CA2075754C - Method of coding 32-kb/s audio signals - Google Patents
Method of coding 32-kb/s audio signalsInfo
- Publication number
- CA2075754C CA2075754C CA002075754A CA2075754A CA2075754C CA 2075754 C CA2075754 C CA 2075754C CA 002075754 A CA002075754 A CA 002075754A CA 2075754 A CA2075754 A CA 2075754A CA 2075754 C CA2075754 C CA 2075754C
- Authority
- CA
- Canada
- Prior art keywords
- values
- magnitude
- frequency
- hearing
- account
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
- G10L19/02—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using spectral analysis, e.g. transform vocoders or subband vocoders
- G10L19/032—Quantisation or dequantisation of spectral components
- G10L19/035—Scalar quantisation
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B1/00—Details of transmission systems, not covered by a single one of groups H04B3/00 - H04B13/00; Details of transmission systems not characterised by the medium used for transmission
- H04B1/66—Details of transmission systems, not covered by a single one of groups H04B3/00 - H04B13/00; Details of transmission systems not characterised by the medium used for transmission for reducing bandwidth of signals; for improving efficiency of transmission
- H04B1/665—Details of transmission systems, not covered by a single one of groups H04B3/00 - H04B13/00; Details of transmission systems not characterised by the medium used for transmission for reducing bandwidth of signals; for improving efficiency of transmission using psychoacoustic properties of the ear, e.g. masking effect
Abstract
For audio signals (50 to 7000 Hz), good speech quality is to be achieved at a data rate of 32 kb/s (also for music signal(s), which requires a considerably greater data reduction than hitherto known.
For a further data reduction of the magnitude values of the coefficients obtained by FFT, the absolute threshold of hearing and a relative threshold of hearing known as masking effect are taken into account in fixing the quantization levels. This is followed by a variable quantization.
For a further data reduction of the magnitude values of the coefficients obtained by FFT, the absolute threshold of hearing and a relative threshold of hearing known as masking effect are taken into account in fixing the quantization levels. This is followed by a variable quantization.
Description
- 1 - 2~ 7S~
Method of Coding 32-kb/s Audio Signals In telecommunications, signals are nowadays mostly trans-mitted in digital form.
To meet the increasing need for connections with existing line capacities necessitates decreasing bit rates with unchanged intelligibility.
For good intelligibility, e.g., during hands-free opera-tion or in video conferences, a present bandwidth of 300 to 3400 Hz at 64 kb/s is already insufficient. The aim is to achieve good speech quality (also for music signals) for future bandwidths of 50 to 7000 Hz with a simultaneous reduction of the data rate to 32 kb/s.
A coding technique for wideband (O to 20 kHz) signals is known (R. Orglmeister, "Transformationscodierung mit fester Bitzuteilung bei Audiosignalen", FREQUENZ 44 (1990) 9-10, pp. 226-232) in which a sequence of sample values (44.1-kHz sampling with 16 bits per sample value) is first divided into blocks of equal size and processed by overlapping windowing (rectangular cosine function) to suppress audible block-boundalry effects.
The signal is then transformed into the frequency domain by means of a first Fourier transform, and decomposed into magnitude and phase values.
The phase values are uniformly quantized, while the magnitude values are subjected to a data reduction.
For the data reduction, magnitude groups are formed such that, instead of the individual magnitude values, only the geometric mean of all magnitudes of a group is transmitted.
These magnitude groups are then routed, according to frequency range, to five different quantizers which quantize nonuniformly, approximating the logarithmic loudness perception.
By this known method, good speech quality is achieved, but due to insufficient data reduction of the magnitude values, a data rate of 32 kb/s is not attainable.
It is the object of the invention to implement a coding method with a considerably higher data reduction and unchanged speech quality, particularly for audio signals with a bandwidth up to 7000 Hz at a data rate of 32 kb/s.
According to the invention, a substantial data reduction of the magnitude values of the coefficients obtained by a discrete Fourier transform and decomposed in magnitude and phase is achieved by taking into account the absolute threshold of hearing and a relative threshold of hearing, known as "masking effect", in fixing the quantization levels, and by a subsequent variable quantization.
Another solution according to the invention, after decomposition of the complex coefficients into magnitude and phase values, first takes into account the relative threshold of hearing, ie., the masking effect, then divides the magnitude values into frequency bands, and takes into account the absolute threshold of hearing. For the values lying below the thresholds, the value zero is used for the magnitude, and the bits for the associated phase values are saved and used for the nonzero magnitude and phase values.
