JP2002278600A - Complete lossless compression method for audio signal data - Google Patents

Abstract

(57) [Summary] [Problem] In a method using a masking property of auditory sense, an original lossy signal cannot be completely restored because the original data is an irreversible compression algorithm in which original data is not completely restored. Further, LHA and ZIP are algorithms of complete lossless compression, and are mainly used as data file compression, and have a low compression ratio for sound signal data.
MLP is a completely lossless compression algorithm for sound signals, but is not widely used because it is not disclosed. The present invention relates to a process of using acoustic signal data as an algorithm of complete lossless compression by converting a linear prediction error into an integer, and encoding the linear prediction error data by dividing the data into two ranges according to the frequency of appearance. The method includes a step of compressing the data amount and a step of calculating the divided range at a high speed from the root mean square value of the waveform data.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a field in which sound data such as audio and noise is handled, in which sound data can be reduced in a state where sound data can be completely restored without distortion. It relates to a completely lossless compression method.

[0002]

2. Description of the Related Art Conventionally, in noise surveys and the like, noise is sometimes recorded as sound pressure waveform data on a hard disk of a computer in order to perform a trial listening or analysis later. Since the size can be several gigabytes, it is useful to compress and record the noise waveform data when recording on media or transferring data through communication circuits. As an algorithm, ATRA as lossy compression
C, MPEG Audio and Twin-VQ, and LHA, ZIP and MLP as complete lossless compression are used.

[0003]

SUMMARY OF THE INVENTION However, ATRA
C, MPEG Audio, and the method using the auditory masking characteristics of Twin-VQ, the compression algorithm is a lossy compression algorithm in which the original data is not completely restored, and accurately restores the original signal of the sound data. I couldn't do that.

[0004] Furthermore, LHA and ZIP are algorithms of complete lossless compression. These are mainly used for compressing data files, and therefore, any of these methods has a compression ratio for sound signal data. It has the problem of being low.

The MLP is a completely lossless compression algorithm for sound signals, but the MLP algorithm is not widely used because it is not disclosed.

[0006]

A completely lossless compression method for audio signal data according to the present invention comprises the steps of using audio signal data as an algorithm for complete lossless compression by converting a linear prediction error into an integer. A step of compressing the amount of data by dividing the data into two ranges according to the frequency and encoding the data; and calculating the range divided into two ranges from the mean square value of the data without calculating the histogram of the waveform data. High-speed calculation.

[0007]

According to the complete lossless compression method for audio signal data of the present invention, the audio signal data is converted into an integer sequence having a finite bit length by integer processing for compression by linear prediction encoding, and the data of the linear prediction error is expressed in frequency. And a complete lossless compression algorithm in combination with the encoding process according to.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A method for completely reversible compression of audio signal data according to the present invention will be specifically described below with reference to the accompanying drawings.

FIG. 1 shows an example of lossy compression in which 1/30 ct. FIG. 2 is a graph showing a level difference of an average band spectrum, and FIG. 2 is a graph showing an analysis result of a narrow band spectrum in an example of lossy compression.
FIG. 4 is a graph showing a level difference of a narrow band spectrum in an example of lossy compression, FIG. 4 is a schematic flowchart of an algorithm of a complete lossless compression method of audio signal data of the present invention, and FIG. FIG. 6 is an explanatory diagram of an example of division of a range according to the appearance frequency of the completely lossless compression method of FIG. 6, FIG. 6 is a graph showing a linear prediction error e (n) before encoding, and FIG. 8A is a graph showing a non-negative integer sequence y (n) of FIG. 8, FIG. 8 is a graph showing a non-negative integer sequence z (n) of a coded sub-block (condition B), and FIG. It is a graph showing various noise waveforms.

The present invention relates to a method for completely reversible compression of audio signal data in the field of handling audio data such as audio and noise, wherein the data amount can be reduced in a state where the audio data can be completely restored without distortion. About
Using the audio signal data as an algorithm for complete lossless compression by converting the linear prediction error into an integer, and compressing the data amount by dividing the linear prediction error data into two ranges according to the frequency of appearance and coding Process and
Calculating the range division divided into the two ranges at high speed from the mean square value of the data without calculating the histogram of the waveform data.

First, an experimental example using an irreversible compression apparatus will be described. Using a railway noise waveform sampled at 44.1 kHz as an original signal, a difference in spectrum from a noise waveform once recorded on an MD and reproduced again is analyzed. The recording / reproduction of the noise waveform is performed using an optical digital input / output terminal.

FIG. 1 shows one of the original waveform and the MD recording waveform.
/ 30ct. It represents the level difference of the average spectrum of Band, and the legend in the figure shows the average time.
About 0.2 dB for 0.125 seconds, 0.1 dB for 1 second
It is understood that a degree of error occurs.

FIG. 2 shows a narrow-band vector of a short-time frequency by FFT of 4096 points.
Represents the level difference of the narrow-band vector of the short-time frequency by the FFT of 4096 points, and it can be seen from FIG. 2 that the recorded waveform of the MD loses the information of the high frequency.
Further, FIG. 3 shows that although the vector information is maintained on average, there is a frequency component in which a difference exceeding 10 dB is generated. It is clear that it is not suitable for compression recording.

