Method and device for arithmetic encoding or arithmetic decoding
Patent Information
- Application Number
- AU2025210849
- Authority / Receiving Office
- AU · AU
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-07-31
- Publication Date
- 2026-08-06
AI Technical Summary
Existing arithmetic coding methods face challenges with high encoding/decoding latency and memory capacity requirements due to the large number of contexts and corresponding probability density functions needed for efficient compression of multimedia data.
The method employs a context class-based approach using a look-up table or hash table to map different context classes to probability density functions, determined from the sum of quantized absolutes of preceding spectral coefficients, reducing the complexity and memory requirements.
This approach achieves efficient compression while decreasing encoding/decoding latency and memory capacity, utilizing context classes to group similar contexts and simplify the mapping process.
Smart Images
Figure 00000046_0000 
Figure 00000047_0000 
Figure 00000048_0000
Abstract
Description
RELATED APPLICATION This application is a divisional of Australian Patent 5 Application No. 2024219410 filed on 5 September 2024, which is a divisional of Australian Patent Application No. 2023226688 filed on 6 September 2023, which is a divisional of Australian Patent Application No. 2021277715 filed on 2 December 2021, which is a divisional of Australian Patent 10 Application No. 2020203538 filed on 29 May 2020, which is a divisional of Australian Patent Application No. 2017272250 filed on 7 December 2017, which is a divisional of Australian Patent Application No. 2015246076 filed on 20 October 2015, which is a divisional of Australian 15 Patent Application No. 2010305548 filed on 1 October 2010 which is the national phase application of PCT Patent Application No. PCT / EP2010 / 064644, and claims priority to EP Provisional Patent Application No. EP09305961.6, filed 1 October 2010, filed 9 October 2009, respectively. The 20 disclosures of the applications listed in this paragraph are incorporated herein by reference in their entirety and for all purposes. TECHNICAL FIELD The disclosure is related to arithmetic encoding and 25 decoding of multimedia data. BACKGROUND Any discussion of the prior art throughout the specification should in no way be considered as an admission that such prior art is widely known or forms part 30 of common general knowledge in the field. 2025210849 31 Jul 2025 Arithmetic coding is a method for lossless compression of data. Arithmetic coding is based on a probability density function (PDF). For achieving a compression effect, the probability density function on which the coding is based 5 has to be identical to or at least resemble -the closer the better- the actual probability density function which the data actually follows. If arithmetic coding is based on a suitable probability density function, it may achieve significant compression 10 resulting in at least almost optimal code. Therefore, arithmetic coding is a frequently used technique in audio, speech or video coding for encoding and decoding of coefficient sequences wherein coefficients are quantized time-frequency-transform of video pixels or audio or speech 15 signal sample values in binary representation. For even improving compression, arithmetic coding may be based on a set of probability density functions, wherein the probability density function used for coding a current coefficient depends on a context of said current 20 coefficient. That is, different probability density functions may be used for coding of a same quantization value in dependency on a context in which the coefficient having said same quantization value appears. The context of a coefficient is defined by the quantization values of 25 coefficients comprised in a neighbourhood of one or more neighbouring coefficients neighbouring the respective coefficient, e.g. a subsequence of one or more already encoded or already decoded coefficients adjacently preceding, in a sequence, the respective coefficient to-be- 30 encoded or to-be-decoded. Each of the different possible appearances the neighbourhood may take defines a different possible context each being mapped onto an associated probability density function. 2025210849 31 Jul 2025 In practice, said compression improvement becomes manifest only if the neighbourhood is sufficiently large. This comes along with a combinatory explosion of the number of different possible contexts as well as a corresponding huge 5 number of possible probability density functions or a correspondingly complex mapping. An example of an context based arithmetic coding scheme can be found in ISO / IEC JTC1 / SC29 / WG11 N10847, July 2009, London, UK, proposing a reference model for Unified Speech 10 and Audio Coding (USAC). According to the proposal, vectors containing 4 spectral bins are jointly coded which results in a tremendous number of contexts to-be-handled. Throughout this specification the word "comprise", or variations such as "comprises" or "comprising", will be 15 understood to imply the inclusion of a stated element, integer or step, or group of elements, integers or steps, but not the exclusion of any other element, integer or step, or group of elements, integers or steps. Any discussion of documents, acts, materials, devices, 20 articles or the like which has been included in the present specification is not to be taken as an admission that any or all of these matters form part of the prior art base or were common general knowledge in the field relevant to the present disclosure as it existed before the priority date 25 of each claim of this application. SUMMARY OF THE INVENTION It is an object of the present invention to substantially overcome or at least ameliorate one or more disadvantages of existing arrangements. 30 Corresponding to the tremendous number of contexts to-behandled there are a tremendous number of probability 2025210849 31 Jul 2025 density functions which need to be stored, retrieved, and handled or at least a correspondingly complex mapping from contexts to probability density functions. This increases at least one of encoding / decoding latency and memory 5 capacity requirements. There is a need in the art for an alternative solution allowing to achieving compression similarly well while decreasing at least one of encoding / decoding latency and memory capacity requirements. One embodiment provides a method for arithmetic decoding of 10 a current spectral coefficient, the method comprising: processing preceding spectral coefficients; determining a context class based on the preceding spectral coefficients, wherein the context class is determined from at least two different context classes, and wherein the context class is 15 based on a sum of quantized absolutes of a processed preceding spectral coefficients; receiving a reset signal, wherein, in response to the reset signal, only preceding spectral coefficients of current frame are used to determine the context class; determining a probability 20 density function based on a mapping from the at least two different context classes to at least two different probability density functions, wherein the mapping is based on a look-up table or a hash table; and arithmetic decoding the current spectral coefficient based on the probability 25 density function, wherein the decoding is based on a look up table which is determined based on the context class. One embodiment provides an apparatus for arithmetic decoding of a current spectral coefficient, the apparatus comprising: a processor configured to process preceding 30 spectral coefficients; a context classifier configured to determine a context class based on a processed preceding spectral coefficients, wherein the context class is determined from at least two different context classes, and 2025210849 31 Jul 2025 wherein the context class is based on a sum of quantized absolutes of the preceding spectral coefficients, a receiver for receiving a reset signal, wherein, in response to the reset signal, only preceding spectral coefficients 5 of current frame are used to determine the context class; a probability density module configured to determine a probability density function, said probability density module being adapted to use a mapping from the at least two different context classes to at least two different 10 probability density functions to determine the probability density function; and an arithmetic decoder configured to arithmetically decode the current spectral coefficient based on the probability density function, wherein the decoding is based on a look up table which is determined 15 based on the context class, wherein the mapping is based on a look-up table or a hash table. One embodiment provides a method for arithmetic decoding of a current spectral coefficient, the method comprising: 20 processing preceding spectral coefficients; determining a context state based on the preceding spectral coefficients, wherein the context state is determined from at least two different context