More particularly, the present invention provides, according to a first aspect a method of coding digitized audio signals, comprising the steps of: dividing an audio signal, consisting of a continuous sequence of sample values, into successive blocks of equal length, and performing overlapping windowing; transforming the blocks into complex Fourier coefficients by means of a discrete Fourier transform; decomposing the complex Fourier coefficients into magnitude values and phase values; quantizing the phase values with a linear quantization characteristic which becomes coarser going from low toward high frequencies; combining the magnitude values into frequency bands which are oriented with regard to predetermined critical bands and become wider toward high frequencies; fixing quantization levels for each frequency band, taking into account a frequency-dependent, absolute threshold of hearing, such that a frequency range lying below the respective frequency-dependent, absolute threshold of hearing is disregarded, and taking into account a relative threshold of hearing, such that frequency ranges in the immediate vicinity of frequencies with large magnitude values are taken into account to a reduced extent; fixing a number of sub-bands consisting of single values of the frequency bands, and determining the greatest single magnitude value of each sub-band, and performing variable quantization of each frequency band by quantizing the greatest magnitude value of a sub-band with a maximum available number of bits and all other magnitude values with a smaller number of bits.
According to a second aspect, the invention provides a method of coding digitized audio signals, comprising the steps of:
dividing an audio signal, consisting of a continuous sequence of sample values, into successive blocks of equal length, and performing overlapping windowing; transforming the blocks into complex Fourier coefficients by means of a discrete Fourier transform; decomposing the complex Fourier coefficients into magnitude values and phase values; taking into account a relative threshold of hearing, such that those magnitude values of coefficients which lie in the immediate vicinity of frequency lines with large magnitude values are either fully taken into account or disregarded; dividing the magnitude values into frequency bands which are oriented with regard to critical bands and become wider toward high frequencies; fixing quantization levels for each frequency band, taking into account a frequency-dependent, absolute threshold of hearing, such that a range of frequencies lying below the respective frequency-dependent, absolute threshold of hearing is disregarded, and performing variable quantization of the magnitude and phase values with a quantization characteristic which becomes coarser going from low toward high frequencies, with the value zero being quantized for those magnitude values of coefficients which lie below the relative threshold of hearing or below the absolute threshold of hearing, no value being quantized for the associated phase values, whereby the bits not needed for said associated phase values are used for a more accurate quantization of the nonzero magnitude and phase values.
Two embodiments of the invention will now be explained with reference to the flowcharts shown in the accompanying drawings.
In both cases (Figs. 1 and 2), the signal to be coded is an audio signal sampled at 16 kHz, quantized with 16 bits, and band-limited to 50-7000 Hz.
In the time domain, this continuous sequence of sample values is divided into blocks of equal length (here 256 values), and overlapping windowing is performed with the aid of a rectangular cosine function (cf. (1) 3a - 4 - Z ~
in the figures). This overlapping window;ng, which is known per se, suppresses audible block-boundary effects.
This is followed by a transformation of the blocks by means of a discrete Fourier transform, particularly a fast Fourier transform FFT (cf. (2)), together with a conventional block normalization.
The resulting 256 conjugate complex coefficients per block are then decomposed into magnitude and phase values (cf. (3)) in order to subsequently perform separate quantization.
Since the resolution of the human ear decreases with increasing frequency, higher frequencies can be quan-tized much more coarsely than low frequencies.
Therefore, the phase values (cf.(4)), here 128 values, since only one half of the 256 conjugate complex values is relevant for reasons of symmetry, are quantized in accordance with a linear quantization characteristic, beginning with preferably 5 bits for the lowest fre-quency and decreasing to 2 bits for the highest fre-quency.
For data reduction, according to the invention, the magnitude values are considerably compressed.
To this end, the magnitude va~ues are divided into frequency bands (cf. (5)) which, oriented with regard to critical bands, become wider toward higher fre-quencies. The principle of critical bands is known, for example, from "Subdivision of the Audible ~requency Range into Critical Bands (Frequenzgruppen)", by E. Zwicker, The Journal of Acoustical Society of America, Vol. 33, No. 2, Feb. 1961, page 248.
For such a magnitude-group formation, instead of the individual magnitude values of the spectral lines, only the geometric mean of aLL magnitudes of a band is computed.
The magnitude-group formation chosen here permits a reduction from initially 128 magnitudes (here, too, only one half of the 256 conjugate complex values is used) to 58 frequency bands. For the lowest frequency range (magnitude group 0 to 31),all magnitude values are used, forming 32 frequency bands; for the subsequent frequency range (magnitudes 32 to 60), every two adjacent magni-tude values are combined, forming 14 frequency bands;
for the next frequency range (magnitudes60 to 104), every four magnitude values are combined, forming 11 fre-quency bands, and for the highest frequency range (magnitudes 104 to 112), every eight adjacent magnitude ~alues are combined, forming 1 frequency band (magni-tudes 112 to 127 are disregarded).
