JP3435310B2 - Voice coding method and apparatus - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice coding method for compressing and coding a voice signal, and more particularly to a process for obtaining a pitch period in a voice signal coding process.

[0002]

2. Description of the Related Art A technique of compressing and encoding a voice signal at a low bit rate with high efficiency is an important technique for effective use of radio waves and reduction of communication cost in mobile communications such as car telephones and in-house communications. .

CELP is used as a voice coding method capable of performing high quality voice synthesis at a bit rate of 8 kbps or less.
(Code Excited Linear Prediction) method is known. The CELP method was developed by MR Schrodeder and BSAtal in "Code-Excited Linear Prediction (CELP) High-
Quality Speech at Very low Bit Rates ”, Proc.ICASS
Since it was announced in P; 1985, pp.937-939 (reference 1), it has attracted attention as a method capable of synthesizing high-quality speech, and various studies have been made to improve quality and reduce the amount of calculation.

An adaptive codebook is one of the constituent elements for implementing the CELP audio coding. The adaptive codebook performs pitch prediction analysis of an input speech signal by closed loop operation or analysis by synthesis. In general, pitch prediction analysis using an adaptive codebook searches a pitch period in a search range of 20 to 147 samples (128 candidates) to find a pitch period that minimizes distortion with respect to a target signal, and obtains information on this pitch period from 7 It is often transmitted as bit encoded data.

In the above-mentioned conventional CELP method, the pitch period is determined by the closed loop operation in units of subframes, so that the calculation amount becomes enormous when the pitch period search range is as large as 128 candidates as described above. Furthermore, in such a direct pitch period search method, pitch period information requires 7 bits per subframe,
If one frame is composed of four subframes, a bit number of 28 bits is required for each frame.

Originally, there are many gradual changes in the pitch period of a voice signal, and it is not necessary to perform a full search for each subframe. It is possible to reduce the amount of calculation and the number of bits by using such a property of the pitch period. From such a viewpoint, a method using a differential pitch expression that limits the search range of the pitch cycle has been reported.

One of them is to search all the candidates for odd subframes and only the candidates near the odd subframes for even subframes in the pitch period search, thereby reducing the calculation amount and the number of bits. A way to reduce it is by JP Campbell, Jr. et al. “An Expandable Erro
r-Protected 4800 bps CELP coder (USFederal Standa
rd 4800 bps Voice Coder) ”, Proc.ICASSP; 1989, pp.73
5-738 (reference 2). According to this method, a full search of 128 candidates for odd subframes,
For an even number of subframes, for example, if the pitch period is limited to 32 candidates, the pitch period information amount can be reduced to 24 bits per frame.

However, according to this method, when a value in which the pitch period obtained in an odd number of sub-frames is significantly different from the actual pitch period is selected, it affects the next sub-frame and the quality deterioration is perceived. The problem arises.
In fact, if the pitch analysis is performed in a short section such as a subframe, it becomes easy to find a pitch cycle different from the actual one.

Further, a method of stably obtaining the pitch period of a frame, expressing each subframe as a variation amount from the frame pitch period, and searching the pitch period only in the vicinity thereof is described in "Efficient Encoding of
the Long-Term Predictor inVector Excitation Coders
, Boston, MA: Kluwer, 1991, pp329-338 (reference 3). According to this method, the pitch period information of a frame is represented by 7 bits, and the pitch period information of a subframe is 5 bits. It is expressed in bits.

The advantage of this method is that the pitch period search for all subframes only needs to be performed for 32 candidates, and the amount of calculation can be greatly reduced. However,
This method has a problem that a large deterioration occurs when the pitch period varies within a frame. This problem will be described with reference to FIG.

As shown in FIG. 1A, when the actual pitch period fluctuates within a frame, the frame pitch period is obtained as an average period among them. The pitch period of each sub-frame has a search range in the vicinity of the frame pitch period. However, if the search range is not large enough, the actual pitch period cannot be obtained as shown in FIG. Further, if the search range is increased, the number of bits required to express the information of the pitch period is correspondingly increased, and the amount of calculation and the effect of bit reduction are reduced.

