JP2958726B2 - Apparatus for coding and decoding a sampled analog signal with repeatability - Google Patents

Abstract

Frequency components are calculated from the STP-filtered speech signal. The amplitudes of these are combined in a manner such that the resultant values are associated with frequencies which are situated equidistantly on a linear Bark scale. Said components are quantised, possibly after scaling. In the decoder, the components are again distributed over the frequency spectrum. In the coder, the fundamental regularity D is determined with an LTP technique, after which it is transmitted. In the decoder the phases of the reconstructed signal at the spacing D in the past are determined. These phases are combined with the amplitudes already present in the frequency spectrum, after which transformation back to the time domain takes place. Inverse STP filtering is then carried out.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a device for encoding a sampled analog signal having repetition, and more particularly to a device for encoding a sampled signal (hereinafter sometimes referred to as a sampled signal) each having a predetermined signal. Divided into consecutive segments containing the number of samples; the short-term prediction analysis is performed on the segment and the coefficients determined in the short-term prediction analysis are transmitted and also supplied to the short-term prediction filter; Is converted to the residual signal
Such a device, wherein the information in the residual signal is encoded and transmitted.

[0002] The present invention is an apparatus for decoding a coded signal by said device, in particular, information signal received has been converted from the residual signal, the inverse short-term prediction filter with the received short-term prediction analysis coefficients Supplied to the output of the filter, wherein a series of samples are produced which reconstruct the sampled analog signal.

[0003] The present invention relates to the above device, wherein
The concept can also be applied to the invention of the method.
Is sometimes referred to as a method. Also terms
“Forecast” is “forecast” and the term “multiplexing” is “multiplication”
May be written.

[0004]

BACKGROUND OF THE INVENTION A highly consistent analog signal, such as a speech signal, can be efficiently used after sampling by performing a number of different transformations on successive segments, each having a particular duration. It is known to be coded. One known transform for this purpose is Linear Predictive Coding (LPC); R. Raviner, R.A. W. Shafer "Digital processing of speech signals"
It is described in Pretis Hall, Chapter 8. That is, LPC is used for signal segments having a specific duration (for example, 20 ms for speech signals) and is considered as short-term coding. In addition, short-term forecasts (S
Not only TP) but also long-term forecasts (LTP)
By combining these two techniques, very efficient coding is achieved. The principle of LTP is described in Berry et al. "Talk coder / decoder for European wireless telephone networks", frequency 42, no. 2-3, 1
988, pages 85-93, with improved LTP
The principle is described in Dutch Patent Application No. 9001985.

[0005]

SUMMARY OF THE INVENTION It is an object of the present invention to provide a very useful method for retrieving information about the human ear in a residual signal remaining after applying the STP principle without impairing the quality of speech reproduced by a decoder at a receiving side. That is, to provide a method for transmitting at a small number of bits / second.

For this purpose, the coding according to the invention is
Transform the residual signal into the frequency domain; combine the amplitudes of at least a number of frequency components obtained in the transform, preferably such that the frequencies associated with the combined signal are equally spaced on a linear Bark scale Done; and
A signal representing the combined signal is transmitted.

The decoding method according to the invention is characterized in that the original amplitude in the frequency domain is recovered from the received combined amplitude value and the information sent as a result of the LTP analysis is used to calculate a phase value for that amplitude, In addition, the calculated phase value is converted to a time domain.

According to the present invention, the residual signal is coded sensibly. This means that only information relevant to the difference in the decoded signal that can be detected by the human ear is sent.

First, since the human ear cannot detect the absolute phase value but only the phase relationship, if the original phase relationship can be reproduced at the receiving end, it should be coded.
It is used due to the known fact that there is in principle no need to send phase information from the residual signal.

Further, the present invention provides that the human hearing functions as a chain of multiple filters with adjacent frequency bands, the filters having different bands called so-called critical bands or Barks. It utilizes the knowledge that the bandwidth of a band is narrower for lower frequencies than for higher frequencies. The frequency scale formed by this finding is called the linear bark scale. To find out more about the principle of the Bark scale, see Scarf, S.M. Booth "Stimulation / Physiology / Threshold"
Recognition and Human Behavior Handbook, Chapter 14, pages 1-43, 1
986 is good.

