JP3454394B2 - Quasi-lossless audio encoding device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a quasi-reversible encoding apparatus for speech which encodes an audio signal with high efficiency at predetermined intervals. 2. Description of the Related Art CDs (Compact Discs) are 1982
More than a dozen years have passed since its appearance in the year, and it has now become established as a digital storage medium through various developments.
Considering the use of audio media, this media having a sampling frequency fs = 44.1 kHz and the number of quantization bits = 16 bits has completely entered the maturity period. Also,
In the studio production side of these years, the number of quantization bits has been reduced to 20 bits and 24 bits, and fs = 88.2 kHz.
High sampling, such as 96 kHz, 96 kHz, etc., is progressing, and there is a movement to create a CD based on a master having higher sound quality. [0003] The reason for this is that the sound quality of the finished CD can be improved by working in a format that has room at the editing stage.
It is considered that fs = 44.1 kHz and a 16-bit information amount tend to be unsatisfactory in the first place. In response to this request, consumer devices have been able to achieve high sound quality by performing a pseudo 16-bit to 20-bit conversion at the time of reproduction, or by adding a super-high-frequency signal in a pseudo manner. I have. On the other hand, systems such as video CDs, MDs (mini-discs), and DCCs (digital compact cassettes) have been realized which improve transmission efficiency by encoding audio signals with high efficiency.
Utilizes the psychoacoustic model and realizes sound quality almost equal to CD in terms of hearing with a code amount of 1/4 of PCM. [0005] Considering next-generation audio media of high sound quality, the required information amount is 20 bits, f
s = 88.2 kHz (hereinafter referred to as “2088”) and the like can be considered. In this case, the transmission rate is 2.5 times that of the CD, and it can be said that it is uneconomical to use a simple PCM as an encoding method. The high-efficiency coding methods that have recently been applied in various fields as an alternative to PCM include: lossless coding (lossless coding, lossless compression, noiseless compression, etc.) lossy coding (lossy coding) Compression, noise compression, etc.). The former is mainly used for text data archivers, etc.
The EG international standard is a typical example. The audio data of "2088" is converted to PCM
In order to transmit data with high efficiency instead of the above, a lossless coding method is desirable if possible. However, the compression ratio in this case greatly differs depending on the music, and it is clear from the information theory that the compression ratio cannot be increased particularly for a noise-like music. The compression ratio of “2088” by the lossless encoding method is about 25% to 50% (the data amount is 75%
(Compression to ５０50%), and the worst value is assumed at the time of design, so the efficiency is increased by about 25%. In this case, there is naturally a frame that cannot be almost instantaneously compressed because the song average is assumed.
It is necessary to adopt a variable transmission rate method. As a result, authoring is complicated and time-consuming. FIG. 5 shows an example of the relationship among each frame bit number, average bit number, and original sound bit number when lossless encoding is performed. On the other hand, in the case of the irreversible coding method, if the psychoacoustic model is optimally used, there is no perceived deterioration even if the compression is made to 1/4, and the compression rate of "2088" is 7
5% (compression to 25% data amount) is sufficiently possible. However, considering that the recording density is likely to be improved in an actual high-quality next-generation medium and that the loss is caused when editing is repeated by irreversible encoding, such a high compression ratio is not necessary. [0008] Considering the next generation media of high sound quality, the following options are conceivable. When the compression rate is possible with "0" → PCM as before Conventional when the compression rate is within 25% and a variable transmission rate is used → A lossless encoding compression rate of about 25% to 50% is required and the variable transmission rate Is used → The usable code amount (the number of bits) of some or all frames is reduced from the lossless encoding state to cope with them (partial or all is irreversible encoding). When the compression rate is required to be 50% or more → Irreversible coding using psychoacoustic model. When performing lossless encoding, the following two methods can be roughly classified. A method of performing linear prediction in the time domain and quantizing and encoding the residual. A method of transforming a signal into the frequency domain, normalizing using the bias of energy, and performing quantization and coding. Since the former method has a limited effect of linear prediction,
It is common to use entropy coding for encoding to increase efficiency, and the encoder is shown in FIG. In the latter case, orthogonal transformation and normalization have a considerable data reduction effect, and encoding is used in an auxiliary role, and its encoder is shown in FIG. In each case, there is no significant difference in the achieved compression ratio. In the encoder for the time domain processing shown in FIG. 6, a linear prediction residual output unit 1 performs linear prediction on a PCM signal in the time domain and outputs the residual. The most effective method is a method of calculating a linear prediction count that minimizes the residual in each frame by a least square method or the like. As an example of linear prediction, in the case of linear prediction, a residual is output by the following equation. d [i] = x [i]-(2 * x [i-1] -x [i-
2]) where x [] is an input signal sequence, d [] is a prediction residual sequence.
