-called first frequency band. low band, an excitation signal and the coefficients of a linear prediction filter. The method is such that it comprises the following steps:

 obtaining an extended signal in at least a second frequency band greater than the first frequency band from an oversampled excitation signal and extended in at least a second frequency band;

scaling the extended signal by a gain defined by subframe according to a ratio of energy per frame and subframe of the audiofrequency signal in the first frequency band; filtering said extended signal scaled by a linear prediction filter whose coefficients are derived from the coefficients of the low band filter.

Thus, taking into account the excitation signal (resulting from the decoding of the low band or from a low band signal extraction) makes it possible to carry out the band extension with a signal model that is more suitable for certain types of band. signals like music signals.

 Indeed, the excitation signal decoded or estimated in the low band has in some cases harmonics, which when they exist, can be transposed in high frequency so that it ensures a certain level of harmonicity in the high band reconstructed.

 The band extension according to the method thus makes it possible to improve the quality for this type of signal.

 In addition, the band extension according to the method is performed by first extending an excitation signal and then applying a synthesis filtering step; this approach exploits the fact that the decoded excitation in the low band is a signal whose spectrum is relatively flat, which avoids the decoded signal whitening treatments that may exist in the known methods of band extension in the frequency domain in the state of the art.

 It will be noted that even if the invention is motivated by the improvement of the quality of the band extension in the context of the interoperable AMR-WB coding, the different embodiments apply to the more general case of the extension of band of an audio signal, in particular in an enhancement device performing an analysis of the audio signal to extract the parameters necessary for the band extension.

 Taking into account the energy at the current frame and that of the subframe in the low band signal (first frequency band) makes it possible to adjust the ratio between the energy per subframe and the subframe. energy per frame in the high band (second frequency band) and thus adjust energy ratios rather than absolute energies. This makes it possible to keep in the high band the same ratio of energy between subframe and frame as in the low band, which is particularly beneficial when the energy of the subframes varies greatly, for example in the case of transient sounds. , of attacks.

 The various particular embodiments mentioned below may be added independently or in combination with each other, to the steps of the extension method defined above.

 In one embodiment, the method further includes an adaptive bandpass filtering step based on the decoding rate of the current frame.

This adaptive filtering makes it possible to optimize the extended bandwidth as a function of the bit rate, and therefore the quality of the reconstructed signal after band extension. Indeed, for low bit rates (typically 6.6 and 8.85 kbit / s for AMR-WB), the overall quality of the decoded signal in low band (by AMR-WB codec or an interoperable version) not being very good, it is better not to extend the decoded band too much and thus to limit the band extension by adapting the frequency response of the filter bandpass associated to cover for example an approximate band of 6 to 7 kHz; this limitation is all the more advantageous that the excitation signal itself is relatively poorly coded and it is preferable not to use a sub-band too wide for the extension of high frequencies. In contrast, for higher rates (12.65 kbit / s and higher for AMR-WB), the quality can be improved with HF synthesis covering a wider band, for example approximately 6 to 7.7 kHz. The high limit of 7.7 kHz (instead of 8 kHz) is an example embodiment, which can be adjusted to values close to 7.7 kHz. This limitation is here justified by the fact that the extension is made in the invention without auxiliary information and that an extension up to 8 kHz (although theoretically possible) could result in artifacts for particular signals. In addition, this limitation at 7.7 kHz takes into account the fact that typically the anti-aliasing filters in analog / digital conversion and resampling filters between 16 kHz and other frequencies are not perfect and they typically introduce a rejection at frequencies below 8 kHz.

 In one possible embodiment, the method comprises a step of transforming the time-frequency of the excitation signal, the step of obtaining an extended signal then taking place in the frequency domain and a reverse time-frequency transforming step. of the extended signal before the scaling and filtering steps.

 The implementation of the band extension (of the excitation signal) in the frequency domain makes it possible to obtain a fineness of frequency analysis which is not available with a temporal approach, and also makes it possible to have a resolution Frequency sufficient to detect harmonics and transpose high harmonics of the signal (in the low band) to improve quality while respecting the structure of the signal.

 In a detailed embodiment, the step of generating an oversampled and extended excitation signal is performed according to the following equation:

U _HBl (k) =

with k the index of the sample, U _H BI (k) the spectrum of the extended excitation signal, U (k) the spectrum of the excitation signal obtained after the transforming step and start_band a predefined variable.

 Thus, this function includes a resampling of the excitation signal by adding samples to the spectrum of this signal.

In the frequency band corresponding to the samples ranging from 200 to 239, the original spectrum is conserved, in order to be able to apply a progressive attenuation response of the high-pass filter in this frequency band and also to not introduce defects. audible during the step of adding the low frequency synthesis to the high frequency synthesis.

 In a particular embodiment, the method comprises a de-emphasis filtering step of the extended signal at least in the second frequency band.

 Thus, the signal in the second frequency band is brought into a domain coherent with the signal in the first frequency band.

 In a particular embodiment, the method further comprises a step of generating a noise signal at least in the second frequency band, the extended signal being obtained by combining the extended excitation signal and the noise signal.

 Indeed, it suffices to have characteristics from the oversampled excitation signal and extended in at least a second frequency band to have a signal model adapted to certain types of signals. This can be combined with another signal, for example a noise generated to obtain the extended signal having a suitable signal pattern.

 In one embodiment, the combining step is performed by adaptive additive mixing with a level equalization gain between the extended excitation signal and the noise signal.

 The application of this equalization gain allows the combining step to adapt to the characteristics of the signal to optimize the relative proportion of noise in the mixture.

