JPH08167878A - Digital audio signal coding device - Google Patents

Abstract

PURPOSE: To provide a digital audio signal coding device which can code the digital audio signals without deteriorating the quality of these signals. CONSTITUTION: The characteristic of an audio signal is analyzed in each frame by a dynamic bit allocation means 2.4 based on the peak energy distribution and the pure sound properties of the audio signal. Based on this analysis result, it is decided to use a non-itelligent bit allocation means 2.5 which allocates the bits based on only the signal energy or an intelligent bit allocation means 2.6. The high audio quality can be secured by means of the means 2.4, and also the measurement performance can be secured for the extremely optimum THD+N (total harmonic distortion plus noise) with extremely low compressibility.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital audio signal encoding apparatus, which is used, for example, in transmission or digital storage media.

[0002]

2. Description of the Related Art Digital audio signal compression algorithms have been remarkably popularized and are widely used in consumer products. Such compression algorithms are used to encode digital audio signals in the 20 kHz bandwidth. For example, digital compact cassette (DC
C), mini discs (MD), and video compact discs (video CD). The dynamic bit allocation method used in this algorithm helps to maintain high quality audio.

FIG. 4 shows a general block diagram of a conventional digital audio signal encoding apparatus. Ie 32
The digital audio signal sampled at kHz, 44.1 kHz, or 48 kHz is first converted into frequency conversion means 3.
At 1, the modified discrete cosine transform, the subband reactor wave analysis, the Wavelet transform, or a mathematical transform of its modification is received, and the transform output coefficient is output. The variants include hybrids of sub-band reactor wave analysis and modified discrete cosine transforms, or switching between different sizes of each modified discrete cosine transform. Then, the transform output coefficient is normalized and quantized by the normalizing means 3.2 and the quantizing means 3.3. The quantization can be either linear or non-linear. The number of bits allocated to quantize each transform coefficient or block of transform coefficients is determined by the perceptual bit allocation means 3.4.

The perceptual bit allocation means removes redundant and irrelevant content in the digital audio signal in such a way that audio quality is not degraded. To do this effectively, knowledge based on psychoacoustic experience of how the human ear perceives sound is used, and a mathematical model based on this knowledge is constructed. The bit allocation method that employs this mathematical model is called the perceptual bit allocation method. An example of this perceptual bit allocation method is ISO / IEC 11172.
-3 document psychoacoustic models I and II.

In the coding apparatus using the psychoacoustic model, a frame or block of a digital audio signal is first converted into a spectral component having a fine resolution, usually by a fast Fourier transform (FFT). The spectral components are grouped into subband sizes that are closely proportional to the critical band and analyzed. Here, the critical band refers to a characteristic portion of a wideband audio spectrum in which specific psychoacoustic regularity such as frequency selectivity / masking threshold is effective.

Pure tone and noise masking are estimated from the spectral components. The characteristics of simultaneous masking with pure tone masking and noise masking have been modeled to obtain a masking threshold at each and every spectral component frequency. Note that the masking threshold may be estimated by the number of frequencies reduced for each critical band in order to reduce calculation complexity. The signal-to-masking threshold ratio for the subbands, each containing a large number of spectral components, can be calculated in succession and used in an iterative bit allocation means.

The coding device serves to achieve a noise-free audio quality. The decoded signal is of such high quality that it is indistinguishable from the original signal for compression rates up to about 8 times. The compression rate is represented by the ratio of the bit rate of the input signal to the bit rate of the decoded bit stream. For example, 44.
When processing a digital audio signal sampled at 1 kHz, the bit rate is 705.6 kbits / second / channel, while DCC is 192 Kbits / second / channe.
l, MD: 146.08 kbits / second / channel, and Video CD audio: 112 kbits / second / cha
It is effectively compressed into nnel and the compression rate is 3.67 each.
5, 4.83, and 6.3 times.