For a further data reduction of the frequency bands, quantization levels are fixed for each frequency band as a preliminary stage for the quantization.
According to the invention, advantage is taken of the fact that below a certain threshold which is frequency-dependent, i.e., below the so-called absolute thresh-old of hearing, the human ear has no perception al-though measurable signals are present. Therefore, th;s - 6 - 2~75~
range below the threshold of hearing is disregarded for the quantization levels, so that the quantization levels can be optimized to form a better image of the audible range (cf. (6)).
In addition, a relative threshold of hearing, which is based on the so-called masking effect, is taken into account. Due to this effect, low tones which lie in the immediate vicinity of a loud tone (large magni-tude values) are covered, i.e., masked. Therefore, these masked frequencies are disregarded, too, or are only taken into account to a reduced extent (cf. (7)).
Since the relative threshold of hearing is frequency-dependent, according to a further advantageous feature of the invention, weighting is introduced in order to assign greater weight to the respective threshoLd having the greater magnitude.
With the aid of the quantization levels thus chosen, a variable quantization of the frequency bands is then carried out (cf. (8, 9)).
To do this, the frequency bands are first divided into a predeterminable number of subbands, here approximately 15, consisting of single values, and the greatest single value of each subband is determined. The greatest magni-tude value is then quantized with the maximum number of bits available, while all other magnitude values are quantized with a smaller number of bits. For the low frequency range, for example, the greatest magnitude 2~7~S~
value is quantized with 7 bits and the other values with 5 bits, and for high frequency ran~es, a quanti-zation from 5 bits to 3 bits is performed~
The magnitude and phase values thus coded are then trans-mitted at 32 kb/s. At the opposite end,.decoding can then take place in an analogous mode.
In a further solution according to the invention (claim 5), after step (3), the relative threshold of hearing is taken into account, as shown in Fig. 2. Those magni-tude values of coefficients which lie in the neighbor-hood of frequency lines with great magnitude values are fully taken into account or disreqarded (steo 10).
In step 11 (cf. Fig. 2), analogously to step (5), the magnitude values are again divided into frequency bands (approx. 60 groups). With increasi'ng frequency, a greater number of values is grouped into one band.
In step (12), analogously to step (6), the absolute threshold of hearing is taken into account, with the quantization levels being fixed for each frequency band in such a way that the magnitude values of a band are div;ded by a frequency-dependent factor specifying the absolute threshold of hearing.
In the following step (13), a variable quantizat;on of the magnitude and phase values is carried out with a quantization characterist;c becoming coarser from low toward high frequenc;es, allowing for the fact that in certain cases, the magnitude values must not be taken into account, i.e., they are assigned the value zero.
- 8 - 2~7~5';
.
That is the case if magnitude values lie below the relative threshold of hearing or below the absolute threshold of hearing. Then it is sufficient to trans mit the magnitude value zero and to disregard the associated phase values, i.e., not to use any bits for the quantiza~ion of these phase values:
According to the invention, the bits thus "saved" are used for the nonzero magnitude and phase values to be quantized.
According to a further advantageous aspect of the in-vention, a so-called spectral roughness can be deter-mined for the magnitude values divided into frequency bands.This allows for the fact that instruments, such as violins, or voices may have a great number of higher harmonics which are contained in particular higher fre-quencies. If, as described in step (5), the geometric mean of all magnitude values of a band were used, unacceptable distortions would occur.
Therefore, the quotient of each magnitude value and the geometric mean of a band is formed as a measure of the spectral roughness. If the maximum magnitude value of a band exceeds the geometric mean by a pre-determinable factor, preferably by the factor 2, only the max;mum magnitude value will be taken into account and the other magnitude values of the band will be set equal to zero.
Otherwise, the geometric mean will be used, as in step (5).
If the maximum magnitude value is used, only a value zero will again be transmitted for the magnitude values set _ 9 ~ 75~
.
equal to zero, and the associated phase vaLues witl not be quantized. The bits thus "saved" can advan.tageousLy be employed for the nonzero magnitude and phase values.
, Since th.e human ear perceives lower frequencies bett.er than higher frequencies, the quantization of the mag~
tude and phase values beg;ns with 4 bits and 3 bits, re-spectively, for the (ower frequencies and uses only 3 bits and 2 bits, respectively~for the higher frequencies.
The above-described coding methods according to the in-vention permit a real-time implementation of a coder/
decoder with a single 32-kb/s digital signal processor,
Method of Coding 32-kb/s Audio Signals In telecommunications, signals are nowadays mostly trans-mitted in digital form.
To meet the increasing need for connections with existing line capacities necessitates decreasing bit rates with unchanged intelligibility.