 The present invention also aims at a frequency band extension device for an audio frequency signal comprising a decoding or extraction stage, in a first so-called low band frequency band, of an excitation signal and of the transmission coefficients. a linear prediction filter. The device is such that it comprises:

a module for obtaining an extended signal (U _H B2 (k), 503) in at least a second frequency band greater than the first frequency band from an oversampled excitation signal and extended in at least a second frequency band (U _H Bi (k));

 - a scaling module (507) of the signal extended by a gain defined by subframe according to a ratio of energy per frame and subframe of the audiofrequency signal in the first frequency band;

 a filtering module (510) of said extended signal scaled by a linear prediction filter whose coefficients are derived from the coefficients of the low band filter.

 This device has the same advantages as the method described above, which it implements.

The invention relates to a decoder comprising a device as described. It is directed to a computer program comprising code instructions for implementing the steps of the tape extension method as described, when these instructions are executed by a processor.

 Finally, the invention relates to a storage medium, readable by a processor, integrated or not to the band expansion device, possibly removable, storing a computer program implementing a band extension method as described above.

 Other features and advantages of the invention will appear more clearly on reading the following description, given solely by way of nonlimiting example, and with reference to the appended drawings, in which:

 FIG. 1 illustrates a part of an AMR-WB decoder implementing frequency band extension steps of the state of the art and as previously described;

 FIG. 2 illustrates a decoder of the interoperable type G.718-LD at 16 kHz according to the state of the art and as described previously;

 FIG. 3 illustrates an interoperable decoder with the AMR-WB coding and integrating a band extension device according to one embodiment of the invention; FIG. 4 illustrates in flowchart form the main steps of a band extension method according to one embodiment of the invention;

 FIG. 5 illustrates a first embodiment in the frequency domain of a band extension device according to the invention;

 FIG. 6 illustrates an example of frequency response of a bandpass filter used in a particular embodiment of the invention;

 FIG. 7 illustrates a second embodiment in the time domain of a band extension device according to the invention; and

 FIG. 8 illustrates a hardware embodiment of a band extension device according to the invention.

FIG. 3 illustrates an exemplary decoder, compatible with the AMR-WB / G.722.2 standard, in which there is a postprocessing similar to that introduced in G.718 and described with reference to FIG. 2 and an improved band extension according to the extension method of the invention, implemented by the band extension device illustrated by block 309.

Unlike AMR-WB decoding which operates with an output sampling frequency of 16 kHz and G.718 decoding which operates at 8 or 16 kHz, a decoder is considered here which can operate with an output signal (synthesis) at the frequency fs = 8, 16, 32 or 48 kHz. Note that it is assumed here that the coding was performed according to the AMR-WB algorithm with an internal frequency of 12.8 kHz for the CELP coding in low band and at 23.85 kbit / s a subframe gain coding at the frequency of 16 kHz; even if the invention is described here at the decoding level, it is assumed here that the coding can also operate with an input signal at the frequency fs = 8, 16, 32 or 48 kHz and adequate resampling operations, beyond the scope of the invention, are implemented at the coding according to the value of fs. It can be noted that when fs = 8 kHz, in the case of a decoding compatible with AMR-WB, it is not necessary to extend the low band 0-6.4 kHz, because the audio band reconstructed at the frequency / s is limited to 0-4000 Hz.

 In Figure 3, the CELP decoding (BF for low frequencies) always operates at the internal frequency of 12.8 kHz, as in AMR-WB and G.718, and the band extension (HF for high frequencies) being the subject of the invention operates at the frequency of 16 kHz, the synthesis BF and HF are combined (block 312) at the frequency fs after adequate resampling (block 306 and internal processing block 311). In variants of the invention, the combination of the low and high bands can be done at 16 kHz, after resampling the low band of 12.8 to 16 kHz, before resampling the extended signal at the frequency fs.

 The decoding according to FIG. 3 depends on the mode (or bit rate) AMR-WB associated with the current frame received. As an indication and without this having an effect on block 309, the decoding of the low band CELP part comprises the following steps:

 • Demultiplexing the coded parameters (block 300) in the case of a correctly received frame (= 0 where is the "bad frame indicator" worth 0 for a received frame and 1 for a lost frame)

 • Decoding of ISF parameters with interpolation and conversion to LPC coefficients (block 301) as described in clause 6.1 of G.722.2.

 • Decoding of the CELP excitation (block 302), with an adaptive and fixed part to reconstruct the excitation (exc or u n)) in each subframe of length 64 to 12.8 kHz:

u \ n) = g _p v (n) + g _c c (n), η = 0, · · ·, 63

by following the notations of clause 7.1.2.1 of G.718 concerning CELP decoding, where v (n) and c (n) are respectively the code words of the adaptive and fixed dictionaries, and g and g _c are the decoded gains associates. This excitation a) is used in the adaptive dictionary of the following subframe; it is then post-processed and one discerns as in G.718 the excitement a) (also noted exc) of its modified post-processed version u (n) (also noted exc2) which serves as input to the synthesis filter, 1 / ((z), in block 303. In variants that can be implemented for the invention, the post-treatments applied to the excitation can be modified (for example, the phase dispersion can be improved) or these post The treatments may be extended (for example, interharmonic noise reduction may be implemented) without affecting the nature of the band extension method of the invention.

 • Synthetic filtering by 1 /  (z) (block 303) where the decoded LPC filter  (z) is of order 16

 • Narrowband post-processing (block 304) according to clause 7.3 of G.718 if fs = 8 kHz.

• Deactivation (block 305) by the filter 1 / (l - 0.68z ^_1 )

 • Low frequency post-processing (block 306) as described in clause 7.14.1.1 of G.718. This processing introduces a delay which is taken into account in the decoding of the high band (> 6.4 kHz).

 • Resampling of the internal frequency from 12.8 kHz to the output frequency fs (block 307). Several achievements are possible. Without loss of generality, we consider here as an example that if fs = 8 or 16 kHz, the resampling described in clause 7.6 of G.718 is repeated here, and if fs = 32 or 48 kHz, filters Finite Impulse Response (FIR) are used.

 • Calculation of the parameters of the "noise gate" (block 308) which is carried out preferentially as described in clause 7.14.3 of G.718.