[] Is normalized for each predetermined frame, and quantized reversibly with required accuracy. In this case, generally, the quantized value is subjected to entropy coding (for example, Huffman code, Lempel-Ziv code, etc.) to further reduce the code amount. Also,
When performing quantization / encoding, it is necessary to increase or decrease the number of quantization bits almost uniformly according to the instruction of the code amount control unit 3 to match the code amount usable in the frame. In such cases, padding bits can be added for adjustment. Format output unit 4
Generally, the prediction method (prediction count) of the linear prediction residual output unit 1, the normalization coefficient (quantization bit number in some cases) of the quantization / encoding unit 2, and the code amount control of the code amount control unit 3 Information and auxiliary information such as a header are added to the information and the data is formatted (bit stream) and transmitted. In the encoder of the frequency domain processing shown in FIG. 7, the buffer 11 buffers PCM signals for frames required when the windowing / orthogonal transformer 12 at the subsequent stage performs orthogonal transform. Then, the windowing / orthogonal transforming unit 12 applies windowing (generally, a windowing such as a Hanning window) to the frame data, performs orthogonal transform using MDCT (Modified Discrete Cosine Transform) or the like, and converts the orthogonal transform coefficient into a plurality of bands. To divide. The normalization unit 13 determines a normalization coefficient (scale factor) for each band, and normalizes the orthogonal transform coefficients in the band. The quantizing / encoding unit 14 reversibly quantizes the normalized coefficient with a required accuracy, and in this case also performs entropy encoding if necessary. However, the effect of entropy coding is generally smaller than in the case of the time domain processing shown in FIG. Further, when performing quantization / encoding, it is necessary to increase or decrease the number of quantization bits almost uniformly according to the instruction of the code amount control unit 15 to match the code amount usable in the frame. If is excessive, it can be adjusted by adding padding bits. The format output unit 16 generally adds a normalization coefficient (in some cases, the number of quantization bits), code amount control information of the code amount control unit 3, and auxiliary information such as a header to perform formatting (bit stream formation). And transmit. FIG. 8 shows a process of normalization and quantization in the frequency domain in the encoder shown in FIG. In this case, the maximum value of each band (a value obtained by quantizing each band in the range of 1 to 2 dB) is defined as a normalization coefficient <S>, and the number of bits required reversibly is determined by the quantization noise level of the assumed PCM original signal ( The noise level <N> is set to be equal to or less than that of white noise (the level should be flat), and requantization is performed with the number of bits satisfying the S / N of each band. The necessary information amount at this time is an area divided by a rectangular area in the figure, and it can be seen that a signal having a bias in energy can be considerably reduced from the original sound information amount. Note that taking the linear prediction residual in the time domain processing is
This corresponds to a process of filtering the signal and averaging the residual spectrum, and is equivalent to roughly realizing the process in the frequency domain shown in FIG. [0016] However, the above 4)
Of the two options, and, which do not cause any particular problem, when (a) when the number of quantization bits is reduced in order to adjust the transmission rate, as shown in FIG. To reduce the excess bits. (B) In particular, priority is given to a frame having excessive bits, and bits are uniformly reduced in the frame. There is an idea. Here, considering the amount of data in lossless encoding, when the signal is close to noise and the entropy is large, the data amount is large (compression rate is low), and when the signal is tone-like and the entropy is small. Has a small amount of data (high compression rate). Conversely, from the psychoacoustic model, it can be said that the noise-like signal has a smaller auditory entropy and a smaller amount of information (the compression ratio may be higher). That is, mathematical entropy and auditory entropy are in inverse proportion. Therefore, in the case of the above (a),
(A) There is a problem that the influence of a frame having a low auditory entropy affects a frame having a high auditory entropy, thereby reducing the amount of data originally required.