states, and wherein the context state is based on a sum of quantized absolutes of a processed 25 preceding spectral coefficients; determining a probability density function based on a mapping from the at least two different context states to at least two different probability density functions; and arithmetic decoding the current spectral coefficient based on the probability 30 density function, wherein the decoding is based on a look up table which is determined based on the context state. One embodiment provides an apparatus for arithmetic decoding of a current spectral coefficient, the apparatus 2025210849 31 Jul 2025 comprising: a processor configured to process preceding spectral coefficients; a context classifier configured to determine a context state based on a processed preceding spectral coefficients, wherein the context state is 5 determined from at least two different context states, and wherein the context state is based on a sum of quantized absolutes of the preceding spectral coefficients, a probability density module configured to determine a probability density function, said probability density 10 module being adapted to use a mapping from the at least two different context states to at least two different probability density functions to determine the probability density function; and an arithmetic decoder configured to arithmetically decode the current spectral coefficient 15 based on the probability density function, wherein the decoding is based on a look up table which is determined based on the context state. One embodiment provides a method for arithmetic decoding of a current spectral coefficient, the method comprising: 20 determining a context state based on preceding spectral coefficients, wherein the context state is determined from at least two different context states, and wherein the context state is based on a sum of quantized absolutes of the preceding spectral coefficients; 25 receiving a reset signal, wherein, in response to the reset signal, only preceding spectral coefficients of current frame are used to determine the context state; determining a probability density function based on the context state and a mapping from the at least two 30 different context states to at least two different probability density functions; and arithmetic decoding the current spectral coefficient based on the probability density function. 2025210849 31 Jul 2025 One embodiment provides a non-transitory storage medium for arithmetic decoding according to the method as herein disclosed. One embodiment provides an apparatus for arithmetic 5 decoding of a current spectral coefficient, the apparatus comprising: a context classifier configured to determine a context state based on preceding spectral coefficients, wherein the context state is determined from at least two different 10 context states, and wherein the context state is based on a sum of quantized absolutes of the preceding spectral coefficients, a receiver for receiving a reset signal, wherein the context classifier, in response to the reset signal, uses 15 only preceding spectral coefficients of current frame to determine the context class; a probability density module configured to determine a probability density function, said probability density module being adapted to use a determined context state and 20 a mapping from the at least two different context states to at least two different probability density functions to determine the probability dens ity function; and an arithmetic decoder configured to arithmetically 25 decode the current spectral coefficient based on the probability density function. One embodiment provides a method for arithmetic decoding of a current spectral coefficient, said method comprising: processing preceding spectral coefficients; 30 determining a context state based on the processed preceding spectral coefficients, wherein the context state is determined from at least two different context states, and wherein the context state is based on a sum of the 2025210849 31 Jul 2025 quantised absolutes of the preceding spectral coefficients; determing a probability density function based on the determined context state and a mapping from the at least two different context states to at least two different 5 probability density functions; and arithmetic decoding the current spectral coefficient based on the determined probability density function. One embodiment provides a method for arithmetic encoding of a current spectral coefficient, said method comprising: 10 processing preceding spectral coefficients; using the processed preceding spectral coefficients for determining a context state being one of at least two different context states, using a sum of the quantised absolutes of the 15 preceding spectral coefficients for determination of the context state; using the determined context state and a mapping from the at least two different context states to at least two different probability density functions for determining the 20 probability density function; and arithmetic encoding the current spectral coefficient based on the determined probability density function. One embodiment provides a non-transitory storage medium 25 carrying arithmetic encoded spectral coefficients arithmetic encoded according to the method as herein disclosed. One embodiment provides a device for arithmetic decoding of a current spectral coefficient, said device comprising: 30 a processor configured to process the preceding spectral coefficients; 2025210849 31 Jul 2025 state based on the processed preceding spectral coefficients, wherein the context state is determined from at least two different context states, and wherein the context state is based on a sum of the quantised absolutes 5 of the preceding spectral coefficients, wherein said context classifier is configured to use a sum of the quantised absolutes of the preceding spectral coefficients in determination of the context state; a probability density module configured to determine a 10 probability density function, said probability density module being adapted to use the determined context state and a mapping from the at least two different context states to at least two different probability density functions to determine the probability density function; 15 and an arithmetic decoder configured to arithmetically decode the current spectral coefficient based on the determined probability density function. One embodiment provides a method for arithmetic decoding of a current spectral coefficient using preceding spectral 20 coefficients, said preceding spectral coefficients being already decoded, wherein said preceding spectral coefficients and said current spectral coefficient are comprised in one or more quantized spectra resulting from quantizing time-frequency-transformed video, audio or 25 speech signal sample values, said method comprising the steps: processing the preceding spectral coefficients; using the processed preceding spectral coefficients for determining a context class out of at least two different context classes, wherein a sum of the quantised absolute 30 values of the preceding spectral coefficients is used for determination of the context class; using the determined context class and a mapping from the at least two different context classes to at least two different probability 2025210849 31 Jul 2025 density functions, for determining one of the at least two different probability density function; arithmetic decoding the current spectral coefficient based on the determined probability density function, wherein said processing of 5 the preceding spectral coefficients comprises non-uniformly quantizing absolute values of the preceding spectral coefficients for use in determination of the context class, receiving at least one of a mode switching signal and a reset signal; using the at least one received signal for 10 controlling the step of determining the context class. One embodiment provides a method for arithmetic encoding of a current spectral coefficient using preceding spectral coefficients, said preceding spectral coefficients being already encoded, wherein said preceding spectral 15 coefficients and said current spectral coefficient are comprised in one or more quantized spectra resulting from quantizing time-frequency-transformed video, audio or speech signal sample values, said method comprising the steps: processing the preceding spectral coefficients; 20 using the processed preceding spectral coefficients for determining a context class out of at least two different context classes, wherein a sum of the quantised absolute values of the preceding spectral coefficients is used for the determination of the context class; using the 25 determined context class and a mapping from the at least two different context classes to at least two different probability density functions, for determining one of the at least two different probability density functions; arithmetic encoding the current spectral coefficient based 30 on the determined probability density function, wherein said processing of the preceding spectral coefficients comprises non-uniformly quantizing absolute values of the preceding spectral coefficients for use in determination of 2025210849 31 Jul 2025 the context class, inserting at least one of a mode switching signal and a reset signal; using the at least one inserted signal for controlling the step of determining the context class. 