JP3465341B2 - Audio signal encoding method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio signal encoding method, and more particularly to an audio signal which is encoded with a compression process for compressing a spectral component of a digital audio signal into data having a fixed number of bits. The present invention relates to an encoding method.

[0002]

2. Description of the Related Art Digital audio, which is widely known as a so-called compact disc or the like, is widely used by consumers because a reproduced signal with high sound quality can be easily obtained. However, uncompressed digital audio signals, such as those found on compact discs, require a large amount of data storage, which is a drawback in media size and / or transmission bandwidth requirements. Therefore, human auditory characteristics,
Alternatively, a compressed digital audio system that deletes redundant data by utilizing psychoacoustic characteristics has been developed. Many recently announced products such as digital audio media and devices use the above compressed digital audio system in order to reduce the amount of audio data.

In such a compressed digital audio system, generally, time-frequency conversion is first performed on audio data. The transform is typically based on band splitting or transform coding schemes or a combination thereof. The converted data is a so-called quantization unit (Quanti
zation Unit). Generally, all data in a single quantizer unit is requantized with the same word length and, if block floating quantization is used, with the same scale factor. Data compression is performed using a word length shorter than the original input data. The word length of each quantization unit is usually variable. At this time, a psychoacoustic technique is used to allocate available bits between the quantization units so as to minimize deterioration of sound quality in the audible range due to requantization, for example, quantization noise.

Regarding the method of grouping spectral coefficients, the applicant of the present invention has previously proposed, for example, Japanese Patent Application No. 5-1.
No. 83322, Japanese Patent Application No. 5-241189, and Japanese Patent Application No. 5-275218. These systems provide a method of extracting the tonality or tonality component of a signal and quantizing them separately. Usually, a small number of spectral coefficients in the frequency domain, such as about 5, are determined as tones. The spectral coefficient of this tone can be accurately quantized without using a large number of bits. The remaining so-called noisy spectrum is quantized with the next lower precision. The signal obtained by synthesizing the tone-like portion and the noise-like portion by the decoder has a particularly high tone characteristic and high sound quality as compared with the conventional system using the same bit rate.

[0005]

By the way, the encoder has to decide how many bits are allocated to the tone part and the noise part of each input signal. If the decision is not correct, the maximum improvement in sound quality will not be realized.

The present invention has been made in view of such circumstances, and provides an audio signal coding method in which the best possible bit allocation is achieved between the tone-like component and the noise-like component of a signal. That is the purpose.

Another object of the present invention is to provide a bit allocation method for an audio encoder that encodes a signal by using both a tone component and a noise component. The present bit allocation method analyzes the input signal for tonal and noisy components using various spectral measurements, and based on these measurements, audio that can allocate available bits to obtain the best possible sound quality. It is to provide a signal coding method.

[0008]

An audio signal encoding method according to the present invention obtains a plurality of spectra by subjecting an input audio signal to a predetermined conversion, for example, time-frequency analysis or orthogonal conversion, to obtain each spectrum. An audio signal encoding method for allocating bits according to auditory characteristics to a spectrum for encoding, wherein tone-like components are identified for the above-mentioned plurality of spectra, and tones are extracted from the identified tone-like components to obtain predetermined bits. Quantize with a number, separate the frequency domain with high power density for the remaining spectrum of the above multiple spectra, quantize with a predetermined number of bits as a power quantization unit, and use the remaining spectrum with the remaining available bits. By using and quantizing, the above-mentioned problems are solved. As for the number of allocated bits, in general, the above-mentioned tone has the largest number, the power quantization unit has the second largest number, and the remaining spectrum has the smallest number.

It is preferable to classify the extracted tones into either normal tones or large tones, and to quantize the large tones with a larger number of bits than the ordinary tones. Here, as will be described later, the large tone means that R _TOT, which is a ratio of tone energy to total energy, is a predetermined value or more (for example, 0.18 or more), and the others are normal tones.

It is preferable that the remaining spectrum is quantized by the remaining bits with the same word length.