The principle of first converting the residual signal into the frequency domain and then transmitting the information has been known for some time. For example, Chang et al., "Fourier Transform Vector Quantization for Talk Coding," IEEE Communication Report, COM35, No. 1, p.
10, 1059-1068.

However, this bulletin does not mention anything about sending amplitude information after conversion using vector quantization.

FIG. 1 is a block diagram of a coding unit, and FIG. 2 is a block diagram of a decoding unit. The analog signal carried from microphone 1 is low-pass
The bandwidth is limited by a filter 2 and converted in an analog-to-digital converter 3 into a series of amplitude and time-division samples representing the analog signal. The output signal of the converter 3 is supplied to the input of the short-term analysis unit 4 and the input of the short-term forecast filter 5. The unit 4 and the filter 5 send short-term forecasts, for example, in segments of 160 samples, and the analysis unit 4 supplies output signals in the form of short-term forecast filter coefficients which are quantized and coded and sent to a decoder unit (FIG. 2). I do. Since the structure and function of the unit 4 and the filter 5 are well known to those skilled in the art and have no further significance for the present invention,
Further explanation is omitted.

The STP filter signal is sent to the LTP analysis unit 6. In this unit 6, LTP is performed twice for every 160 sample segments, for example in a manner as described in Dutch patent application no.
An analysis is made. In such an LTP analysis, the signal segment to be coded is searched according to a particular search strategy for a segment as similar as possible to the segment of the preceding signal with a particular connection time, and the initial and code of the found segment A signal is sent representing the number D of samples placed between the beginning of the segment to be segmented.

The output signal of the STP filter 5 is called a residual signal. According to the invention, the coded transmission of this residual signal is such that only perceivable information is transmitted. For this purpose, the segment of 160 samples in the residual signal is divided in circuit 7 into 8 sub-segments of 30 samples. This completes the supplied segment first by dividing it into eight sub-segments of 20 samples, then ending them with the last 10 samples of the preceding sub-segment. This also means that the last 10 samples in each segment
Indicates that it must be stored (to complete the first subsegment of the subsequent segment). Each subsegment of 30 samples is then multiplexed in circuit 8 by a window function (eg, a cosine function).
The window function is selected such that the sum of the squares of the two multiplexing factors is the same at the overlap of the subsegments for each sample. The reason for this requirement for squaring is that multiplexing by the window function occurs in both the coding and decoding units. A discrete Fourier transform (DFT) is performed on the windowed sub-segments in circuit 9 and 16 different frequency components are obtained for each sub-segment.
The circuit 10 calculates the amplitude A of the components 1 to 13 out of the 16 components 0 to 15. Component 0,1
4, 15 are 300 to 3,400 Hz selected for speech communication
Can be ignored because it is outside the frequency band of. If a wider or narrower frequency band is appropriate, the number of amplitude components to be considered can be adjusted accordingly. Starting from the 13 components, four so-called Bark amplitude components are
Is calculated. These are the amplitudes associated with equally spaced frequencies on the linear bark scale. The bark amplitude components B _{1 to} B ₄ are, for example, DFT amplitudes A _{1 to} A
₁₃ from the following equation.

[0016]

(Equation 5)

If desired, the gain factor G is calculated as a scaling value in circuit 12 from the four bark amplitude components as follows:

[0018]

(Equation 6)

The application of the scaling value G has the advantage that the scaled amplitude can be coded more efficiently. G
Is quantized in circuit 13 and sent to the decoding unit. Once the scale factor G has been calculated, all the bark components are divided in circuit 14 by the quantization gain factor G. The result of the division is quantized in circuit 15 and coded and sent to a decoding unit.

If no scaling value is used, circuit 1
2, 13 and 14 can be omitted and the four calculated values for the bark amplitude component can be sent directly after quantization in circuit 15.

After decoding by the decoder unit circuit 16, the four scaled bark amplitude components are multiplied by the gain factor G decoded by the circuit 17 by the multiplier 18, so that the reproduced bark amplitude components B ₁ to B ₁ are output.
B ₄ is obtained. This is, of course, not applicable unless a scaling fighter is used in the coding unit. In the circuit 19, the amplitudes A _{1 to} A in the frequency domain
₁₃ (equal intervals on the frequency scale) is calculated by the following equation.

[0022]

(Equation 7)

IDF in inverse DFT (IDFT) circuit
The amplitude and phase are required so that T allows the 13 frequency components considered by the coder to be returned to the time domain.