For good intelligibility, e.g., during hands-free opera-tion or in video conferences, a present bandwidth of 300 to 3400 Hz at 64 kb/s is already insufficient. The aim is to achieve good speech quality (also for music signals) for future bandwidths of 50 to 7000 Hz with a simultaneous reduction of the data rate to 32 kb/s.
A coding technique for wideband (O to 20 kHz) signals is known (R. Orglmeister, "Transformationscodierung mit fester Bitzuteilung bei Audiosignalen", FREQUENZ 44 (1990) 9-10, pp. 226-232) in which a sequence of sample values (44.1-kHz sampling with 16 bits per sample value) is first divided into blocks of equal size and processed by overlapping windowing (rectangular cosine function) to suppress audible block-boundalry effects.
The signal is then transformed into the frequency domain by means of a first Fourier transform, and decomposed into magnitude and phase values.
The phase values are uniformly quantized, while the magnitude values are subjected to a data reduction.
For the data reduction, magnitude groups are formed such that, instead of the individual magnitude values, only the geometric mean of all magnitudes of a group is transmitted.
These magnitude groups are then routed, according to frequency range, to five different quantizers which quantize nonuniformly, approximating the logarithmic loudness perception.
By this known method, good speech quality is achieved, but due to insufficient data reduction of the magnitude values, a data rate of 32 kb/s is not attainable.
It is the object of the invention to implement a coding method with a considerably higher data reduction and unchanged speech quality, particularly for audio signals with a bandwidth up to 7000 Hz at a data rate of 32 kb/s.
According to the invention, a substantial data reduction of the magnitude values of the coefficients obtained by a discrete Fourier transform and decomposed in magnitude and phase is achieved by taking into account the absolute threshold of hearing and a relative threshold of hearing, known as "masking effect", in fixing the quantization levels, and by a subsequent variable quantization.
Another solution according to the invention, after decomposition of the complex coefficients into magnitude and phase values, first takes into account the relative threshold of hearing, ie., the masking effect, then divides the magnitude values into frequency bands, and takes into account the absolute threshold of hearing. For the values lying below the thresholds, the value zero is used for the magnitude, and the bits for the associated phase values are saved and used for the nonzero magnitude and phase values.
More particularly, the present invention provides, according to a first aspect a method of coding digitized audio signals, comprising the steps of: dividing an audio signal, consisting of a continuous sequence of sample values, into successive blocks of equal length, and performing overlapping windowing; transforming the blocks into complex Fourier coefficients by means of a discrete Fourier transform; decomposing the complex Fourier coefficients into magnitude values and phase values; quantizing the phase values with a linear quantization characteristic which becomes coarser going from low toward high frequencies; combining the magnitude values into frequency bands which are oriented with regard to predetermined critical bands and become wider toward high frequencies; fixing quantization levels for each frequency band, taking into account a frequency-dependent, absolute threshold of hearing, such that a frequency range lying below the respective frequency-dependent, absolute threshold of hearing is disregarded, and taking into account a relative threshold of hearing, such that frequency ranges in the immediate vicinity of frequencies with large magnitude values are taken into account to a reduced extent; fixing a number of sub-bands consisting of single values of the frequency bands, and determining the greatest single magnitude value of each sub-band, and performing variable quantization of each frequency band by quantizing the greatest magnitude value of a sub-band with a maximum available number of bits and all other magnitude values with a smaller number of bits.
According to a second aspect, the invention provides a method of coding digitized audio signals, comprising the steps of:
dividing an audio signal, consisting of a continuous sequence of sample values, into successive blocks of equal length, and performing overlapping windowing; transforming the blocks into complex Fourier coefficients by means of a discrete Fourier transform; decomposing the complex Fourier coefficients into magnitude values and phase values; taking into account a relative threshold of hearing, such that those magnitude values of coefficients which lie in the immediate vicinity of frequency lines with large magnitude values are either fully taken into account or disregarded; dividing the magnitude values into frequency bands which are oriented with regard to critical bands and become wider toward high frequencies; fixing quantization levels for each frequency band, taking into account a frequency-dependent, absolute threshold of hearing, such that a range of frequencies lying below the respective frequency-dependent, absolute threshold of hearing is disregarded, and performing variable quantization of the magnitude and phase values with a quantization characteristic which becomes coarser going from low toward high frequencies, with the value zero being quantized for those magnitude values of coefficients which lie below the relative threshold of hearing or below the absolute threshold of hearing, no value being quantized for the associated phase values, whereby the bits not needed for said associated phase values are used for a more accurate quantization of the nonzero magnitude and phase values.
Two embodiments of the invention will now be explained with reference to the flowcharts shown in the accompanying drawings.
