Patent Application
20210035592
The present invention generally relates to digital audio coding and more precisely to coding techniques for audio signals containing components of different characters.
A widespread class of coding method for audio signals containing speech or singing includes code excited linear prediction (CELP) applied in time alternation with different coding methods, including frequency-domain coding methods especially adapted for music or methods of a general nature, to account for variations in character between successive time periods of the audio signal. For example, a simplified Moving Pictures Experts Group (MPEG) Unified Speech and Audio Coding (USAC; see standard ISO/IEC 23003-3) decoder is operable in at least three decoding modes, Advanced Audio Coding (AAC; see standard ISO/IEC 13818-7), algebraic CELP (ACELP) and transform-coded excitation (TCX), as shown in the upper portion of accompanying
The various embodiments of CELP are adapted to the properties of the human organs of speech and, possibly, to the human auditory sense. As used in this application, CELP will refer to all possible embodiments and variants, including but not limited to ACELP, wide- and narrow-band CELP, SB-CELP (sub-band CELP), low- and high-rate CELP, RCELP (relaxed CELP), LD-CELP (low-delay CELP), CS-CELP (conjugate-structure CELP), CS-ACELP (conjugate-structure ACELP), PSI-CELP (pitch-synchronous innovation CELP) and VSELP (vector sum excited linear prediction). The principles of CELP are discussed by R. Schroeder and S. Atal in Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), vol. 10, pp. 937-940, 1985, and some of its applications are described in references 25-29 cited in Chen and Gersho, IEEE Transactions on Speech and Audio Processing, vol. 3, no. 1, 1995. As further detailed in the former paper, a CELP decoder (or, analogously, a CELP speech synthesizer) may include a pitch predictor, which restores the periodic component of an encoded speech signal, and a pulse codebook, from which an innovation sequence is added. The pitch predictor may in turn include a long-delay predictor for restoring the pitch and a short-delay predictor for restoring formants by spectral envelope shaping. In this context, the pitch is generally understood as the fundamental frequency of the tonal sound component produced by the vocal chords and further coloured by resonating portions of the vocal tract. This frequency together with its harmonics will dominate speech or singing. Generally speaking, CELP methods are best suited for processing solo or one-part singing, for which the pitch frequency is well-defined and relatively easy to determine.
To improve the perceived quality of CELP-coded speech, it is common practice to combine it with post filtering (or pitch enhancement by another term). U.S. Pat. No. 4,969,192 and section II of the paper by Chen and Gersho disclose desirable properties of such post filters, namely their ability to suppress noise components located between the harmonics of the detected voice pitch (long-term portion; see section IV). It is believed that an important portion of this noise stems from the spectral envelope shaping. The long-term portion of a simple post filter may be designed to have the following transfer function:
where T is an estimated pitch period in terms of number of samples and a is a gain of the post filter, as shown in
S
E(z)=S(z)−αS(z)PLT(z)HLP(z),
where
and S is the decoded signal which is supplied as input to the post filter.
The long-term portion described in the previous paragraph may be used alone. Alternatively, it is arranged in series with a noise-shaping filter that preserves components in frequency intervals corresponding to the formants and attenuates noise in other spectral regions (short-term portion; see section III), that is, in the ‘spectral valleys’ of the formant envelope. As another possible variation, this filter aggregate is further supplemented by a gradual high-pass-type filter to reduce a perceived deterioration due to spectral tilt of the short-term portion.
Audio signals containing a mixture of components of different origins—e.g., tonal, non-tonal, vocal, instrumental, non-musical—are not always reproduced by available digital coding technologies in a satisfactory manner. It has more precisely been noted that available technologies are deficient in handling such non-homogeneous audio material, generally favouring one of the components to the detriment of the other. In particular, music containing singing accompanied by one or more instruments or choir parts which has been encoded by methods of the nature described above, will often be decoded with perceptible artefacts spoiling part of the listening experience.
In order to mitigate at least some of the drawbacks outlined in the previous section, it is an object of the present invention to provide methods and devices adapted for audio encoding and decoding of signals containing a mixture of components of different origins. As particular objects, the invention seeks to provide such methods and devices that are suitable from the point of view of coding efficiency or (perceived) reproduction fidelity or both.
The invention achieves at least one of these objects by providing an encoder system, a decoder system, an encoding method, a decoding method and computer program products for carrying out each of the methods, as defined in the independent claims. The dependent claims define embodiments of the invention.