5 One embodiment provides a device for arithmetic decoding of a current spectral coefficient using preceding spectral coefficients, said preceding spectral coefficients being already decoded, wherein said preceding spectral coefficients and said current spectral coefficient are 10 comprised in one or more quantized spectra resulting from quantizing time-frequency-transformed video, audio or speech signal sample values, said device comprising: processing means for processing the preceding spectral coefficients, first means for determining a context class 15 out of at least two different context classes, said first means being adapted for using the processed preceding spectral coefficients for determining the context class, wherein a sum of the quantised absolute values of the preceding spectral coefficients is used for the 20 determination of the context class; second means for determining a probability density function, said second means being adapted for using the determined context class and a mapping from the at least two different context classes to at least two different probability density 25 functions for determining the probability density function; an arithmetic decoder for arithmetic decoding the current spectral coefficient based on the determined probability density function, wherein said processing means are adapted for non-uniformly quantizing absolute values of the 30 preceding spectral coefficients for use in determination of the context class, wherein said first means is adapted to receive at least one of a mode switching signal and a reset signal and wherein said first means is adapted to use the 2025210849 31 Jul 2025 at least one received signal to control the step of determining the context class. One embodiment provides a device for arithmetic encoding of a current spectral coefficient using preceding spectral 5 coefficients, said preceding spectral coefficients being already encoded, wherein said preceding spectral coefficients and current spectral coefficient are comprised in one or more quantized spectra resulting from quantizing time-frequency-transformed video, audio or speech signal 10 sample values, said device comprising: processing means for processing the preceding spectral coefficients, first means for determining a context class out of at least two different context classes, said first means being adapted for using the processed preceding spectral coefficients for 15 determining the context class, wherein a sum of the quantised absolute values of the preceding spectral coefficients is used for the determination of the context class; second means for determining a probability density function, said second means being adapted for using the 20 determined context class and a mapping from the at least two different context classes to at least two different probability density functions for determining the probability density function; an arithmetic encoder for arithmetic encoding the current spectral coefficient based 25 on the determined probability density function, wherein said processing means are adapted for non-uniformly quantizing absolute values of the preceding spectral coefficients for use in determination of the context class, wherein said first means is adapted to insert at least one 30 of a mode switching signal and a reset signal and wherein said first means is adapted to use the at least one inserted signal to control the step of determining the 2025210849 31 Jul 2025 In one embodiment, a non-transitory storage medium carries arithmetic encoded spectral coefficients wherein a current spectral coefficient is encoded using preceding spectral coefficients, said preceding spectral coefficients being 5 already encoded, and both, said preceding and current spectral coefficients are comprised in one or more quantised spectra resulting from quantising time-frequency-transform of video, audio or speech signal sample values, wherein the preceding spectral coefficients are processed; 10 the processed preceding spectral coefficients are used for determining a context class being one of at least two different context classes, a sum of the quantised absolutes of the preceding spectral coefficients is used for determination of the context class, the determined context 15 class and a mapping from the at least two different context classes to at least two different probability density functions are used for determining the probability density function; the current spectral coefficient is arithmetically encoded based on the determined probability 20 density function, and wherein the preceding spectral coefficients are processed comprising non-uniformly quantising absolutes of the preceding spectral coefficients for use in determination of the context class, wherein at least one of a mode switching signal and a reset signal is 25 present for controlling determining of the context class. The features of further proposed embodiments are specified in the dependent claims. Said method for arithmetic encoding, or decoding, respectively, uses preceding spectral coefficients for 30 arithmetic encoding or decoding, respectively, of a current spectral coefficient, wherein said preceding spectral coefficients are already encoded, or decoded, respectively. Both, said preceding spectral coefficients and said current 2025210849 31 Jul 2025 spectral coefficient, are comprised in one or more quantized spectra resulting from quantizing time-frequency-transform of video, audio or speech signal sample values. Said method further comprises processing the preceding 5 spectral coefficients, using the processed preceding spectral coefficients for determining a context class being one of at least two different context classes, using the determined context class and a mapping from the at least two different context classes to at least two different 10 probability density functions for determining the probability density function, and arithmetic encoding, or decoding, respectively, the current spectral coefficient based on the determined probability density function. It is a feature of the method that processing the preceding 15 spectral coefficients comprises non-uniformly quantizing the preceding spectral coefficients. The use of context classes as alternative to contexts for determining the probability density function allows for grouping two or more different contexts which result into 20 different but very similar probability density functions into a single context class being mapped onto a single probability density function. The grouping is achieved by using non-uniformly quantized preceding spectral coefficients for determining the context class. 25 For instance, there is an embodiment in which processing the preceding spectral coefficients comprises determining absolutes of the preceding spectral coefficients. Similarly, there is an embodiment of the device for arithmetic encoding as well as a corresponding embodiment 30 of the device for arithmetic decoding in which the processing means are adapted for determining absolutes of the preceding spectral coefficients. 2025210849 31 Jul 2025 In further embodiments of the devices, the processing means may be adapted such that processing the preceding spectral coefficients further comprises a first quantization in which the absolutes of the preceding spectral coefficients 5 are quantized according a first quantization scheme, a variance determination in which variance of the absolutes of the preceding spectral coefficients quantized according the first quantization scheme is determined, usage of the determined variance for selection of one of at least two 10 different non-linear second quantization schemes, and a second quantization in which the absolutes of the preceding spectral coefficients quantized according the first quantization scheme are further quantized according to the selected non-linear second quantization scheme. 15 Variance determination may comprise determination of a sum of the absolutes of the preceding spectral coefficients quantized according the first quantization scheme and comparison of the determined sum with at least one threshold. 20 In further embodiments, the processing means of each of the devices may be adapted such that processing either results in a first outcome or at least a different second outcome. Then, determination of the context class further comprises determination of a number of those preceding spectral 25 coefficients for which processing resulted in the first outcome, and usage of the determined number for determination of the context class. Each of the devices may comprise means for receiving at least one of a mode switching signal and a reset signal 30 wherein devices are adapted for using the at least one received signal for controlling the determination of the context class. 2025210849 31 Jul 2025 The at least two different probability density functions may be determined beforehand using a representative set of data for determining the at least two different probability density functions and the mapping may be realized using a 5 look-up table or a hash table. One embodiment provides a device for arithmetic decoding of a current spectral coefficient using preceding spectral coefficients, said preceding spectral coefficients being already decoded, wherein said preceding spectral 10 coefficients and said current spectral coefficient are comprised in one or more quantized spectra resulting from quantizing time-frequency-transformed video, audio or speech signal sample values, said device comprising processing means for processing the preceding spectral 15 coefficients; first means for determining a context class out of at least two different context classes, said first means being adapted for using the processed preceding spectral coefficients for determining the context class, wherein a sum of the quantised absolute values of the 20 preceding spectral coefficients is used for determination of the context class; second means for determining a probability density function, said second means being adapted for using the determined context class and a mapping from the at least two different context classes to 25 at least two different probability density functions for determining the probability density function; an arithmetic decoder for arithmetic decoding the current spectral coefficient based on the determined probability density function, wherein said processing means are adapted for 30 non-uniformly quantising absolute values of the preceding spectral coefficients for use in determination of the context class, wherein said second means is configured to 2025210849 31 Jul 2025 use a representative set of data to determine the at least two different probability density functions. One embodiment provides a device for arithmetic encoding of a current spectral coefficient using preceding spectral 5 coefficients, said preceding spectral coefficients being already encoded, wherein said preceding spectral coefficients and said current spectral coefficient are comprised in one or more quantized spectra resulting from quantizing time-frequency-transformed video, audio or 10 speech signal sample values, said device comprising processing means for processing the preceding spectral coefficients; first means for determining a context class out of at least two different context classes, said first means being adapted for using the processed preceding 15 spectral coefficients for determining the context class, wherein a sum of the quantised absolute values of the preceding spectral coefficients is used for determination of the context class; second means for determining the probability density function, said second means being 20 adapted for using the determined context class and a mapping from the at least two different context classes to at least two different probability density functions for determining the probability density function; an arithmetic encoder for arithmetic encoding the current spectral 25 coefficient based on the determined probability density function, wherein said processing means are adapted for non-uniformly quantising absolute values of the preceding spectral coefficients for use in determination of the context class, wherein said second means is configured to 30 use a representative set of data to determine the at least two different probability density functions. Another embodiment provides a method for arithmetic decoding of a current spectral coefficient, the method 2025210849 31 Jul 2025 comprising: processing preceding spectral coefficients; determining a context class based on the preceding spectral coefficients, wherein the context class is determined from at least two different context classes, and wherein the 5 context class is based on a sum of quantized absolutes of a processed preceding spectral coefficients; determining a probability density function based on a mapping from the at least two different context classes to at least two different probability density functions; and arithmetic 10 decoding the current spectral coefficient based on the probability density function, wherein the decoding is based on a look up table which is determined based on the context class. Another embodiment provides a non-transitory storage medium 15 for arithmetic decoding according to the invention. A further embodiment provides an apparatus for arithmetic decoding of a current spectral coefficient, the apparatus comprising: a processor configured to process preceding spectral coefficients; a context classifier configured to 20 determine a context class based on a processed preceding spectral coefficients, wherein the context class is determined from at least two different context class, and wherein the context class is based on a sum of quantized absolutes of the preceding spectral coefficients, a 25 probability density module configured to determine a probability density function, said probability density module being adapted to use a mapping from the at least two different context classes to at least two different probability density functions to determine the probability 30 density function; and an arithmetic decoder configured to arithmetically decode the current spectral coefficient based on the probability density function, wherein the 2025210849 31 Jul 2025 decoding is based on a look up table which is determined based on the context class. A further embodiment provides a method for arithmetic decoding of a current spectral coefficient of a current 5 frame using preceding spectral coefficients of the current frame and / or a previous frame, said preceding spectral coefficients being already decoded, wherein said preceding spectral coefficients and said current spectral coefficients, are comprised in one or more quantised 10 spectra resulting from quantising time-frequency-transformed video, audio or speech signal sample values, said method comprising the steps: - performing a first quantising in which the absolute values of preceding spectral coefficients are quantised 15 according a first quantisation scheme; - determining a variance of the absolute values of the preceding spectral coefficients quantised according to the first quantisation scheme, comprising determining a sum of the absolute values of the preceding spectral 20 coefficients quantised according the first quantisation scheme and comparing the determined sum with at least one threshold; - using the determined variance for selecting one of at least two different non-linear second quantisation 25 schemes; - performing a second quantising in which the absolute values of the preceding spectral coefficients quantised according the first quantisation scheme are further quantised according to the selected non-linear second 30 quantisation scheme; 2025210849 31 Jul 2025 - determining a context class out of at least two different context classes, wherein the context class is determined based on an output of the second quantising; - determining the probability density function using the 5 determined context class and a mapping from the at least two different context classes to at least two different probability density functions, wherein the mapping is realised using a look-up table or a hash table; and - arithmetic decoding the current spectral coefficient 10 based on the determined probability density function. A further embodiment provides a method for arithmetic encoding of a current spectral coefficient of a current frame using preceding spectral coefficients of the current frame and / or a previous frame, said preceding spectral 15 coefficients being already encoded, wherein said preceding spectral coefficients and said current spectral coefficients, are comprised in one or more quantised spectra resulting from quantising time-frequency-transformed video, audio or speech signal sample values, 20 said method comprising the steps: - performing a first quantising in which the absolute values of the preceding spectral coefficients are quantised according a first quantisation scheme; - determining a variance of the absolute values of the 25 preceding spectral coefficients quantised according to the first quantisation scheme, comprising determining a sum of the absolute values of the preceding spectral coefficients quantised according the first quantisation scheme and comparing the determined sum with at least 30 one threshold; 2025210849 31 Jul 2025 - using the determined variance for selecting one of at least two different non-linear second quantisation schemes; and - performing a second quantising in which the absolute 5 values of the preceding spectral coefficients quantised according the first quantisation scheme are further quantised according to the selected non-linear second quantisation scheme; - determining a context class out of at least two 10 different context classes, wherein the context class is determined based on an output of the second quantising; - determining