In both cases (Figs. 1 and 2), the signal to be coded is an audio signal sampled at 16 kHz, quantized with 16 bits, and band-limited to 50-7000 Hz.
In the time domain, this continuous sequence of sample values is divided into blocks of equal length (here 256 values), and overlapping windowing is performed with the aid of a rectangular cosine function (cf. (1) 3a - 4 - Z ~
in the figures). This overlapping window;ng, which is known per se, suppresses audible block-boundary effects.
This is followed by a transformation of the blocks by means of a discrete Fourier transform, particularly a fast Fourier transform FFT (cf. (2)), together with a conventional block normalization.
The resulting 256 conjugate complex coefficients per block are then decomposed into magnitude and phase values (cf. (3)) in order to subsequently perform separate quantization.
Since the resolution of the human ear decreases with increasing frequency, higher frequencies can be quan-tized much more coarsely than low frequencies.
Therefore, the phase values (cf.(4)), here 128 values, since only one half of the 256 conjugate complex values is relevant for reasons of symmetry, are quantized in accordance with a linear quantization characteristic, beginning with preferably 5 bits for the lowest fre-quency and decreasing to 2 bits for the highest fre-quency.
For data reduction, according to the invention, the magnitude values are considerably compressed.
To this end, the magnitude va~ues are divided into frequency bands (cf. (5)) which, oriented with regard to critical bands, become wider toward higher fre-quencies. The principle of critical bands is known, for example, from "Subdivision of the Audible ~requency Range into Critical Bands (Frequenzgruppen)", by E. Zwicker, The Journal of Acoustical Society of America, Vol. 33, No. 2, Feb. 1961, page 248.
For such a magnitude-group formation, instead of the individual magnitude values of the spectral lines, only the geometric mean of aLL magnitudes of a band is computed.
The magnitude-group formation chosen here permits a reduction from initially 128 magnitudes (here, too, only one half of the 256 conjugate complex values is used) to 58 frequency bands. For the lowest frequency range (magnitude group 0 to 31),all magnitude values are used, forming 32 frequency bands; for the subsequent frequency range (magnitudes 32 to 60), every two adjacent magni-tude values are combined, forming 14 frequency bands;
for the next frequency range (magnitudes60 to 104), every four magnitude values are combined, forming 11 fre-quency bands, and for the highest frequency range (magnitudes 104 to 112), every eight adjacent magnitude ~alues are combined, forming 1 frequency band (magni-tudes 112 to 127 are disregarded).
For a further data reduction of the frequency bands, quantization levels are fixed for each frequency band as a preliminary stage for the quantization.
According to the invention, advantage is taken of the fact that below a certain threshold which is frequency-dependent, i.e., below the so-called absolute thresh-old of hearing, the human ear has no perception al-though measurable signals are present. Therefore, th;s - 6 - 2~75~
range below the threshold of hearing is disregarded for the quantization levels, so that the quantization levels can be optimized to form a better image of the audible range (cf. (6)).
In addition, a relative threshold of hearing, which is based on the so-called masking effect, is taken into account. Due to this effect, low tones which lie in the immediate vicinity of a loud tone (large magni-tude values) are covered, i.e., masked. Therefore, these masked frequencies are disregarded, too, or are only taken into account to a reduced extent (cf. (7)).
Since the relative threshold of hearing is frequency-dependent, according to a further advantageous feature of the invention, weighting is introduced in order to assign greater weight to the respective threshoLd having the greater magnitude.
With the aid of the quantization levels thus chosen, a variable quantization of the frequency bands is then carried out (cf. (8, 9)).
To do this, the frequency bands are first divided into a predeterminable number of subbands, here approximately 15, consisting of single values, and the greatest single value of each subband is determined. The greatest magni-tude value is then quantized with the maximum number of bits available, while all other magnitude values are quantized with a smaller number of bits. For the low frequency range, for example, the greatest magnitude 2~7~S~
value is quantized with 7 bits and the other values with 5 bits, and for high frequency ran~es, a quanti-zation from 5 bits to 3 bits is performed~
The magnitude and phase values thus coded are then trans-mitted at 32 kb/s. At the opposite end,.decoding can then take place in an analogous mode.
In a further solution according to the invention (claim 5), after step (3), the relative threshold of hearing is taken into account, as shown in Fig. 2. Those magni-tude values of coefficients which lie in the neighbor-hood of frequency lines with great magnitude values are fully taken into account or disreqarded (steo 10).
In step 11 (cf. Fig. 2), analogously to step (5), the magnitude values are again divided into frequency bands (approx. 60 groups). With increasi'ng frequency, a greater number of values is grouped into one band.