The phase is decoded by the circuit 23 and is obtained as follows with the help of the LTP information consisting of the sample interval D.

Reproduction STP as it appears at the output of circuit 22
The 120 latest samples of the residuals are stored in each case. In circuit 24, a past subsegment that is D away from the current subsegment is determined, and this subsegment is multiplied in circuit 25 by the same window function used in circuit 8 of the coder. Next, after the DFT is applied to this subsegment in circuit 26, 1
The phases of the three components are calculated by the circuit 27. Using the phase thus determined and the amplitude already calculated, the IDF
T is made by the circuit 20 and A ₀ , A ₁₄ , A ₁₅ , A ₁₆
Is set to zero.

At the output of circuit 20, a 30-sample long subsegment reproduction is available, but modified by a window function performed by the coder. Thus, the playback subsegment is again multiplied by the window function in circuit 21. In the case of the first 10 samples of the sub-segment multiplied twice with the window function, it is multiplied twice with the window function, so that the last 10 samples of the previous segment stored are added in circuit 22. As a result, the sum of the multiplication elements in the ten samples of the synthesis is equal to one.

The last 10 samples of this subsegment are stored. The first 20 samples are part of the playback of the segment of the STP residual . After the eight sub-segments have been reconstructed and combined, a complete reconstructed segment of the STP residual is obtained, and the STP analysis is placed on the past 10 samples than the segment made in the coding unit.

In a manner known per se with the help of the received STP coefficients , the inverse STP filter is
Irutaringu is made in this segment, the filter coefficients from the previous <br/> line segment at this time is used relative to the beginning of the 10 samples.

The output signal of the filter 28 is converted to an analog signal in a digital-to-analog converter 29. This analog signal is sent to the speaker 31 via the low-pass filter 30. This speaker reproduces the speech signal applied to the microphone 1 with high fidelity and can transmit a speech signal encoded with a small number of bits by the method according to the present invention.

If desired, a circuit 23 'can be provided between circuits 23 and 24 to provide more operations to the value of D received by the decoder in order to obtain an optimum value of D for the reproduction of the speech signal. it can. These are the following three successive operations.

1) If the sequence of values of D received shows a trend, and if the current D is out of the acceptable range for the trend, then the current D is determined by the value within the trend. Be replaced. Algorithms that determine the trend in a sequence of continuous values and determine the replacement of values for signals that deviate from the trend are themselves well known to those skilled in the art.

2) The three intermediate values I ₁ , I ₂ , I ₃ are obtained by interpolation with the aid of the algorithm
Is calculated between _two consecutive values D ₁ and D ₂ . For example,
It is performed as follows.

[0033]

(Equation 8)

Since the interval D is determined twice per segment in the coding unit, an interpolation method is performed. Without interpolation, the decoding of four consecutive sub-segments is performed for the same value of D. If there is no basic regularity in the coding unit signal, regularity will be erroneously provided in the decoder during the four sub-segments. This problem is overcome by interpolation.

If there is a basic regularity in the speech signal, the signal repetition interval generally changes slowly. Due to the interpolation method, the change in the value of D has a smooth property in the decoder.

3) If necessary, make the value of D equal by calculating the replacement value, and after performing the interpolation method, make the calculated interval D as good as possible the actual repetition interval D in the signal. Matches. However, if the spacing D is less than 30, D is multiplied by an integer chosen to minimize the product. This is necessary because all samples of the sub-segment have not yet been regenerated at intervals less than 30 for the current segment and cannot be used to calculate phase.

Nevertheless, the reason that an interval D of less than 30 is sent is that the fundamental regularity of the signal, around a large number of samples less than 30, is a value that is an unequal multiple of the actual repetition interval, This prevents the decoded interval D from being taken. As a result, the equalization algorithm has no opportunity to detect trends.

[Brief description of the drawings]

FIG. 1 is a block diagram of one embodiment of a coding unit of an apparatus according to the present invention.

FIG. 2 is a block diagram of one embodiment of a decoding unit of the device according to the invention.