[0012]

As described above, in the CELP system, which is a conventional speech coding method, the pitch period is obtained by a closed loop search in subframe units, so that the amount of calculation required to obtain the pitch period is small. There is a problem that the number of bits becomes huge and the number of bits of pitch period information increases.

Further, in the method for obtaining the pitch period by limiting the search range as described in the reference 2, the amount of calculation for obtaining the pitch period and the number of bits of the pitch period information are reduced, but the pitch period is set correctly. There is a problem that it is difficult to ask.

Further, the method of searching the pitch period of the sub-frame as the variation amount from the pitch period of the frame only in the vicinity of the pitch period of the frame as described in Document 3 can significantly reduce the calculation amount. However, there is a problem that the pitch period of the sub-frame cannot be correctly obtained when the pitch period varies within the frame.

The present invention has been made in order to solve the above-mentioned problems of the prior art. The frame period of a voice signal can be correctly obtained with a small amount of calculation, and the pitch period can be expressed with a small amount of information. An object of the present invention is to provide a speech coding method capable of performing the speech coding.

[0016]

In order to solve the above problems, the present invention is a speech coding including a process of dividing an input speech signal into frames of a predetermined length and obtaining a pitch period of the input speech signal. The process is characterized by including a process of obtaining a pitch cycle of a future frame in time with respect to a current frame to be coded, and a process of coding this pitch cycle.

Further, according to the present invention, a voice code including a process of dividing an input voice signal into frames of a predetermined length, further dividing the voice signal of each frame into subframes, and obtaining a pitch period of the voice signal. In the coding process, the pitch period prediction value of the subframe in the current frame is obtained by using the pitch periods of at least two frames of the past frame and the future frame for the current frame to be encoded and the current frame. The pitch period prediction value is used to obtain the pitch period of the subframe in the current frame.

As described above, according to the present invention, in order to obtain the pitch period of the future frame with respect to the present frame, the pitch period of the present frame and the pitch period of the past frame are used to interpolate the subframes in the present frame. Even if the pitch cycle varies within a frame, the pitch cycle prediction value is calculated, and the pitch cycle of the subframe in the current frame is calculated using this pitch cycle prediction value. It can be calculated well and with a small amount of calculation, and can be expressed with a small amount of information.

Explaining with reference to FIG. 1B, as shown in FIG. 1A, the actual pitch period changes within the frame as in FIG. 1A. In the present invention, for example, interpolation is performed with the pitch period of the current frame and the pitch period of the past frame, the subframe pitch period prediction value is obtained for each subframe, and the search range of the pitch period of the subframe is determined in the vicinity thereof. . Therefore, the actual pitch period is included in the search range, and the pitch period can be accurately searched.

Since the predicted value of the subframe pitch period approximates the actual pitch period quite accurately, there is no problem even if the search range of the pitch period of the subframe is narrowed to, for example, 8 candidates. If the search range of the subframe pitch period is 8 candidates, the pitch period of the frame is represented by 7 bits and 3 bits per subframe. Therefore, if 4 subframes are 1 frame, the pitch period of the subframe is In contrast to 28 bits required for each frame, the amount of information can be represented by 7 bits + 3 bits * 4 = 19 bits. Further, since the search range of the subframe pitch period is as small as 8 candidates, it greatly contributes to the reduction of the calculation amount.

In the present invention, the pitch period of the subframe in the current frame obtained as described above may be encoded, or a pitch filter for emphasizing the pitch period component of the input voice signal may be provided. The transfer function of the pitch filter may be determined using the pitch period of the subframe in the current frame obtained as described above. The pitch filter is known as one component of, for example, a perceptual weighting filter or a post filter.

Further, the present invention has an adaptive codebook in which a plurality of adaptive vectors generated by repeating a past drive signal sequence in a cycle included in a predetermined range are stored, and is extracted from this adaptive codebook. In a voice encoding process including a process of searching an adaptive vector having a period in which an error between a signal obtained by passing an adaptive vector through a predetermined filter and a target vector is a predetermined search range, an input voice signal is predetermined. At least two of the current frame to be encoded and the past frame and future frame after the audio signal of each frame is further divided into subframes. The pitch period prediction value of the subframe in the current frame is calculated using this pitch period, and this pitch period prediction value is used to calculate the current frame. And determines the search range for the sub-frame in the arm.