In step (12), analogously to step (6), the absolute threshold of hearing is taken into account, with the quantization levels being fixed for each frequency band in such a way that the magnitude values of a band are div;ded by a frequency-dependent factor specifying the absolute threshold of hearing.
In the following step (13), a variable quantizat;on of the magnitude and phase values is carried out with a quantization characterist;c becoming coarser from low toward high frequenc;es, allowing for the fact that in certain cases, the magnitude values must not be taken into account, i.e., they are assigned the value zero.
- 8 - 2~7~5';
.
That is the case if magnitude values lie below the relative threshold of hearing or below the absolute threshold of hearing. Then it is sufficient to trans mit the magnitude value zero and to disregard the associated phase values, i.e., not to use any bits for the quantiza~ion of these phase values:
According to the invention, the bits thus "saved" are used for the nonzero magnitude and phase values to be quantized.
According to a further advantageous aspect of the in-vention, a so-called spectral roughness can be deter-mined for the magnitude values divided into frequency bands.This allows for the fact that instruments, such as violins, or voices may have a great number of higher harmonics which are contained in particular higher fre-quencies. If, as described in step (5), the geometric mean of all magnitude values of a band were used, unacceptable distortions would occur.
Therefore, the quotient of each magnitude value and the geometric mean of a band is formed as a measure of the spectral roughness. If the maximum magnitude value of a band exceeds the geometric mean by a pre-determinable factor, preferably by the factor 2, only the max;mum magnitude value will be taken into account and the other magnitude values of the band will be set equal to zero.
Otherwise, the geometric mean will be used, as in step (5).
If the maximum magnitude value is used, only a value zero will again be transmitted for the magnitude values set _ 9 ~ 75~
.
equal to zero, and the associated phase vaLues witl not be quantized. The bits thus "saved" can advan.tageousLy be employed for the nonzero magnitude and phase values.
, Since th.e human ear perceives lower frequencies bett.er than higher frequencies, the quantization of the mag~
tude and phase values beg;ns with 4 bits and 3 bits, re-spectively, for the (ower frequencies and uses only 3 bits and 2 bits, respectively~for the higher frequencies.
The above-described coding methods according to the in-vention permit a real-time implementation of a coder/
decoder with a single 32-kb/s digital signal processor,
Claims (9)
1. A method of coding digitized audio signals, comprising the steps of: dividing an audio signal, consisting of a continuous sequence of sample values, into successive blocks of equal length, and performing overlapping windowing; transforming the blocks into complex Fourier coefficients by means of a discrete Fourier transform; decomposing the complex Fourier coefficients into magnitude values and phase values; quantizing the phase values with a linear quantization characteristic which becomes coarser going from low toward high frequencies; combining the magnitude values into frequency bands which are oriented with regard to predetermined critical bands and become wider toward high frequencies; fixing quantization levels for each frequency band, taking into account a frequency-dependent, absolute threshold of hearing, such that a frequency range lying below the respective frequency-dependent, absolute threshold of hearing is disregarded, and taking into account a relative threshold of hearing, such that frequency ranges in the immediate vicinity of frequencies with large magnitude values are taken into account to a reduced extent;
fixing a number of sub-bands consisting of single values of the frequency bands, and determining the greatest single magnitude value of each sub-band, and performing variable quantization of each frequency band by quantizing the greatest magnitude value of a sub-band with a maximum available number of bits and all other magnitude values with a smaller number of bits.
fixing a number of sub-bands consisting of single values of the frequency bands, and determining the greatest single magnitude value of each sub-band, and performing variable quantization of each frequency band by quantizing the greatest magnitude value of a sub-band with a maximum available number of bits and all other magnitude values with a smaller number of bits.
2. A method as claimed in claim 1, wherein the step of dividing the audio signal into blocks and performing windowing comprises: dividing the continuous sequence of sample values into blocks of 256 values, and using a rectangular cosine function for the overlapping windowing.
3. A method as claimed in claim 2, wherein the step of quantizing the phase values comprises: quantizing the phase values, beginning with 5 bits for the lowest frequency and linearly decreasing down to 2 bits for the highest frequency.
4. A method as claimed in claim 1, wherein the step of quantizing the phase values comprises: quantizing the phase values, beginning with 5 bits for the lowest frequency and linearly decreasing down to 2 bits for the highest frequency.
5. A method as claimed in claim 1, wherein the step of fixing quantization levels comprises: weighting the lower threshold of hearing and the relative threshold of hearing to fix the quantization levels, such that the threshold having the greater magnitude value is assigned a greater weight.
