WO1998057436A2 - Source coding enhancement using spectral-band replication - Google Patents

Abstract

The present invention proposes a new method and apparatus for the enhancement of source coding systems. The invention employs bandwidth reduction (101) prior to or in the encoder (103), followed by spectral-band replication (105) at the decoder (107). This is accomplished by the use of new transposition methods, in combination with spectral envelope adjustments. Reduced bitrate at a given perceptual quality or an improved perceptual quality at a given bitrate is offered. The invention is preferably integrated in a hardware or software codec, but can also be implemented as a separate processor in combination with a codec. The invention offers substantial improvements practically independent of codec type and technological progress.

Description

SOURCE CODING ENHANCEMENT USING SPECTRAL-BAND REPLICATION

TECHNICAL FIELD

In source coding systems, digital data is compressed before transmission or storage to reduce the required bitrate or storing capacity The present mvention relates to a new method and apparatus for the improvement of source coding systems by means of Spectral Band Replication (SBR) Substantial bitrate reduction is achieved while maintaining the same perceptual quality or conversely, an improvement in perceptual quality is achieved at a given bitrate This is accomplished by means of spectral bandwidth reduction at the encoder side and subsequent spectral band replication at the decoder, whereby the mvention exploits new concepts of signal redundancy in the spectral domain

BACKGROUND OF THE INVENTION

Audio source coding techniques can be divided into two classes natural audio coding and speech coding Natural audio coding is commonly used for music or arbitrary signals at medium bitrates, and generally offers wide audio bandwidth Speech coders are basically limited to speech reproduction but can on the other hand be used at very low bitrates, albeit with low audio bandwidth Wideband speech offers a major subjective quality improvement over narrow band speech Increasing the bandwidth not only improves intelligibility and naturalness of speech, but also facilitates speaker recognition Wideband speech coding is thus an important issue in next generation telephone systems Further, due to the tremendous growth of the multimedia field, transmission of music and other non-speech signals at high quality over telephone systems is a desirable feature

A high-fidelity lmear PCM signal is very inefficient in terms of bitrate versus the perceptual entropy The CD standard dictates 44 1 kHz sampling frequency, 16 bits per sample resolution and stereo This equals a bitrate of 1411 kbit/s To drastically reduce the bitrate, source codmg can be performed usmg split-band perceptual audio codecs These natural audio codecs exploit perceptual irrelevancy and statistical redundancy in the signal Using the best codec technology, approximately 90% data reduction can be achieved for a standard CD-format signal with practically no perceptible degradation Very high sound quality in stereo is thus possible at around 96 kbit/s, I e a compression factor of approximately 15 1 Some perceptual codecs offer even higher compression ratios To achieve this, it is common to reduce the sample-rate and thus the audio bandwidth It is also common to decrease the number of quantization levels, allowing occasionally audible quantization distortion, and to employ degradation of the stereo field, through intensity coding Excessive use of such methods results m annoying perceptual degradation Current codec technology is near saturation and further progress m coding ga is not expected In order to improve the coding performance further, a new approach is necessary

The human voice and most musical instruments generate quasistationary signals that emerge from oscillating systems Accordmg to Founer theory, any periodic signal may be expressed as a sum of sinusoids with the frequencies If, 3 4/j 5/etc where/is the fundamental frequency The frequencies form a harmonic senes A bandwidth limitation of such a signal is equivalent to a truncation of the harmonic series Such a truncation alters the perceived timbre, tone colour, of a musical instrument or voice, and yields an audio signal that will sound "muffled" or "dull", and intelligibility may be reduced The high frequencies are thus important for the sub_jective impression of sound quality Pπor art methods are mainly mtended for improvement of speech codec performance and in particular intended for High Frequency Regeneration (HFR), an issue m speech coding Such methods employ broadband lmear frequency shifts, non-linearities or aliasing [U S Pat 5,127,054] generating intermodulation products or other non-harmonic frequency components which cause severe dissonance when applied to music signals Such dissonance is referred to m the speech codmg literature as "harsh" and "rough" sounding Other synthetic speech HFR methods generate sinusoidal harmonics that are based on fundamental pitch estimation and are thus limited to tonal stationary sounds [U S Pat 4,771,465] Such pπor art methods, although useful for low-quality speech applications, do not work for high quality speech or music signals A few methods attempt to improve the performance of high quality audio source codecs One uses synthetic noise signals generated at the decoder to substitute noise-like signals in speech or music previously discarded by the encoder ["Improving Audio Codecs by Noise Substitution" D Schultz, JAES, Vol 44, No 7/8, 1996] This is performed withm an otherwise normally transmitted highband at an intermittent basis when noise signals are present Another method recreates some missing highband harmonics that were lost in the coding process ["Audio Spectral Coder" A J S Ferrerra, AES Prepnnt 4201, 100^th Convention, May 11-14 1996, Copenhagen] and is agam dependent on tonal signals and pitch detection Both methods operate at a low duty- cycle basis offering comparatively limited codmg or performance gain

SUMMARY OF THE INVENTION

The present mvention provides a new method and an apparatus for substantial improvements of digital source coding systems and more specifically for the improvements of audio codecs The objective includes bitrate reduction or improved perceptual quality or a combination thereof The invention is based on new methods exploiting harmomc redundancy, offenng the possibility to discard passbands of a signal pnor to transmission or storage No perceptual degradation is perceived if the decoder performs high quality spectral replication accordmg to the mvention The discarded bits represent the codmg gam at a fixed perceptual quality Alternatively, more bits can be allocated for encoding of the lowband information at a fixed bitrate, thereby achieving a higher perceptual quality

The present mvention postulates that a truncated harmomc senes can be extended based on the direct relation between lowband and highband spectral components This extended senes resembles the onginal in a perceptual sense if certain rules are followed First, the extrapolated spectral components must be harmonically related to the truncated harmomc senes, m order to avoid dissonance-related artifacts The present mvention uses transposition as a means for the spectral replication process, which ensures that this cnteπon is met It is however not necessary that the lowband spectral components form a harmomc senes for successful operation, since new replicated components, harmonically related to those of the lowband, will not alter the noise-like or transient nature of the signal A transposition is defined as a transfer of partials from one position to another on the musical scale while maintaining the frequency ratios of the partials Second, the spectral envelope, I e the coarse spectral distribution, of the replicated highband, must reasonably well resemble that of the onginal signal The present mvention offers two modes of operation, SBR-1 and SBR-2, that differ m the way the spectral envelope is adjusted

SBR-1, mtended for the improvement of intermediate quality codec applications, is a single-ended process which relies exclusively on the information contamed m a received lowband or lowpass signal at the decoder The spectral envelope of this signal is determined and extrapolated, for instance using polynomials together with a set of rules or a codebook. This information is used to continuously adjust and equalise the replicated highband. The present SBR- 1 method offers the advantage of post-processing, i.e. no modifications are needed at the encoder side. A broadcaster will gain in channel utilisation or will be able to offer improved perceptual quality or a combination of both. Existing bitstream syntax and standards can be used without modification.

SBR-2, intended for the improvement of high quality codec applications, is a double-ended process where, in addition to the transmitted lowband signal according to SBR-1, the spectral envelope of the highband is encoded and transmitted. Since the variations of the spectral envelope has a much lower rate than the highband signal components, only a limited amount of information needs to be transmitted in order to successfully represent the spectral envelope. SBR-2 can be used to improve the performance of current codec technologies with no or minor modifications of existing syntax or protocols, and as a valuable tool for future codec development.

Both SBR-1 and SBR-2 can be used to replicate smaller passbands of the lowband when such bands are shut down by the encoder as stipulated by the psychoacoustic model under bit-starved conditions. This results in improvement of the perceptual quality by spectral replication within the lowband in addition to spectral replication outside the lowband. Further, SBR-1 and SBR-2 can also be used in codecs employing bitrate scalability, where the perceptual quality of the signal at the receiver varies depending on transmission channel conditions. This usually implies annoying variations of the audio bandwidth at the receiver. Under such conditions, the SBR methods can be used successfully in order to maintain a constantly high bandwidth, again improving the perceptual quality.

The present invention operates on a continuous basis, replicating any type of signal contents, i.e. tonal or non-tonal (noise-like and transient signals). In addition, the present spectral replication method creates a perceptually accurate replica of the discarded bands from available frequency bands at the decoder. Hence, the SBR method offers a substantially higher level of coding gain or perceptual quality improvement compared to prior art methods. The invention can be combined with such prior art codec improvement methods; however, no performance gain is expected due to such combinations.

The SBR-method comprises the following steps:

- encoding of a signal derived from an original signal, where frequency bands of the signal are discarded and the discarding is performed prior to or during encoding, forming a first signal, during or after decoding of the first signal, transposing frequency bands of the first signal, forming a second signal,

- performing spectral envelope adjustment, and

- combining the decoded signal and the second signal, forming an output signal.

The passbands of the second signal may be set not to overlap or partly overlap the passbands of the first signal, and may be set in dependence of the temporal characteristics of the original signal and/or the first signal, or transmission channel conditions. The spectral envelope adjustment is performed based on estimation of the original spectral envelope from said first signal or on transmitted envelope information of the original signal. The present invention includes to basic types of transposers: multiband transposers and time-variant pattern search prediction transposers, having different properties. A basic multiband transposition may be performed according to the present invention by the following: filtering the signal to be transposed through a set of N ≥ 2 bandpass filters with passbands comprising the frequencies \f_\,...JΗ] respectively, forming bandpass signals, shifting the bandpass signals in frequency to regions comprising the frequencies M [f ...,f_N] where M ≠ 1 is the transposition factor, and

- combining the shifted bandpass signals, forming the transposed signal.