the probability density function using the determined context class and a mapping from the at least two different context classes to at least two different 15 probability density functions, wherein the mapping is realised using a look-up table or a hash table; and - arithmetic encoding the current spectral coefficient based on the determined probability density function. A further embodiment provides a device for arithmetic 20 encoding of a current spectral coefficient of a current frame using preceding spectral coefficients of the current frame and / or a previous frame, said preceding spectral coefficients being already encoded, wherein said preceding spectral coefficients and said current spectral 25 coefficients, are comprised in one or more quantised spectra resulting from quantising time-frequency-transformed video, audio or speech signal sample values, said device comprising: - processing means adapted for: 2025210849 31 Jul 2025 10 o performing a first quantising in which the absolute values of the preceding spectral coefficients are quantised according a first quantisation scheme; o determining a variance of the absolute values of the preceding spectral coefficients quantised according to the first quantisation scheme, comprising determining a sum of the absolute values of the preceding spectral coefficients quantised according the first quantisation scheme and comparing the determined sum with at least one threshold; o using the determined variance for selecting one of at least two different non-linear second 15 quantisation schemes; and o performing a second quantising in which the absolute values of the preceding spectral coefficients quantised according the first quantisation scheme are further quantised 20 according to the selected non-linear second quantisation scheme, and - first means for determining a context class out of at least two different context classes, said first means being adapted for determining the context class based on 25 an output of the second quantising; - second means for determining the probability density function, said second means being adapted for determining the probability density function using the determined context class and a mapping from the at least 30 two different context classes to at least two different 2025210849 31 Jul 2025 probability density functions, wherein the mapping is realised using a look-up table or a hash table; and - an arithmetic encoder for arithmetic encoding the current spectral coefficient based on the determined 5 probability density function. A further embodiment provides a device for arithmetic decoding of a current spectral coefficient of a current frame using preceding spectral coefficients of the current frame and / or a previous frame, said preceding spectral 10 coefficients being already decoded, wherein said preceding spectral coefficients and said current spectral coefficients, are comprised in one or more quantised spectra resulting from quantising time-frequency-transformed video, audio or speech signal sample values, 15 said device comprising: - processing means adapted for: o performing a first quantising in which the absolute values of the preceding spectral coefficients are quantised according a first 20 quantisation scheme; o determining a variance of the absolute values of the preceding spectral coefficients quantised according to the first quantisation scheme, comprising determining a sum of the absolute 25 values of the preceding spectral coefficients quantised according the first quantisation scheme and comparing the determined sum with at least one threshold; 2025210849 31 Jul 2025 o using the determined variance for selecting one of at least two different non-linear second quantisation schemes; and o performing a second quantising in which the 5 absolute values of the preceding spectral coefficients quantised according the first quantisation scheme are further quantised according to the selected non-linear second quantisation scheme, and 10 - first means for determining a context class out of at least two different context classes, said first means being adapted for determining the context class based on an output of the second quantizing; - second means for determining the probability density 15 function, said second means being adapted for determining the probability density function using the determined context class and a mapping from the at least two different context classes to at least two different probability density functions, wherein the mapping is 20 realised using a look-up table or a hash table; and - an arithmetic decoder for arithmetic decoding the current spectral coefficient based on the determined probability density function. A further embodiment provides a method for arithmetic 25 decoding of a current spectral coefficient, said method comprising: processing preceding spectral coefficients; determining a context state based on the processed preceding spectral coefficients, wherein the context state is determined from at least two different context states, 30 wherein the context state is based on a sum of the quantised absolutes of the preceding spectral coefficients, 2025210849 31 Jul 2025 and wherein the context state determination is based on using at least one of a reset signal and a mode switching signal; determining a probability density function based on the determined context state and a mapping from the at 5 least two different context states to at least two different probability density functions; and arithmetic decoding the current spectral coefficient based on the determined probability density function. A further embodiment provides a device for arithmetic 10 decoding of a current spectral coefficient, said device comprising: a processor configured to process the preceding spectral coefficients; a context classifier configured to determine a context state based on the processed preceding spectral coefficients, wherein the context state is 15 determined from at least two different context states, and wherein the context state is based on a sum of the quantised absolutes of the preceding spectral coefficients and wherein the context state determination is based on using at least one of a reset signal and a mode switching 20 signal; a probability density module configured to determine a probability density function, said probability density module being adapted to use the determined context state and a mapping from the at least two different context states to at least two different probability density 25 functions to determine the probability density function; and an arithmetic decoder configured to arithmetically decode the current spectral coefficient based on the determined probability density function. Unless the context clearly requires otherwise, throughout 30 the description and the claims, the words “comprise”, “comprising”, and the like are to be construed in an inclusive sense as opposed to an exclusive or exhaustive 2025210849 31 Jul 2025 sense; that is to say, in the sense of “including, but not limited to”. BRIEF DESCRIPTION OF THE DRAWINGS Examples are illustrated in the drawings and are explained 5 in more detail in the following description. The examples are explained only for elucidating the invention, but not limiting the invention’s scope and spirit defined in the claims. In the figures: 10 Fig. 1 exemplarily depicts an embodiment of the inventive encoder, Fig. 2 exemplarily depicts an embodiment of the inventive decoder, Fig. 3 exemplarily depicts a first embodiment of a 15 context classifier for determining a context class, Fig. 4 exemplarily depicts a second embodiment of a context classifier for determining a context class, 20 Fig. 5a exemplarily depicts a first neighbourhood of preceding spectral bins preceding a current spectral bin to-be-encoded or to-be-decoded in frequency domain mode, Fig. 5b exemplarily depicts a second neighbourhood of 25 preceding spectral bins preceding a current spectral bin to-be-encoded or to-be-decoded in weighted linear prediction transform mode, 2025210849 31 Jul 2025 Fig. 6a exemplarily depicts a third neighbourhood of preceding spectral bins preceding a current lowest frequency spectral bin to-be-encoded or to-be-decoded in frequency domain mode, 5 Fig. 6b exemplarily depicts a fourth neighbourhood of preceding spectral bins preceding a current second lowest frequency spectral bin to-be-encoded or to-be-decoded in frequency domain mode, 10 Fig. 7a exemplarily depicts a fifth neighbourhood of preceding spectral bins preceding a current lowest frequency spectral bin to-be-encoded or to-be-decoded in weighted linear prediction transform mode, 15 Fig. 7b exemplarily depicts a sixth neighbourhood of preceding spectral bins preceding a current second lowest frequency spectral bin to-be-encoded or to-be-decoded in weighted linear prediction transform mode, 20 Fig. 7c exemplarily depicts a seventh neighbourhood of preceding spectral bins preceding a current third lowest frequency spectral bin to-be-encoded or to-be-decoded in weighted linear prediction transform mode, 25 Fig. 7d exemplarily depicts an eighth