6. A method of coding digitized audio signals, comprising the steps of: dividing an audio signal, consisting of a continuous sequence of sample values, into successive blocks of equal length, and performing overlapping windowing; transforming the blocks into complex Fourier coefficients by means of a discrete Fourier transform; decomposing the complex Fourier coefficients into magnitude values and phase values; taking into account a relative threshold of hearing, such that those magnitude values of coefficients which lie in the immediate vicinity of frequency lines with large magnitude values are either fully taken into account or disregarded; dividing the magnitude values into frequency bands which are oriented with regard to critical bands and become wider toward high frequencies; fixing quantization levels for each frequency band, taking into account a frequency-dependent, absolute threshold of hearing, such that a range of frequencies lying below the respective frequency-dependent, absolute threshold of hearing is disregarded, and performing variable quantization of the magnitude and phase values with a quantization characteristic which becomes coarser going from low toward high frequencies, with the value zero being quantized for those magnitude values of coefficients which lie below the relative threshold of hearing or below the absolute threshold of hearing, no value being quantized for the associated phase values, whereby the bits not needed for said associated phase values are used for a more accurate quantization of the nonzero magnitude and phase values.
7. A method as claimed in claim 6, further comprising the steps of: determining a spectral roughness for the magnitude values grouped into frequency bands by forming a quotient of each magnitude value and a geometric mean of the respective frequency band, and if a maximum magnitude value in a respective frequency band exceeds the geometric mean by a predetermined factor, taking into account only this maximum magnitude value and the position of this maximum magnitude value within the respective frequency band, and if the maximum magnitude value in a respective frequency band does not exceed the geometric means by the predetermined factor, taking into account a value representative of the geometric mean of all magnitude values in the frequency band.
8. A method as claimed in claim 7, further comprising the steps of: if only the maximum magnitude value of a frequency band is taken into account, taking into account the value zero for all other magnitude values of the frequency band, and using bits not needed for associated phase values for the nonzero magnitude and phase values.
9. A method as claimed in claim 7, wherein: a value of 2 is used as the predeterminable factor.
Applications Claiming Priority (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| DEP4126581.5 | 1991-08-12 | ||
| DE4126581 | 1991-08-12 | ||
| DEP4212339.9 | 1992-04-13 | ||
| DE4212339A DE4212339A1 (en) | 1991-08-12 | 1992-04-13 | CODING PROCESS FOR AUDIO SIGNALS WITH 32 KBIT / S |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CA2075754A1 CA2075754A1 (en) | 1993-02-13 |
| CA2075754C true CA2075754C (en) | 1998-02-17 |
Family
ID=25906266
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA002075754A Expired - Fee Related CA2075754C (en) | 1991-08-12 | 1992-08-11 | Method of coding 32-kb/s audio signals |
Country Status (5)
| Country | Link |
|---|---|
| US (1) | US5303346A (en) |
| EP (1) | EP0527374A3 (en) |
| JP (1) | JP3081378B2 (en) |
| CA (1) | CA2075754C (en) |
| DE (1) | DE4212339A1 (en) |
Families Citing this family (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5787387A (en) * | 1994-07-11 | 1998-07-28 | Voxware, Inc. | Harmonic adaptive speech coding method and system |
| US5774837A (en) * | 1995-09-13 | 1998-06-30 | Voxware, Inc. | Speech coding system and method using voicing probability determination |
| US5745392A (en) * | 1995-10-05 | 1998-04-28 | Chevron U.S.A. Inc. | Method for reducing data storage and transmission requirements for seismic data |
| JP3265962B2 (en) * | 1995-12-28 | 2002-03-18 | 日本ビクター株式会社 | Pitch converter |
| US6044162A (en) * | 1996-12-20 | 2000-03-28 | Sonic Innovations, Inc. | Digital hearing aid using differential signal representations |
| GB2326572A (en) * | 1997-06-19 | 1998-12-23 | Softsound Limited | Low bit rate audio coder and decoder |
| US6026348A (en) * | 1997-10-14 | 2000-02-15 | Bently Nevada Corporation | Apparatus and method for compressing measurement data correlative to machine status |
| US6507804B1 (en) | 1997-10-14 | 2003-01-14 | Bently Nevada Corporation | Apparatus and method for compressing measurement data corelative to machine status |
| KR100297832B1 (en) * | 1999-05-15 | 2001-09-26 | 윤종용 | Device for processing phase information of acoustic signal and method thereof |