Further, in extracting the tone, first, a predetermined frequency range or a predetermined number of spectra, for example, 25
A local peak for each spectrum is obtained, and when there are two local peaks within a predetermined number of spectral ranges, for example, within 4 spectra, a single tone having a larger power is extracted as a tone. Further, the tone extraction may be performed based on the ratio of the average value / median value of the spectrum, and the tone extraction may be performed based on the tonality of the spectrum.

The invention also provides an input digital audio signal, a time-frequency transform for transforming the signal into coefficients that are subdivided into time and frequency, a number representing the maximum number of bits that can be assigned to the set of coefficients, and Provided is an audio signal coding method having an optimum bit allocation method in which a coefficient is divided into a tone part and a noise part and bit allocation is performed between these two parts.

[0013]

The tonal component of the spectrum of the input audio signal is extracted and quantized, and then the high power frequency region is separated as a power quantization unit with a bit number such that quantization noise is relatively inaudible. By quantizing and quantizing the last remaining spectrum with the number of remaining bits, psychoacoustically, encoding with efficient bit allocation such that many bits are sequentially allocated from the part that is important in terms of hearing can be performed. It is possible to realize high quality audio signal encoding at a low bit rate.

By dividing the tones into normal tones and large tones and allocating more bits to the large tones, the efficiency can be further improved and the sound quality can be improved.

Also, the bit allocation process can be simplified by quantizing the remaining spectrum with the same word length.

Furthermore, by extracting only a single tone that does not overlap as a tone, highly accurate tone detection can be performed. Also, by extracting the tone based on the ratio of the average value / median value of the spectrum, or by extracting the tone based on the tonality of the spectrum,
False detection of tones can be prevented and appropriate tones can be detected.

[0017]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Some preferred embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a diagram schematically showing a general configuration of an audio encoder / decoder system to which an audio signal coding method according to the present invention is applied.

In FIG. 1, the input audio signal supplied to the input terminal 10 is the audio encoder 11
This audio encoder 11 operates in accordance with the psychoacoustic principle to minimize the deterioration of the audible sound quality. The encoded data, which is the output signal from the audio encoder 11, is transmitted or recorded / reproduced via a transmission or recording medium, and is sent to the audio decoder 12. The audio decoder 12 reconstructs an audio signal from the input encoded data and outputs it from the output terminal 16.

FIG. 2 is a block diagram showing a concrete example of the psychoacoustic audio encoder 11.

In FIG. 2, first, the time-frequency analysis block 21 divides the input audio signal into time-frequency coefficients. Next, these frequency components are quantized by the spectrum quantization block 23 using a block floating algorithm. The word length and scale factor of each block of the time-frequency coefficient is determined by the bit allocation algorithm of the bit allocation control block 22. Finally, the quantized spectrum coefficient from the spectrum quantization block 23 is taken out from the output terminal 13, the parameter of the bit allocation table from the bit allocation control block 22 is taken out from the output terminal 14, and is written in, for example, a recording medium. Or transmit.

In order to ensure that the quantization noise generated by the spectrum quantization block 23 remains inaudible or minimally audible to the human ear, the bit allocation algorithm of the bit allocation control block 22 is , The minimum audible curve from the psychoacoustic model output block 24,
It is based on a psychoacoustic model with equal loudness curves, simultaneous masking characteristics, temporal masking characteristics, etc.

Here, the time-frequency analysis block 21
Is generally based on either band coding, transform coding or a combination of these two types. Figure 3
Shows an example of the time-frequency analysis block 21 in the case of such a combination of two types (hybrid).

In FIG. 3, first, a quadrature mirror filter (hereinafter referred to as QMF) is used.
, 31 divides the input signal from the input terminal 10 into a high frequency band and a low frequency band. After that, the low frequency band is further divided by the second QMF 32. Second QMF3
The high frequency band is delayed by the delay circuit 33 for the purpose of compensating for the processing delay due to 2. The range of each of these bands is, for example, 11 to 22 kHz, 5.5 to 11 kHz in order from the high frequency side.
The frequency may be set to Hz, 0 to 5.5 kHz. The signals in each of these bands are converted into spectral coefficients in the frequency domain by the improved discrete cosine transform (MDCT) transform blocks 34, 35 and 36, and are taken out through output terminals 37, 38 and 39, respectively.