The inventors have realized that some artefacts perceived in decoded audio signals of non-homogeneous origin derive from an inappropriate switching between several coding modes of which at least one includes post filtering at the decoder and at least one does not. More precisely, available post filters remove not only interharmonic noise (and, where applicable, noise in spectral valleys) but also signal components representing instrumental or vocal accompaniment and other material of a ‘desirable’ nature. The fact that the just noticeable difference in spectral valleys may be as large as 10 dB (as noted by Ghitza and Goldstein, IEEE Trans. Acoust., Speech, Signal Processing, vol. ASSP-4, pp. 697-708, 1986) may have been taken as a justification by many designers to filter these frequency bands severely. The quality degradation by the interharmonic (and spectral-valley) attenuation itself may however be less important than that of the switching occasions. When the post filter is switched on, the background of a singing voice sounds suddenly muffled, and when the filter is deactivated, the background instantly becomes more sonorous. If the switching takes place frequently, due to the nature of the audio signal or to the configuration of the coding device, there will be a switching artefact. As one example, a USAC decoder may be operable either in an ACELP mode combined with post filtering or in a TCX mode without post filtering. The ACELP mode is used in episodes where a dominant vocal component is present. Thus, the switching into the ACELP mode may be triggered by the onset of singing, such as at the beginning of a new musical phrase, at the beginning of a new verse, or simply after an episode where the accompaniment is deemed to drown the singing voice in the sense that the vocal component is no longer prominent. Experiments have confirmed that an alternative solution, or rather circumvention of the problem, by which TCX coding is used throughout (and the ACELP mode is disabled) does not remedy the problem, as reverb-like artefacts appear.
Accordingly, in a first and a second aspect, the invention provides an audio encoding method (and an audio encoding system with the corresponding features) characterized by a decision being made as to whether the device which will decode the bit stream, which is output by the encoding method, should apply post filtering including attenuation of interharmonic noise. The outcome of the decision is encoded in the bit stream and is accessible to the decoding device.
By the invention, the decision whether to use the post filter is taken separately from the decision as to the most suitable coding mode. This makes it possible to maintain one post filtering status throughout a period of such length that the switching will not annoy the listener. Thus, the encoding method may prescribe that the post filter will be kept inactive even though it switches into a coding mode where the filter is conventionally active.
It is noted that the decision whether to apply post filtering is normally taken frame-wise. Thus, firstly, post filtering is not applied for less than one frame at a time. Secondly, the decision whether to disable post filtering is only valid for the duration of a current frame and may be either maintained or reassessed for the subsequent frame. In a coding format enabling a main frame format and a reduced format, which is a fraction of the normal format, e.g., ⅛ of its length, it may not be necessary to take post-filtering decisions for individual reduced frames. Instead, a number of reduced frames summing up to a normal frame may be considered, and the parameters relevant for the filtering decision may be obtained by computing the mean or median of the reduced frames comprised therein.
In a third and a fourth aspect of the invention, there is provided an audio decoding method (and an audio decoding system with corresponding features) with a decoding step followed by a post-filtering step, which includes interharmonic noise attenuation, and being characterized in a step of disabling the post filter in accordance with post filtering information encoded in the bit stream signal.
A decoding method with these characteristics is well suited for coding of mixed-origin audio signals by virtue of its capability to deactivate the post filter in dependence of the post filtering information only, hence independently of factors such as the current coding mode. When applied to coding techniques wherein post filter activity is conventionally associated with particular coding modes, the post-filtering disabling capability enables a new operative mode, namely the unfiltered application of a conventionally filtered decoding mode.
In a further aspect, the invention also provides a computer program product for performing one of the above methods. Further still, the invention provides a post filter for attenuating interharmonic noise which is operable in either an active mode or a pass-through mode, as indicated by a post-filtering signal supplied to the post filter. The post filter may include a decision section for autonomously controlling the post filtering activity.
As the skilled person will appreciate, an encoder adapted to cooperate with a decoder is equipped with functionally equivalent modules, so as to enable faithful reproduction of the encoded signal. Such equivalent modules may be identical or similar modules or modules having identical or similar transfer characteristics. In particular, the modules in the encoder and decoder, respectively, may be similar or dissimilar processing units executing respective computer programs that perform equivalent sets of mathematical operations.
In one embodiment, encoding the present method includes decision making as to whether a post filter which further includes attenuation of spectral valleys (with respect to the formant envelope, see above). This corresponds to the short-term portion of the post filter. It is then advantageous to adapt the criterion on which the decision is based to the nature of the post filter.