 It can be noted that the use of blocks 306, 308, 314 is optional.

 Note also that the decoding of the low band described above assumes a current frame called "active" with a rate between 6.6 and 23.85 kbit / s. In fact, when the DTX (Continuous Transmission in French) mode is activated, some frames can be coded as "inactive" and in this case you can either transmit a silence descriptor (on 35 bits) or not transmit anything. In particular, it is recalled that the SID frame describes several parameters: ISF parameters averaged over 8 frames, average energy over 8 frames, dithering flag for the non-stationary noise reconstruction. In all cases, at the decoder, we find the same decoding model as for an active frame, with a reconstruction of the excitation and an LPC filter for the current frame, which makes it possible to apply the invention itself to inactive frames. The same applies for the decoding of "lost frames" (or FEC, PLC) in which the LPC model is applied.

Unlike decoding AMR-WB or G.718, the decoder according to the invention makes it possible to extend the decoded low band (50-6400 Hz while taking into account the 50 Hz high-pass filtering at the decoder, 0-6400 Hz in the decoder. the general case) to an extended band whose width varies, ranging from approximately 50-6900 Hz to 50-7700 Hz depending on the mode implemented in the current frame. We can talk about a first frequency band from 0 to 6400Hz and a second frequency band from 6400 to 8000Hz. In reality, in the preferred embodiment, the extension of the excitation is carried out in the frequency domain in a band of 5000 to 8000 Hz, to allow bandpass filtering of width 6000 to 6900 or 7700 Hz.

 In a preferred embodiment, at 23.85 kbit / s, as in the G.718 decoder described with reference to FIG. 2, the gain correction information HF (0.8 kbit / s) transmitted at 23.85 kbit / s is here ignored. Thus in FIG. 3, no specific block at 23.85 kbit / s is used.

 The high band decoding part is produced in the block 309 representing the band extension device according to the invention and which is detailed in FIG. 5 in a first embodiment and in FIG. 7 in a second embodiment.

This device comprises at least one module for obtaining an extended signal in at least one second frequency band greater than the first frequency band from an oversampled excitation signal and extended in at least a second band. of frequency (U _H Bi (k)), a scaling module of the signal extended by a gain defined by subframe according to a ratio of energy per frame and subframe of the audiofrequency signal in the first frequency band and a filtering module of said extended signal scaled by a linear prediction filter whose coefficients are derived from the coefficients of the low band filter.

 In order to align the decoded low and high bands, a delay (block 310) is introduced in the first embodiment to synchronize the outputs of the blocks 306 and 307 and the high band synthesized at 16 kHz is resampled from 16 kHz to the frequency fs (block output 311). For example, when fs = lb kHz the delay Γ = 30 samples, which corresponds to the resampling delay of 12.8 to 16 kHz of 15 samples + delay of the low frequency post-processing of 15 samples. The value of the delay T will have to be adapted for the other cases {fs = 32, 48 kHz) according to the treatments implemented. Remember that when fs = 8 kHz, it is not necessary to apply the blocks 309 to 311 because the output signal band of the decoder is limited to 0-4000 Hz.

It should be noted that the extension method of the invention implemented in block 309 according to the first embodiment introduces preferentially no additional delay with respect to the low band reconstructed at 12.8 kHz; however, in variants of the invention (for example using a time / frequency transformation with overlap), a delay may be introduced. Thus, in general the value of in block 310 will have to be adjusted according to the specific implementation. For example, in the case where the post-processing of low frequencies (block 306) is not used, the delay to be introduced for fs = 16 kHz can be set at Γ = 15 samples; likewise if the invention is carried out according to the variant of the embodiment described in FIG. 7, the value of T is reduced to compensate for the delay introduced by the post-processing of the low frequencies (block 306) if it is used. The low and high bands are then combined (added) in block 312 and the resulting synthesis is post-processed by high-order 50 Hz (type IIR) high-pass filtering whose coefficients depend on the frequency fs (block 313) and output post-processing with optional noise gate application similar to G.718 (block 314).

 The band extension device according to the invention, illustrated by block 309 according to the embodiment of the decoder of FIG. 3, implements a band extension method described now with reference to FIG. 4.

 This extension device may also be independent of the decoder and may implement the method described in FIG. 4 to perform a band extension of an existing audio signal stored or transmitted to the device, with an analysis of the audio signal to extract it an excitation and an LPC filter.

 This device receives as input an excitation signal in a first so-called low-band frequency band u (n) in the case of an implementation in the time domain or U (k) in the case of an implementation. in the frequency domain for which a time-frequency transform step is then applied.

 In the case of an application in a decoder, this received excitation signal is a decoded signal.

 In the case of an enhancement device independent of the decoder, the low band excitation signal is extracted by analysis of the audio signal.

 In a possible embodiment, the low band audio signal is resampled before the excitation extraction step, so that the excitation extracted from the audio signal by linear prediction estimated from the low band signal (or parameters LPC associated with the low band) is already resampled. An exemplary embodiment in this case consists in taking a sampled low band signal at 12.8 kHz, which has a low-band LPC filter describing the short-term spectral envelope for the current frame, oversampling it at 16 kHz, and filter it by an LPC prediction filter obtained by extrapolating the LPC filter. Another embodiment is to take a low band signal sampled at 12.8 kHz which is not available LPC model, the oversampler at 16 kHz, perform an LPC analysis on this signal at 16 kHz, and filter this signal by a LPC prediction filter obtained by this analysis.

A step E401 for generating an extended _oversampled excitation signal (e _xt (n) or U _HB1 (k)) in a second frequency band greater than the first frequency band is performed. This generation step may comprise both a resampling step and an extension step or simply an extension step depending on the excitation signal obtained at the input.

This step is detailed later in the embodiments described with reference to FIGS. 5 and 7. This extended oversampled excitation signal is used to obtain an extended signal (U _H B2 (k)) in a second frequency band. This extended signal then has a signal model adapted to certain types of signals thanks to the characteristics of the extended excitation signal.