In the present invention, when the pitch period of the frame is obtained, the analysis position of the pitch period may be adaptively determined for each frame. More specifically, the pitch period analysis position is determined by the magnitude of the short-term power of the voice signal, the prediction residual signal, or the prediction residual signal after passing the low pass filter. By doing so, the pitch period can be obtained more accurately, and the quality of the decoded speech can be improved.

Further, a method of obtaining the pitch period of the subframe in the current frame may be selected according to the continuity of the pitch period. For example, when it is determined that the pitch period is continuously changing, the subframe pitch period prediction value is obtained, and the subframe pitch period is obtained by searching this neighborhood. On the contrary, when it is determined that the change of the pitch period is discontinuous, the subframe pitch period is obtained by the full search. By such an adaptive process, the optimum search method for the subframe pitch period is selected according to the continuity of the pitch period, so that the quality of the decoded speech is improved.

Further, the apparatus further comprises a relative pitch pattern codebook in which a plurality of relative pitch patterns representing fluctuations in the pitch cycle of a plurality of subframes are stored, and a change in the pitch cycle of the subframes is predetermined from the relative pitch pattern codebook. It may be represented by one relative pitch pattern selected on the basis of the index, thereby further reducing the number of bits of the information representing the subframe pitch period.

That is, the relative pitch pattern codebook is
For example, a relative pitch pattern having a high appearance frequency is stored as a vector. This is matched with a vectorized pitch cycle of a plurality of sub-frames, and the pitch cycle of a plurality of sub-frames is represented by the most suitable relative pitch pattern. For example, if 3 bits are required for each subframe to individually represent the pitch period of the subframe, 12 bits are required for every 4 subframes. However, this 4D vector requires a size of 7 bits in the relative pitch pattern codebook. If expressed with one relative pitch pattern having a certain length, it leads to a bit reduction of 5 bits per frame.

Further, according to the present invention, the input voice signal is divided into frames of a predetermined length, and a computer readable recording a program for performing a voice encoding process including a process for obtaining a pitch period of the input voice signal. It is a possible recording medium, and records a program for executing a process of obtaining a pitch period of a temporally future frame with respect to a current frame to be encoded and a process of encoding this pitch period. A recording medium is provided.

Further, according to the present invention, the input voice signal is divided into frames of a predetermined length, the voice signal of each frame is further divided into subframes, and a voice including a process of obtaining a pitch period of the voice signal is included. A computer-readable recording medium in which a program for performing an encoding process is recorded, and a current frame to be encoded and a pitch of at least two frames of a past frame and a future frame with respect to the current frame. Calculate the pitch period prediction value of the subframe in the current frame using the period,
There is provided a recording medium recording a program for executing a process of obtaining a pitch period of a subframe in a current frame using the pitch period prediction value.

Furthermore, the present invention has an adaptive codebook that stores a plurality of adaptive vectors generated by repeating the past drive signal sequence in a cycle included in a predetermined range, and extracts the adaptive codebook from this adaptive codebook. A program for performing a voice coding process including a process for searching an adaptive vector of a period in which the error between the target vector and the signal obtained by passing the adaptive vector through a predetermined filter from a predetermined search range is recorded. A computer-readable recording medium, in which an input audio signal is divided into frames of a predetermined length, the audio signal of each frame is further divided into subframes, and then the current frame and the current frame to be encoded For the current frame using the pitch period of at least two of the past and future frames Determine the pitch period prediction value of the sub-frame, a recording medium recording a program for executing processing for determining the search range for the sub-frame of the current frame by using the pitch period prediction value is provided.

[0030]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of a speech coding method according to the present invention will be described below.

(First Embodiment) FIG. 2 shows a first embodiment of the present invention.
6 is a block diagram for explaining a basic operation of pitch period analysis in the speech coding method according to the embodiment of FIG. In the figure, digitized audio signals (hereinafter referred to as input audio signals) are sequentially input to an input terminal 11. This input audio signal is divided by the frame configuration unit 12 into units called frames each having a predetermined length (frame formation). The pitch cycle analysis section 13 analyzes the pitch cycle of the voice signal framed by the frame configuration section 12, and the pitch cycle information is output from the output terminal 14.