Alternatively, this basic multiband transposition may be performed according to the invention by the following: - bandpass filtering the signal to be transposed signal using an analysis filterbank or transform of such a nature that real- or complex-valued subband signals of lowpass type are generated,

- an arbitrary number of channels k of said analysis filterbank or transform are connected to channels Mk, M ≠ 1, in a synthesis filterbank or transform, and

- the transposed signal is formed using the synthesis filterbank or transform. An improved multiband transposition according to the invention incorporates phase adjustments, enhancing the performance of the basic multiband transposition.

The time-variant pattern search prediction transposition according to the present invention may be performed by the following: - performing transient detection on the first signal, determining which segment of the first signal to be used when duplicating/discarding parts of the first signal depending on the outcome of the transient detection, adjusting statevector and codebook properties depending on the outcome of the transient detection, and

- searching for synchronisation points in chosen segment of the first signal, based on the synchronisation point found in the previous synchronisation point search.

The SBR methods and apparatuses according to the present invention offer die following features:

1. The methods and apparatuses exploit new concepts of signal redundancy in the spectral domain.

2. The methods and apparatuses are applicable on arbitrary signals. 3. Each harmomc set is individually created and controlled.

4. All replicated harmonics are generated in such a manner as to form a continuation of the existing harmonic series.

5. The spectral replication process is based on transposition and creates no or imperceptible artifacts.

6. The spectral replication can cover multiple smaller bands and/or a wide frequency range. 7. In the SBR-1 method, the processing is performed at the decoder side only, i.e. all standards and protocols can be used without modification.

8. The SBR-2 method can be implemented in accordance with most standards and protocols with no or minor modifications.

9. The SBR-2 method offers the codec designer a new powerful compression tool. 10. The coding gain is significant. The most attractive application relates to the improvement of vanous types of low bitrate codecs, such as MPEG 1/2 Layer I/II/III [U S Pat 5,040,217], MPEG 2/4 AAC, Dolby AC-2/3, NTT TwinVQ [U S Pat 5,684,920], AT&T/Lucent PAC etc The mvention is also useful m high-quality speech codecs such as wide-band CELP and SB-ADPCM G 722 etc to improve perceived quality The above codecs are widely used in multimedia, m the telephone industry, on the Internet as well as in professional applications T-DAB (Terrestnal Digital Audio

Broadcastmg) systems use low bitrate protocols that will gain in channel utilisation by using the present method, or improve quality m FM and AM DAB Satellite S-DAB will gam considerably, due to the excessive system costs mvolved, by using the present method to increase the number of programme channels m the DAB multiplex Furthermore, for the first tune, full bandwidth audio real-time streaming over the Internet is achievable using low bitrate telephone modems

BRIEF DESCRIPTION OF THE DRAWINGS

The present mvention will now be described by way of illustrative examples, not limiting the scope or sprπt of the mvention, with reference to the accompanying drawings, in which Fig 1 illustrates SBR incorporated in a codmg system according to the present invention,

Fig 2 illustrates spectral replication of upper harmonics according to the present invention,

Fig 3 illustrates spectral replication of inband harmonics according to the present invention,

Fig 4 is a block diagram for a time-domain implementation of a transposer accordmg to the present mvention,

Fig 5 is a flow-chart representing a cycle of operation for the pattern-search prediction transposer according to the present mvention,

Fig 6 is a flow-chart representing the search for synchronisation point according to the present invention,

Fig 7a - 7b illustrates the codebook positioning dunng transients accordmg to the present mvention,

Fig 8 is a block diagram for an implementation of several time-domam transposers in connection with a suitable filterbank, for SBR operation accordmg to the present invention, Fig 9a - 9c are block diagrams representing a device for STFT analysis and synthesis configured for generation of 2^nd order harmonics accordmg to the present mvention,

Fig 10a - 10b are block diagrams of one sub-band with a lmear frequency shift in the STFT device according to the present mvention,

Fig 11 shows one sub-band usmg a phase-multiplier according to the present invention, Fig 12 illustrates how 3^rd order harmonics are generated according to the present invention,

Fig 13 illustrates how 2^nd and 3^rd order harmonics are generated simultaneously accordmg to the present mvention,

Fig 14 illustrates generation of a non-overlappmg combination of several harmonic orders according to the present mvention, Fig 15 illustrates generation of an mterleaved combination of several harmomc orders accordmg to the present mvention,

Fig 16 illustrates generation of broadband lmear frequency shifts,

Fig 17 illustrates how sub-harmonics are generated accordmg to the present invention,

Fig 18a - 18b are block diagrams of a perceptual codec, Fig 19 shows a basic structure of a maximally decimated filterbank, Fig. 20 illustrates generation of 2^nd order harmonics in a maximally decimated filterbank according to the present invention;

Fig. 21 is a block diagram for the improved multiband transposition in a maximally decimated filterbank operating on subband signals according to the present invention; Fig. 22 is a flow-chart representing the improved multiband transposition in a maximally decimated filterbank operating on subband signals according to the present invention;

Fig. 23 illustrates subband samples and scalefactors of a typical codec;

Fig. 24 illustrates subband samples and envelope information for SBR-2 according to the present invention;

Fig. 25 illustrates hidden transmission of envelope information in SBR-2 according to the present invention; Fig. 26 illustrates redundancy coding in SBR-2 according to the present invention;

Fig. 27 illustrates an implementation of a codec using the SBR-1 method according to the present invention; and

Fig. 28 illustrates an implementation of a coded using the SBR-2 method according to the present invention;

Fig. 29 is a block diagram of a "pseudo-stereo" generator according to the present invention.

DESCRIPTION OF PREFERRED EMBODIMENTS

Throughout the explanation of the embodiments herein, emphasis is given to natural audio source coding applications. However, it should be understood that the present invention is applicable on a range of source coding applications other than that of encoding and decoding audio signals.

Transposition basics

Transposition as defined according to the present invention, is the ideal method for spectral replication, and has several major advantages over prior art, such as: no pitch detection is required, equally high performance for single- pitched and polyphonic programme material is obtained, and the transposition works equally well for tonal and non- tonal signals. Contrary to other methods, the transposition according to the invention can be used in arbitrary audio source coding systems for arbitrary signal types.

An exact transposition a factor M of a discrete time signal x(n) in the form of a sum of cosines with time varying amplitudes, is defined by the relation

JV-l x(n) = ∑e_l (n) cos(2πf,n/f_s +a, ) → (1)

1=0

y(n) = ∑e. («) cosQaMf. n l f, +β, ) (2) ι=0 where N is the number of sinusoids, hereafter refeπed to as partials, f„ e,(n), a, are the individual input frequencies, time envelopes and phase constants respectively, β, are the arbitrary output phase constants and^ is the sampling frequency, and 0 ≤Mf, ≤fJ2.

In Fig. 2, the generation of At* order harmonics, where M is an integer > 2, is shown. The term Λ ¹ order harmonics is used for simplicity, albeit the process generates A/* order harmonics to all signals in a certain frequency region, which in most cases are themselves harmonics of unknown order. The input signal with the frequency domain representation^/) is bandlimited to the range 0 tof_max, 201. The signal contents in the range f_max IM to Of_malM, where Q is the desired bandwidth expansion factor 1 < Q ≤M, is extracted by means of a bandpass filter, forming a bandpass signal with spectrum _Bp(/) 203. The bandpass signal is transposed a factor M, forming a second bandpass signal with spectrum Xj f) covering the range f_max to Qf_maχ, 205. The spectral envelope of this signal is adjusted by means of a programme-controlled equaliser, forming a signal with spectrum X_E(f) 207. This signal is then combined with a delayed version of the input signal in order to compensate for the delay imposed by the bandpass filter and transposer, whereby an output signal with spectrum Y(f) covering the range 0 to Qfmax is formed 209. Alternatively, bandpass filtering may be performed after the transposition M, using cut-off frequencies /_max and Qf_m!a. By using multiple transposers. simultaneous generation of different harmonic orders is of course possible. The above scheme may also be used to "fill in" stopbands within the input signal, as shown in Fig. 3, where the input signal lias a stopband extending from/₀ to Qf₀301. A passband [f₀/M,Qf₀/M] is then extracted 303, transposed a factor Mto [ ό ,<2/o] 305, envelope adjusted 307 and combined with the delayed input signal forming the output signal with spectrum Y(f) 309.