 This extended signal can be obtained after combining the oversampled and extended excitation signal with another signal, for example a noise signal.

Thus, in one embodiment, a step E402 for generating a noise signal (u _HB (n) or U _HB (k)) at least in the second frequency band is performed. The second frequency band is for example a high frequency band ranging from 6000 to 8000 Hz. For example, this noise can be generated pseudo-randomly by a linear congruential generator. In variants of the invention, this noise generation can be replaced by other methods, for example a signal of constant amplitude (of arbitrary value such as 1) could be defined and random signs applied to each frequency line. generated.

The extended excitation signal is then combined with the noise signal in step E403 to obtain the extended signal which may also be called a combined signal (u _HB1 (n) or U _HB2 (k)) in the frequency band range corresponding to the entire frequency band including the first and the second frequency band. Thus, the combination of these two types of signals makes it possible to obtain a combined signal with characteristics more adapted to certain types of signals such as musical signals.

 Indeed, the decoded or estimated excitation signal in the low band in some cases has harmonics closer to the musical signals than the noise signal alone.

Low-frequency harmonics, if they exist, can thus be transposed into high frequency so that their mixing with noise makes it possible to ensure a certain level of harmonicity or relative level of noise or spectral flatness ("spectral flatness"). in English) in the reconstructed high band.

 The band extension according to the method improves the quality for this type of signals compared to AMR-WB.

The combined (or extended) signal is then filtered at E404 by a linear prediction filter whose coefficients are derived from the coefficients of the low band filter (((z)) decoded or obtained by analysis and extraction from the low band signal or an oversampled version of it. The band extension according to the method is therefore achieved by first extending an excitation signal and then applying a linear prediction synthesis filtering step (LPC); this approach exploits the fact that the LPC decoded excitation in the low band is a signal whose spectrum is relatively flat, which avoids additional processing of whitening of the decoded signal in the band extension. Advantageously, the coefficients of this filter can for example be obtained from the decoded parameters of the linear band prediction filter (LPC). If the LPC filter used in the high sampled 16 kHz band is of the form \ l A {zl γ), where 1 / Â (z) is the low band decoded filter, and γ is a weighting factor, the frequency response of the IIA filter (zl γ) corresponds to a spread of the frequency response of the decoded low band filter. In a variant, the filter 1 / A (z) can be extended to a higher order (as to 6.6 kbit / s in the block 111) to avoid such spreading.

 Preferably, but additionally, additional steps of adaptive bandpass filtering at E405 and / or scaling at E406 and E407 can be performed to improve the quality of the extension signal according to the bit rate. decoding and on the other hand to ensure to keep the same energy ratio between a subframe and a combined signal frame as in the low frequency band.

 These steps will be explained in more detail in the embodiments of the figures

5 and 7.

 In a first embodiment, the band extension device is described now with reference to FIG. 5. This device implements the band extension method described above with reference to FIG. 4.

 Thus, at the input of this device, a low band excitation signal decoded or estimated by analysis is received (u (n)). The band extension here uses the decoded excitation at 12.8 kHz (exc2 or u (n)) at the output of block 302.

 Note that in this embodiment, the generation of the oversampled and extended excitation is carried out in a frequency band ranging from 5 to 8 kHz including a second frequency band (6.4-8kHz) greater than the first band of frequency (0-6.4 kHz).

 Thus, the generation of the extended excitation signal is effected at least on the second frequency band but also on a part of the first frequency band.

 Of course, the values defining these frequency bands may be different depending on the decoder or the processing device in which the invention applies.

 For this exemplary embodiment, this signal is transformed to obtain an excitation signal spectrum U (k) by the time-frequency transformation module 500.

In a particular embodiment, the transform uses a DCT-IV (for "Discrete Cosine Transform" - Type IV in English) (block 500) on the current frame of 20 ms (256 samples), without windowing, which amounts to directly transform u (n) with n = 0, · · ·, 255 according to the following formula:

where N = 256 and & = 0, · · ·, 255.

 We note here that the transformation without windowing (or equivalently with a rectangular window implicit in the length of the frame) is possible because the processing is performed in the field of excitation, and not the domain of the signal, so that no artifact (block effects) is audible, which is an important advantage of this embodiment of the invention.

In this embodiment, the DCT-IV transformation is implemented by FFT according to the "Evolved ZXT (EDCT)" algorithm described in the article by D. M. Zhang, HT. Li, A Low Complexity Transform - Evolved DCT, IEEE 14th International Conference on Computational Science and Engineering (CSE), Aug. 2011, pp. 144-149, and implemented in ITU-T G.718 Annex B and G.729.1 Annex E.

 In variants of the invention and without loss of generality, the DCT-IV transformation may be replaced by other short-term time-frequency transformations of the same length and in the field of excitation, such as an FFT (for Fast Fourier Transform "in English) or DCT-II (Discrete Cosine Transform - Type II). Alternatively, it will be possible to replace the DCT-IV on the frame by a recovery-addition and windowing transformation of length greater than the length of the current frame, for example by using an MDCT (for "Modified Discrete Cosine Tranform"). ). In this case, the delay T in block 310 of FIG. 3 will have to be adjusted (reduced) adequately as a function of the additional delay due to the analysis / synthesis by this transform.

 The DCT spectrum, U (k), of 256 samples covering the band 0-6400 Hz (at 12.8 kHz), is then extended (block 501) into a spectrum of 320 samples covering the band 0-8000 Hz (at 16 kHz) in the following form:

 0 £ = 0, · · ·, 199

t * ⁾ = U (k) £ = 200, · · -, 239

 U (k + start _ band - 240) k = 240, · · ·, 319

where preferentially start_band = 160 is used.

 Block 501 functions as a module for generating an oversampled and extended excitation signal and performs step E401 with resampling of 12.8 to 16 kHz in the frequency domain, adding ¼ samples (k = 240, · · ·, 319) to the spectrum, the ratio between 16 and 12.8 being 5/4.

In addition, the block 501 carries out an implicit high-pass filtering in the 0-5000 Hz band since the first 200 samples of U _HB1 (k) are set to zero; as explained more later, this high-pass filtering is also completed by a progressive attenuation part of the spectral values of indices k = 200, · · -, 255 in the band 5000-6400 Hz, this progressive attenuation is implemented in the block 504 but could be performed separately outside the block 504. Equivalently and in variants of the invention, the implementation of high-pass filtering separated into blocks of index coefficients & = 0, · · ·, 199 set to zero, with coefficients k = 200, · · -, 255 attenuated, in the transformed domain, can therefore be carried out in a single step.

In this exemplary embodiment and according to the definition of U _HB1 (k), it is noted that the 5000-6000 Hz band of U _HB1 (k) (which corresponds to the indices k = 200, · · -, 239) is copied from of the 5000-6000 Hz U (k) band. This approach preserves the original spectrum in this band and avoids introducing distortions in the 5000-6000 Hz band during the addition of HF synthesis with BF synthesis - particularly the signal phase (implicitly represented in the DCT-IV domain) in this band is preserved.

The 6000-8000 Hz band of U _HB1 (k) is here defined by copying the 4000-6000 Hz band of U (k) since the value of start_band is preferably fixed at 160.

 In a variant of the embodiment, the value of start_band can be made adaptive around the value of 160, without changing the nature of the invention. The details of the adaptation of the value start_band are not described here because they go beyond the scope of the invention without changing the scope.

For some broadband signals (sampled at 16 kHz), the high band (> 6 kHz) may be noisy, harmonic or have a mixture of noise and harmonics. In addition, the level of harmonicity in the 6000-8000 Hz band is generally correlated with that of the lower frequency bands. Thus, in a particular embodiment, the noise generation block 502 implements the step E402 of FIG. 4 and generates a noise generation in the frequency domain, ½ _BN () for k = 240, · · ·, 319 (80 samples) corresponding to a second so-called high frequency frequency band in order to then combine this noise with the spectrum U _HB1 (k) in the block 503.

In a particular embodiment, the noise (in the 6000-8000 Hz band) is generated pseudo-randomly with a 16-bit linear congruent generator

with the convention that U _HBN {2 9) in the current frame corresponds to the value U _HBN Q \ 9) of the previous frame. In variants of the invention, this noise generation can be replaced by other methods. The combination block 503 can be realized in different ways. In a privileged way, we consider an adaptive additive mix of the form:

U _HB2 (k) = βυ _ΗΒΙ (k) + G _HBN U _HBN (k), k = 240, · · ·, 319

where G _HBN is a normalization factor for equalizing the energy level between the two signals,

 319

Σ u _HB1 (k) ² + ε

 G = 240

 HBN 319

Σ U _HBN (k) ² + ε

 ¾ = 240

with ε = 0.01, and the coefficient (between 0 and 1) is adjusted according to parameters estimated from the decoded low band and the coefficient β (between 0 and 1) depends on a.

 In a preferred embodiment, the energy of the noise is calculated in three bands: 2000-4000 Hz, 4000-6000 Hz and 6000-8000 Hz, with

^ N2-4 Σ U 'k)

 eN (80.159)

^ N4-6 Σ U ' ² (k)

 teN 4 ((116600,, 223399))

Σ U ' ² (k)

 Ten (240.319)

or

and N (^, ^ ₂ ) is the set of indices k for which the index coefficient est is classified as being associated with noise. This set can be obtained for example by detecting the local peaks in U and in whereas these lines are not associated with noise, ie (by applying the negation of the previous condition):

N (a, b) = {a≤k≤b \\ U + 1) |}

 It may be noted that other methods of calculating the noise energy are possible, for example by taking the median value of the spectrum on the band in question or by applying a smoothing to each frequency line before calculating the energy per band.

The ratio of the energy of the noise in the bands 4-6 kHz and 6-8 kHz is set so that between the bands 2-4 kHz and 4-6 kHz: = P - E,

Σ U ² (k) - E or

 El

E _ _{6 N4} = MZX _(N4 _ E _6, E _{N 2} _ _4), p = p = max (p, E _N6 _ _&)

 'N2-4 where max (.,.) Is the function that gives the maximum of both arguments.

 In variants of the invention, the calculation of a may be replaced by other methods. For example, in one variant, it will be possible to extract (calculate) different parameters (or "features" in English) characterizing the low band signal, including a "tilt" parameter similar to that calculated in the AMR-WB codec, and it will be estimated the factor a according to a linear regression from these different parameters by limiting its value between 0 and 1. The linear regression could for example be estimated in a supervised manner by estimating the factor a by giving itself the original high band in a learning base. Note that the calculation method does not limit the nature of the invention.

In a preferred embodiment, we take

to preserve the energy of the extended signal after mixing.

 In a variant, the factors β and a may be adapted to take account of the fact that noise injected into a given band of the signal is generally perceived as stronger than a harmonic signal at the same energy in the same band. Thus we can modify the factors β and gold as follows:

β ^ β.ί ()

a <- af {a) where f {a) is a decreasing function of, for example f () = b - a Jcc, = 1.1, a = 1.2, f (oc) limited from 0.3 to 1. It should be noted that after multiplication by f (cc ), a ² + β ² <1 so that the signal energy U _HB2 (k) = βυ _HB1 (k) + ocG _HBN U _HBN (k) is lower than the energy of U _HB1 (k) ( the energy difference depends on a, the more noise is added, the more the energy is attenuated).

 In other variants of the invention, it will be possible to take:

β = \ - α

which preserves the amplitude level (when the combined signals are of the same sign); however, this variant has the disadvantage of resulting in a global energy (at U _HB2 (k)) which is not monotonic as a function of.