[0008]

In a digital audio coding apparatus having a perceptual bit allocation means with a compression rate of less than 5, not only high noise-free quality but also relatively good total harmonic distortion plus noise ( The measurement result of THD + N) is shown. (The THD + N measurement refers to the sum of broadband noise, AC power related hum, and some other disturbing signal below and above the frequency at which the measurement is made.)
This is because an individual or block of large valued coefficients that requires a large number of bits can utilize sufficient bits to maintain high quality. This shows that securing sufficient bits is effective in achieving high quality in high addition coefficients of individual or blocks that require a large amount of bits. This is independent of the bit allocation criteria being based on the signal-to-masking threshold ratio rather than the signal energy. However, as the compression ratio increases, the total number of bits available for quantizing the spectral transform coefficients decreases. Nevertheless, good audio quality can be achieved, but the THD + N measurement is significantly reduced. Although the evaluation of the sound quality of the encoding device provided with the perceptual bit allocation means should be based on the subjective sensation audition as the main evaluation criterion, the measured value of THD + N still indicates an important quality of the encoding performance. It is treated as an index to confirm. This is especially true for professional audio or studio applications.

At high compression ratios, the criterion for the signal-to-masking threshold ratio used by the perceptual bit allocation means is often not allocable to the required bit units, resulting in good results. THD +
It becomes impossible to secure the measurement result of N. here,
A series of THs measured for digital encoding of audio signals of video CDs with perceptual bit allocation means
The measurement results of D + N are shown in Table 1.

[0010]

[Table 1]

Higher THD + N measurements should be discussed as input test signals used to measure THD + N belonging to the class of non-critical audio signals.

FIG. 5 shows the frequency coefficient of the test tone. (Compare this to the frequency spectrum of any audio signal in FIG. 6.) This class of non-critical signals is
It has high energy levels in a few units and very low energies in other units. Although the energy is spread across more units, there are also non-critical signals that do not require perceptual bit allocation to ensure high quality. Such a non-critical signal does not require the high bit rate used for coding to ensure a high quality of audio, while at the same time obtaining good THD + N measurements.

In view of the above problems of the prior art, the present invention provides a digital audio signal having a dynamic bit allocation means for deciding whether to use the perceptual bit allocation means or the non-perceptual bit allocation means. It is an object of the present invention to provide an encoding device.

[0014]

According to the present invention of claim 1, there is provided a frequency conversion means for frequency-converting each frame of a digital audio signal into a spectrum component, and a group for grouping the spectrum component into a predetermined bandwidth or unit. A means of division,
Perceptual bit allocation means for allocating bits or bandwidths based on human auditory characteristics; non-perceptual bit allocation means for dynamic or dynamical allocation of bandwidths or units based on signal energy; A digital audio signal coding apparatus, comprising: a dynamic bit allocating unit that determines whether to use the bit allocating unit or the non-perceptual bit allocating unit according to a predetermined standard.

[0015]

The frequency conversion means for converting the frame of the audio signal converts it into a spectral component by a mathematical filter conversion such as a fast Fourier transform, a modified discrete cosine transform, or a wavelet transform, and outputs it. Since high resolution is essential for these spectral components, the number of points used for conversion is typically 512 or more.

The spectral components need to be grouped into uniform unit sizes, such as uniform subband coding. Alternatively, it needs to be grouped into units of different lengths, such as non-uniform subband coding or Wavelet coding.

The peaks of the contained spectral components of each and every unit are extracted and compared to a set of thresholds. These thresholds are often set in the Quiet mode. Alternatively, for simplicity, it may be set to a constant value close to the minimum of the set of threshold values in the Quiet mode. The number of units with peaks above said threshold represents an estimate of the energy of the examined audio signal.

Next, the unit pureness index for peaks above the threshold is calculated. The calculation of this index is only possible for units relating to the number of spectral components above the lower limit. The index may be either binary, such as 1 or 0, or a fraction. Fractions are obtained by calculating a measure of spectral flatness for that unit. The pure tone index indicates whether the spectral characteristic of the unit is pure tone.

A stronger masking effect is obtained from the units that are pure tones. From the number or percentage of units of the spectral component where the peak is above the Quiet mode threshold, it is possible to identify whether it is a non-critical audio signal. If this number of units is lower than a predetermined value, it indicates that the audio signal contains energy in very few units. Even with high spectral energy, the demand for bit allocation is not necessarily high. If the number of units exceeds the predetermined value and is below the level above which signals are considered wideband, a further check on the estimated requirements for bit allocation is made by the calculation of the pure tone index. If the sum of the pure tone indexes is lower than a predetermined threshold, the requirement for bit allocation is small and the frame of the audio signal can be considered non-critical. For frames of the audio signal identified as non-critical, conventional dynamic bit allocation based on signal energy is applied. In other cases, perceptual bit allocation applies.