Next, the operation of the frame forming section 12 will be described with reference to FIG. For convenience, the operation of the subframe configuration unit in the frame / subframe configuration unit 22 (FIG. 4) in the second embodiment described later will also be described here.

FIG. 3 shows a frame structure and a subframe structure of an input voice signal in this embodiment. In Figure 3,
The symbol f (m) is added to identify the frame. Here, a frame of m = 0 represents a frame to be encoded (hereinafter referred to as a current frame), and m ≦ −1.
Frame represents a past frame that has already been encoded. Further, fr represents a frame that is temporally future with respect to the current frame (this is referred to as a prefetch section).

Letting s (n) be the input voice signal and s (0) being the voice signal located at the beginning of the current frame, the following is given between s (n) and f (m) and fr: Relationship is established.

F (m) = {s (n); n = m * NF to (m + 1) * NF-1} (1) (m ≦ 0) fr = {s (n); n = NF to NF + ND-1} (2) Here, NF represents the frame length, and ND represents the length of the prefetch section. In this embodiment, NF = 160 and ND = 120 will be described.

Also, the subframe of the frame f (0) is s
If f (k), the following relationship holds between s (m) and sf (k). sf (k) = {s (n); n = k * NSF to (k + 1) * NSF-1} (0≤k≤K-1) (3) where NSF is the subframe length and K is Each represents the number of subframes. In this embodiment, NSF = 40, K = 4
Will be described as. Further, n is a variable representing a sample, m
Is a variable that represents a frame, and k is a variable that represents a subframe.

After the frame and sub-frames are constructed in this way, the pitch cycle analysis unit 13 performs pitch cycle analysis. The pitch period analysis unit 13 is characterized in that the pitch period is analyzed with the analysis centered on the look-ahead unit fr. The definition of the center of analysis will be described later.

Existing techniques can be applied to the analysis of the pitch period. For example, some pitch analysis methods are introduced in "Digital signal processing of voice", published by Corona (Reference 4). In this embodiment, the following pitch period analysis method is used.

First, in an LPC analysis section (not shown),
The input speech signal s (n) is analyzed to find the LPC coefficient, and the LPC coefficient is used to construct a prediction filter of the transfer function A (z) represented by the following equation.

[0040]

[Equation 1]

Here, α (i) represents the LPC coefficient, and IP represents the analysis order. In this embodiment, the description will proceed with IP = 10. The prediction residual signal e (n) is obtained according to the following expression by passing the speech signal s (n) through the prediction filter of this transfer function A (z).

[0042]

[Equation 2]

Next, a Hamming window is applied to the prediction residual signal e (n) to obtain a windowing signal u (n), and this windowing signal u (n) is obtained.
Then, the pitch period is extracted using the correlation value ρ (t) shown in the following equation.

[0044]

[Equation 3]

Here, NW represents the pitch period analysis length, and NC represents the pitch period analysis position. In this embodiment,
The description will be made assuming that NW = 160 and NC = 200. Also,
The analysis range of the correlation value ρ (t) is {20 ≦ t ≦ 147}.

Then, t when the correlation value ρ (t) becomes maximum is output terminal 14 as the information of the pitch period T of the frame.
Is output from. The information on the pitch period T is encoded, for example, and transmitted to a decoder (not shown).

When a symmetrical window such as a Hamming window or a rectangular window is used as the window function, the pitch analysis position can be considered as the center of the window function. Also, when an asymmetric window having different left and right window shapes is used, the analysis position cannot be considered to be the central portion like the symmetrical window. In this case, in the present embodiment, the position where the amplitude value of the window function is maximum is regarded as the center.

As described above, in the present embodiment, the pitch period of the future frame is calculated with respect to the present frame. Therefore, this is used together with the pitch period of the current frame and the pitch period of the past frame to interpolate, as will be described later. Obtaining the pitch period prediction value of the subframe in the current frame, by obtaining the pitch period of the subframe in the current frame using this pitch period prediction value, even if the pitch period varies in the frame, It is possible to obtain the pitch period of a subframe with high accuracy and a small amount of calculation, and to represent it with a small amount of information.

(Second Embodiment) FIG. 4 shows a second embodiment of the present invention.
6 is a block diagram for explaining a basic operation of subframe pitch extraction in the speech encoding method according to the embodiment of FIG. In the figure, the digitized input audio signal to the input terminal 21 is the frame / subframe configuration section 22.
Are divided into frames, and each frame is further divided into subframes. This frame / subframe construction part 2
The speech signal framed in 2 and the speech signal subframed are input to the pitch cycle analysis unit 23 and the subframe pitch cycle extraction unit 24. The pitch cycle analysis unit 23 analyzes the pitch cycle T, and the information of the pitch cycle T is input to the subframe pitch cycle extraction unit 24.

In the subframe pitch period extraction section 24,
Based on the pitch cycle T calculated by the pitch cycle analysis unit 23, the pitch cycle ST (k) for each subframe is calculated,
Information on the subframe pitch period ST (k) is output from the output terminal 25.

Next, the subframe pitch period extraction section 24, which is a characteristic part of this embodiment, will be described with reference to FIG. FIG. 5 is a block diagram showing the configuration of the subframe pitch period extraction unit 24. Using the information of the pitch period T from the pitch period analysis unit 23 of FIG. 4 input to the input terminal 101, the subframe pitch period prediction is performed. Value calculation unit 102
Then, a subframe pitch cycle prediction value is calculated for each subframe. The subframe pitch period calculation unit 104 receives the subframe pitch period prediction value calculated by the subframe pitch period prediction value calculation unit 102 from the frame / subframe configuration unit 22 of FIG. 4 which is input to the input terminal 103. The subframe pitch period ST (k) is calculated for each subframe using the audio signal, and the information of the subframe pitch period ST (k) is output to the output terminal 105.

The method of calculating the subframe pitch period will be described in more detail with reference to FIG. First, the pitch cycle analysis unit 23 sets the pitch cycle found in the current frame f (0) to T (0), and the pitch cycle found in the past frame f (-1) to T (-1). Subframe pitch period extraction unit 24
Then, a subframe pitch period prediction value STP (k) is obtained for each subframe using these two pitch periods. In the present embodiment, a method of obtaining a subframe pitch period prediction value STP (k) from T (-1) and T (0) by interpolation will be described. Due to the nature of the audio signal, the original subframe pitch period ST (k) can be considered to be in the vicinity of the subframe pitch period predicted value STP (k). The reason is that the pitch period of the audio signal has a small temporal variation.

Next, the subframe pitch period prediction value ST
A method of obtaining the subframe pitch period ST (k) using P (k) will be described with reference to FIG. As described above, since it is assumed that the subframe pitch period ST (k) is in the vicinity of the subframe pitch period prediction value STP (k), only the neighborhood can be considered as the range for pitch period extraction. . In FIG. 7, the 0th sub-frame sf
Subframe pitch period prediction value STP (0) at (0)
And the correlation value calculation range NR for extracting the subframe pitch period ST (0).

As described above, in this embodiment, the range of ± NR / 2 around the subframe pitch period prediction value STP (0) is set as the correlation value calculation range for obtaining the subframe pitch period ST (0). . That is, STP (0)-
Only for the period included in the range of NR / 2 ≦ t ≦ STP (0) + NR / 2−1, the correlation value ρ (t) defined by the equation (6) shown above is calculated, and this is the maximum value. S when taking
Information on the relative pitch period ΔT (0) with respect to TP (0) is output from the output terminal 25. Therefore, subframe sf (0)
The subframe pitch period ST (0) of can be obtained as the sum STP (0) + ΔT (0) of the subframe pitch period predicted value STP (0) and the relative pitch period ΔT (0).
Similarly, the relative pitch period ΔT (k) of the other subframe sf (k) is obtained, and this information is output from the output terminal 25. In this embodiment, NR = 8.

According to this embodiment, the following effects can be obtained. First, since the range of correlation value calculation for obtaining the subframe pitch period can be set to NR, a significant reduction in the amount of calculation can be realized. If the sub-frame pitch period is calculated by the equation (6) for each sub-frame, about 40,000 product-sum operations are required to calculate all the pitch period candidates (128 candidates). On the other hand, when the present embodiment is applied, the correlation value calculation range is N
If R = 3, the sum of products operation will be about 4,000 times, and about 9
The amount of calculation can be reduced by 0%. However, this is the result of calculation under the pitch period analysis length NW = 160.

In addition, conventionally, the amount of information required to express the subframe pitch period is such that, if one subframe is composed of four subframes when the entire range of 128 candidates is calculated for each subframe. 7 bits * 4 = 28 bits were needed. On the other hand, in the present embodiment, the frame pitch period T (0) is represented by 7 bits, and the relative pitch period ΔT (k) is represented by 3 bits per subframe. Therefore, the subframe pitch period is represented. The amount of information required for each frame is 7 bits + 3
Bit * 4 = 19 bits, which leads to a reduction in the number of bits by about 35%. The pitch period T (-
Since 1) can be substituted with the pitch period T (0) in the processing one frame before, it is not necessary to newly represent this with a bit.

Next, the processing procedure for calculating the sub-frame pitch period in this embodiment will be described with reference to the flowchart of FIG. First, in step S11, the input audio signal is framed and subframed. Next, step S12
With the center of analysis in the current frame, pitch period T (0)
To calculate. Next, in step S13, the subframe pitch period prediction value STP (k) for each subframe is obtained using the pitch period T (0) of the current frame and the pitch period T (-1) of the past frame. Then, after setting the counter k to 0 in step S14, the relative pitch period ΔT (k) of the sub-frame sf (k) is calculated in step S15, and the counter k is incremented by 1 in step S16. In step S17, the counter k is K (K is the number of subframes)
It is determined whether or not If k does not match K, the process returns to step S15 to continue the process. If they match, the process ends.

(Third Embodiment) A third embodiment of the present invention will be described with reference to FIG. The difference between the present embodiment and the second embodiment is that the subframe pitch period prediction value STP (k)
The point is that the pitch period Tr of the prefetching unit fr is used when obtaining

The center of the sub-frame is the past frame f (-
1) between the analysis position of the pitch period T (-1) and the analysis position of the pitch period T (0) of the current frame f (0), the subframe pitch period prediction value STP (k) of the subframe
Is calculated using T (-1) and T (0). Similarly, the center of the subframe is the pitch period T (0) of the current frame f (0).
, And the analysis position of the pitch period Tr of the prefetching unit fr, the subframe pitch period prediction value STP (k) of the subframe is obtained using T (0) and Tr. In the present embodiment, the subframe pitch period prediction value STP (k) is calculated using linear interpolation.

After the subframe pitch period predicted value STP (k) is obtained in this way, the relative pitch period ΔT (k) is obtained for each subframe in the same manner as in the second embodiment.

The information necessary to represent the sub-frame pitch period ST (k) in this embodiment is the pitch period of the look-ahead part fr obtained by processing the pitch period T (0) of the current frame one frame before. Since Tr can be used as a substitute, only the pitch period Tr of the look-ahead unit fr and the relative pitch period ΔT (k) for each subframe are sufficient.

Next, the processing procedure in this embodiment will be described with reference to the flow chart shown in FIG. The processing of steps S21, S24, S25, S26 and S27 of FIG. 10 is performed by steps S11, S14, S15 and S1 of FIG.
6 and S17, which are the same as those in S6 and S17, respectively, and the description thereof is omitted here. The characteristic part of the processing procedure of FIG. 10 is that the pitch cycle Tr is calculated with the analysis centered on the look-ahead section fr in step S22, and the sub-frame pitch cycle predicted value STP (k) is read in the step S23. The pitch period Tr of fr and the pitch period T (0) of the current frame f (0) are used to determine the pitch period T (0) of the current frame f (0) and the past frame f (-1). One of the points is to use one of the pitch periods T (-1).

In this embodiment, any subframe pitch cycle prediction value STP (k) has a pitch cycle T (-1) of the past frame f (-1) and a pitch cycle T (0 of the current frame f (0). )
, Or T (0) and an interpolated value of the pitch period Tr of the prefetching unit fr, a more accurate predicted value can be obtained.

For example, when the actual pitch period fluctuates as shown in FIG. 11, the pitch period T (-1) of the past frame and the pitch period T (0 of the present frame are changed as in the second embodiment. ) Subframe pitch period prediction value STP
The method of obtaining (k) may cause a problem that the reliability of the subframe pitch period prediction value STP (k) is lowered. On the other hand, in the present embodiment, the prefetch unit f
Since the pitch period Tr of r accurately represents the actual pitch period, the interpolation value thereof can also accurately represent the actual pitch period.