[Explanation of symbols]

 Reference Signs List 1 microphone 2 low-pass filter 3 analog-to-digital converter 4 short-term forecast analysis unit 5 short-term forecast filter 6 long-term forecast analysis unit 7 to 17 circuit 18 multiplier 19 to 27 circuit 28 filter 29 digital-to-analog converter 30 low-pass filter 31 speaker

Continuation of the front page (72) Inventor Robertus Lambertas Adrianas van Lavestein The Netherlands 2275 Tibby Bobberg Hawkway 46 (56) References JP-A-1-296300 (JP, A) (58) Fields investigated (Int. Cl. ^6, DB name) G10L 3/00 - 3/02 G10L 7/00 - 7/04 G10L 9/00 - 9/18

Claims (14)

(57) [Claims] 1. An apparatus for coding an iterative analog signal, comprising: performing a short-term prediction analysis on a quantized and sampled analog signal; and calculating coefficients determined in the short-term prediction analysis. first means for providing an output, means for dividing received a sampled analog signal, and short-term prediction filter for generating a segmented residual signal, the segmented residual signal in sub-segments, the Transform the subsegments from the time domain to the frequency domain,
Then, for each sub-segment , the frequency component is obtained and the
When the wave number component is converted to the frequency domain,
Means configured to have an amplitude, and combining a number of said amplitudes to produce several new frequency components.
And calculate the amplitude of some new amplitudes
Less than the amplitude and the new frequency component
Means for providing the amplitude of the second output to the second output. 2. The method of claim 1, further comprising: performing a long-term prediction analysis on the sub-segments of the segmented residual signal; and providing the coefficients determined in the long-term prediction analysis to a third output. apparatus. 3. The apparatus of claim 2 including means for calculating a gain function as a scaling value, dividing each of the new amplitudes with the gain function, and providing the gain function to a fourth output. 4. The apparatus of claim 3 including means for multiplexing each of the subsegments with a window function. 5. The apparatus of claim 4 including means for quantizing the new amplitude. 6. The amplitudes A ₁ -A ₁₃ of the thirteen frequency components are combined to form a new amplitude B ₁ -B ₄ of _four ; And the gain function G is given by the equation: Apparatus according to claim 5, calculated according to: 7. An apparatus for decoding a coded signal, comprising: a first input receiving coefficients determined in a short-term predictive analysis, calculated by combining the amplitudes of a number of frequency components.
A second input receiving some new amplitudes, a third receiving the phase values determined in the long-term forecast analysis
Means for calculating the amplitude of a frequency component that is an amplitude of some new frequency components from a number of new amplitudes and the number of the new amplitudes is less than the amplitude of the some new frequency components; Means for transforming the amplitude of the new frequency component from the frequency domain to the time domain into a new subsegment; and a first filter input coupled to the first input and receiving the coefficients; The phase value of the third input and the phase value of the second input
An inverse short-term filter having a second filter input coupled to the inverse transform means for inverse transforming using some amplitude and receiving the new subsegment, and producing a series of samples representing the sampled analog signal The above device, comprising: a prediction filter. 8. A third input for receiving the coefficients determined in the long-term prediction analysis, coupled to the third input and the inverse transform means, for determining a past sub-segment that is D samples away from the current sub-segment. Means for determining, means for transforming the determined subsegment from the time domain to the frequency domain, and calculating a phase from the transformed and determined subsegment,
8. The apparatus of claim 7 including means for supplying said phases to said inverse transform means. 9. The apparatus of claim 8, further comprising a fourth input receiving the gain function as a scaling value, and means for multiplexing each of the received new amplitudes with the gain function. 10. A means for multiplexing each of the new subsegments with a window function, and means for multiplexing each of the determined subsegments with a window function.
Equipment. 11. The apparatus of claim 10 further comprising means coupled to said second input for decoding a new amplitude. 12. The amplitude A ′ ₁ of the thirteen new frequency components.
-A _'13 new amplitude B of 4' expression from ₁ -B _'4: Apparatus according to claim 11, calculated according to: 13. The apparatus of claim 8 including means for equalizing values D samples apart according to a predetermined algorithm. 14. Two consecutive D's₁And D₂sample
Intermediate value I of 3 for values D samples away between values₁, I
₂And I₃With the formula: I₁= 0.75 * D₁+ 0.25 * D₂  I₂= 0.50 * D₁+ 0.50 * D₂  I₃= 0.25 * D₁+ 0.75 * D₂  9. The apparatus of claim 8 including means for calculating according to: