JPS5936280B2 - Adaptive transform coding method for audio - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an adaptive transform encoding method that transforms an audio signal into a frequency domain and adaptively changes its quantization.

<Prior art> This type of audio encoding method is known, for example, from Japanese Patent Application Laid-Open No. 55-579.
No. 00 "Audio Signal Processing Circuit".

As shown in FIG. 1, in this method, input audio from an input terminal 11 is sampled, for example, at KH2, and each sample value is input as a digital signal to an orthogonal transform section 12.
The orthogonal transform unit 12 converts a certain number of input voice samples 5 shown in FIG.
... is converted into fn (FIG. 2B) and sent to the adaptive quantization section 13. On the other hand, the input voice of the terminal 11 is input to the spectral envelope extraction section 14. The spectral envelope of the input voice is estimated by linear predictive analysis, and the spectral envelope and pitch period are supplied to the adaptive information allocation section 15. The adaptive information allocation unit 15 receives frequency domain signals F, f2.
. . . Depending on the instantaneous level of the spectral envelope for each Fn, the quantization unit 13 assigns more bits if the level is large, and decreases the assigned bits if it is small, taking into account the pitch period. The quantization bits for each signal Flf2...Fn are adaptively changed. The information quantized in this way,
The information indicating the bit allocation is combined by a combining circuit 16 and sent out as an encoded output. By this method, 8KH
A z-sampled audio signal can be efficiently encoded with an information amount of about 16 Kbps, and high-quality audio can be obtained.

However, since bit allocation information requires an amount of information of about 2 Kbps, when encoding with a total amount of information of 9.6 Kbps or less (one of the commonly used transmission speeds), the signals F, f2 . . . It is necessary to quantize Fn to less than 1 bit/sample. At this time, almost no information can be assigned to sections of low intensity among frequency components, resulting in significant deterioration of voice quality. <Summary of the Invention> This invention converts an input audio signal into the frequency domain, divides the converted spectrum into block units, and adaptively performs vector quantization in each block according to spectrum envelope information. For example, the encoding speed is 9.6Kb
An object of the present invention is to provide an adaptive conversion encoding method for audio that reduces deterioration in audio quality even at frequencies below ps.

Furthermore, vector quantization can be efficiently performed by flattening the spectrum before dividing it into blocks. <First Embodiment> FIG. 3 shows an embodiment of the speech encoding system according to the present invention.

The input signal from the terminal 11 is converted into a frequency domain signal, that is, a spectrum, by orthogonal transformation such as discrete Fourier transform (DFT) or discrete cosine transform (DCT) in units of one frame in orthogonal transform section 12. The spectrum is separately determined and globally flattened using quantized spectrum envelope information in the spectrum smoothing unit IT. That is, the spectral envelope of the input voice at the terminal 11 is estimated by linear predictive analysis in the spectral envelope extraction section 14, and this spectral envelope information and voice power are quantized as auxiliary information in the quantization section 18, and this quantized output is locally decoded. Part 19
The decoded auxiliary information is used to divide the spectra F, f2 . . . Fn by the spectrum smoothing unit IT. This flattened spectrum is divided into p consecutive blocks by the block dividing unit 21 as shown in FIG.
2 (divided into F2l.Each component element Ii of the spectrum
(i=l...S, j=l...p)
each consists of a real part R (Fij) and an imaginary part I (Fij), and for each block there is a vector R (Fsp)) whose elements are these real parts, and a vector R (Fsp) whose elements are each imaginary part, and s vectors whose elements are each imaginary part. A vector I(Fi)=(I(FiJ)) is created. The vector quantization unit 22 detects which standard vector in a dictionary prepared in advance these vectors most closely corresponds to, and performs vector quantization. In other words, it stores multiple predicted standard vectors as a dictionary, detects which standard vector the input speech vector is close to, and outputs a code such as a number indicating the matching or similar standard vector. . Therefore, the intensity of each spectral component can be encoded using fewer bits than when quantized. Moreover, the bit allocation for this vector quantization is adaptively changed. That is, the number of bits is assigned to each block according to the decoded output of the auxiliary information which is the output of the local decoding section 19.