 It should therefore be noted here that the block 503 realizes the equivalent of the block 101 of FIG. 1 to normalize the white noise as a function of an excitation which is on the other hand here in the frequency domain, already extended at the rate of 16 kHz; in addition, the mix is limited to the band 6000-8000 Hz.

In a simple variant, it is possible to consider an embodiment of block 503, where the spectra, U _HB1 (k) or G _HBN U _HBN (k), are selected (switched) adaptively, which amounts to allowing only the values 0 or 1 for a; this approach amounts to classifying the type of excitation to be generated in the 6000-8000 Hz band

 The block 504 optionally carries out a dual operation of application of bandpass filter frequency response and deemphasis filtering (or deemphasis) in the frequency domain.

 In a variant of the invention, the deemphasis filtering may be performed in the time domain, after block 505 or even before block 500; however, in this case, bandpass filtering performed in block 504 may leave some low frequency components of very low levels which are amplified by de-emphasis, which may slightly discern the decoded low band. For this reason, it is preferred here to perform the deemphasis in the frequency domain. In the preferred embodiment, the index coefficients & = 0, · · ·, 199 are set to zero, so the deemphasis is limited to the higher coefficients.

 The excitation is first de-emphasized according to the following equation:

U _{HB 2} 55

19 where G _{deem ≠} (k) is the frequency response of the filter l / (l - 0.68z ^_1 ) over a restricted discrete frequency band. Taking into account the discrete (odd) frequencies of the DCT-IV, we define here G _{deem ≠} (k) as:

G _{→ h} (k) = ¹ , fc = 0.

e ^k - 0.68

or

256 -80 + £ + -

In the case where a transformation other than the DCT-IV is used, the definition of 6 _k can be adjusted (for example for even frequencies).

It is noted that the de-emphasis is applied in two phases for k = 200, · · -, 255 corresponding to the frequency band 5000-6400 Hz, where the response l / (l - 0.68z ^_1 ) is applied as at 12.8 kHz, and for k = 256, · · -, 319 corresponding to the 6400-8000 Hz frequency band, where the response is extended from 16 kHz here to a constant value in the 6.4-8 kHz band.

 It may be noted that in the AMR-WB coding the HF synthesis is not de-emphasized. In the embodiment presented here, the high frequency signal is on the contrary de-emphasized so as to bring it back to a domain coherent with the low frequency signal (0-6.4 kHz) coming out of block 305. This is important for the estimation and the subsequent adjustment of the energy of the HF synthesis.

In a variant of the embodiment, in order to reduce the complexity, it is possible to fix G _deemph (k) at a constant value independent of k, for example by taking

G _deemph (k) = 0.6 which corresponds approximately to the average value of G _deemph (k) for k = 200, · · -, 319 under the conditions of the embodiment described above.

 In another variant of the embodiment of the extension device, the de-emphasis can be performed in an equivalent way in the time domain after inverse DCT. Such an embodiment is implemented in Figure 7 described below.

 In addition to de-emphasis, band-pass filtering is applied with two separate parts: one fixed high-pass, the other adaptive low-pass (flow-rate function).

This filtering is carried out in the frequency domain, and its frequency response is illustrated in FIG. 6. The 3 dB cutoff frequencies are 6000 Hz for the low part and for the high part approximately 6900, 7300, 7600 Hz at 6.6, 8.86 and at rates greater than 8.85 kbit / s (respectively). In the preferred embodiment, the partial low-pass filter response in the frequency domain is calculated as follows:

G _ln (k) = 1-0.999 ^k where N _ln = 60 to 6.6 kbit / s, 40 to 8.85 kbit / s, at rates> 8.85 bit / s.

 Then we apply a bandpass filter in the form:

 0 £ = 0, ···, 199

G _hp (k-200) _{HB HB2} \ k) = = 200, ···, 255

U _HB3 (k)

U HB2 ()) = = 256, ---, 319-N, _r

[G _lp (k -320- N _lv) U _HB2 \ k) * = _320-lp N, -, 319

The definition of G _h (k), k = 0, ···, 55, is given for example in Table 1 below.

 Table 1

It will be noted that in variants of the invention the values of G _hp (k) may be modified while keeping a gradual attenuation. Similarly, the variable bandwidth low-pass filtering, G _lp (k), may be adjusted with different values or frequency support, without changing the principle of this filtering step.

It will also be noted that the bandpass filtering example illustrated in FIG. 6 can be adapted by defining a single filtering step combining the high-pass and low-pass filterings. In another embodiment, the bandpass filtering may be performed equivalently in the time domain (as in block 112 of FIG. 1) with different filter coefficients according to the bit rate, after an inverse DCT step. Such an embodiment is implemented in Figure 7 described below. However, it will be noted that it is advantageous to carry out this step directly in the frequency domain because the filtering is carried out in the field of LPC excitation and therefore the problems of circular convolution and edge effects are very limited in this field. .

The inverse transform block 505 performs an inverse DCT on 320 samples to find the high frequency excitation sampled at 16 kHz. Its implementation is identical to block 500, because the DCT-IV is orthonormed, except that the length of the transform is 320 instead of 256, and we obtain:

where N _l6k = 320 and & = 0, · · ·, 319.

This excitation sampled at 16 kHz is then optionally scaled by gains defined by subframe of 80 samples (block 507).

In a preferred embodiment, a gain g _H Bi (m) per subframe is first calculated (block 506) by energy ratios of the subframes such that in each subframe the frame common:

or

 63

Σu (n + 64m) ² +

 79

e ₂ (m) Σu _HB (n + 80m) ² + ε

 319

Σu _HB (n) ² +

e ₃ (m) 255

u (n) ² + £ with ε = 0.01. We can write the gain by _subframe g _HBl (m) in the form:

which shows that the signal u _{HB has} the same ratio between energy per subframe and energy per frame as in the signal u (n).