An approximation of an exact transposition may be used. According to the present invention, the quality of such approximations is determined using dissonance theory. A criterion for dissonance is presented by Plomp ["Tonal Consonance and Critical Bandwidth" R. Plomp, W. J. M. Levelt JASA , Vol 38, 1965], and states that two partials are considered dissonant if the frequency difference is within approximately 5 to 50% of the bandwidth of the critical band in which the partials are situated. For reference, the critical bandwidth for a given frequency can be approximated by c2>(/) = 25 + 75 (l + 1.4 (-^-)² )⁰⁶⁹ (3)

1000 with/and cb in Hz. Further, Plomp states that the human auditory system can not discriminate two partials if they differ in frequency by approximately less than five percent of the critical bandwidth in which they are situated. The exact transposition in Eq. 2 is approximated by

y_apProX(n) = ∑e, («) cos(2 r( / ± Δ/ / /, + β. ) , (4)

(=0 where Af, is the deviation from the exact transposition. If the input partials form a harmonic series, a hypothesis of the invention states that the deviations from the harmonic series of the transposed partials must not exceed five percent of the critical bandwidth in which they are situated. This would explain why prior art methods give unsatisfactory "harsh" and "rough" results, since broad band linear frequency shifts yields a much larger deviation than acceptable. When prior art methods produce more than one partial for only one input partial, the partials must nevertheless be within the above stated deviation limit, as to be perceived as one partial. This again explains the poor results obtained with prior art methods using nonlinearities etc, since they produce intermodulation partials not within the limit of deviation.

When using the above transposition based method of spectral replication according to the present invention, the following important properties are achieved: - Normally, no frequency domain overlap occur between replicated harmonics and existing partials. - The replicated partials are harmonically related to the partials of the input signal and will not give rise to any annoying dissonance or artifacts.

- The spectral envelope of the replicated harmonics forms a smooth continuation of the input signal spectral envelope, perceptually matching the original envelope.

Transposition based on time-variant pattern search prediction

Various ways to design the required transposers exist. Typical time-domain implementations expand die signal in time by duplicating signal segments based on the pitch-period. This signal is subsequently read out at a different rate. Unfortunately such methods are strictly dependent on pitch-detection for accurate time splicing of the signal segments. Furthermore, the constraint to work on pitch-period based signal segments makes them sensitive to transients. Since the detected pitch-period can be much longer than the actual transient, the risk of duplicating the entire transient rather than just expanding it in time is obvious. Anotiier type of time domain algorithms obtains time expansion/compression of speech signals by utilising pattern search prediction of the output signal ["Pattern Search Prediction of Speech" R. Bogner, T. Li, Proc. ICASSP '89, Vol. 1, May 1989, "Time-Scale Modification of Speech based on a nonlinear Oscillator Model" G. Kubin, W. B. Kleijn, IEEE, 1994]. This is a form of granular synthesis, where the input signal is divided into small parts, granules, used to synthesise the output signal. This synthesis is usually done by performing coπelation of signal segments in order to determine the best splicing points. This means that the segments used to form the output signal are not dependent on die pitch period, and thus the non-trivial task of pitch detection is not required. Nevertheless, problems with rapidly changing signal amplitudes remain in tiiese methods, and high quahty transposition tends to raise high computational demands. However, an improved time- domain pitch shifter/transposer is now presented, where die use of transient detection and dynamic system parameters produces a more accurate transposition for high transposition factors during both stationary (tonal or non-tonal) and transient sounds, at a low computational cost.

Refeπing to the drawings wherein like numerals indicate like elements, tiiere is shown in Fig. 4 nine separate modules: a transient-detector 401, a window position adjuster 403, a codebook generator 405, a synchronisation signal selector 407, a synchronisation position memory 409, a minimum difference estimator 411, an output segment memory 413, a mix unit 415, and a down sampler 417. The input signal is fed to both the codebook generator 405 and the transient-detector 401. If a transient is detected, its position is sent to the window position module 403. This module stipulates die size and position of the window that is multiphed with die input signal when creating the codebook. The codebook generator 405 receives a synchronisation position from the sync select module 407, provided it has been connected to another transposer. If this synchronisation position is within the codebook, it is used and an output segment is produced. Otherwise the codebook is sent to the minimum difference estimator 411 which returns a new synchronisation position. The new output segment is windowed together with the previous output segment in the mix module 415 and subsequently down sampled in module 417.

In order to clarify the explanation, a state space representation is introduced. Here, the state vectors, or granules, represent the input and output signals. The input signal is represented by a statevector x(n): x(«) = [x(n), x(n - D), x(n - 2D) x(n - (N - Ϊ)D)] (5) which is obtained from N delayed samples of the input signal, where N is die dimension of the state vector and D is the delay between the mput samples used to build die vector The granular mapping yields the sample x(n) following each statevector x(n-l) This gives Eq 6, where a(.) is the mappmg

In die present method the granular mapping is used to determine the next output based on the former output, usmg a state transition codebook The codebook of lengtii L is continuously rebuilt containing die statevectors and the next sample following each statevector Each statevector is separated from its neighbour by K samples, this enables die system to adjust the time resolution depending on die characteπstics of die cuπently processed signal, where K equal to one represents the finest resolution The mput signal segment used to build die codebook is chosen based on die position of a possible transient and die synchromsation position in die previous codebook

This means that die mappmg a(«), theoretically, is evaluated for all transitions included in die codebook x(κ - ) x(n -L + l) x(n - L +K) x(n -L + K +\)

(7)

X(H - 1) x(ri

With this transition codebook, die new output y(n) is calculated by searching for die statevector m die codebook most similar to the current statevector y(n-l) This nearest-neighbour search is done by calculating the minimum difference and gives the new output sample y(n) = a(y(n -D) (8)

However, the system is not limited to work on a sample by sample basis, but is preferably operated on a segment by segment basis The new output segment is windowed and added, mixed, widi die previous output segment, and subsequently down sampled The pitch transposition factor is determined by the ratio of d e mput segment length represented by the codebook and d e output segment lengdi read out of the codebook

Returning to the drawings, m Fig 5 and Fig 6 flowcharts are presented, displaying the cycle of operation of the transposer In 501 die mput data is represented, a transient detection 503 is performed on a segment of the mput signal, t e search for transients is performed on a segment lengdi equal to the output segment lengdi If a transient is found 505, the position of the transient is stored 507 and die parameters L (representing the codebook length), K (representing die distance m samples between each statevector), and D (representing die delay between samples in each statevector) are adjusted 509 The position of the transient is compared to die position of the previous output segment 511, m order to determine whetiier the transient has been processed If so 513, the position of the codebook (window L), and die parameters K, L, and D are adjusted 515 After the necessary parameter adjustments, based on die outcome of the transient detection, die search for a new synchromsation, or splicing point takes place 517 This procedure is displayed m Fig 6 First a new synchronisation pomt is calculated based on die previous 601, accordmg to

Sync pos = Sync_pos_old + SM - S, (9) where Sync_pos and Sync JJOS old are the new and old synchromsation positions respectively, S is die lengdi of the mput segment being processed, and Mis die transposition factor This synchromsation pomt is used to compare die accuracy of the new splicing point with die accuracy of the old splicing point 603 If the match is as good as or better than die previous 605, tins new synchronisation point is returned 607 provided it is witiiin the codebook If not, a new synchromsation point is searched for in die loop 609 This is performed widi a siπulanty measure, m tius case a minimum difference function 611, however, it is also possible to use coπelation in the time- or frequency- domain If die position yields a better match than that of the previous position found 613 the synchromsation position is stored 615 When all positions are tned 617 die system returns 619 to die flowchart in Fig 5 The new synchromsation point obtamed is stored 519 and a new segment is read out from die codebook 521 starting at die given synchromsation pomt This segment is windowed and added to the previous 523, down sampled by the transposition factor 525, and stored in the output buffer 527

In Fig 7 the behaviour of the system under transient conditions regarding die position of die codebook is illustrated Pnor to the transient, die codebook 1 representing the mput segment 1 is positioned "to d e left" of segment 1 Coπelation segment 1 represents a part of the previous output and is used to find synchromsation pomt 1 in codebook 1 When die transient is detected, and die pomt of the transient is processed, die codebook is moved accordmg to Fig 7a and is stationary untd the mput segment currently being processed is once again "to die right" of the codebook This makes it impossible to duplicate die transient smce die system is not allowed to search for synchromsation pomts pπor to the transient

Most pitch transposers, or time expanders, based on pattern search prediction give satisfactory results for speech and single-pitched matenal However, their performance detenorates rapidly for high complexity signals, like music, in particular at large transposition factors The present mvention offers several solutions for unproved performance, therefore producing excellent results for any type of signal Contrary to other designs, die system is time-vanant and die system parameters are based on die properties of the mput signal, and die parameters used duπng die previous operation cycle The use of a transient detector controlling not only the codebook size and position, but also the properties of the statevectors included, is a very robust and computationally efficient method to avoid audible degradation duπng rapidly changing signal segments Furthermore, alteration of the length of the signal segment being processed, which would raise higher computational demands, is not required Also, die present mvention utilises a refined codebook search based on die results from the preceding search This means that contrary to an ordinary correlation of two signal segments, as is usually done in time-domain systems based on pattern search prediction, the most likely synchromsation positions are tned first mstead of trying all positions consecutively This new method for reducing die codebook search drastically reduces die computational complexity of the system Further, when usmg several transposers, synchromsation position information can be shared among die transposers for further reduction of the computational complexity, as shown m the following implementation