Also in the case of (b), since (b) bit reduction does not follow the psychoacoustic model, there is a problem that it is not an optimal method for hearing. In each case, since the variable transmission rate method is used, (c) it takes time to perform processing when the number of bits becomes insufficient in the entire music piece during authoring (authoring must be performed in the worst two passes). There is a problem that processing related to code amount control becomes complicated, for example, a certain head management and a certain read buffer are required at the time of reproduction. In view of the above problems, the present invention can prevent reduction of the originally required data amount, prevent deterioration of sound quality in audibility, and prevent complexity of code amount control. It is an object of the present invention to provide a quasi-lossless audio encoding device. In order to achieve the above object, the present invention is based on a psychoacoustic model when the code amount required for quantization in a reversible manner is less than the available code amount. To perform requantization in an irreversible manner. That is, according to the present invention, means for framing an audio signal for each predetermined section length ,
The coefficient obtained by orthogonal transformation is divided into multiple bands and
Means for normalizing, and the normalized
Calculate the amount of code needed to encode the number in a lossless manner,
A code amount control means for comparing the predetermined code amount of available frames, and psychoacoustic analysis means for analyzing the normalized said coefficients in said frame psychoacoustic model, lossless scheme
If the code amount required for encoding is equal to or less than a predetermined code amount that can be used, the normalized coefficient in the frame is quantized in a lossless manner, and necessary to encode in a lossless manner. quantizing irreversible manner on the basis of a normalized the coefficients in the <br/> frame to the output of the psychoacoustic analysis means when br /> code amount exceeds a predetermined code amount of available Quantizing means and auxiliary information necessary for decoding are output to the output of the quantizing means.
Information to a fixed bit rate bit stream.
Quasi-lossless audio encoding apparatus comprising: means for formatting . According to the present invention, if the code amount necessary for quantization in the lossless method is not less than the usable code amount, quantization is performed in the lossless method as it is, so that the originally required data amount is reduced. Can be prevented. In addition, since the quantization is re-quantized in an irreversible manner based on the psychoacoustic model when insufficient, it is possible to prevent the sound quality from being deteriorated in auditory sense. The complexity of the quantity control can be prevented. Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of a quasi-lossless audio coding apparatus according to the present invention, FIG. 2 is a flowchart for explaining psychoacoustic analysis and code amount adjustment processing in FIG. 1, and FIG. FIG. 4 is an explanatory diagram showing a comparative example of the requantization noise level when the code amount is insufficient between the quasi-reversible encoding device and the conventional example, and FIG. 4 shows a comparative example of audible sound quality between the quasi-reversible encoding device of FIG. 1 and the conventional example. FIG. In the apparatus shown in FIG. 1, first, similarly to the encoder in the frequency domain processing shown in FIG. 7, the buffer 21 is a PCM signal for a frame necessary when the windowing / orthogonal transformation unit 22 in the subsequent stage performs orthogonal transformation. Buffering and windowing
The orthogonal transform unit 22 applies a window to the frame data (generally, a window such as a Hanning window), performs an orthogonal transform using an MDCT (Modified Discrete Cosine Transform) or the like, and divides the orthogonal transform coefficient into a plurality of bands. The normalization unit 23 determines a normalization coefficient (scale factor) for each band, and normalizes the orthogonal transform coefficients in the band. The quantizing / encoding unit 24 reversibly quantizes the normalized coefficient with necessary accuracy, and in this case also performs entropy encoding if necessary. However,
The effect of entropy coding is generally less than in the case of the time domain processing shown in FIG. In this embodiment, the psychoacoustic analyzer 2
5 and the code amount control unit 26 and the quantization / encoding unit 24
The following processing is performed. In FIG. 2, first, the first quantization bit number (Bit [i]) of the coefficient normalized by the quantization / encoding unit 24 is determined, the code amount is estimated, and the total code amount (Total bit) is calculated. It is calculated (step S1). Next, the available code amount (Avail bit) of the frame is confirmed or calculated (step S2), and then the total code amount (Tota) is calculated.
l bit) is compared with the available code amount (Avail bit) to check whether the code amount is insufficient (step S3). If the code amount is insufficient (Total bi
t> Avail bit) includes band power p in consideration of the masking effect of the psychoacoustic model and the minimum audible characteristic.