One embodiment is directed to an encoder particularly adapted for speech coding. As some of the problems motivating the invention have been observed when a mixture of vocal and other components is coded, the combination of speech coding and the independent decision-making regarding post filtering afforded by the invention is particularly advantageous. In particular, such a decoder may include a code-excited linear prediction encoding module.
In one embodiment, the encoder bases its decision on a detected simultaneous presence of a signal component with dominant fundamental frequency (pitch) and another signal component located below the fundamental frequency. The detection may also be aimed at finding the co-occurrence of a component with dominant fundamental frequency and another component with energy between the harmonics of this fundamental frequency. This is a situation wherein artefacts of the type under consideration are frequently encountered. Thus, if such simultaneous presence is established, the encoder will decide that post filtering is not suitable, which will be indicated accordingly by post filtering information contained in the bit stream.
One embodiment uses as its detection criterion the total signal power content in the audio time signal below a pitch frequency, possibly a pitch frequency estimated by a long-term prediction in the encoder. If this is greater than a predetermined threshold, it is considered that there are other relevant components than the pitch component (including harmonics), which will cause the post filter to be disabled.
In an encoder comprising a CELP module, use can be made of the fact that such a module estimates the pitch frequency of the audio time signal. Then, a further detection criterion is to check for energy content between or below the harmonics of this frequency, as described in more detail above.
As a further development of the preceding embodiment including a CELP module, the decision may include a comparison between an estimated power of the audio signal when CELP-coded (i.e., encoded and decoded) and an estimated power of the audio signal when CELP-coded and post-filtered. If the power difference is larger than a threshold, which may indicate that a relevant, non-noise component of the signal will be lost, and the encoder will decide to disable the post filter.
In an advantageous embodiment, the encoder comprises a CELP module and a TCX module. As is known in the art, TCX coding is advantageous in respect of certain kinds of signals, notably non-vocal signals. It is not common practice to apply post-filtering to a TCX-coded signal. Thus, the encoder may select either TCX coding, CELP coding with post filtering or CELP coding without post filtering, thereby covering a considerable range of signal types.
As one further development of the preceding embodiment, the decision between the three coding modes is taken on the basis of a rate—distortion criterion, that is, applying an optimization procedure known per se in the art.
In another further development of the preceding embodiment, the encoder further comprises an Advanced Audio Coding (AAC) coder, which is also known to be particularly suitable for certain types of signals. Preferably, the decision whether to apply AAC (frequency-domain) coding is made separately from the decision as to which of the other (linear-prediction) modes to use. Thus, the encoder can be apprehended as being operable in two super-modes, AAC or TCX/CELP, in the latter of which the encoder will select between TCX, post-filtered CELP or non-filtered CELP. This embodiment enables processing of an even wider range of audio signal types.
In one embodiment, the encoder can decide that a post filtering at decoding is to be applied gradually, that is, with gradually increasing gain. Likewise, it may decide that post filtering is to be removed gradually. Such gradual application and removal makes switching between regimes with and without post filtering less perceptible. As one example, a singing episode, for which post-filtered CELP coding is found to be suitable, may be preceded by an instrumental episode, wherein TCX coding is optimal; a decoder according to the invention may then apply post filtering gradually at or near the beginning of the singing episode, so that the benefits of post filtering are preserved even though annoying switching artefacts are avoided.
In one embodiment, the decision as to whether post filtering is to be applied is based on an approximate difference signal, which approximates that signal component which is to be removed from a future decoded signal by the post filter. As one option, the approximate difference signal is computed as the difference between the audio time signal and the audio time signal when subjected to (simulated) post filtering. As another option, an encoding section extracts an intermediate decoded signal, whereby the approximate difference signal can be computed as the difference between the audio time signal and the intermediate decoded signal when subjected to post filtering. The intermediate decoded signal may be stored in a long-term prediction buffer of the encoder. It may further represent the excitation of the signal, implying that further synthesis filtering (vocal tract, resonances) would need to be applied to obtain the final decoded signal. The point in using an intermediate decoded signal is that it captures some of the particularities, notably weaknesses, of the coding method, thereby allowing a more realistic estimation of the effect of the post filter. As a third option, a decoding section extracts an intermediate decoded signal, whereby the approximate difference signal can be computed as the difference between the intermediate decoded signal and the intermediate decoded signal when subjected to post filtering. This procedure probably gives a less reliable estimation than the two first options, but can on the other hand be carried out by the decoder in a standalone fashion.