The time-domain transposers as explamed above are used to implement d e SBR-1 and SBR-2 systems according to die following, dlustrative but not limiting, example In Fig 8 three time expansion modules are used in order to generate second, tinrd and fourth order harmonics Smce, m this example, each time domam expansion /transposer works on a wideband signal, it is beneficial to adjust die spectral envelope of the source frequency range pπor to transposition, considering tiiat tiiere will be no means to do so after the transpositions, widiout adding a separate equaliser system The spectral envelope adjusters, 801, 803 and 805, each work on several filterbank channels The gam of each channel m die envelope adjusters must be set so that die sum, 813, 815, 817, at die output, after transposition, yields die desired spectral envelope The transposers 807, 809 and 811 are interconnected in order to share synchromsation position information This is based on die fact tiiat under certain conditions, a high correlation will occur between die synchromsation positions found m die codebook duπng coπelation m die separate transposmg umts Assume, as an example and agam not limiting the scope of the invention, die fourth order harmomc transposer works on a tune frame basis half of diat of die second order harmomc transposer but at twice the duty cycle Assume further, diat die codebooks used for die two expanders are the same and diat die synchromsation positions of the two time-domain expanders are labelled syncjpos4 and sync_pos2, respectively This yields the following relation sync _ pos2 = sync _ posA - n • 4 S - sync _offset , for n= 1,2,3 ,4 , (10) where sync _offset = sync _ posA - sync _ pos2 , for «=0, (11) and S is die lengdi of the mput segment represented by the codebook This is valid as long as neitiier of the synchromsation position pomters reaches the end of die codebook Duπng normal operation n is mcreased by one for each tune-frame processed by the second order harmonic transposer, and when die codebook end inevitably is reached, by either of die pomters, die counter n is set to «=0, and sync_pos2 and sync posA are computed individually Similar results are obtamed for the tinrd order harmomc transposer when connected to die fourth order harmomc transposer

The above-presented use of several interconnected time-domain transposers, for the creation of higher order harmonics, introduces substantial computational reduction Furthermore, die proposed use of time-domain transposers m connection with a suitable filterbank, presents the opportunity to adjust die envelope of die created spectrum wlule maintaining die simplicity and low computational cost of a time domam transposer, smce these, more or less, may be implemented using fixed point aπdimetic and solely additive/subtractive-operations

Other, illustrative but not limiting, examples of the present invention are

- the use of a time domam transposer within each subband m a subband filter bank, thus reducmg die signal complexity for each transposer the use of a time domam transposer in combination with a frequency domain transposer, thus enablmg die system to use different metiiods for transposition dependmg on the charactenstics of the mput signal being processed d e use of a time domam transposer in a wideband speech codec, operating on for instance the residual signal obtamed after lmear prediction It should be recogmsed diat die metiiod outiined above may be advantageously used for timescale modffication only, by simply omitting the sample rate conversion Further it is understood, diat altiiough die outlined mediod focuses on pitch transposmg to a higher pitch, l e time expansion, die same prmciples apply when transposmg to a lower pitch, l e time compression, as is obvious to those skilled m the art Filter bank based transposition

Various new and innovative filter bank based transposition techniques will now be described. The signal to be transposed is divided into a series of BP- or subband signals. The subband signals are then transposed, exact or approximately, which is advantageously accomplished by a reconnection of analysis- and synthesis subbands, hereinafter refeπed to as a "patch". The metiiod is first demonstrated using a Short Time Fourier Transform, STFT.

The N-point STFT of a discrete-time signal x(ri) is defined by

X_k (n) = ∑x(p) h(n - p) e ^]ω>y (12)

where k = 0,1,... N-l and ω^ 2π£/Λ^rand /2(w) is a window. If die window satisfies the following conditions

an inverse transform exists and is given by

. N-l

* = -77 ∑ * (O e^ya*^B - (14)

Jt=0

The direct transform may be interpreted as an analyser, see Fig. 9a, consisting of a bank of NBP-filters with impulse responses A(«)exp(/^'e%«) 901 followed by a bank of N multipliers witii carriers expi-jeotn) 903 which shift die BP- signals down to regions around 0 Hz, forming die N analysis signals Xk(n). The window acts as a prototype LP-filter. Xάμ) have small bandwidtiis and are normally downsampled 905. Eq. 12 need tiius only be evaluated at n = rR, where R is die decimation factor and r is the new time variable. Xy ) can be recovered lιoτa.XfrR) by upsampling, see Fig 9b, i.e. insertion of zeros 907 followed by LP-filtering 909. The inverse transform may be interpreted as a synthesiser consisting of a bank of N multipliers witii carriers ( l/TV)exp(/ fi% ri) 911 that shift the signals Xii ) up to tiieir original frequencies, followed by stages 913, Fig. 9c, that add die contributions y^ri) from all channels. The STFT and ISTFT may be reananged in order to use die DFT and IDFT, which makes the use of FFT algoritiims possible ["Implementation of die Phase Vocoder using the Fast Fourier Transform" M. R. Portnoff, IEEE ASSP, Vol. 24, No. 3, 1976].

Fig. 9c shows a patch 915 for generation of second harmonics, M = 2, with TV = 32. For the sake of simplicity, only channels 0 through 16 are shown. The centre frequency of BP 16 equals the Nyqvist frequency, channels 17 through 31 conespond to negative frequencies. The blocks denoted P 917 and die gain blocks 919 will be described later and should presendy be considered shorted out. The input signal is in this example bandlimited so that only channels 0 dirough 7 contain signals. Analyser channels 8 through 16 are thus empty and need not be mapped to die syndiesiser. Analyser channels 0 through 7 are connected to synthesiser channels 0 through 7, conesponding to an input signal delay path. Analysis channels k where 4 < k < 1 are also connected to syntiiesis channels Mk, M= 2, which shift die signals to frequency regions at two times die centre-frequencies of BP filters k. Hence, die signals are upshifted to tiieir original ranges as well as transposed one octave up. To explore the harmomc generation in terms of real-valued filter responses and modulators the negative frequencies must also be considered, see die lower branch of Fig. 10a. Hence, die combined output of the remapping k → k 1001 and N-k - N-Mk 1003 where 4< k <7 must be evaluated. This yields y(n) = — [x(n) * h(n)cos(ω_kn)]cos((M - \)ω_kn)) +

^N ₂ (15) [x(») * h(n)sin(ω_kn)]sin((M - \)ω_kn)

where M = 2. Eq. 15 may be interpreted as a BP-filtering of the input signal, followed by a linear frequency shift or Upper Side Band (USB) modulation, i.e. single side band modulation using die upper side band, see Fig. 10b, where 1005 and 1007 form a Hubert transformer, 1009 and 1011 are multipliers with cosine and sine carriers and 1013 is a difference stage which selects the upper sideband. Clearly, such a multiband BP and SSB method may be implemented explicitiy, i.e. witiiout filterbank patching, in the time or frequency domain, allowing arbitrary selection of individual passbands and oscillator frequencies.

According to Eq. 15, a sinusoid witii die frequency ω. within the passband of analysis channel k yields a harmonic at the frequency Mo%+(<2 - o^). Hence the method, refeπed to as basic multiband transposition, only generates exact harmonics for input signals with frequencies ω, =

where A ≤ k ≤ l. However, if die number of filters is sufficiently large, die deviation from an exact transposition is negligible, see Eq. 4. Further, the transposition is made exact for quasi-stationary tonal signals of arbitrary frequencies by inserting die blocks denoted P 917 (Fig. 9c), provided every analysis channel contains maximum one partial. In this case X_k(rR) are complex exponentials with frequencies equal to the differences between the partial frequencies ω, and the centre frequencies O_k of the analysis filters. To obtain the exact transposition M, these frequencies must be increased by a factor M, modifying die above frequency relationship to ω, -» Mω7-M((θι - <θk) - Mω,. The frequencies of X_k(rR) are equal to die time derivatives of their respective unwrapped phase angles and may be estimated using first order differences of successive phase angles. The frequency estimates are multiplied by M and syntiiesis phase angles are calculated using tiiose new frequencies. However, die same result, aside from a phase constant, is obtained in an simplified way by multiplying die analysis arguments by M directly, eliminating die need for frequency estimation. This is described in Fig. 11, representing the blocks 917. Thus X^rR), where 4 < k ≤ 7 in this example, are converted from rectangular to polar coordinates, illustrated by die blocks R → P, 1101. The arguments are multiplied by M = 2 1103 and die magnitudes are unaltered. The signals are then converted back to rectangular coordinates (P -> R) 1105 forming the signals Y_MkifR) and fed to syndiesiser channels according to Fig. 9c. This improved multiband transposition method thus has two stages: The patch provides a coarse transposition, as in the basic metiiod, and die phase-multipliers provide fine frequency corrections. The above multiband transposition methods differ from traditional pitch shifting techniques using the STFT, where lookup-table oscillators are used for die syntiiesis or, when die ISTFT is used for die synthesis the signal is time-stretched and decimated, i.e. no patch is used.