neighbourhood of preceding spectral bins preceding a current third lowest frequency spectral bin to-be-encoded or to-be-decoded in weighted linear prediction transform mode, 30 Fig. 8 exemplarily depicts neighbourhoods of different spectral bins to-be-encoded or to-be-decoded, 2025210849 31 Jul 2025 said different spectral bin being comprised in a first spectrum to-be-encoded or to-be-decoded after initiation of encoding / decoding or occurrence of a reset signal in frequency domain 5 mode, and Fig. 9 exemplarily depicts further neighbourhoods of different spectral bins to-be-encoded or to-be-decoded in weighted linear prediction transform mode, said different spectral bin being comprised 10 in a second spectrum to-be-encoded or to-be- decoded after initiation of encoding / decoding or occurrence of a reset signal in weighted linear prediction transform mode. EXEMPLARY EMBODIMENTS 15 The invention may be realized on any electronic device comprising a processing device correspondingly adapted. For instance, the device for arithmetic decoding may be realized in a television, a mobile phone, or a personal computer, an mp3-player, a navigation system or a car audio 20 system. The device for arithmetic encoding may be realized in a mobile phone, a personal computer, an active car navigation system, a digital still camera, a digital video camera or a Dictaphone, to name a few. The exemplary embodiments described in the following are 25 related to encoding and decoding of quantized spectral bins resulting from quantization of time-frequency transform of multimedia samples. The disclosure is based on the way the already transmitted quantized spectral bins, e.g. preceding quantized spectral 30 bins preceding a current quantized spectral bin BIN in a sequence, are used to determine the probability density 2025210849 31 Jul 2025 function PDF to be used for arithmetic encoding and decoding, respectively, of the current quantized spectral bin BIN. The described exemplary embodiments of the methods and 5 devices for arithmetic encoding or arithmetic decoding comprise several steps or means, respectively, for nonuniform quantization. All steps or means, respectively, together offer the highest coding efficiency, but each step or means, respectively, alone already realizes the 10 inventive concept and provides advantages regarding encoding / decoding latency and / or memory requirements. Therefore, the detailed description shall be construed as describing exemplary embodiments realizing only one of the steps or means, respectively, described as well as 15 describing exemplary embodiments realizing combinations of two or more of the steps or means described. A first step which may but need not to be comprised in an exemplary embodiment of the method is a switching step in which it is decided which general transform mode shall be 20 used. For instance, in USAC Noiseless Coding Scheme the general transform mode may be either Frequency Domain (FD) mode or weighted Linear Prediction Transform (wLPT) mode. Each general mode might use a different neighbourhood, i.e. a different selection of already encoded or decoded, 25 respectively, spectral bins for the determination of the PDF. After that, the context of a current spectral bin BIN may be determined in module context generation COCL. From the determined context, a context class is determined by 30 classifying the context wherein, prior to classification, the context is processed by preferably but not necessarily non-uniform quantization NUQ1 of the spectral bins of the 2025210849 31 Jul 2025 context. Classification may comprise estimating a variance VES of the context and comparing the variance with at least one threshold. Or, the variance estimate is determined directly from the context. The variance estimate is then 5 used for controlling a further quantization NUQ2 which is preferably but not necessarily non-linear. In the encoding process exemplarily depicted in Fig. 1, a suited probability Density Function (PDF) is determined to encode the current quantized spectral bin BIN. For this 10 purpose only information can be used that is also already known at the decoder side. That is, only preceding encoded or decoded quantized spectral bins can be used. This is done in context classifier block COCL. There, selected preceding spectral bins define a neighbourhood NBH which is 15 used to determine the actual context class. The context class may be symbolized by a context class number. The context class number is used to retrieve the corresponding PDF from a PDF-memory MEM1 via a mapping MAP, e.g. via a table look-up or a hash table. The determination of the 20 context class may depend on a general mode switch GMS which allows to use different neighbourhoods in dependence on the selected mode. As mentioned above, for USAC there may be two general modes (FD mode and wLPT mode). If a general mode switch GMS is realized at the encoder side, a mode 25 change signal or a current general signal has to be comprised in the bitstream, so that it is also known in the decoder. For instance, in the reference model for Unified Speech and Audio Coding (USAC) proposed by ISO / IEC JTC1 / SC29 / WG11 N10847, July 2009, London, UK there are WD 30 table 4.4 coremode and table 4.5 coremode0 / 1 proposed for transmission of the general mode. After determination of a suitable PDF for encoding of the current quantized spectral bin BIN by arithmetic encoder 2025210849 31 Jul 2025 AEC, the current quantized spectral bin BIN is fed to neighbourhood memory MEM2, i.e. the current bin BIN becomes a preceding bin. The preceding spectral bins comprised in neighbourhood memory MEM2 may be used by block COCL for 5 coding the next spectral bin BIN. During, before or after memorizing the current spectral bin BIN, said current bin BIN is arithmetic encoded by arithmetic encoder AEC. The output of arithmetic encoding AEC is stored in bit buffer BUF or is written in the bitstream directly. 10 The bitstream or the content of buffer BUF may be transmitted or broadcasted via cable or satellite, for instance. Or, the arithmetic encoded spectral bins may be written on a storage medium like DVD, hard disc, blue-ray disk or the like. PDF-memory MEM1 and neighbourhood memory 15 MEM2 may be realized in a single physical memory. Reset switch RS may allow for restarting encoding or decoding from time to time at dedicated frames at which the encoding and decoding may be started without knowledge of the preceding spectra, the dedicated frames being known as 20 decoding entry points. If a rest switch RS is realized at the encoder side, a reset signal has to be comprised in the bitstream, so that it is also known in the decoder. For instance, in the reference model for Unified Speech and Audio Coding (USAC) proposed by ISO / IEC JTC1 / SC29 / WG11 25 N10847, July 2009, London, UK there is a arithresetflag in WD table 4.10 and table 4.14. The corresponding neighbourhood based decoding scheme is exemplarily depicted in Fig. 2. It comprises similar blocks as the encoding scheme. The determination of the PDF to be 30 used for the arithmetic decoding is identical with the encoding scheme to guarantee that in both, the encoder and decoder, the determined PDF is the same. The Arithmetic 2025210849 31 Jul 2025 10 15 20 25 decoding gets the Bits form the bit buffer BUF or the bitstream directly and uses the determined PDF to decode the current quantized spectral bin BIN. Afterwards the decoded quantized spectral bin is fed to neighbourhood memory MEM2 of the Determination of context class number block COCL and may be used for decoding the next spectral bin. Fig. 3 exemplarily depicts a first embodiment of context classifier COCL for determining a context class in detail. Before storing current quantized spectral bin BIN in the spectra memory MEM2 it may be non-uniformly quantized in block NUQ1. This has two advantages: first, it allows a more efficient storage of the quantized bins, which are usually 16Bit signed integer values. Second, the number of values each quantized bin could have is reduced. This allows an enormous reduction of possible context classes in the context class determination process in block CLASS. Further more, as in the context class determination the sign of the quantized bins may be discarded, the calculation of the absolute values may be included in the non-uniform quantization block NUQ1. In Table Table 1 is shown exemplary non-uniform quantization as it may be performed by block NUQ1. In the example, after non-uniform quantization three different values are possible for each bin. But in general, the only constraint for the nonuniform quantization is that it reduces the number of values a bin may take. Absolute value of quantized spectral bins 0 1 2 3 4 5 6 7 8 > 8 Non-uniform quantization 0 1 2 2025210849 31 Jul 2025 Table 1 Exemplary non-uniform quantization step including the calculation of absolute values The non-uniform quantized / mapped spectral bins are stored 5 in the spectral memory MEM2. According to the selected general mode selection GMS, for the context class determination CLASS for