The approximate difference signal thus obtained is then assessed with respect to one of the following criteria, which when settled in the affirmative will lead to a decision to disable the post filter:
a) whether the power of the approximate difference signal exceeds a predetermined threshold, indicating that a significant part of the signal would be removed by the post filter;
b) whether the character of the approximate difference signal is rather tonal than noise-like;
c) whether a difference between magnitude frequency spectra of the approximate difference signal and of the audio time signal is unevenly distributed with respect to frequency, suggesting that it is not noise but rather a signal that would make sense to a human listener;
d) whether a magnitude frequency spectrum of the approximate difference signal is localized to frequency intervals within a predetermined relevance envelope, based on what can usually be expected from a signal of the type to be processed; and
e) whether a magnitude frequency spectrum of the approximate difference signal is localized to frequency intervals within a relevance envelope obtained by thresholding a magnitude frequency spectrum of the audio time signal by a magnitude of the largest signal component therein downscaled by a predetermined scale factor. When evaluating criterion e), it is advantageous to apply peak tracking in the magnitude spectrum, that is, to distinguish portions having peak-like shapes normally associated with tonal components rather than noise. Components identified by peak tracking, which may take place by some algorithm known per se in the art, may be further sorted by applying a threshold to the peak height, whereby the remaining components are tonal material of a certain magnitude. Such components usually represent relevant signal content rather than noise, which motivates a decision to disable the post filter.
In one embodiment of the invention as a decoder, the decision to disable the post filter is executed by a switch controllable by the control section and capable of bypassing the post filter in the circuit. In another embodiment, the post filter has variable gain controllable by the control section, or a gain controller therein, wherein the decision to disable is carried out by setting the post filter gain (see previous section) to zero or by setting its absolute value below a predetermined threshold.
In one embodiment, decoding according to the present invention includes extracting post filtering information from the bit stream signal which is being decoded. More precisely, the post filtering information may be encoded in a data field comprising at least one bit in a format suitable for transmission. Advantageously, the data field is an existing field defined by an applicable standard but not in use, so that the post filtering information does not increase the payload to be transmitted.
In other embodiments, an audio decoder for decoding an audio bitstream is disclosed. The decoder includes a first decoding module adapted to operate in a first coding mode and a second decoding module adapted to operate in a second coding mode, the second coding mode being different from the first coding mode. The decoder further includes a pitch filter in either the first coding mode or the second coding mode, the pitch filter adapted to filter a preliminary audio signal generated by the first decoding module or the second decoding module to obtain a filtered signal. The pitch filter is selectively enabled or disabled based on a value of a first parameter encoded in the audio bitstream, the first parameter being distinct from a second parameter encoded in the audio bitstream, the second parameter specifying a current coding mode of the audio decoder.
In some embodiments, a pitch filter for filtering a preliminary audio signal generated from an audio bitstream is disclosed. The pitch filter has an operating mode selected from one of either: (i) an active mode where the preliminary audio signal is filtered using filtering information to obtain a filtered audio signal, and (ii) an inactive mode where the pitch filter is disabled. The preliminary audio signal is generated in an audio encoder or audio decoder having a coding mode selected from at least two distinct coding modes, and the pitch filter is capable of being selectively operated in either the active mode or the inactive mode while operating in the coding mode based on control information.
It is noted that the methods and apparatus disclosed in this section may be applied, after appropriate modifications within the skilled person's abilities including routine experimentation, to coding of signals having several components, possibly corresponding to different channels, such as stereo channels. Throughout the present application, pitch enhancement and post filtering are used as synonyms. It is further noted that AAC is discussed as a representative example of frequency-domain coding methods. Indeed, applying the invention to a decoder or encoder operable in a frequency-domain coding mode other than AAC will only require small modifications, if any, within the skilled person's abilities. Similarly, TCX is mentioned as an example of weighted linear prediction transform coding and of transform coding in general.
Features from two or more embodiments described hereinabove can be combined, unless they are clearly complementary, in further embodiments. The fact that two features are recited in different claims does not preclude that they can be combined to advantage. Likewise, further embodiments can also be provided by the omission of certain features that are not necessary or not essential for the desired purpose.
Embodiments of the present invention will now be described with reference to the accompanying drawings, on which:
As a variation, the decoder system of
s
ORIG(n)−SE(n)=SORIG(n)−(SDEC(n)−α[SDEC*pLT*hLP](n)),
where α is the post filter gain. By studying the total energy, low-band energy, tonality, actual magnitude spectrum or past magnitude spectra of this signal, as disclosed in the Summary section and the claims, the control section may find a basis for the decision whether to activate or deactivate the pitch enhancement module 740.