The harmonic patch of Fig. 9c is easϋy modified for other transposition factors than two. Fig. 12 shows a patch 1203 for generation of 3^rd order harmonics, where 1201 are the analysis channels and 1205 are die syntiiesis channels. Different harmonic orders may be created simultaneously as shown in Fig. 13, where 2^nd and 3^rd order harmonics are used. Fig. 14 illustrates a non-overlapping combination of 2^nd, 3^rd and 4^th order harmonics. The lowest possible harmonic number is used as high in frequency as possible. Above the upper limit of the destination range of harmonic M, harmonic M+l is used. Fig. 15 demonstrates a method of mapping all syndiesiser channels (N= 64, channels 0 - 32 shown). All highband channels with non prime-number indices are mapped according to die following relation between source and destmation channel number k_desl = Mk_∞urce where M is the smallest mteger >2 that satisfies the condition diat k_∞mce hes m die lowband and k^ m die highband Hence, no synthesiser channel receives signal from more than one analysis channel Prime-number highband channels may be mapped to k_source = 1 or lowband channels k_source > 1 diat yield good approximations of the above relation (Only non-prime number connections with M= 2, 3, 4, 5 are shown in Fig 15)

It is also possible to combine amplitude and phase information from different analyser channels The amplitude signals |AΑ /vft)| may be connected according to Fig 16, whereas die phase signals arg{X/ ,rR)} are connected accordmg to die principle of Fig 16 In this way the lowband frequencies will still be transposed, whereby a penodic repetition of die source region envelope is generated mstead of the stretched envelope that results from a transposition accordmg to Eq 2 Gatmg or otiier means may be incorporated in order to avoid amphfication of "empty" source channels Fig 17 illustrates another application, the generation of sub-harmonics to a lughpass filtered or bass limited signal by using connections from higher to lower subbands When using the above transpositions it may be beneficial to employ adaptive switching of patch based on die characteπstics of the signal

In the above descnption it was assumed diat die highest frequency contamed in die input signal was sigmficantly lower than the Nyqvist frequency Thus, it was possible to perform a bandwidtii expansion wtthout an mcrease m sample rate This is however not always die case, why a preceding upsampling may be necessary When using filter bank metiiods for transposition, it is possible to integrate upsampling in the process

Most perceptual codecs employ maximally decimated filter banks m die time to frequency mappmg ["Introduction to Perceptual Codmg" K Brandenburg, AES, Collected Papers on Digital Audio Bitrate Reduction, 1996] Fig 18a shows the basic structure of a perceptual encoder system The analysis filter bank 1801 sphts the mput signal mto several subband signals The subband samples are individually quantised 1803, using a reduced number of bits, where the number of quantization levels are determined from a perceptual model 1807 which estimates die minimum masking threshold The subband samples are normalised, coded with optional redundancy coding metiiods and combined witii side information consisting of die normalisation factors, bit-allocation information and otiier codec specrfic data 1805, to form the senal bit stream The bit stream is then stored or transmitted In the decoder, Fig 18b, die coded bitstream is demultiplexed 1809, decoded and die subband samples are re-quantised to the equal number of bits 1811 A syntiiesis filter bank combines the subband samples m order to recreate die onginal signal 1813 Implementations using maximally decimated filter banks will drastically reduce computational costs In the following descnptions, there is a focus on cosine modulated filter banks It should be appreciated however, that the mvention can be implemented usmg other types of filter banks or transforms, mcludmg filter bank interpretations of the wavelet transform, other non-equal bandwidtii filter banks or transforms and multi-dimensional filter banks or transforms

In the illustrative, but not limiting, descnptions below it is assumed tiiat an L-channel cosine modulated filter bank splits the mput signal x(ri) mto L subband signals The geneπc structure of a maximally decimated filter bank is shown m Fig 19 The analysis filters are denoted H^z) 1901, where k = 0, 1, , L-l The subband signals v_k(n) are maximally decimated 1903, each of sampling frequency fJL, where/ is the sampling frequency of x(ri) The synthesis section reassembles the subband signals after interpolation 1905 and filtering 1907 to produce (ri) The synthesis filters are denoted F_k(z) In addition, the present invention performs a spectral replication on x(n) , giving an enhanced signal y(n)

Synthesising the subband signals witii a C?£-channel filter bank, where only the L lowband channels are used and die bandwidtii expansion factor Q is chosen so that QL is an mteger value, will result in an output bit stream with sampling frequency Qf_s Hence, the extended filter bank will act as if it is an -channel filter bank followed by an upsampler Smce, in this case, the L(Q-Y) highband filters are unused (fed witii zeros), the audio bandwidtii will not change - the filter bank will merely reconstruct an upsampled version of x(n) If, however, the L subband signals are patched to die highband filters, the bandwidtii of x(n) will be increased by a factor Q, producing («) This is the maximally decimated filter bank version of the basic multiband transposer, according to the invention Using tins scheme, the upsampling process is integrated m the synthesis fdteπng as explained earlier It should be noted tiiat any size of the synthesis filter bank may be used, resulting m different sample-rates of the output signal, and hence different bandwidtii expansion factors Performing spectral replication on x(n) according to the present invention of the basic multiband transposition method witii an integer transposition factor M, is accomplished by patching die subband signals as

VJW* (") = *_Aft (») (-V) ^M-^])kn v_k (n) , (16) where k e [OJ-l] and chosen so thatMfe e [L,QL-\], e_Mk(n) is die envelope correction and (-1)^{(M X)k}" is a correction factor for spectral inverted subbands Spectral mversion results from decimation of subband signals, and die inverted signals may be reinverted by changing sign on every second sample m those channels Referring to Fig 20, consider an 16-channel synthesis filter bank, patched 2009 for a transposition factor M= 2, with Q = 2 The blocks 2001 and 2003 denote the analysis filters H^z) and die decimators of Fig 19 respectively Sinulariy, 2005 and 2007 are the interpolators and syntiiesis filters Fk(z) Eq 16 then simplifies to patching of the four upper frequency subband signals of the received data into every second of the eight uppermost channels m the synthesis filter bank Due to spectral mversion, every second patched subband signal must be frequency inverted before die synthesis Additionally, the magnitudes of the patched signals must be adjusted 2011 according to die pπnciples of SBR-1 or SBR-2

Using die basic multiband transposition metiiod accordmg to die present mvention, the generated harmonics are m general not exact multiples of the fundamentals All frequencies but the lowest in every subband drffers in some extent from an exact transposition Further, the replicated spectrum contams zeros since the target mterval covers a wider frequency range than the source mterval Moreover, the alias cancellation properties of the cosine modulated filter bank vanishes, smce the subband signals are separated in frequency in die target mterval That is, neighbouring subband signals do not overlap m the high-band area However, ahasmg reduction metiiods, known by those skilled m the art, may be used to reduce this type of artifacts Advantages of tins transposition method are ease of implementation, and die very low computational cost

To achieve perfect transposition of sinusoids, an effective maximally decimated filter bank solution of die improved multiband transposition method is now presented The system uses an additional modified analysis filter bank, wlule the syntiiesis filter bank is cosine modulated as descπbed by Vaidyanathan ["Multirate Systems and Filter Banks" P P Vaidyanatiian, Prentice Hall, Englewood ClnTs, New Jersey, 1993, ISBN 0-13-605718-7] The steps for operation, using the improved multiband transposition method according to the present invention, based on maximally decimated filter banks, are shown schematically in Fig. 21 and in die flow-chart of Fig. 22 and are as follows:

1. The L received subband signals are synthesised with a g -channel filter bank 2101, 2201, 2203, where the

L(Q-\) upper channels are fed with zeros, to form signal x_\(n), which is tiius oversampled by the bandwidtii expansion factor Q.

2. x_\( ) is downsampled by a factor O, to form signal *₂(«') 2103, 2205, i. e. xι(ri) = x_\(Qri).

3. An integer-value K is chosen as the size of a synthesis filter bank, constrained so diat T = KMIQ is an integer, where T is the size of the modified analysis filter bank and M is die transposition factor 2207, 2209, 2211. K should preferably be chosen large for stationary (tonal) signals, and smaller for dynamic (transient) signals.

4. x₂(«') is filtered through a /"-channel modified analysis filter bank 2107, 2213, where the ^analysis filters are exponentially modulated, producing a set of complex-valued subband signals. The subband signals are downsampled by a factor TIM, giving subband signals V_k ^m(n"), k = 0, l,..., T-l. Hence, the filter bank will be oversampled by a factor M

5. The samples v_k ^m(n") are converted to a polar representation (magnitude and phase-angle). The phase-angles are multiphed by die factor M, and the samples are converted back to a rectangular representation according to the scheme of Fig. 11. The real parts of the complex-valued samples are taken, giving the signals s^i ") 2109, 2215. After this operation, the signals s_k ^m(n") are critically sampled. 6. The gains ofthe signals s_k (n") are adjusted according to die principles of SBR-1 or SBR-2 2111, 2217.

7. The subband signals s_k ^(M)(n"), where k e [TIM, mnU , 7)-l], are synthesised with an ordinary cosine modulated AT-channel filterbank, where the channels 0 through 77M-1 are fed with zeros 2105, 2221. This produces the signal X₃ ^(M(n).

8. X₃ ^<M («) is finally added to

to give > «) 2223, which is the desired spectral replicated signal.

Steps 3 to 6 may be repeated for different values ofthe transposition factor M, thus adding multiple harmonics to *](«). This mode of operation is Ulustrated by the dotted figures of Fig. 21, and in Fig. 22, by iterating the loop over boxes 2211 - 2219. In this case, K is chosen as to make T integer-valued for all choices of M - for integer valued Ms; preferably select K as to make KIQ a positive integer. All subband

, where ι = l, 2,..., m, and /M is the number of transposition factors, are added according to

for every apphcable k. In the first iteration of the loop of Fig. 22, the signals s_k(n") may be considered to be subband samples of zeros only, where k = 0, 1,...,ΛT-1. In every loop, the new samples are added 2219 to S (n") as

where k = KIQ, KIQ+l, ..., min^ ,)-! . The subband signals s^n") are synthesised once with a ^-channel filter bank according to step 7. The modified analysis filter bank of step 4, is derived through the theory of cosine modulated filter banks, where the modulated lapped transform (MLT) ["Lapped Transforms for Efficient Transfoπn/Subband Coding" H S Malvar, IEEE Trans ASSP, vol 38, no 6, 1990] is a special case The impulse responses h_k(n) of the filters in a ^-channel cosine modulated filter bank may be wntten π N- L h_k (n) = C p₀ (n)cos (2k +ϊ)(n - IT -) + Φ* (19)

where k = 0, 1, , T-l, N is die length ofthe lowpass prototype filter p₀(n), C is a constant and Φ_k is a phase-angle that ensures alias cancellation between adjacent channels The constraints on Φ* is

7 7 ^φo = ± > Φr-i ^■■ +— and Φ_k = Φ_k , ±— (20a-c)

4 2 which may be simplified to die closed form expression

With tins choice of Φ_k , perfect reconstruction systems or approximate reconstruction systems (pseudo QMF systems) may be obtamed using syntiiesis filter banks with impulse responses as π N -l

Λ («) = C ₀ («) cos — (2* + !)(/! — — ) - Φ* (22)

Consider die filters

h_k' (n) = C p₀ (n) ₊ Φ_k (23)

where h'_k(n) are sme-modulated versions ofthe prototype filter p₀(n) The filters H (z) and H_k(z) have identical passband supports, but the phase responses differ The passbands of die filters are actually Hilbert transforms of each other (this is not valid for frequencies close to ω = 0 and ω = π) Combming Eq 19 and Eq 23 according to iπ N - l

K («) = ") + X (") = ^c _Po (n) exp (24)

yields filters that have the same shape ofthe magmtude responses as H_k(z) for positive frequencies but are zero for negative frequencies Using a filter bank with unpulse responses as in Eq 24 gives a set of subband signals tiiat may be interpreted as die analytic (complex) signals coπesponding to the subband signals obtamed from a filter bank with impulse responses as m Eq 19 Analytic signals are suitable for manipulation, since the complex-valued samples may be wntten m a polar form, that is z(ri) = r(ri)+j ι(n) = |z(«)|exp{/ arg(z(«))} However, when usmg the complex filter bank for transposition, the constraint on Φ_k has to be generalised to retam the alias cancellation property The new constraint on Φ_k , to ensure alias cancellation m combination with a synthesis filter bank witii impulse responses as in Eq 22 is π ^φ* =±(-ιr (25) ^~AM which simplifies to Eq 21 when = 1 With this choice, transposed partials will have the same relative phases as they would have whenM= 1 (no transposition) Combimng Eq 24 and Eq 25 results in

which are the filters used m die modified filter bank of step 4, accordmg to die present invention

Some clarifications concerning step 5 downsampling die complex-valued subband signals by a factor TIM makes them oversampled by M, which is an essential cntenon when the phase-angles subsequently are multiplied by the transposition factor M The oversamplmg forces the number of subband samples per bandwidth, after transposition to the target range, to equal that ofthe source range The individual bandwidtiis ofthe transposed subband signals are M times greater than those m the source range, due to die phase-multiplier This makes the subband signals critically sampled after step 5, and additionally, there will be no zeros in the spectrum when transposing tonal signals

In order to avoid tπgonometπc calculations, that is, having to compute the new subband signals as

where |v ,^W, (n")\ is the absolute value of v_k ⁽ n"), the following tngonometnc relationship is used cos(Mα) = cos^M (a) - ( ) sin² (α) cos^M~2 (a) + ( ) sin⁴ (a) cos^M~4 (a) - ... (28)

Letting

and noting that

and

the computations of step 5 may be accomplished witiiout tngonometnc calculations, reducing computational complexity When usmg transpositions where Mis even, obstacles with the phase-multiplier may arise, depending on die charactenstics ofthe lowpass prototype filter p₀(n) All applicable prototype filters have zeros on the unit circle in the z-plane A zero on the unit circle imposes a 180° shift in the phase response of the filter For M even, the phase- multiplier translates these shifts to 360° shifts, I e the phase-shifts vanish The partials so located in frequency that such phase-shifts vamsh will give πse to aliasmg m the synthesised signal The worst case scenano is when a partial is located at a point m frequency coπesponding to the top of the first side lobe of an analysis filter Depending on the rejection of tins lobe in the magnitude response, the aliasing will be more or less audible As an example, the first side lobe of the prototype filter used m the ISO/MPEG layer 1 and 2 standard is rejected 96 dB, while die rejection is only 23 dB for die first side lobe ofthe sine-wmdow used in die MDCT scheme of die ISO/MPEG layer 3 standard It is clear, tiiat tins type of aliasing, usmg the sme-window, will be audible A solution to tius problem will be presented, and is refeπed to as relative phase locking

The filters h^a _k(ri) all have linear phase responses The phase-angles Φ_k introduce relative phase differences between adjacent channels, and the zeros on the unit circle introduce 180° phase-shifts at locations in frequency that may differ between channels By monitonng die phase-drfference between neighbouπng subband signals, before the phase-multiplier is activated, it is easy to detect the channels that contain phase-inverted information Considering tonal signals, die phase-difference is approximately τz/2M, according to Eq 25, for non-inverted signals, and consequently approximately .I-1/2M) for signals, where either ofthe signals is inverted The detection of inverted signals may be accomplished simply by computmg die dot product of samples m adjacent subbands as v[^M )

(/?")} +

(32)

If the product inEq 32 is negative, the phase-difference is greater than 90°, and a phase-mversion condition is present The phase-angles of the complex-valued subband signals are multiplied by M, according to die scheme of step 5, and finally, die inversion-tagged signals are negated The relative phase locking method thus forces the 180° shifted subband signals to retain this shift after the phase-multiplication, and hence maintain die aliasing cancellation properties

Spectral envelope adiustment

Most sounds, like speech and music, are charactensed as products of slowly varying envelopes and rapidly varying earners with constant amplitude, as desenbed by Stockham ["The Application of Generalized Lmeanty to Automatic Gam Control" T G Stockham, Jr, IEEE Trans on Audio and Electroacoustics, Vol AU-16, No 2, June 1968] and Eq 1

In split-band perceptual audio coders, die audio signal is segmented into frames and split into multiple frequency bands usmg subband filters or a time-to-frequency domam transform In most codec types, die signal is subsequently separated mto two major signal components for transmission or storage, the spectral envelope representation and the normalised subband samples or coefficients Throughout the following desenption, die term "subband samples" or "coefficients" refers to sample values obtamed from subband filters as well as coefficients obtamed from a time-to-frequency transform The term "spectral envelope" or "scale factors" represent values of the subbands on a time-frame basis, such as the average or maximum magmtude m each subband, used for normalisation of die subband samples However, die spectral envelope may also be obtamed usmg lmear prediction LPC, [U.S. Pat. 5,684,920]. In a typical codec, die normalised subband samples require coding at a high bitrate (using approximately 90% ofthe available bitrate), compared to die slowly varying temporal envelopes, and tiius the spectral envelopes, that may be coded at a much-reduced rate (using approximately 10% ofthe available bitrate).

Accurate spectral envelope of the replicated bandwidtii is important if the timbral qualities of the original signal are to be preserved. The perceived timbre of a musical instrument, or voice, is mainly determined by the spectral distribution below a frequency /„-, located in the highest octaves of hearing. The spectral details above m, are thus of less importance, and consequently die highband fine structures obtained by the above transposition methods require no adjustment, while the coarse structures generally do. In order to enable such adjustment, it is useful to filter the spectral representation ofthe signal to separate the envelope coarse structure from the fine structure.

In the SBR-1 implementation according to the present invention, the highband coarse spectral envelope is estimated from the lowband infoπnation available at die decoder. This estimation is performed by continuously monitoring the envelope ofthe lowband and adjusting die highband spectral envelope according to specific rules. A novel metiiod to accomplish the envelope estimation uses asymptotes in a logarithmic frequency-magnitude space, which is equivalent to curve fitting with polynomials of varying order in the linear space. The level and slope of an upper portion ofthe lowband spectrum are estimated, and die estimates are used to define the level and slope of one or several segments representing die new highband envelope. The asymptote intersections are fixed in frequency and act as pivot points. However not always necessary, it is beneficial to stipulate constraints to keep the highband envelope excursions within reahstic boundaries. An alternative approach to estimation of the spectral envelope is to use vector quantization, VQ, of a large number of representative spectral envelopes, and store these in a lookup- table or codebook. Vector quantization is performed by training the desired number of vectors on a vast amount of training data, in tiiis case audio spectral envelopes. The training is usually done with die Generalised Lloyd Algorithm ["Vector Quantization and Signal Compression" A. Gersho, R. M. Gray, Kluwer Academic Publishers, USA 1992, ISBN 0-7923-9181-0], and yields vectors tiiat optimally cover the contents ofthe training data. Considering a VQ codebook consisting of A spectral envelopes trained by B envelopes (B » A), tiien the A envelopes represent the A most likely transitions from the lowband envelope to the highband envelope, based on B observations of a wide variety of sounds. This is, theoretically, the A optimum rules for predicting die envelope based on the B observations. When estimating a new highband spectral envelope, the original lowband envelope is used to search the codebook and the highband part of the best matching codebook entry is applied to create die new highband spectrum.