each bin to be coded a selected neighbourhood NBH of spectral bins is selected. Fig. 5a exemplarily depicts a first exemplary neighbourhood 10 NBH of a spectral bin BIN to-be-encoded or to-be-decoded. In this example only spectral bins of the actual or current spectrum (frame) and spectral bins of one preceding spectrum (frame) define the neighbourhood NBH. It is, of course, possible to use spectral bins from more than one 15 preceding spectrum as par of the neighbourhood, which results in a higher complexity, but may also offer a higher coding efficiency in the end. Note, from the actual spectrum only already transmitted bins may be used to define the neighbourhood NBH, as they also have to be 20 accessible at the decoder. Here as well as in the following examples, the transmission order from low to high frequencies for the spectral bins is assumed. The selected neighbourhood NBH is then used as input in the context class determination block COCL. In the following, 25 first the general idea behind the context class determination and a simplified version is explained, before a special realization is described. The general idea behind the context class determination is to allow a reliable estimation of the variance of the bin 30 to be coded. This predicted variance, again, can be used to get an estimation of the PDF of the bin to be coded. For 2025210849 31 Jul 2025 variance estimation it is not necessary to evaluate the sign of the bins in the neighbourhood. Therefore the sign can already be discarded in the quantization step before storage in the spectral memory MEM2. A very simple context 5 class determination may look like as follows: the neighbourhood NBH of spectral bin BIN may look like in Fig. 5a and consists of 7 spectral bins. If exemplarily the nonuniform quantization shown in Table Table is used each bin can have 3 values. This results in 37 = 2187 possible 10 context classes. To further reduce this number of possible context classes the relative position of each bin in the neighbourhood NBH may be discarded. Therefore, only the number of bins is counted, which have the value 0, 1 or 2, respectively, 15 wherein, the sum of the number of 0-bins, the number of 1-bins and the number of 2-bins equals the overall number of bins in the neighbourhood, of course. In the neighbourhood NBH comprising n bins of which each may take one out of three different values there are 0.5*(n2+3*n+2) context 20 classes. For instance, in a neighbourhood of 7 bins there are 36 possible context classes and a neighbourhood of 6 bins there are 28 possible context classes. A more complex but still quite simple context class determination takes into account that research has shown 25 the spectral bin of the preceding spectrum at the same frequency being of special importance (the spectral bin depicted by a dotted circle in the Fig. 5a, 5b, 6a, 6b, 7a, 7b, 7c, 8 and 9). For the other bins in the neighbourhood, those depicted as horizontally striped circles in the 30 respective figures, the relative position is less relevant. So the bin at same frequency in the preceding spectrum is used explicitly for context class determination, whereas for the other 6 bins only the number of 0-bins, the number 2025210849 31 Jul 2025 of 1-bins and the number of 2-bins are counted. This results in 3 x 28 = 84 possible context classes. Experiments have shown that such context classification is very efficient for the FD mode. 5 The context class determination may be extended by a variance estimation VES, which controls a second nonuniform quantization NUQ2. This allows a better adaptation of the context class generation COCL to a higher dynamic range of the predicted variance of the bin to be coded. The 10 corresponding block diagram of the extended context class determination is exemplarily shown in 4. In the example shown in fig. 4, non-uniform quantization is separated in two steps of which a preceding step provides finer quantization (block NUQ1) and a subsequent step 15 provides coarser quantization (block NUQ2). This allows for adaptation of the quantization to e.g. the variance of the neighbourhood. The variance of the neighbourhood is estimated in the variance estimation block VES wherein the variance estimation is based on said preceding finer 20 quantization of bins in the neighbourhood NBH in block NUQ1. The estimation of the variance need not to be precise but can be very rough. For example, it is sufficient for USAC application to decide whether the sum of the absolute values of the bins in the neighbourhood NBH after said 25 finer quantization meets or exceeds a variance threshold or not, that is, a switch between high and low variance is sufficient. The 2-step non-uniform quantization may look as shown in Table Table 2. In this example the low variance mode 30 corresponds to the 1-step quantization shown in Table Table 2. 2025210849 31 Jul 2025 Absolute value of quantized spectral bins 0 1 2 3 4 5 6 7 8 > 8 Finer quantization step 1 (6 values) 0 1 2 3 4 5 Coarser quantization step 2 (low variance) (3 values) 0 1 2 Coarser quantization step 2 (high variance (3 values) 0 1 2 Table 2 depicts an exemplary 2-step non-uniform quantization; the second or subsequent step quantizes differently in dependence on whether variance has been estimated as being high or low The final context class determination in block CLASS is the same as in the simplified version of Fig. 3. It is possible to use different context class determinations according to the variance mode. It is also possible to use more than two 10 variance modes, which of course results in an increase in the number of context classes and an increase in complexity. For the first bins in a spectrum a neighbourhood like it is shown in Fig. 5a or 5b is not applicable, because for the 15 first bins none or not all lower frequency bins exist. For each of these special cases an own neighbourhood may be defined. In a further embodiment, the non-existing bins are filled with a predefined value. For the exemplary neighbourhood given in Fig. 5a the defined neighbourhoods 20 for the first bins to be transmitted in a spectrum are 2025210849 31 Jul 2025 shown in Fig. 6a and Fig. 6b. The idea is to expand the neighbourhood to higher frequency bins in order to allow for using the same context class determination function as for the rest of the spectrum. This means also the same 5 context classes and at last the same PDFs can be used. This would not be possible, if the size of the neighbourhood is just reduced (of course this is also an option). Resets usually occur before a new spectrum is coded. As already mentioned, this is necessary to allow dedicated 10 starting points for decoding. For example, if the decoding process shall start from a certain frame / spectrum, in fact the decoding process has to start from the point of the last reset to successively decode the preceding frame until the desired starting spectrum. This means, the more resets 15 occur, the more entry points for the decoding exits. However, the coding efficiency is smaller in a spectrum after a reset. After a reset occurred no preceding spectrum is available for the neighbourhood definition. This means only preceding 20 spectral bins of the actual spectrum may be used in the neighbourhood. However, the general procedure may not be changed and the same “tools” can be used. Again, the first bins have to be treated differently as already explained in the previous section. 25 In Fig. 8 an exemplary reset neighbourhood definition is shown. This definition may be used in case of reset in the FD mode of USAC. The number of additional context classes as shown in the example in Fig. 8 (using the quantization of Table Table 30 with finally 3 possible quantized values or 6 values if values after quantization step 1 are used) are as follows: the handling for the very first bin adds 1 context class, 2025210849 31 Jul 2025 2nd bin adds 6 (value after quantization step 1 is used), 3rd bin adds 6 and 4th bin adds 10 context classes. If additionally consider two (low and high) variance modes this number of context classes is almost doubled (only for 5 the first bin, where no information is available and for the second bin, where the value for the bin after quantization step 1 is used are not doubled). This results in this example in 1 + 6 + 2x6 + 2x10 = 39 additional context classes for the handling of the resets. 10 Mapping block MAP takes the context classification determined by block COCL, e.g. a determined context class number, and selects the corresponding PDF from PDF-memory MEM1. In this step it is possible to further reduce the amount of necessary memory size, by using a single PDF for 15 more than one context class. That is, context classes which have a similar PDF may use a joint PDF. These PDFs may be predefined in a training phase using a sufficiently large representative set of data. This training may include an optimization phase, where context classes corresponding to 20 similar PDFs are identified and the corresponding PDFs are merged. Depending on the statistics of the data this can result in a rather small number of PDFs which have to be stored in the memory. In an exemplary experiment version for USAC a mapping from 822 context classes to 64 PDFs was 25 successfully applied. The realization of this mapping function MAP may be a simple table look-up, if the number of context classes is not too large. If the number gets larger a hash table search may be applied for efficiency reasons. 