A masking curve m [i] is calculated from [i] (= normalized value ² = scale [i] ² ) (step S4). In this case, the masking curve m [i] is obtained by performing a convolution operation on the reference curve curve [i] and the band power p [i]. Next, the standard noise level N [i] of each band is calculated from the minimum audibility and the masking curve (step S5), and then the bit is reduced one bit at a time from the band having the higher standard noise level N [i]. The insufficient code amount is allocated to each band. However, every time one bit is reduced in band i, 6.0 is subtracted from N [i] so that the bit reduction becomes similar to the standard noise level N [i] (step S6). Then, the quantization / encoding unit 24 performs requantization and encoding using the number of quantization bits finally determined for each band (step S7). If the code amount is not insufficient at step S3, the surplus bits are assigned or padded to each band (step S8), and the quantization / encoding unit 24 requantizes the surplus bits with the number of quantization bits. And encoding (step S7). The format output unit 26 generally includes
A normalization coefficient (the number of quantization bits in some cases), the code amount control information of the code amount control unit 26, and auxiliary information such as a header are added thereto to format (bit stream) and transmit. FIG. 3 shows a case where the above embodiment is compared with a setting example of the requantization noise level when the code amount is insufficient in the encoder shown in FIG. According to the above embodiment,
The re-quantization noise is shaped according to the psychoacoustic model, so that even if the amount of noise is the same, it is possible to obtain an effect equivalent to a case where the noise level is reduced in terms of audibility. Therefore, it is possible to perform quasi-reversible encoding while minimizing the deterioration of sound quality in the sense of hearing. FIG. 4 shows a comparison example of the sound quality between the case where the lossy encoding of the conventional example (FIG. 7) is performed and the case of the above embodiment.
FIG. 4A shows a case where a part of a frame is irreversible.
(B) shows a case where most of the frame is irreversible.
In the irreversible section as shown in the figure, the present invention shown by a thick line can improve the sound quality more than the conventional example shown by a thin line, and therefore, can obtain a stable sound quality as a whole encoding. Further, according to the present invention, sound quality when irreversible coding is performed can be sufficiently ensured, so that the available code amount of each frame is a fixed “fixed transmission rate”.
Therefore, even if the number of irreversible frames is greatly increased, no sound quality problem occurs. As a result, the processing related to the authoring and the code amount control on the reproduction apparatus side can be greatly simplified. As described above, according to the present invention, if the code amount required for quantization in the lossless method is not less than the usable code amount, quantization is performed in the lossless method as it is. It is possible to prevent the amount of data originally required from being reduced, and in the case of shortage, requantization is performed in an irreversible manner based on the psychoacoustic model, so that it is possible to prevent the sound quality from being perceived from deteriorating. Since transmission can be performed at a fixed transmission rate, the complexity of code amount control can be prevented.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing an embodiment of a quasi-lossless audio coding apparatus according to the present invention. FIG. 2 is a flowchart for explaining psychoacoustic analysis and code amount adjustment processing in FIG. 1; FIG. 3 is an explanatory diagram showing a comparative example of a requantization noise level when the code amount is insufficient in the quasi-reversible encoding device of FIG. 1 and a conventional example. FIG. 4 is an explanatory diagram showing a comparative example of audible sound quality between the quasi-reversible encoding device of FIG. 1 and a conventional example. FIG. 5 is an explanatory diagram showing an example of the relationship among the number of frame bits, the average number of bits, and the number of original sound bits when lossless encoding is performed. FIG. 6 is a block diagram illustrating a conventional lossless encoding method encoder for time domain processing. FIG. 7 is a block diagram showing a conventional frequency-domain processing lossless encoding encoder. FIG. 8 is an explanatory diagram showing frequency domain processing in the encoder shown in FIG. 7; [Description of Code] 22 Windowing / Orthogonal Transformation Unit 23 Normalization Unit 24 Quantization / Encoding Unit (Quantization Unit, Requantization Unit) 25 Psychoacoustic Analysis Unit (Requantization Unit) 26 Code Amount Control Unit ( Requantization means) 27 Formatted output unit

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-6-69811 (JP, A) JP-A-63-15559 (JP, A) JP-A-2-288739 (JP, A) JP-A-7- 50589 (JP, A) WO 90/010993 (WO, A1) (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 19/00 G10L 19/02

Claims (1)

(57) [Claim 1] A means for framing an audio signal for each predetermined section length, and a coefficient obtained by orthogonally transforming a signal in the frame.
Means for dividing into a number of bands and normalizing, and calculating a code amount necessary for coding the normalized coefficients in the frame in a lossless manner, and a predetermined code amount usable in the frame. a code amount control means for comparing, the psychoacoustic analysis means for analyzing the normalized said coefficients in said frame psychoacoustic model, where the available amount of code required for encoding in a reversible manner
If the code amount is less than the specified code amount, the normalized
The coefficients are quantized in a lossless manner and encoded in a lossless manner.
Quantizing means for quantizing an irreversible manner on the basis of a normalized the coefficients in the frame when the code amount needed to exceed the predetermined code amount that can be used for the output of the psychoacoustic analysis means And the auxiliary information necessary for decoding in the output of the quantization means.
To a fixed bit rate bit stream.
Quasi-reversible encoding apparatus for audio having a matting unit.