Generally, a large number of bits are assigned to a program that includes a strong spectrum, and when a large number of bits are assigned to the weak spectrum quantization unit 22, a dictionary with a large number of standard vectors to be compared is referred to, and a small number of standard vectors are referred to. When a large number of bits is allocated, refer to a dictionary with a small number of standard vectors. Since the number p of standard vector elements is constant, the standard vectors stored in a dictionary with a large number of standard vectors also display minute patterns, and the standard vectors stored in a dictionary with a small number of standard vectors are stored in a dictionary with a small number of standard vectors. It can be said that the standard vectors shown only show a large pattern. This adaptive information allocation (bit allocation) is performed for the purpose of maximizing the signal-to-noise ratio of the input signal and the output signal for each frame.

Since the S/N ratio remains unchanged even after orthogonal transformation, it is best to minimize the distortion between the output of the spectral smoothing unit IT of the encoder 24 and the spectral reproduction output of the decoder 25 on the receiving side. is the Euclidean distance. The distortion D per l frame is given by the following equation. Also all samples (spectra)
The number is p·s, and the average amount of information per sample (average number of quantized bits) R is.

Since B is held constant, the number of quantization bits Bj that minimizes the distortion D is given by the following equation. Quantization is performed by converting this Bj into an integer value and selecting the one with the minimum distortion from a dictionary consisting of 2bj pieces.

Note that the quantization in the quantization unit 18 can also be vector quantization.

The quantized output of this spectral envelope, that is, the auxiliary information,
The waveform information output from the vector quantization unit 22 is combined and sent to the decoder 25 as an encoded output. In the decoder 25, the input waveform information is passed through the smoothing spectrum reproducing unit 26 and the vector quantizing unit 22 in the encoder 24.
Using the same dictionary used in step 1, the standard vector is read out using the quantization code of each block, and the smoothed spectrum is reproduced.

The spectrum envelope of the input auxiliary information is reproduced by the spectrum envelope reproduction section 2T, and the spectrum envelope is multiplied by the power by the reproduced smoothed spectrum in the spectrum reproduction section 28 to reproduce the spectrum. The reproduced spectrum is inversely transformed into the time domain by the inverse transformer 29 to obtain a reproduced audio signal at the output terminal 31. <Second Embodiment> In the above description, spectrum smoothing was performed after orthogonal transformation, but orthogonal transformation may be performed after passing the input audio through an inverse filter.

For example, as shown in FIG. 5, in which parts corresponding to those in FIG. On the other hand, the spectral envelope of the input audio signal is analyzed by the linear prediction analyzer 33, the analyzed prediction coefficients are vector quantized by the quantizer 18, the quantized output is decoded by the local decoder 19, and the decoded output is , that is, the filter constant of the inverse filter 32 is controlled by the linear prediction coefficient. The output of this inverse filter 32 is a residual signal, which is orthogonally transformed, encoded in the same manner as described above, and sent out. In the decoder 25, the vector quantized code is decoded by the spectrum reproducing unit 26 to reproduce the spectrum of the residual signal, which is inversely transformed into the time domain and sent to the linear prediction synthesis filter unit 34. The filter constant of this synthesis filter section 34 is controlled by the prediction coefficients reproduced by the spectrum envelope reproduction section 2T, and the audio signal is reproduced from the filter section 34. Figure 6A shows the waveform Al of the input audio signal and the waveform A of the real part of the orthogonal transform output.
2, the waveform A3 of the imaginary part is shown, and FIG. 6B shows the waveform A3 of the input audio signal. The residual signal waveform B after passing through the inverse filter section 32 is the real part waveform B2 of the orthogonal transform output of this residual signal.
and the imaginary part waveform B3 are respectively shown.

The spectral envelope B4 of the audio input waveform A and the allocated bits B5 for each block are shown. Tazushi p=
6. This is an example of B = 1.0. In the above, the efficiency of quantization can be further improved by adapting the dimension P of the vector serving as the unit of quantization to the pitch frequency of the input audio and making the length of one frame an integral multiple of the pitch period.

In this case, since the pitch frequency changes over time, the pitch frequency is also included in the auxiliary information. It is also possible to process vectors as units of complex numbers without making the real and imaginary parts independent. Further, each of the above-mentioned units can be processed by an independent computer or by a common computer. <Effects> As explained above, quantization efficiency can be improved by dividing a signal flattened in the frequency domain into blocks and adaptively allocating information. It is possible to reproduce audio with a higher signal-to-noise ratio than the adaptive transform coding method.