Here, the spectrum quantization block 23 of the audio encoder system to which this embodiment is applied.
Needs to quantize the tonal and other parts of the supplied spectral coefficients. These two parts are a tonality part that consists of a small number of small groups of spectrum and whose position in the frequency domain can fluctuate, and a set of larger groups whose bands and areas are fixed in the frequency domain ( Quantization unit).

FIG. 4 shows the spectrum quantization block 23.
The block diagram of one specific example is shown. In FIG. 4, the tone component extraction circuit 41 extracts the tone component from the spectrum coefficient supplied from the input terminal 40. Next, the extracted tones are quantized by the tone characteristic component quantization circuit 42, and the remaining spectrum is quantized by the noise characteristic component quantization circuit 43 in units of quantization units.

Next, specific examples of the tone component and the noise component in the above-mentioned spectral coefficient will be described with reference to FIG. FIG. 5 shows the spectrum distribution of the audio signal transformed into the frequency domain by the MDCT transform blocks 34, 35 and 36.

In the example of FIG. 5, the spectral coefficients indicated by broken lines are tone-like portions TC _A , TC _B , TC _C , T.
C _D , and the spectral coefficient indicated by the solid line indicates the noisy portion. Since these tonal components are concentrated and distributed in a small number of spectrum signals as in the example of FIG. 5, even if these components are quantized accurately, a large number of bits as a whole is not required. .

For the noise component shown by the solid line in FIG. 5, the band of each quantization unit, for example, five bands b
1 to b5, since the tonality component of the broken line is removed from the original spectrum, the normalization coefficient in each quantization unit has a small value, and therefore, the quantization noise generated even with a small number of bits is reduced. be able to. In practice, the entire spectrum of the audio signal is divided into bands of, for example, about 25 bands, and these bands are so-called critical bandwidths in which the higher the band, the wider the band in consideration of human auditory characteristics. It is divided.

Next, the bit allocation control block 22 shown in FIG. 2 can allocate bits to all the assignable bits to the tone-like component and the noise-like component in a psychoacoustically meaningful manner. Needed.

FIG. 6 is a flow chart schematically showing a main part of an example of a bit allocation method used in the audio signal encoding method of this embodiment. This flow chart shows the order in which bits are assigned to each part.

First, in step S62, tone components of the signal are identified, tones are extracted and quantized. These tones are called local tones because they are detected using the local energy reference in units of a local frequency range defined by a given number of spectra. A local tone is defined as a group of spectra that contains most of the signal energy within the local frequency range. Tones are classified as either normal tones or loud tones. Large tones have more energy than normal tones and are therefore quantized with extra bits per spectrum. The tonality of the signal is also calculated, and the signal is considered to be a tone if the total energy contained within the tone component comprises most of the total energy of the signal. When considering a signal as a tone, the tone is assigned an extra bit.

Once the tone components have been detected and extracted,
The remaining important spectral components are recognized by separating the high power frequency domain of the signal and quantizing it with reasonable accuracy. This power can be quantized as a tone or as a so-called power quantization unit. In this algorithm, first, the power tone is detected in step S63, and then the power quantization unit is detected in step S64. To quantize accurately and more accurately quantize the residual spectrum in the quantisation unit if a single tone, which was not detected as a tone in the previous method, contains a significant proportion of the signal power. Therefore, further extraction may be performed.
If a single quantisation unit contains a high power density, i.e. power per spectrum, it is classified as a power quantisation unit and quantized with a sufficient number of bits so that relatively no quantization noise is heard.

Finally, in step S65, the remaining spectrum is quantized as accurately as possible. Note that at this point only the unmasked spectrum is considered. Normally, the spectrum is quantized by simply assigning the remaining bits to the quantization unit in the order of power strength, but in some cases it is possible to extract the tones from the quantization unit containing multiple spectra. This reduces the scale factor in the rest of the spectrum, so it will quantize more accurately with the same word length. Such tones are called scale factor tones.

FIG. 7 is a flow chart showing a concrete example of the main steps of the bit allocation method used in this embodiment.