In Fig. 23, the normalised subband samples are represented by 2301 and die spectral envelopes are represented by the scalefactors 2305. For illustrative purposes the transmission to decoder 2303 is shown in parallel form. In die SBR-2 metiiod Fig. 24, the spectral envelope information is generated and transmitted according to Fig. 23, whereby only me lowband subband samples are transmitted. Transmitted scalefactors thus span the full frequency range while die subband samples only span a restricted frequency range, excluding die highband. At die decoder die lowband subband samples 2401 are transposed 2403 and combined with the received highband spectral envelope information 2405. In this way the synthetic highband spectral envelope is identical to that ofthe original, while maintaining a significant bit rate reduction. In some codecs, it is possible to transmit the scalefactors for the entire spectral envelope while omitting the highband subband samples, as shown m Fig 24 Other codec standards stipulate that scalefactors and subband samples must cover the same frequency range, l e scale-factors cannot be transmitted rf die subband samples are omitted In such cases, there are several solutions, the highband spectral envelope information can be transmitted in separate frames, where the frames have their own headers and optional eπor protection, followed by the data

Regular decoders, not taking advantage of die present mvention, will not recognise die headers and therefore discard die extra frames In a second solution, die highband spectral envelope information is transmitted as auxdiary data withm the encoded bitstream However, the available auxdiary data field must be large enough to hold die envelope information In cases where none ofthe first two solutions are adaptable, a third solution, where the highband spectral envelope information is hidden as subband samples, may be applied Subband scalefactors cover a large dynamic range, typically exceeding 100 dB It is tiius possible to set an arbitrary number of subband scalefactors, 2505 in Fig 25, to very low values, and to transmit the highband scalefactors "camouflaged" as subband samples, 2501 This way of transmitting the highband scale factors to the decoder 2503 ensures compatibility with the bitstream syntax Hence, arbitrary data may be transmitted in this fashion A related metiiod exists where information is coded into die subband sample stream [U S Pat 5,687,191] A fourth solution, Fig 26, can be applied when a codmg system uses Huffman- or other redundancy coding 2603 The subband samples for die highband is then set to zero 2601 or a constant value as to achieve a high redundancy

Transient response improvements Transient related artifacts are common problems m audio codecs, and similar artifacts occur m the present mvention In general, patching generates spectral "zeros" or notches, coπesponding to time domain pre- and post-echoes, I e spunous transients before and after "true" transients Albeit the P-blocks "fill in the zeros" for slowly varying tonal signals, the pre- and post-echoes remain The improved multiband method is mtended to work on discrete smusoids, where die number of sinusoids is restπcted to one per subband Transients or noise in a subband can be viewed as a large number of discrete smusoids witiun that subband Ηus generates mtermodulation distortion These artifacts are considered as additional quantization-noise sources connected to die rephcated highband channels dunng transient intervals Traditional metiiods to avoid pre- and post-echo artifacts m perceptual audio coders, for example adaptive window switching, may hence be used to enhance the subjective quahty ofthe improved multiband metiiod By usmg the transient detection provided by the codec or a separate detector and reducing die number of channels under transient conditions the "quantization noise" is forced not to exceed the time-dependent masking threshold A smaller number of channels is used duπng transient passages whereas a larger is used dunng tonal passages Such adaptive window switching is commonly used in codecs in order to trade frequency resolution for time resolution Different methods may be used in applications where the filterbank size is fixed One approach is to shape the "quantization noise" m time via lmear prediction m die spectral domam The transposition is then performed on the residual signal, which is the output of the lmear prediction filter Subsequently, an inverse prediction filter is applied to die onginal- and spectral replicated channels simultaneously Another approach employs a compander system 1 e dynamic amplitude compression of die transient signal pπor to transposition or coding, and a complementary expansion after transposition It is also possible to switch between transposition methods m a signal dependent manner, for example, a high resolution filterbank transposition method is used for stationary signals, and a time- vaπant pattern search prediction method is employed for transient signals Practical implementations

Using a standard signal-processor or a powerful PC, real-tune operation of a SBR-enhanced codec is possible The SBR enhanced codec may also be hard-coded on a custom chip It may also be implemented in various kinds of systems for storage or transmission of signals, analogue or digital, using arbitrary codecs, Fig 27 and Fig 28 The SBR-1 method may be integrated in a decoder or supplied as an add-on hardware or software post-processing module The SBR-2 method needs additional moddication of die encoder In Fig 27 die analogue input signal is fed to die A/D-converter 2701, formmg a digital signal which is fed to die an arbitrary encoder 2703, where source coding is performed The signal fed mto the system may be of such a low-pass type that spectral bands witiun the auditory range already have been discarded, or spectral bands are discarded in die arbitrary encoder The resultmg lowband signals are fed to the multiplexer 2705, formmg a senal bitstream which is transmitted or stored 2707 The de-multiplexer 2709 restores the signals and feeds them to an arbitrary decoder 2711 The spectral envelope information 2715 is estimated at the decoder 2713 and fed to die SBR-1 unit 2713 which transposes the lowband signal to a highband signal and creates an envelope adjusted wideband signal Finally, die digital wideband signal is converted 2717 to an analogue output signal

The SBR-2 method needs additional modification ofthe encoder In Fig 28 the analogue mput signal is fed to die A D-converter 2801, formmg a digital signal which is fed to die an arbitrary encoder 2803, where source coding is performed The spectral envelope information is extracted 2805 The resultmg signals, lowband subband samples or coefficients and wideband envelope information, are fed to the multiplexer 2807, formmg a seπal bitstream which is transmitted or stored 2809 The de-multiplexer 2811 restores the signals, lowband subband samples or coefficients and wideband envelope information, and feeds them to an arbitrary decoder 2815 The spectral envelope information 2813 is fed from die de-multiplexer 2811 to the SBR-2 umt 2817 which transposes the lowband signal to a highband signal and creates an envelope adjusted wideband signal Finally, die digital wideband signal is converted 2819 to an analogue output signal

When only very low bitrates are available, (Internet and slow telephone modems, AM-broadcasting etc ) mono coding ofthe audio program matenal is unavoidable In order to improve the perceived quality and make die programme more pleasant sounding, a simple "pseudo-stereo" generator, Fig 29, is obtamed by the introduction of a tapped delayhne 2901 This may feed 10ms and 15ms delayed signals at approximately -6dB 2903 to each output channel m addition to the onginal mono signal 2905 The pseudo-stereo generator offers a valuable perceptual improvement at a low computational cost

The above-described embodiments are merely illustrative for the prmciples ofthe present mvention for audio source coding improvement It is understood that modifications and vaπations of the arrangements and die details descnbed herem will be apparent to others skilled m the art It is the intent, therefore, to be limited only by the scope ofthe impendmg patent claims and not by the specific details presented by way of descnption and explanation of die embodiments herein

Claims

1. A method for enhancement of a source coding system where said source coding system compnses an encoder representing all operations performed pnor to storage or transmission, and a decoder representing all operations performed after storage or transmission, characterised by at said encoder, discarding frequency band(s) of an original signal, forming a first signal, at said decoder, by means of transposition performed on said first signal, replicating frequency band(s) of said onginal signal, forming a second signal, and combining said first and said second signal, forming an output signal, whereby reduced bitrate at a given perceptual quahty or unproved perceptual quahty at a given bitrate is obtained

2. A method accordmg to claim 1, characterised in at passband(s) of said second signal are set not to overlap or only partly overlap passband(s) of said first signal

3. A method accordmg to claims 1 - 2, characterised in that spectral envelope adjustment is performed based on estimation of the spectral envelope of said discarded frequency band(s) of said origmal signal usmg said first signal

4. A method accordmg to claims 1 - 2, characterised in tiiat spectral envelope adjustment is performed based on transmitted envelope information of said discarded frequency band(s) of said onginal signal

5. A method accordmg to claim 4, characterised in that said spectral envelope information is transmitted as subband samples m an arbitrary number of subband channels where die gains of said subband channels are set to a low level, whereby compatibility with standardised decoders is ensured

6. A method accordmg to claim 4, characterised in that said envelope information is transmitted as scale factors without transmission ofthe coπespondmg subband samples

7. A method accordmg to claim 4, characterised in that said envelope information is transmitted as scale factors and the coπespondmg subband samples are set to zero or a constant value, whereby the entropy ofthe subband samples is reduced

8. A method according to claims 1 - 7, characterised in that said output signal, when monophomc audio, is spht mto two signals each compnsed of said output signal and delayed versions of the same to obtain a pseudo-stereo signal

9. A method accordmg to claims 1 - 7, where said transposition is characterised by filtering a signal through a set of N ≥ 2 bandpass filters with passbands compnsmg die frequencies [/i, f_N] respectively, formmg TV bandpass signals, shifting said bandpass signals m frequency to regions compnsmg the frequencies M f ,f_N] where M ≠ 1 is the transposition factor, and formmg a transposed signal by combining said shifted bandpass signals

10. A method according to claim 9, characterised in tiiat said frequency shifting is obtained tiu-ough upper side band (USB) modulation.

11. A method for transposition by a factor M, characterised by: bandpass filtering a signal using an analysis filter bank or transfoπn of such a nature that real- or complex- valued subband signals of lowpass type are generated; patching an arbitrary number of channels k of said analysis filter bank or transform to channels Mk, M ≠ 1, in a synthesis filter bank or transform; and forming a transposed signal using said synthesis filter bank or transform.

12. A method according to claim 11, characterised in that said filter bank is maximally decimated and said patching is performed according to die relation

VMk (.ny 7V) ^M-^{X kn}v_k n) , where (-1) is a coπection factor, v_k(ri) the subband signal of channel k, and v_m(n) die subband signal of channel Mk; whereby compensation of spectral inverted subband signals is obtained.

13. A method according to claims 11 - 12, characterised by: patching the phases ofthe subband signals from channels k of said analysis filter bank or transform as the phases of the subband signals associated witii synthesis channels Mk, MΓëá l; and patching the magnitudes of die subband signals from consecutive channels / of said analysis filter bank or transform as the magnitudes ofthe subband signals associated with consecutive synthesis channels l+S, where S is an integer Γëá 1.

14. A method according to claims 11 - 13, characterised in tiiat the phases of said subband signals of said channels & are multiphed by said factor M before using said synthesis filter bank or transform.

15. A method according to claims 11 - 14, characterised in that = Kr , where K is an integer > 1.

16. A method according to claims 11 - 15, characterised in that said patching employs multiple values of said transposition factor M

17. A method for transposition by a factor M, characterised by: filtering a signal through a parallel bank of L filters with impulse responses as h_k (n) = K

where k = 0, l,..., L-l, K is a constant, and p₀(n) is a lowpass prototype filter of length N, producing a set of L complex-valued signals; downsampling said L signals with a factor L/M, producing a set of L complex-valued subband signals; multiplying the phase-angles of said complex-valued subband signals by M, giving a new set of subband signals; selecting the real part of said new set of subband signals, resulting in a set of L real-valued subband signals; upsampling a subset of said real-valued subband signals witii a factor L', producing a set of real-valued signals; filtering said real-valued signals through a parallel bank of 11 filters with impulse responses as

where k = 0, 1, ..., L'-l, K' is a constant and p'o(n) is a lowpass prototype filter of lengtii TV, forming a set of I! filtered signals; and adding said I! filtered signals to produce a transposed signal.

18. A method according to claim 17, characterised in that said multiplication of said phase-angles and said selecting ofthe real part, is computed by: writing said complex-valued subband signals as Z_k (n) = R_k (n) + jI_k (n) , where Rif i) and li ji) are the real and imaginary parts of Z_k(ri) respectively; calculating said real-valued subband signals W_k( ) as

W_k (n) = \Z_k

where

andMis a positive integer transposition factor, using the trigonometric identity cos(M ) = cos^M (α) -( ) sinyα)cos^M^ (α) + (7 ) si7(α)cos^{M →} (α) -.. where α= arctan{I/f,n)/R_k(n)}, and d e relations

whereby reducing computational complexity by elimination of all trigonometric calculations.

19. A method accordmg to claim 17, characterised by on a block basis, extracting information conveyed by the phase-difference of an adjacent pair of said complex-valued subband signals, performing said multiplication of said phase-angles by M. forming a pair of said new subband signals, and negatmg one of said new subband signals, on a condition provided by said information, whereby 180┬░ phase- shifts ofthe subband signals are retained when employing an even integer-valued transposition factor M

20. A method accordmg to claim 19, characterised in tiiat said information is given by the dot-product of said complex-valued subband signals Z_k(n and Z_k+ι(n) accordmg to Z_k («) o Z_k+ (n) = R_k (n)R_M ( ) + I_k (n)I_k+l (n) , where R,(n) and I,(n) are the real and imaginary parts of Z,(n) respectively, / = k, k+ 1 , and one of said new subband signals is negated provided said dot-product is negative

21. A method for transposition, by expanding or compressmg a first signal in time and duplicating or discarding arbitrary long segments of said first signal, subsequently down- or up-sample said first signal, characterised by performing transient detection on said first signal, determining which segment of said first signal to be used when duplicating or discarding parts of said first signal depending on the outcome ofthe transient detection, adjusting the length L of said signal segment depending on the outcome of the transient detection, adjusting the number of samples TV used for each statevector depending on die outcome of the transient detection, adjusting the delay D between samples in said statevector depending on the outcome ofthe transient detection, adjustmg the number of samples AT between each statevector dependmg on the outcome ofthe transient detection, and searching for synchromsation pomts in chosen segment of said first signal, based on die synchromsation point found m the previous synchromsation point search

22. A method accordmg to claim 21, characterised in that several transposers are mterconnected m order to share synchromsation pomt information, whereby reduced computational complexity is achieved

23. A method accordmg to claims 21 - 22, characterised in that said transposers are connected to a suitable filter bank, and the signals fed mto each of said transposers are filtered as to obtam an arbitrary spectral envelope ofthe new signal bemg the sum of said signals bemg processed by said transposers

24. A system for enhanced decoding of a source coded signal derived from an original signal, characterised by transposition means for transposing frequency band(s) of said source coded signal, formmg a first signal, estimation means operating on said source coded signal, for estimation ofthe spectral envelope of said onginal signal, adjustmg means for adjustmg the spectral envelope of said first signal, based on said estimation, combining means for combining said source coded signal and said adjusted first signal, whereby reduced bitrate at a given perceptual quahty or unproved perceptual quality at a given bitrate is achieved

25. An apparatus accordmg to claim 24, operating when said output signal is monophomc audio characterised by delaying means for delaying and attenuating means for attenuating said output signal formmg a first delayed signal, delaying means for delaying and attenuating means for attenuating said output signal, usmg different parameters, formmg a second delayed signal, means for addmg said output and said first delayed signal, forming a left-channel output signal, and means for addmg said output and said second delayed signal, formmg a right-channel output signal, whereby obtaining a pseudo stereophonic signal

26. A system for enhanced source coding where said system compnses an encoder representing all umts precedmg a storage media or transmission channel, and a decoder representing all umts following said storage media or transmission channel, characterised by discarding means at said encoder for discarding frequency band(s) of an onginal signal, forming a first signal, extracting means at said encoder for extracting spectral envelope information of said onginal signal, forming a second signal, means at said encoder for coding of said first and second signals, transposition means at said decoder for transposmg frequency band(s) of said first signal, formmg a third signal, adjusting means at said decoder for spectral envelope adjustment of said third signal, based on said second signal, and combining means at said decoder for combining said first and said adjusted third signal, whereby reduced bitrate at a given perceptual quahty or unproved perceptual quality at a given bitrate is achieved

27. An apparatus accordmg to claim 26, operating when said output signal is monophomc audio characterised by delaying means for delaying and attenuating means for attenuating said output signal formmg a first delayed signal, delaying means for delaying and attenuating means for attenuating said output signal, usmg different parameters, formmg a second delayed signal, means for addmg said output and said first delayed signal, formmg a left-channel output signal, and means for addmg said output and said second delayed signal, formmg a nght-channel output signal, whereby obtaining a pseudo stereophomc signal

28. An apparatus for transposition a factor M, characterised by: filtering a signal by means of an analysis filter bank or transform of such a nature that real- or complex- valued subband signals of lowpass type are generated; means for patching an arbitrary number of channels k of said analysis filter bank or transform to channels Mk, MΓëá 1, in a syntiiesis filter bank or transform; and forming a transposed signal by means of said syntiiesis filter bank or transform.

29. An apparatus according to claim 28, characterised in that when = 7^Λ , where K is an integer > 1, the phases of the subband signals of said channels k are multiplied by M before using said synthesis filter bank or transform.

30. An apparatus for transposition by a factor M, characterised by: filtering means for filtering a signal through a parallel bank of L filters with impulse responses as

where k = 0, l,..., L-l, K is a constant, p₀(n) is a lowpass prototype filter of length TV, and M is the transposition factor, producing a set of L complex-valued signals; means for downsampling said L signals witii a factor LIM, producing a set of L complex-valued subband signals; means for multiplying the phase-angles of said complex-valued subband signals by M, giving a new set of subband signals; means for selecting the real part of said new set of subband signals, resulting in a set of L real-valued subband signals; means for upsampling a subset of said real-valued subband signals with a factor 11 , producing a set of real- valued signals; filtering means for filtering said real-valued signals through a parallel bank of I' filters with impulse responses as

where k = 0, l,..., L'-l, K is a constant andp'₀(┬½) is a lowpass prototype filter of lengtii TV, forming a set of L' filtered signals; and means for adding said L' filtered signals to produce the transposed signal.

31. An apparatus for transposition, by expanding or compressing a first signal in time by duplicatmg or discarding arbitrary long segments of said first signal, and subsequently down- or up-sample said first signal, characterised by detection means for performing transient detection on said first signal, means for using the position of a possible transient when determming which segment of said first signal to be used when duplicating or discarding parts of said first signal, in order to obtain said transposition, adjustmg means for adjustmg the length (L) of said signal segment dependmg on the output from the transient detector, adjusting means for adjusting the number of samples (TV) used for each statevector depending on the output from the transient detector, adjusting means for adjusting the delay (D) between samples m said statevector depending on the output from the transient detector, adjusting means for adjustmg die number of samples (K) between each statevector dependmg on the output from the transient detector, and searching means for searching for synchromsation points in chosen segment of said first signal, based on the synchromsation pomt found m the previous synchromsation pomt search

32. An apparatus accordmg to claim 31, operating on subband signals characterised by means for sharing synchromsation information between multiple instances of said transposer, means for formmg subsets of said subband signals, means for amplitude adjustment of every channel withm each of said subsets, synthesis filter bank means for formmg, from each of said subsets, an mput signal to each instance of said transposers, processing said mput signals by said transposers, and summation means for acquiring a new signal by summation of said processed signals, whereby an arbitrary spectral envelope can be obtained