| AU2003202975A1 (en) * | 2002-01-15 | 2003-07-30 | University Of Miami | Coding a masked data channel in a radio signal |
| US6943716B2 (en) * | 2003-03-28 | 2005-09-13 | Ess Technology, Inc. | Variable rate sigma delta modulator |
| US7668715B1 (en) * | 2004-11-30 | 2010-02-23 | Cirrus Logic, Inc. | Methods for selecting an initial quantization step size in audio encoders and systems using the same |
| US7764634B2 (en) * | 2005-12-29 | 2010-07-27 | Microsoft Corporation | Suppression of acoustic feedback in voice communications |
| CN102099855B (en) * | 2008-08-08 | 2012-09-26 | 松下电器产业株式会社 | Spectral smoothing device, encoding device, decoding device, communication terminal device, base station device, and spectral smoothing method |
| US8457976B2 (en) | 2009-01-30 | 2013-06-04 | Qnx Software Systems Limited | Sub-band processing complexity reduction |
| EP2355094B1 (en) * | 2010-01-29 | 2017-04-12 | 2236008 Ontario Inc. | Sub-band processing complexity reduction |
| PT3624119T (en) * | 2011-10-28 | 2022-05-16 | Fraunhofer Ges Forschung | Encoding apparatus and encoding method |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100220861B1 (en) * | 1989-01-27 | 1999-09-15 | 쥬더, 에드 에이. | Low time delay transform encoder, decoder and encoding/decoding method for high quality audio |
| US5109417A (en) * | 1989-01-27 | 1992-04-28 | Dolby Laboratories Licensing Corporation | Low bit rate transform coder, decoder, and encoder/decoder for high-quality audio |
| EP0386418B1 (en) * | 1989-03-06 | 1994-12-21 | Robert Bosch Gmbh | Method for data reduction of digital audio signals and for approximate recovery of same |
| US5040217A (en) * | 1989-10-18 | 1991-08-13 | At&T Bell Laboratories | Perceptual coding of audio signals |
-
1992
- 1992-04-13 DE DE4212339A patent/DE4212339A1/en not_active Withdrawn
- 1992-07-25 EP EP19920112745 patent/EP0527374A3/en not_active Ceased
- 1992-08-05 US US07/924,922 patent/US5303346A/en not_active Expired - Fee Related
- 1992-08-11 CA CA002075754A patent/CA2075754C/en not_active Expired - Fee Related
- 1992-08-12 JP JP04215121A patent/JP3081378B2/en not_active Expired - Fee Related
Also Published As
| Publication number | Publication date |
|---|---|
| JPH05206865A (en) | 1993-08-13 |
| EP0527374A2 (en) | 1993-02-17 |
| JP3081378B2 (en) | 2000-08-28 |
| EP0527374A3 (en) | 1994-05-25 |
| DE4212339A1 (en) | 1993-02-18 |
| CA2075754A1 (en) | 1993-02-13 |
| US5303346A (en) | 1994-04-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CA2075754C (en) | Method of coding 32-kb/s audio signals | |
| US4972484A (en) | Method of transmitting or storing masked sub-band coded audio signals | |
| KR100242864B1 (en) | Digital signal coder and the method | |
| KR970008640B1 (en) | System for subband coding of a digital audio signal | |
| US7752041B2 (en) | Method and apparatus for encoding/decoding digital signal | |
| US5054075A (en) | Subband decoding method and apparatus | |
| EP0700032B1 (en) | Methods and apparatus with bit allocation for quantizing and de-quantizing of transformed voice signals | |
| US5634082A (en) | High efficiency audio coding device and method therefore | |
| US5632003A (en) | Computationally efficient adaptive bit allocation for coding method and apparatus | |
| DE60110679T3 (en) | Perceptual coding of audio signals using separate reduction of irrelevance and redundancy | |
| KR100295217B1 (en) | High efficiency encoding and/or decoding device | |
| EP1533789A1 (en) | Sound encoding apparatus and sound encoding method | |
| JP3186292B2 (en) | High efficiency coding method and apparatus | |
| AU7364294A (en) | Computationally efficient adaptive bit allocation for codingmethod and apparatus | |
| JPH07336232A (en) | Method and device for coding information, method and device for decoding information and information recording medium | |
| US7634400B2 (en) | Device and process for use in encoding audio data | |
| JPH0846518A (en) | Information coding and decoding method, information coder and decoder and information recording medium | |
| EP0612160A2 (en) | A bit allocation method for transform coder | |
| US20020173969A1 (en) | Method for decompressing a compressed audio signal | |
| JP3088580B2 (en) | Block size determination method for transform coding device. | |
| Boland et al. | High quality audio coding using multipulse LPC and wavelet decomposition | |
| JP2993324B2 (en) | Highly efficient speech coding system | |
| EP0720146A1 (en) | A method for measuring speech masking properties | |
| KR0144841B1 (en) | The adaptive encoding and decoding apparatus of sound signal | |
| Teh et al. | Subband coding of high-fidelity quality audio signals at 128 kbps |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| EEER | Examination request | ||
| MKLA | Lapsed |