30 As stated above, general mode switch GMS allows for switching between frequency domain mode (FD) and weighted linear prediction transform mode (wLPT). In dependency on 2025210849 31 Jul 2025 the mode, different neighbourhoods may be used. The exemplary neighbourhoods depicted in Fig. 5a, Fig. 6a and 6b and Fig. 8 has been shown in experiments as sufficiently large for FD mode. But for wLPT mode, larger neighbourhoods 5 as exemplarily depicted in Fig. 5b, Fig. 7a, 7b and 7c and Fig. 9 has been found to be advantageous. That is, exemplary reset handling in wLPT mode is depicted in Fig. 9. Exemplary neighbourhoods in wLPT mode for the lowest, the second lowest, the third lowest and the fourth 10 lowest bin in a spectrum are depicted in Fig. 7a, 7b, 7c and 7d, respectively. And, an exemplary neighbourhood in wLPT mode for all other bins in a spectrum is depicted in Fig. 5b. The number of context classes resulting from the exemplary 15 neighbourhood depicted in Fig. 5b is 3 x 91 = 273 context classes. The factor 3 results from the special handling of the one bin at the same frequency as the one currently tobe-encoded or currently to-be-decoded. According to the formula given above, there are 0.5*((12*12)+3*12+2) = 91 20 combinations of number of bins with value 2, 1 or 0 for the remaining 12 bins in the neighbourhood. In an embodiment which differentiates context classes in dependency on whether variance of the neighbourhood meets or exceeds a threshold, the 273 context classes are doubled. 25 An exemplary reset handling as shown in Fig. 9 may also add a number of context classes. In a tested exemplary embodiment which yielded good results in experiments, there are 822 possible context classes, which are broken down in the following Table ITable—1. 30 Mode Low variance mode High variance mode FD mode 84 84 2025210849 31 Jul 2025 FD mode reset after 39 wLPT mode 273 273 wLPT mode reset after 69 Table 1 Broken down possible context classes of the MPEG USAC CE proposal In a tested exemplary embodiment, these 822 possible 5 context classes are mapped onto 64 PDFs. The mapping is determined in a training phase, as described above. The resulting 64 PDFs have to be stored in ROM tables e.g. in 16Bit accuracy for a fixpoint arithmetic coder. Here another advantage of the proposed scheme is revealed: in 10 the current working draft version of the USAC standardization mentioned in the background section, quadruples (vectors containing 4 spectral bins) are jointly coded with a single codeword. This results in very large codebooks even if the dynamic range of each component in 15 the vector is very small (e.g. each component may have the values [-4,...,3] ^ 84 = 4096 possible different vectors). Coding of scalars, however, allows a high dynamic range for each bin with a very small codebook. The codebook used in the tested exemplary embodiment has 32 entries offering a 20 dynamic range for the bin form -15 to +15 and an Esccodeword (for the case, that the value of a bin lies outside this range). This means that only 64 x 32 16Bit values have to stored in ROM tables. Above, a method for arithmetic encoding of a current 25 spectral coefficient using preceding spectral coefficients has been describe wherein said preceding spectral coefficients are already encoded and both, said preceding and current spectral coefficients, are comprised in one or more quantized spectra resulting from quantizing time- 2025210849 31 Jul 2025 frequency-transform of video, audio or speech signal sample values. In an embodiment, said method comprises processing the preceding spectral coefficients, using the processed preceding spectral coefficients for determining a context 5 class being one of at least two different context classes, using the determined context class and a mapping from the at least two different context classes to at least two different probability density functions for determining the probability density function, and arithmetic encoding the 10 current spectral coefficient based on the determined probability density function wherein processing the preceding spectral coefficients comprises non-uniformly quantizing the preceding spectral coefficients. In another exemplary embodiment, the device for arithmetic 15 encoding of a current spectral coefficient using preceding, already encoded spectral coefficients comprises processing means, first means for determining a context class, a memory storing at least two different probability density functions, second means for retrieving the probability 20 density, and an arithmetic encoder. Then, the processing means are adapted for processing the preceding, already encoded spectral coefficients by non- uniformly quantizing them and said first means are adapted for using the processing result for determining the context 25 class as being one of at least two different context classes. The memory stores at least two different probability density functions and a mapping from the at least two different context classes to the at least two different probability density functions which allows for 30 retrieving the probability density function which corresponds to the determined context class. The second means are adapted for retrieving, from the memory, the probability density which corresponds to the determined 2025210849 31 Jul 2025 context class, and the arithmetic encoder is adapted for arithmetic encoding of the current spectral coefficient based on the retrieved probability density function. There is a corresponding another exemplary embodiment of 5 the device for arithmetic decoding of a current spectral coefficient using preceding, already decoded spectral coefficients which comprises processing means, first means for determining a context class, a memory storing at least two different probability density functions, second means 10 for retrieving the probability density, and an arithmetic decoder. Then, the processing means are adapted for processing the preceding, already decoded spectral coefficients by non-uniformly quantizing them and said first means are adapted 15 for using the processing result for determining the context class as being one of at least two different context classes. The memory stores at least two different probability density functions and a mapping from the at least two different context classes to the at least two 20 different probability density functions which allows for retrieving the probability density function which corresponds to the determined context class. The second means are adapted for retrieving, from the memory, the probability density which corresponds to the determined 25 context class, and the arithmetic decoder is adapted for arithmetic decoding of the current spectral coefficient based on the retrieved probability density function.
Claims
1. A method for arithmetic decoding of a currentspectral coefficient, said method comprising: processing preceding spectral coefficients;5 determining a context state based on the processedpreceding spectral coefficients, wherein the context stateis determined from at least two different context states,wherein the context state is based on a sum of thequantised absolutes of the preceding spectral coefficients,10 and wherein the context state determination is based onusing at least one of a reset signal and a mode switchingsignal;determining a probability density function based on the determined context state and a mapping from the at-15 least two different context states to at least two different probability density functions; andarithmetic decoding the current spectral coefficient based on the determined probability density function.20 2. A device for arithmetic decoding of a current spectralcoefficient, said device comprising:a processor configured to process the preceding spectral coefficients;a context classifier configured to determine a context 25 state based on the processed preceding spectralcoefficients, wherein the context state is determined fromat least two different context states, and wherein thecontext state is based on a sum of the quantised absolutesof the preceding spectral coefficients and wherein the30 context state determination is based on using at least one of a reset signal and a mode switching signal;a probability density module configured to determine a probability densityfunction, said probability density module being2025210849 31 Jul 2025adapted to use the determined context state and a mapping from the at least two different context states to at leasttwo different probability density functions to determinethe probability density function; and5 an arithmetic decoder configured to arithmeticallydecode the current spectral coefficient based on the determined probability density function.