In the first step S72, the masking threshold is calculated before allocating the bits. The masking threshold is calculated by adding the powers within each band of interest and then applying a masker expansion function. The expansion function determines the masking derived from the adjacent critical band, and an example of the expansion function is shown in FIG.
It is not necessary to encode any of the expanded components of the signal that fall below the final masking function (hatched areas in the figure), since they are not heard.

After the masking calculation, this algorithm
The process proceeds to so-called local tone detection step S73.
Local tones are defined as a group of spectra that contains most of the local energy of a signal. The tone is 5
Defined as a group of spectra, the local frequency range is defined as 25 spectra, for example. These two
The ratio of the signal powers in the two ranges is calculated, and this value is called the local rate R _loc . That is, in FIG. 9, the local power P _loc for 25 spectra is calculated in the first step S92, and the tone power P _tone for 5 spectra is calculated in the next step S93,
In the final step S94, the local rate R _loc (= P _tone / P _loc ) as these ratios is calculated.

The tone threshold value is set to 0.75, for example, and a group consisting of, for example, 5 spectra containing 75% of the energy in the local frequency range is flagged in advance as a tone. It has to be noted, however, that it is possible to classify several neighboring spectra as tones due to the effect of a single high energy spectrum. Therefore, only non-overlapping tones are allowed, ie tones whose center spectra are separated by at least 4 spectra. If two or more tones are located within 4 spectra of each other, then only the tones with the highest local rate R _loc are classified as tones.

The above algorithm depends only on the ratio of powers and does not consider the power distribution in the tone spectrum. Thus, a tone does not necessarily include a single tone peak, but may include five approximately equal-sized spectra, each containing, for example, 15% of the power in the local frequency range. In this case, the spectrum should be quantized as a normal quantization unit rather than the actual tone.

To detect this condition, the mean / median rate R _{mean_med} is calculated in the flowchart shown in FIG. This is the ratio of the mean value of the local spectrum to the median value of the local spectrum. Here, the median is a so-called median, which is the value of the median size among the values of the quantization unit. Step S1 of FIG.
In 02, for example, the median value of 25 local spectra is calculated, and when 25 spectra are arranged in order from the smallest value, the median value is the 13th value. In the next step S103, the average value is calculated,
This average value is, for example, the arithmetic average of 25 local spectra. Generally, a total arithmetic mean is used. In the next step S104, the above steps S102 and S10 are performed.
The average value / median value is calculated using the median value and the average value obtained in 3.

Therefore, the average / median rate R
_{mean_med} will be defined as the ratio of the local mean to the local median. An average / median rate R _{mean_med} is calculated for each tone selected in the power-based method described above. In the case of a tone whose average / median rate R _{mean_med} has a value of 2.5 or less, for example, it is regarded as a false detection tone and included in the tonal component of quantization.

Returning to FIG. 7 again, step S73 is completed.
Since the local tone detection has been performed in step S74, the tonality of the signal is calculated in the next step S74. Specifically, if the sum of the energies in the local tones is greater than 98% of the total energy, then the signal is considered to be tonality.

Next, in step S75, bits are assigned to local tones. The number of allocated bits depends on the tonality of the signal calculated above and the ratio of the total energy contained in each tone. More specifically, the total energy is calculated as the sum of the energy of all spectra. The tone energy is the same as above. The total rate R _tot is defined as the ratio of tone energy to total energy.

The calculation of the total rate R _tot will be described with reference to FIG. In FIG. 11, the total power P _tot of all spectra is calculated in step S112, the tone power P _tone of 5 spectra is calculated in next step S113, and the total rate R _tot is P _tone / P _tot in next step S114. Calculated and calculated.

When the total rate R _tot thus obtained is 0.18 (18% of the total energy) or more, the tone is regarded as a large tone, and otherwise the tone is regarded as a normal tone. For example, the following number of bits is assigned to each of these tones. Large tone Tone component: 5 bits Non-tone component: 4 bits Normal tone Tone component: 4 bits Non-tone component: 3 bits .

Next, the power tone is searched in step S76 of FIG. A power tone is defined as a tone that contains, for example, 40% or more of the total energy of the signal. The total energy is calculated as the sum of the energy of all spectra. Tone energy is calculated over the 5 spectra as described above.
The total rate R _{tot2 at} this time is defined as the ratio of the tone energy to the total energy. This is similar to the above test performed between the large tone and the normal tone.

It should be noted that such energy calculation applies only to the spectrum that the local tone detection method has already determined as the tone spectrum. For example, a spectrum having a total rate R _tot2 of 0.4 or more is detected as a power tone in advance. As mentioned above, it is possible to classify several closely spaced spectra as tones due to the effect of a single high energy spectrum, so that only non-overlapping power tones, ie the central spectra are separated by at least 4 spectra, for example. Accept only tones. If two or more tones lie within four spectra of each other, then only the tones with the maximum total rate R _tot2 are classified as power tones.

In the next step S77, bits are assigned to the power tones thus detected.
The power tone is always quantized as a large tone. If any power tone is detected, the tonality detection described above flags the block as non-tone, but the local tone must contain, for example, 98% of the signal energy for the tonality measurement. Please note. Therefore, the power tone is quantized into 4 bits, for example.

Next, according to the present algorithm, the process proceeds to step S78, a so-called power quantization unit is detected, and bits are assigned in step S79. The quantisation unit contains the non-quantized major signal components as tones. Quantization units that are not quantized as power quantization units generally contain insignificant spectral or masked signal components. The algorithm seeks to quantize the power quantisation unit with, for example, at least 3 or 4 bits.

As shown in FIG. 7, the power quantization unit selection is performed iteratively. In the present algorithm, as shown in step S80, until no more power quantization units are found or there are not enough bits to allocate to another power quantization unit,
Iteratively detect power quantization units and allocate as many bits as needed.

At each iteration, the power intensity of each quantisation unit is calculated, except for the power quantisation units already detected. The power density D _QU of the quantization unit is calculated by adding the power P _QU of all spectra of the quantization unit except the tone spectrum and dividing by the total number of spectra in the quantization unit. The power density is calculated for the rest of the spectrum and defined as the total spectrum other than the previously detected power quantization unit and the spectrum classified as a tone.

FIG. 12 shows a flowchart showing the calculation of the power density D _rem of the remaining spectrum.

In FIG. 12, the first step S1
In step 22, the total number of spectra is set to N _SPEC, and step S1
23, S124, and S125, the remaining power _Prem is set to 0.0 as the initial value setting, and the spectrum number i
Is set to 0, and the count value n of the remaining spectrum is set to 0. In the next step S126, the i-th spectrum sp
It is determined whether or not ec [i] is classified into the above power quantization unit. If YES, the process proceeds to step S130, and if NO, the process proceeds to step S127. In step S127, it is determined whether or not the i-th spectrum spec [i] is classified into the above tone. If YES, the process proceeds to step S130, and if NO, to step S128. In step S128, the remaining power P _rem is replaced with the sum of the remaining power P _rem up to that point and the power spec [i] × spec [i] of the current spectrum, and the process proceeds to step S129 to count value n of the remaining spectrum.
Is being incremented. In the next step S130, it is determined whether or not i has reached the total number N _SPEC . If NO, the process proceeds to step S132, i is incremented and the process returns to step S126, and if YES, the process proceeds to step S126. Proceeding to S131, the power density D _rem of the remaining spectrum is obtained by P _rem / n and output.

That is, D _rem is obtained by calculating the total power P _rem of these remaining spectra and then dividing by the number of spectra n included in the power calculation. Note that this number n is generally not equal to the total number of spectra N _SPEC .

In the present algorithm, each quantisation unit is checked to see if it meets the power quantisation unit requirements. Allocate bits to the power quantization unit according to the inspection result.

FIG. 13 is a flow chart showing a concrete example of the above detection and bit allocation.

In FIG. 13, first, in step S14
In 2, the density rate R _dens , that is, the ratio of the power density D _QU of the quantization unit to the remaining spectral power density D _rem is calculated. Further, in the next step S143, the total rate R _tot2 is calculated. It is defined as the ratio of the energy P _QU in the quantization unit to the total energy P _tot of the whole spectrum, including all tones and power quantization units.

In the next step S144, it is determined whether or not the density rate R _dens is larger than 14, for example, and if YES, the process proceeds to step S145, and if NO, the process proceeds to step S147. Step S1
At 45, the total rate R _tot2 is, for example, 0.18.
If YES, the process proceeds to step S147 to allocate 4 bits to the quantization unit, and if NO, proceeds to step S148 to allocate 3 bits to the quantization unit. Step S14
In 6, it is judged whether or not the total rate R _tot2 is larger than 0.35, for example, and if YES, proceed to step S148 to allocate 3 bits to the quantization unit, and if NO, proceed to step S149. It is progressing. In step S149, the remaining power rate R _{rem is set} to P _QU /
It is calculated by P _{rem, and} this R _rem
Is greater than 0.9, for example, the process proceeds to step S148 if YES and 3 bits are assigned to the quantization unit, and if NO, the process ends.

That is, in the specific example of FIG. 13, when the density rate R _dens is larger than 14 and the total rate R _tot2 is larger than 0.18, 4 bits are allocated to the quantization unit. Otherwise, if R _dens is greater than 14 and R _tot2 is 0.18 or less, or if R _dens is 14 or less and R _tot2 is greater than 0.35, allocate 3 bits. . Finally, if none of the above examples apply,
The above R _rem is calculated by P _QU / P _rem , and if this R _rem is larger than 0.9, 3 bits are assigned to the quantization unit, otherwise, the next quantum is not detected as a power quantization unit. Inspect the conversion unit. As shown in FIG. 7, this iterative loop is repeated until no quantization unit is detected as the power quantization unit.

Then, if there are remaining bits, the remaining bits are used to quantize the remaining unmasked spectrum in a possible manner. Here it is assumed that the significant spectral components of the signal have been identified as tones or power quantization units. In the present algorithm, as shown in step S81, the goal is to allocate, for example, at least 2 bits to the remaining unmasked spectrum. However, there may not be enough bits left to do this, so the algorithm allocates bits to each quantization unit according to the order of the power density of the quantization units.

The power density is calculated by first calculating the unmasked quantizer unit power and then dividing the power by the number of spectra in the quantizer unit. The power calculation includes only unmasked spectra that were not detected as tone spectra. The unmasked spectrum consists of a spectrum whose magnitude is greater than the masking threshold of its quantization unit. Further, the quantization units are rearranged in order of decreasing power intensity. This algorithm has a quantization unit power of 0.
For each quantization unit that has the above unmasked spectrum and has not been allocated bits so far, that is, has not been allocated bits as a power quantization unit, an allocation of 2 bits is attempted.

If there are more bits left, the algorithm attempts to improve the quantization of the non-power quantisation unit. This is done by adding the remainder bits (2 to 3 bits) to the quantisation unit, or by assigning so-called scale factor tones, as shown in step S82. Scale factor tones are used to reduce the scale factor in the quantisation unit by eliminating the maximum. That is, the remaining values will be quantized more accurately without adding more bits. This algorithm determines whether to add a surplus bit or to assign a scale factor tone according to the number of bits required to perform one of the operations, and which operation produces less quantization error. It is done depending on whether to generate.

Finally, in the present algorithm, step S
At 83, the remaining bits, if any, are assigned to the remaining quantization units. It is first assigned to a 3-bit power quantization unit (higher quality to 4 bits), then to another quantization unit with an unmasked spectrum, and finally to all remaining quantization units.

It is unlikely that the algorithm will reach the final stage with any remaining bits, and this part of the algorithm is not particularly essential and will generally only be reached in the case of extremely tonal signals. .

The present invention is not limited to the above-mentioned embodiment, and for example, the number of divided bands, the number of spectra for tone characteristic identification, the number of allocated bits, etc. are limited to the above specific numerical values. Of course, it is needless to say that it may be appropriately set in consideration of the bit rate, sound quality, and the like.

[0066]

According to the audio signal encoding method of the present invention, the tonal component in the spectrum of the input audio signal is extracted and quantized, and then the high power frequency region is separated as a power quantization unit. Quantization is performed by the number of bits so that the relative quantization noise is not heard, and the last remaining spectrum is quantized by the number of remaining bits, so that many bits are allocated in order from the part that is important for hearing. Encoding can be done with efficient bit allocation,
High quality audio signal coding can be realized at a low bit rate.