The unit may be bandwidth.

[0021]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below with reference to the drawings showing its embodiments.

FIG. 3 is a block diagram of an embodiment of the digital audio signal encoding apparatus of the present invention. That is, the frequency conversion means 2.1 frequency-converts each frame of the digital audio signal into spectral components. The normalizing means 2.2 normalizes the values of the spectral components grouped into a predetermined bandwidth or unit within a range of ± 1. The quantizing means 2.3 performs general uniform linear quantization. The non-perceptual bit allocation means 2.5 performs dynamic bit allocation based on signal energy for the bandwidth or unit. The perceptual bit allocation means 2.6 allocates bits to the bandwidth or unit based on human auditory characteristics. The dynamic bit allocation means 2.4 is for deciding whether to perform the non-perceptual bit allocation means or the perceptual bit allocation means.
The repetitive bit allocation means 2.7 performs allocation based on the determination of the dynamic bit allocation means.

Next, the operation of the above embodiment will be described.

When the frame of the digital audio signal S2.1 is input to the frequency conversion means 2.1, the fast Fourier transform, the modified discrete cosine transform, or the Wa
By a mathematical filter transformation such as the velet transformation,
The spectrum component is converted into the spectrum component S2.2.
Is output as Since high resolution is indispensable for the spectral components, the number of points used for the frequency conversion is usually 512 or more.

Prior to normalization and quantization, the spectral components must be further grouped on the frequency or time axis based on their resolution. That is,
The spectral components need to be grouped into uniform bandwidths or unit sizes, such as uniform subband coding. Alternatively, they need to be grouped into bandwidths or units of different length, such as non-uniform subband coding or Wavelet coding. For example, if the spectral component is a 512-point MDCT,
When the resolution is high, the components are grouped on the frequency axis. In other cases, component grouping in the time domain is performed like uniform 32 band subband block coding. The spectral components, which may be uniform or non-uniform as described above, must be grouped to be the same size as the bandwidth or unit for which the quantized bits are to be allocated separately. . (Hereinafter, the bandwidth or unit is simply referred to as a unit.) Next, the peaks of the spectral components contained in each and all of said units are extracted and compared with a set of thresholds. As the threshold value, a threshold value in Quiet mode is often set. Alternatively, for simplicity, it may be set to a constant value close to the minimum of the set of threshold values in the Quiet mode. Here, the threshold of the Quiet mode of a certain frequency refers to the minimum value of pure tones with respect to the audible frequency. The number of units for peaks above said threshold represents an estimate of the energy of the examined audio signal.

A unit pureness index for peaks above the threshold is calculated. This measure can only be calculated in units of the number of spectral components above the lower limit. The index may be either binary, such as 1 or 0, or a fraction. The fraction is
It is obtained by calculating a measure of the flatness of the spectrum for that unit. The pure tone index indicates whether the spectral characteristic of the unit is pure tone. Here, tonality refers to the sinusoidal nature of an audio signal.

A more powerful masking effect is obtained from the units that are pure tones. That is, it is possible to identify whether or not it is a non-critical audio signal from the number or percentage of units of the spectral component whose peak exceeds the threshold of the Quiet mode. If the number of units is lower than a predetermined value,
The audio signal is shown to contain energy in very few units. Even with high spectral energy, the demand for bit allocation is not necessarily high. If the number of units is above the predetermined value and below the level above the signal considered wideband, a further check on the estimated demand for bit allocation is made by calculation of the pureness index. If the sum of the pureness indices is lower than a predetermined threshold, the requirement for bit allocation is small and the frame of the audio signal can be considered non-critical. For frames of the audio signal identified as non-critical, the non-perceptual bit allocation means 2.5 based on signal energy is applied. For frames of the audio signal identified as critical,
The perceptual bit allocation means 2.6 is applied.

It is the dynamic bit allocation means 2.4 that makes the decision whether to perform the non-perceptual bit allocation means or the perceptual bit allocation means.

When the non-perceptual bit allocation means is selected and the spectral component S2.2 is not of sufficiently high resolution, no request is made to perform parallel frequency conversion on the input audio signal. At this time, the repetitive bit allocating means 2.7 needs to allocate based on only the signal energy or the fluctuation. It is not necessary to perform complicated analysis on the signal content. The signal energy or variation is S by the non-perceptual bit control means 2.5.
It is output as 2.5.