 Block 507 scales the combined (or extended) signal (step E406 of FIG. 4) according to the following equation:

^U _HB ^N) = S _HBI ^{(M) U} _HB = ⁸ ° m, · · ·, 80 (m + 1) - 1

 Note that the embodiment of the block 506 differs from that of the block 101 of Figure 1, because the energy at the current frame is taken into account in addition to that of the subframe. This makes it possible to have the ratio of the energy of each sub-frame with respect to the energy of the frame. Energy ratios (or relative energies) are compared rather than the absolute energies between low band and high band.

 Thus, this scaling step makes it possible to keep in the high band the energy ratio between the subframe and the frame in the same way as in the low band.

 Optionally, block 509 then scales the signal (step E407 of FIG. 4) according to the following equation:

^U _HB "(") = S _HBI ( ^M ) ^U _HB '(") / ^N = ⁸ ° m, · · ·, 80 (m + 1) - 1

where the gain g _HB2 (m) is obtained from the block 508 by executing the blocks 103, 104 and 105 of the AMR-WB coding (the input of the block 103 being the decoded excitation in the low band, u (n)) . Blocks 508 and 509 are useful for adjusting the level of the LPC synthesis filter (block 510), here depending on the tilt of the signal. Other methods of calculating the gain g _HB2 (m) are possible without changing the nature of the invention.

Finally, the excitation, u _HB n) or u _HB "(n), is filtered (step E404 of FIG. 4) by the filtering module 510 which can be realized here by taking as transfer function 1 / A (zJ Y), where ^ = 0.9 to 6.6 kbit / s and ^ = 0.6 at other rates, which limits the order of the filter to order 16.

In a variant, this filtering can be done in the same way as described for the block 111 of FIG. 1 of the AMR-WB decoder, however the order of the filter goes to 20 at the rate of 6.6, which does not change. not significantly the quality of the synthesized signal. In another variant, it will be possible to perform LPC synthesis filtering in the domain frequency, after calculating the frequency response of the filter implemented in block 510.

 In alternative embodiments of the invention, the coding of the low band (0-6.4 kHz) may be replaced by a CELP coder other than that used in AMR-WB, for example the CELP coder in G.718 to 8. kbit / s. Without loss of generality other encoders in wide band or operating at frequencies higher than 16 kHz, in which the coding of the low band operates at an internal frequency at 12.8 kHz could be used. Moreover, the invention can be obviously adapted to other sampling frequencies than 12.8 kHz, when a low frequency encoder operates at a sampling frequency lower than that of the original or reconstructed signal. When the low band decoding does not use a linear prediction, it does not have an excitation signal to be extended, in this case it will be possible to carry out an LPC analysis of the reconstructed signal in the current frame and calculate an LPC excitation. so as to be able to apply the invention.

 Finally, in another variant of the invention, the excitation (u (n)) is resampled, for example by linear interpolation or cubic "spline", from 12.8 to 16 kHz before transformation (for example DCT-IV) of length 320. This variant has the defect of being more complex, because the transform (DCT-IV) of the excitation is then calculated over a greater length and the resampling is not carried out in the field of the transform.

Moreover, in variants of the invention, all the calculations necessary for the estimation of the gains {G _HBN , g _HB1 (m), g _HB2 (m), g _HBN , ...) can be carried out in a field logarithmic.

 Referring to Fig. 7, a second embodiment of the tape extender is now described. This embodiment operates in the time domain.

 As in the embodiment of FIG. 5, the principle of the embodiment with mixing of a signal extended to 16 kHz and of a noise signal is retained, however this mixture is this time realized in the time domain and this once the main generation of the excitation is done by subframe and not by frame.

The excitation signal u (n), η = 0, · · ·, 255, resulting from the low frequency decoding in the current frame is first resampled without delay (step E401 of FIG. 4) at 16 kHz ( block 700) and in a particular embodiment, a linear interpolation is used to obtain the extended excitation signal in a second frequency band, u _ext (n), η = 0, · · ·, 319. In an alternative embodiment, it will be possible to use other resampling methods, for example by "splines" or by multi-rate filtering.

It is ensured that the energy of the signal u _ext (n) has a level similar to the excitation u (n) with the blocks 701 and 702 as follows:

In an alternative embodiment it is possible to multiply u ' _ext (n) by 5/4 to compensate the attenuation by the ratio 12.8 / 16, caused by different signal sampling frequencies u _ext (n) and u (n).

The noise generator in the block 703 implements the step E402 of FIG. 4 and can be made as in the block 502 described in FIG. 5, except that the output signal corresponds to a temporal _subframe , u _HBN ( _FIG. n), η = 0, · · ·, 319

 The combination block 704 can be realized in different ways. In a privileged way, we consider an adaptive additive mix by subframe of the form:

u _HB1 (n + 80m) = βιι _{€ Χί} (n + 80m) + ag _HBN u _HBN (n + 80m), n = 0, · · ·, 79

where § ^{is a} f ^ac t ^eur Standards for equalizing the level of harmonicity of the two combined signals

+ £

 k = 0

 > HBN

^U HB _N ( ^{n + ε}

 k = 0 and m is the index of the subframe and the factors and β are calculated as in the first embodiment. It should be noted here that the block 704 realizes the equivalent of the block 101 of FIG. 1. Moreover, the calculation of the factor obliges to calculate the transform of the decoded excitation signal (or the decoded signal itself according to the domain of FIG. calculation of the relative level of noise or flatness spectral flatness "spectral flatness" in English) in low band if this calculation rests on the spectral flatness; in variants, including the use of a linear regression described above, such a transform is not necessary.