That is, the algorithm of the method for completely reversible compression of audio signal data of the present invention performs linear prediction and code allocation according to the frequency of appearance, which have been conventionally used for compression of audio signals, as shown in the schematic flowchart of FIG. Focusing on the method, the original signal is set to an integer value, and the prediction error is converted to an integer to obtain a complete lossless compression algorithm.

That is, the present invention uses the conventional energy reduction by linear prediction and compression by encoding. According to the following equation (1), linear prediction is performed by setting the target signal to an integer x (n). First, a coefficient a (k) that minimizes the energy of the linear prediction error e (n) in Equation (1) is obtained, where 1 is the prediction order.

[0016]

Next, in order to convert the prediction value and the prediction error into integers, the coefficients calculated by the equation (1) are used. Here, fix ｛｝ indicates the integer part of the number in ｛｝, and the coefficients a (k) and

[0018] Since the energy is small, it can be recorded with low bits.

[0019]

Next, as shown in FIG. 5, the encoding is to adjust the code length according to the appearance probability to reduce the average information amount.
Break it down into two other blocks, each with a short bit and a long bit Non-negative integer sequence y of encoded LK and LV bits
(N) and z (l) are calculated by the following equation (3), where K is the number of bits to be reduced in the main block (condition A), and V is other (condition B and condition B). Bits) It is determined.

[0021] (N), and similarly, (K) in equation (4) is the number in that case.

[0022]

[0023]

Further, at the time of compression, L, K, V, y
(N) and z (n) are recorded, and the one shown in FIG. 6 represents the linear prediction error e (n) before encoding, which is 14 bits, 44100 points, and FIG. 8 shows a non-negative integer sequence y (n) of the main block (condition A), which is 12 bits and 44100 points.
Is the non-negative integer sequence z of the encoded sub-block (condition B)
(N), 7 bits, 2745 points,
That is, when the reduced K and V are calculated with the bit length L = 14, K = 2 in the main block (condition A),
14-2 = 12 bits, and the sub-block (condition B)
Is V = 7, and 14−7 = 7 bits.
44100 × 14 = 617400 before encoding, and 44100 × 12 + 2745 × 7 = 54841 after encoding.
5 is obtained.

The restoration is performed by the following equation (5).

[0026]

Next, an experiment and a result of the completely lossless compression method of the audio signal data according to the present invention will be described. FIG. 9 shows various types of noise waveforms used as a sound source, their types, sampling frequencies and lengths. This is a graph showing the numerical values, and Table (1) described later shows the numerical values, each of which is data of one channel and 16 bits.

[0028]

Next, Table 2 shows the number of bytes of the file before and after the waveform file was compressed by the completely lossless compression method for acoustic signal data of the present invention and the file compression software LHA.

[0030]

As shown in Table (2) above, the present invention is effective because the compression ratio of all sound sources is higher than that of compression by the compression software LHA.

[0032]

As described above, the completely lossless compression method for audio signal data according to the present invention uses the linear prediction, which has been conventionally irreversible compression, to focus on the encoding of the prediction error. It enables lossless compression of data.
It is a revolutionary and highly practical invention that can achieve a compression of about 65% to 45%.

[Brief description of the drawings]

FIG. 1 shows an example of irreversible compression, in which 1/30 ct. 5 is a graph showing a level difference of an average spectrum of Band.

FIG. 2 is a graph showing an analysis result of a narrow band spectrum in an example of lossy compression.

FIG. 3 is a graph showing a level difference of a narrow band spectrum in an example of lossy compression.

FIG. 4 is a schematic flowchart of an algorithm of a completely lossless compression method for audio signal data according to the present invention.

FIG. 5 is an explanatory diagram of an example of division of a range based on the appearance frequency of the complete lossless compression method for audio signal data of the present invention.

FIG. 6 is a graph showing a linear prediction error e (n) before encoding.

FIG. 7 is a graph showing a non-negative integer sequence y (n) of an encoded main block (condition A).

FIG. 8 is a graph showing a non-negative integer sequence z (n) of an encoded sub-block (condition B).

FIG. 9 is a graph showing various noise waveforms used as a sound source.

 ────────────────────────────────────────────────── ─── Continuing from the front page (72) Inventor Mikio Higashiyama 2665-1, Nakanocho, Hachioji-shi, Tokyo F-term in Kogakuin University 5D045 DA02 DA20 5J064 AA03 BB03 BC21 BC28 BD01

Claims (1)

[Claims] A step of using the acoustic signal data as an algorithm of complete lossless compression by converting a linear prediction error into an integer, and encoding the linear prediction error data by dividing the data into two ranges according to the frequency of appearance. An audio signal comprising a step of compressing a data amount, and a step of calculating the range division divided into the two ranges at a high speed from a root mean square value of data without calculating a histogram of waveform data. Complete lossless compression method for data.