By flattening the frequency domain, the number of standard vectors for vector quantization can be reduced. Further, the amount of information allocated per l block only needs to be an integer, and the amount of information per sample is allocated finely in units of l/P bits. This makes it possible to avoid auditory deterioration caused by the existence of frequency components to which the amount of information cannot be fully allocated, which is a drawback of the conventional method, and which is adaptively changed. Next, an experimental example will be described.

The sampling frequency is 8 KHz) The analysis order of the linear prediction analysis unit 33 is 8th, the analysis length (conversion length) is 26 to 31 ms,
Figure T shows the S/N ratio for the amount of information B (bits/sample) when the analysis overlap is 2 ms (connected by trapezoidal windows) and the number of vector dimensions is 6 to 12 (pitch adaptation). In FIG. 6, curve 41 is uniform quantization and each sample is encoded, curve 42 is uniform quantization and 6-dimensional fixed vector encoding is performed, and curve 43 is uniform quantization and the vector dimension is pitched. When encoding is performed by changing it according to the frequency, curve 44 is when each sample is encoded by adaptive quantization (conventional method), and curve 45 is encoded by six-dimensional fixed vector quantization using the method of the present invention. If the curve 46
This is a case where the method of the present invention is used to adaptively change the dimension of a vector according to the pitch frequency for encoding. From these results, adaptive quantization (curves 44 to 46) is better than uniform quantization (curves 41 to 43), and even in adaptive quantization, the inventive method (curves 45 and 46) is better than the conventional method (curve 44).
It is understood that this is better. 0.5~1.1
In the bit/sample domain, this improvement in SN ratio was obtained even outside of the training samples by 2.5 dB for female voices and 1.0 dB for male voices.

By vector quantizing the spectrum envelope, the auxiliary information including pitch, power, etc. can be estimated at about 800 bps, so if the information amount B per residual signal l sample is 0.5, it is 4.8 Kbps) 1.1 at 9.6Kbps
It is possible to encode

[Brief explanation of the drawing]

Figure 1 is a block diagram showing a conventional adaptive transform encoding method.
FIG. 2 is a diagram for explaining its operation, FIG. 3 is a block diagram showing an example of the adaptive transform encoding method according to the present invention, FIG. 4 is a diagram showing an example of block division, and FIG. FIG. 6 is a diagram showing an example of its operation, and FIG. T is a diagram showing the relationship between the SN ratio and the amount of information B of various encoding methods. 11: Audio input, 12: Orthogonal transform unit, 14: Svector envelope extractor, IT: Svector smoothing unit, 18: Vector quantizer, 19: Local decoder, 21: Block division unit, 22: Vector quantum Converter, 23: Adaptive information allocation unit, 2
4: Encoder.

Claims (1)

[Claims] 1. In an encoding method in which a fixed number of sample value sequences of an audio signal are set as one frame, a spectrum is obtained by orthogonal transformation for each frame, and adaptively quantized, the envelope of the spectrum is a spectral envelope extraction means for calculating and quantizing the power and encoding it as auxiliary information; a local decoding means for decoding the auxiliary information; and a spectral envelope extraction means for decoding the auxiliary information. a block dividing means that divides the flattened spectrum decoded sequence into blocks on the frequency axis; 1. An adaptive transform encoding method for speech, comprising an adaptive information allocation means for allocating information, and a vector quantization means for vector quantizing the divided spectral signal sequence according to the allocation. 2 In a coding method that analyzes and encodes a sample value sequence of an audio signal in units of one frame, linear prediction analyzes the audio signal, determines its spectral envelope, quantizes it together with power, and encodes it as auxiliary information. an analysis means, a local decoding means for decoding the auxiliary information, a filter constant is controlled by a linear prediction coefficient of the decoded auxiliary information, and a sample value sequence of the audio signal is input, and a residual signal is output. an orthogonal exchange means for orthogonally exchanging the residual signal for each frame to obtain a spectrum; a block dividing means for dividing the spectrum into blocks on the frequency axis; Adaptive information allocation means for adaptively allocating information using the auxiliary information, and vector quantization means for vector quantizing the divided spectral residual signal sequence according to the allocation. Adaptive transform coding method.