Then the time signal is deemphasized (block 705) by a filter of the form ^^ / (! -O- ⁶⁸ ^ ¹ ) ^where 8 deemph ^{is calculated so as to} extend the filter 1 / (l - 0.68z ^_1 ) ( defined at 12.8 kHz) at the sampling frequency of 16 kHz = | (l - 0.68e ^ ² " ^6000/16000 ) / (l - o.68 _e ^ ² " ^6000/12800 ) | , then processed by bandwidth filtering variable bandwidth (block 706) whose order is fixed (value 30) but the coefficients change depending on the decoding rate of the current frame. An exemplary embodiment of such type FIR adaptive bandpass filtering is given in the tables below defining the impulse response of the FIR filter according to the bit rate.

 Table 2c (bitrates> 8.85 kbit / s)

The scaling step (E407 in FIG. 4) is performed by blocks 508 and 509 identical to FIG. The filtering step (E404 of FIG. 4) is performed by the filtering module (block 510) identical to that described with reference to FIG. 5.

 It is not useful here to implement a scaling step as performed in the embodiment of Fig. 5 by blocks 506 and 507 since the excitation is generated by subframes. The coherence of the energy ratio at the level of the frame is already assured.

 In variations of the invention, the low band excitation u (n) and the LPC 1 / λ filter (z) will be estimated per frame, by LPC analysis of a low band signal whose band must be extended. The low band excitation signal is then extracted by analyzing the audio signal.

 In a possible embodiment of this variant, the low band audio signal is resampled before the excitation extraction step, so that the excitation extracted from the audio signal (by linear prediction) is already resolved. sampled.

 The invention illustrated in FIG. 5, or alternatively in FIG. 7, applies in this case to a low band which is not decoded but analyzed.

FIG. 8 represents an exemplary hardware embodiment of a band extension device 800 according to the invention. This may be an integral part of an audio-frequency signal decoder or equipment receiving decoded or non-decoded audio signals.

 This type of device comprises a PROC processor cooperating with a memory block

BM having a memory storage and / or working MEM.

 Such a device comprises an input module E able to receive a decoded or extracted excitation audio signal in a first so-called low band frequency band (u (n) or U (k)) and the parameters of a filter of linear prediction synthesis (A (z)). It comprises an output module S adapted to transmit the synthesized high frequency signal (HF_syn) for example to a delay application module such as block 310 of FIG. 3 or to a resampling module such as module 311 .

The memory block may advantageously comprise a computer program comprising code instructions for implementing the steps of the band extension method in the sense of the invention, when these instructions are executed by the processor PROC, and in particular the steps for obtaining an extended signal in at least a second frequency band greater than the first frequency band from an oversampled and extended excitation signal in at least a second frequency band, scaling of the signal extended by a subframe defined gain based on a frame and subframe energy ratio and filtering said extended signal scaled by a linear prediction filter whose coefficients are derived from the coefficients of the low band filter. Typically, the description of FIG. 4 repeats the steps of an algorithm of such a computer program. The computer program can also be stored on a memory medium readable by a reader of the device or downloadable in the memory space thereof.

 The memory MEM generally records all the data necessary for the implementation of the method.

 In a possible embodiment, the device thus described may also comprise the low band decoding functions and other processing functions described for example in FIG. 3 in addition to the band extension functions according to the invention.

Claims

A method of extending the frequency band of an audiofrequency signal during a decoding or improvement process comprising a decoding or extraction step, in a first so-called low band frequency band, of a signal of excitation and coefficients of a linear prediction filter, the method being characterized in that it comprises the following steps:

obtaining an extended signal (U _H B2 (k), E403) in at least one second frequency band greater than the first frequency band from an oversampled and extended excitation signal in at least a second frequency band (U _H Bi (k), E401);

 scaling (E406) the extended signal by a gain defined by subframe according to a ratio of energy of a frame and a subframe;

 filtering (E404) said extended signal scaled by a linear prediction filter whose coefficients are derived from the coefficients of the low band filter.

Method according to claim 1, characterized in that it further comprises an adaptive bandpass filtering step (E405) as a function of the decoding rate of the current frame.

Method according to Claim 1, characterized in that it comprises a step of transforming the time-frequency of the excitation signal, the step of obtaining an extended signal then taking place in the frequency domain and a transforming step inverse time-frequency of the extended signal before the scaling and filtering steps.

Method according to claim 3, characterized in that the step of generating an oversampled and extended excitation signal is performed according to the following equation:

U _HBl (k) =

with k the index of the sample, U _H BI (k) the spectrum of the extended excitation signal, U (k) the spectrum of the excitation signal obtained after the transforming step and start_band a predefined variable.

5. Method according to one of claims 1 to 4, characterized in that it comprises a de-emphasis filtering step of the extended signal at least in the second frequency band.

6. Method according to claim 1, characterized in that it further comprises a step of generating (E402) a noise signal at least in the second frequency band the extended signal (U _H B2 (k)) being obtained by combining (E403) the extended excitation signal and the noise signal.

7. Method according to claim 6, characterized in that the combining step is carried out by adaptive additive mixing with a level equalization gain between the extended excitation signal and the noise signal.

8. A frequency band extension device for an audiofrequency signal comprising a decoding or extraction stage, in a first so-called low-band frequency band, of an excitation signal and coefficients of a radio frequency filter. linear prediction, the device being characterized in that it comprises:

a module for obtaining an extended signal (U _H B2 (k), 503) in at least a second frequency band greater than the first frequency band from an oversampled excitation signal and extended in at least a second frequency band (U _H Bi (k));

 - a scaling module (507) of the signal extended by a gain defined by subframe according to a ratio of energy per frame and subframe of the audiofrequency signal in the first frequency band;

 a filtering module (510) of said extended signal scaled by a linear prediction filter whose coefficients are derived from the coefficients of the low band filter.

9. Audio frequency signal decoder characterized in that it comprises a frequency band extension device according to claim 8.

Computer program comprising code instructions for implementing the steps of the frequency band extension method according to one of claims 1 to 7, when these instructions are executed by a processor.

11. Storage medium readable by a frequency band extension device on which is recorded a computer program comprising code instructions for performing the steps of the frequency band extension method according to one of claims 1 to 1. at 7.