When the perceptual bit allocation means 2.6 is selected, a masking threshold needs to be calculated.
For this purpose, spectral components with sufficiently high frequency resolution must be used. The signal-to-masking threshold ratio is output from the perceptual bit allocation means 2.6 as S2.6. At this time, the repetitive bit allocation means 2.
7 prioritizes and assigns all the bits that are available iteratively to the unit of the spectrum with the highest signal change or signal to masking threshold ratio. The set of bits assigned to each and every spectral unit is output as S2.7.

The normalizing means 2.2 normalizes the values of the spectral components belonging to the same unit to within ± 1 before the quantization. The normalized spectral component is referred to as a normalized spectral component S2.3. The normalization is performed by dividing the values of all spectral components of the unit by the value of the spectral component of maximum magnitude or of a scale equal to or only slightly greater than the maximum magnitude. .

The quantizing means 2.3 performs general uniform linear quantization, but it may be non-linear quantization. If uniform linear quantization is performed, reduction of quantization noise can be easily realized. Alternatively, if the additional bits are used for quantization, then an improved SNR can be easily achieved. Using more bits to quantize the normalized spectral component reduces the quantization noise.

The operation of the dynamic bit allocating means 2.7 will now be described more specifically step by step with reference to FIGS. 1 and 2. Regarding the spectral components,
Although it is possible to group into units of different sizes or bandwidths, for the sake of explanation, we will group into units of equal size and the units are dB.

The grouping of audio frames into non-critical or critical audio signals includes analysis of peak energy distribution and pureness. The unit size is generally uniform or proportional to the critical band size. Due to the size of the critical band, lower numbered units will have fewer spectral components than higher numbered units.

The peak or maximum magnitude in each and every frequency unit is extracted. In the analysis of peak energy distribution, the peak or maximum magnitude of all units is first extracted (1.1). For example, comparison of these peak values to a predetermined threshold, such as the Quiet mode threshold, is performed to identify "audible units."
For every "audible unit" detected, the unit counter is incremented or the spectral component counter is incremented by the number of spectral components in that unit. The incrementing of the unit counter is done when uniformly sized units are used (1.4).

The tonality index is calculated as 1.5 by the tonality determining means. The index is calculated to give a binary or a fraction. For binary pureness index calculations, the window size is 7 spectral components. This means that the units immediately before and after the unit being processed are
It means that it is necessary for calculation. X (k) -X (k +
If the spectral component X (k) with j) ≧ 6 dB can be found, the pure tone index is set to 1. Here, k is a spectral component index for the unit, and j is j = −
3, -2, + 2, + 3. Instead of this, the formula (1
-SFM) may be used to use the Fractional Tonicity Index. Here, SFM is a measure of the flatness of the spectrum and is defined as the ratio of the arithmetic mean of the power spectrum to the geometric mean of the power spectrum. The window size for SFM calculation is at least 16 spectral components. For units below this size, spectral lines from adjacent units can be used to give the appropriate number. The spectral lines of the unit under examination should form the central component of the window for the SFM calculation.

From the analysis of the peak energy distribution, Qui
The ratio of the spectral components exceeding the threshold of the et mode is 10
If it is less than%, the audio signal is said to be completely non-critical. If it is more than 20%, the signal is said to be completely critical. If the proportion is between 10% and 20%, the number of units with a pureness index above 0.5 is investigated, as indicated by 1.10. If more than two units have a high pureness index, it may not be possible to base bit allocation solely on signal energy, as this may complicate the calculation of masking thresholds. It means that.

Next, FIGS. 1 and 2 will be described step by step. Frame of digital audio signal is FFT, MD
Transform into spectral components using CT or other transforms. The spectral components are grouped into units of equal size. Identify the peak or maximum magnitude of the spectral components of each and every unit (1.1). Initialize the peak counter p to 0 (1.
2). The peak value is compared with the minimum Quiet mode threshold value η dB for the unit (1.3), and if the peak value exceeds η, the peak counter p is incremented (1.
4). A pure tone index is calculated (1.5). Binary or fractional pureness index calculation methods can be used. When the peak value check and the pure tone index calculation have been performed for all N units, the peak counter p value is checked (1.6). If it is less than or equal to χ, proceed to the implementation of non-perceptual bit allocation means, i.e. allocation based on the signal energy of each and every unit (1.13). However, χ = 0.1 × N. If p is larger than χ, it is checked whether p is γ or less (1.7). However, γ = 0.2 × N. If it is not true, proceed to the perceptual bit allocation means, i.e. performing the signal-to-masking threshold based allocation of each and every unit (1.14). If p is less than or equal to γ, the tonality counter q is initialized to 0 (1.9), and the tonality index of p units having a peak value exceeding η is sequentially checked for whether they are greater than ξ. Yes (1.10). However, ξ = 0.5. When it is true, the pure tone counter q is incremented (1.11). If q is less than or equal to M, proceed to the non-perceptual bit allocation means (1.13). Here, it is assumed that M = 2 for the original audio bandwidth exceeding 4 kHz, and M = 1 in other cases. Otherwise, go to Perceptual Bit Allocation Means (1.14). Finally, regarding the conventional digital audio signal encoding device and the encoding device of the present invention,
The D + N measurement results are compared.

First, regarding the conventional encoding device, a video CD operating at 224 kbits / second for stereo L / R will be compared. When the structure is a digital audio encoding device having the perceptual bit allocation means shown in FIG. 4, the measurement result of THD + N is shown in Table 1 above.
Becomes From this, various frequencies, especially 2018Hz,
It can be seen that the measurement results are undesirable at 15005Hz, 16004Hz, and 17003Hz. The main reason for this is that the perceptual bit allocation means is always working, independent of all signal formats.

On the other hand, as shown in FIG. 3, Table 2 shows the measurement results of THD + N obtained by the digital audio signal coding apparatus of the present invention provided with the dynamic bit allocation means.

[0041]

[Table 2]

From this, it can be seen that the measurement result of THD + N is significantly improved for the frequencies whose results are not desirable in Table 1. At the same time, no noticeable deterioration in sound quality due to the audition was found.

[0043]

As is clear from the above description, the present invention enables the dynamic bit allocation means to operate at a high compression rate or a low bit rate and identify a non-critical audio signal. By this, by applying a non-perceptual bit allocation means, it is possible to encode without impairing the quality of the digital audio signal,
Then, there is an effect that the measurement result of THD + N can be maximized at all frequencies.

[Brief description of drawings]

FIG. 1 is a flow chart showing the operation of a dynamic bit allocation means according to an embodiment of the present invention.

FIG. 2 is a flowchart showing the operation of the dynamic bit allocation means of the embodiment of the present invention.

FIG. 3 is a configuration diagram of a digital audio signal encoding device according to the present embodiment.

FIG. 4 is a block diagram of a conventional digital audio signal encoding device.

FIG. 5: Graph of frequency coefficient of 1 kHz sinusoidal test input

FIG. 6 is a graph of frequency coefficients of an arbitrary audio signal.

[Explanation of symbols]

 2.1 Frequency conversion means 2.2 Normalization means 2.3 Quantization means 2.4 Dynamic bit allocation means 2.5 Non-perceptual bit allocation means 2.6 Perceptual bit allocation means 2.7 Iterative bit allocation means

─────────────────────────────────────────────────── ─── Continuation of front page (51) Int.Cl. ⁶ Identification number Office reference number FI Technical indication location H03M 7/30 A 9382-5K H04L 29/02 (72) Inventor Kee Nung Phat Singapore Truro Road 28

Claims (2)

[Claims] 1. A frequency conversion means for frequency-converting each frame of a digital audio signal into a spectrum component, a grouping means for grouping the spectrum component into a predetermined bandwidth or unit, and the bandwidth or unit of a human being. Perceptual bit allocation means for performing bit allocation based on auditory characteristics; non-perceptual bit allocation means for performing dynamic bit allocation based on signal energy in the bandwidth or unit; and the perceptual bit allocation means or the non-perceptual bit allocation means. A digital audio signal coding apparatus, comprising: a dynamic bit allocating means for determining which of the bit allocating means is to be used according to a predetermined standard. 2. The determination comprises extracting spectral component values of peaks contained in the bandwidth or unit and calculating a ratio of the bandwidth or units having a spectral component value of the peak smaller than a predetermined threshold value to the total number of units. , Determining the tonality of the signal contained in each and all of said bandwidths or units, calculating the total tonality of the digital audio signal, and performing based on said ratio and said total tonality. The digital audio signal encoding device according to claim 1.