Preferably, the decision module 820 bases its decision on an approximate difference signal computed from an intermediate decoded signal Si_DEC, which can be subtracted from the encoding module 810. The intermediate decoded signal represents an intermediate stage in the decoding process, as discussed in preceding paragraphs, but may be extracted from a corresponding stage of the encoding process. However, in the encoder system 800 the original audio signal SORIG is available so that, advantageously, the approximate difference signal is formed as:
S
ORIG(n)−(Si_DEC(n)−α[(Si_DEC*pLT)*hLP](n)).
The approximation resides in the fact that the intermediate decoded signal is used in lieu of the final decoded signal. This enables an appraisal of the nature of the component that a post filter would remove at decoding, and by applying one of the criteria discussed in the Summary section, the decision module 820 will be able to take a decision whether to disable post filtering.
As a variation to this, the decision module 820 may use the original signal in place of an intermediate decoded signal, so that the approximate difference signal will be [(Si_DEC* pLT)*hLP](n). This is likely to be a less faithful approximation but on the other hand makes the presence of a connection line 816 between the decision module 820 and the encoding module 810 optional.
In such other variations of this embodiment where the decision module 820 studies the audio signal directly, one or more of the following criteria may be applied:
In all the described variations of the encoder structure shown in
A 6-person listening test has been carried out, during which music samples encoded and decoded according to the invention were compared with reference samples containing the same music coded while applying post filtering in the conventional fashion but maintaining all other parameters unchanged. The results confirm a perceived quality improvement.
Further embodiments of the present invention will become apparent to a person skilled in the art after reading the description above. Even though the present description and drawings disclose embodiments and examples, the invention is not restricted to these specific examples. Numerous modifications and variations can be made without departing from the scope of the present invention, which is defined by the accompanying claims.
The systems and methods disclosed hereinabove may be implemented as software, firmware, hardware or a combination thereof. Certain components or all components may be implemented as software executed by a digital signal processor or microprocessor, or be implemented as hardware or as an application-specific integrated circuit. Such software may be distributed on computer readable media, which may comprise computer storage media (or non-transitory media) and communication media (or transitory media). As is well known to a person skilled in the art, computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by a computer. Further, it is well known to the skilled person that communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media.
This application is a continuation of U.S. patent application Ser. No. 16/351,133 filed on Mar. 12, 2019, which in turn is a continuation of U.S. patent application Ser. No. 15/792,589, filed Oct. 24, 2017 (now U.S. Pat. No. 10,236,010, issued Mar. 19, 2019), which in turn is a divisional of U.S. patent application Ser. No. 15/086,409, filed Mar. 31, 2016 (now U.S. Pat. No. 9,858,940, issued Jan. 2, 2018), which in turn is a continuation of U.S. patent application Ser. No. 14/936,408, filed Nov. 9, 2015 (now U.S. Pat. No. 9,343,077, issued May 17, 2016), which in turn is a continuation of U.S. patent application Ser. No. 13/703,875, filed Dec. 12, 2012 (now U.S. Pat. No. 9,224,403, issued Dec. 29, 2015), which in turn is the 371 National Stage of International Application No. PCT/EP2011/060555 having an international filing date of Jun. 23, 2011. PCT/EP2011/060555 claims priority to U.S. Provisional Patent Application No. 61/361,237, filed Jul. 2, 2010. The entire contents of U.S. Ser. Nos. 16/351,133, 15/792,589 (now U.S. Pat. No. 10,236,010), U.S. Ser. No. 15/086,409 (now U.S. Pat. No. 9,858,940), U.S. Pat. No. 14/936,408 (now U.S. Pat. No. 9,343,077), U.S. Ser. No. 13/703,875 (now U.S. Pat. No. 9,224,403), PCT/EP2011/060555 and U.S. 61/361,237 are hereby incorporated by reference in their entirety.
| Number | Date | Country | |
|---|---|---|---|
| 61361237 | Jul 2010 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 15086409 | Mar 2016 | US |
| Child | 15792589 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 16351133 | Mar 2019 | US |
| Child | 17073228 | US | |
| Parent | 15792589 | Oct 2017 | US |
| Child | 16351133 | US | |
| Parent | 14936408 | Nov 2015 | US |
| Child | 15086409 | US | |
| Parent | 13703875 | Dec 2012 | US |
| Child | 14936408 | US |