Further, by classifying the tones into normal tones and large tones and allocating more bits to the large tones, the coding efficiency can be further improved, and the sound quality can be enhanced at the same bit rate.

Further, the bit allocation process can be simplified by quantizing the remaining spectrum with the same word length.

Furthermore, when there are two local peaks in the predetermined number of spectral ranges, by extracting the single tone with the larger power as a tone, duplicate tone detection can be avoided. Further, by extracting the tone based on the ratio of the average value / median value of the spectrum and by extracting the tone based on the tonality of the spectrum, it is possible to prevent erroneous detection of the tone and perform appropriate tone detection.

[Brief description of drawings]

FIG. 1 is a block diagram showing a schematic configuration of an audio encoder / decoder system to which an embodiment of an audio signal encoding method according to the present invention is applied.

FIG. 2 is a block diagram showing an example of an audio encoder to which the embodiment of the audio signal encoding method according to the present invention is applied.

3 is a block circuit diagram showing an example of a structure of a time-frequency analysis block 21 in the encoder of FIG.

FIG. 4 is a block circuit diagram showing a spectrum quantization block used in an embodiment of an audio signal coding method according to the present invention.

FIG. 5 is a diagram for explaining an example of a tone characteristic portion and a noise characteristic portion of the spectrum of the audio signal in the audio signal encoding method according to the embodiment of the present invention.

FIG. 6 is a flowchart showing an outline of a bit allocation method used in an embodiment of the present invention.

FIG. 7 is a flowchart showing an example of the operation of the bit allocation method used in the embodiment of the present invention.

FIG. 8 is a graph showing a typical masking expansion function.

9 is a flowchart showing an example of local rate calculation for tone measurement in the operation of the bit allocation method of FIG.

10 is a flowchart showing an example of calculating an average / median rate for tone measurement in the operation of the bit allocation method of FIG. 7;

FIG. 11 is a flowchart showing an example of total rate calculation for tone measurement in the operation of the bit allocation method of FIG.

12 is a flowchart showing an example of power density calculation of a residual spectrum used for power quantization unit detection in the operation of the bit allocation method of FIG.

13 is a flowchart showing a specific example of bit allocation according to various inspection results in the operation of the bit allocation method of FIG.

[Explanation of symbols]

11 audio encoder 21 time-frequency analysis block 22-bit allocation control block 23 Spectral quantization block 24 Psychoacoustic model output block 41 Tone component extraction circuit 42 Tone component quantization circuit 43 Noise component quantization circuit

─────────────────────────────────────────────────── ─── Continuation of front page (58) Fields surveyed (Int.Cl. ⁷ , DB name) G10L 19/02

Claims (6)

(57) [Claims] 1. An audio signal coding method for coding a plurality of spectra by subjecting an input audio signal to a predetermined conversion, and allocating bits corresponding to auditory characteristics to each spectrum for coding. Tone components are identified in the spectrum, and tones are extracted from the identified tone components to obtain a predetermined bit.
Quantize betting amount, the remaining spectrum of the plurality of spectrum, high
A frequency domain having a high power density is separated, quantized by a predetermined number of bits as a power quantization unit, and the remaining spectrum is quantized using the remaining available bits. Audio signal encoding method. 2. The extracted tone is a normal tone or
It is classified as either a large tone, the large tone audio signal encoding method according to claim 1, wherein the quantized with larger number of bits than the normal tone. 3. The audio signal encoding method according to claim 1, wherein the quantization by the remaining available bits is performed by the same word length. 4. When extracting the tones , a local peak is obtained, and when the two obtained local peaks are within a predetermined number of spectrum ranges , the one having a larger power is used.
Audio signal encoding method according to claim 1, wherein the extracting as tone. Wherein the tone extraction, an audio signal encoding method according to claim 1, wherein the performed have specific based Dzu the mean / median of the spectrum. Wherein the tone extraction, an audio signal encoding method according to claim 1, wherein the performed based on the tonality of the spectrum.