(Fourth Embodiment) A fourth embodiment of the present invention will be described with reference to FIG. The difference between the present embodiment and the third embodiment is that the subframe pitch period prediction value STP is
The point is that the pitch period Tr of the prefetching unit fr and the pitch period T (0) of the current frame f (0) are used when obtaining (k).

The processing procedure of this embodiment will be described with reference to the flowchart shown in FIG. The processing of steps S31 and S34 to S37 in FIG. 13 is the same as the processing of steps S11 and S14 to S17 in FIG.

In step S32, the pitch period Tr is calculated with the analysis centered on the look-ahead part fr. next,
In step S33, the subframe pitch cycle prediction value STP (k) is calculated using the pitch cycle Tr of the prefetching unit fr and the pitch cycle T (0) of the current frame f (0).

In the present embodiment, the subframe pitch period prediction value STP (k) can be obtained by interpolation from the pitch period Tr of the prefetching unit fr and the pitch period T (0) of the current frame f (0). Compared with the third embodiment, there is an advantage that the length of the prefetching unit fr can be shortened and therefore the delay can be reduced.

(Fifth Embodiment) A fifth embodiment of the present invention will be described with reference to FIG. The difference between the present embodiment and the fourth embodiment is that the subframe pitch period prediction value STP
The point is that the pitch period Tr of the prefetching unit fr and the pitch period T (-1) of the past frame f (-1) are used when obtaining (k).

According to this embodiment, the following effects can be obtained. Accurate analysis of pitch period is generally difficult, and there is no established method even now. In the present embodiment, the method based on the correlation analysis represented by the equation (6) is described, but in this case as well, the correct pitch period cannot always be obtained, and the pitch period which is an integer fraction of the actual pitch period is There is a case where the pitch period is obtained or, conversely, an integral multiple pitch period is obtained.

As shown in FIG. 14, the current frame f
When the analysis of the pitch period T (0) of (0) fails and is largely deviated from the actual pitch period, the pitch period Tr of the look-ahead part fr and the pitch period T (-of the past frame f (-1) are
Interpolation with 1) and subframe pitch period prediction value ST
By determining P (k), the pitch period can be efficiently expressed as in the case where the pitch period T (0) of the current frame f (0) is accurately determined.

The processing procedure of this embodiment will be described with reference to the flowchart shown in FIG. The processes of steps S41, S42, S45, S47, S48, S49 and S50 in FIG. 15 are the same as steps S21, S in FIG.
Since it is the same as the processing of S22, S23, S24, S25, S26 and S27, the description thereof is omitted here. The characteristic part of the present embodiment is the processing of steps S43, S44, and S46, so these processings will be described.

In step S43 and S44, the average value of the pitch cycle Tr obtained by the look-ahead section and the pitch cycle T (-1) of the past frame is calculated in step S43, and the average value and the pitch cycle T () of the current frame are calculated. And 0) are compared according to the following equation (7) (step S44).

[0074]

[Equation 4]

If the expression (7) is satisfied, the current frame f
Assuming that the pitch period T (0) of (0) is reliable, the pitch period T (0) of the current frame f (0) is calculated in order to calculate the subframe pitch period prediction value STP (k) as shown in step S45. To use. On the contrary, when the expression (7) is not established, as shown in step S46, only the pitch period Tr of the look-ahead part fr and the pitch period T (-1) of the past frame f (-1) are used to determine the sub-frame pitch. The cycle prediction value STP (k) is calculated.

(Sixth Embodiment) The sixth embodiment of the present invention will be described with reference to FIG. 16, the input terminal 201, the frame / subframe configuration unit 202, and the pitch period analysis unit 203 are denoted by reference numerals 101 to 101 in FIG.
Since it is the same as the element denoted by 103, its explanation is omitted here.

A feature of this embodiment is that the present invention is applied to a speech coding method using an adaptive codebook. The adaptive codebook refers to a codebook that stores a plurality of adaptive vectors generated by repeating a past drive signal sequence in a cycle included in a predetermined range.

Subframe pitch period prediction value calculation unit 20
In 4, the subframe pitch cycle prediction value STP (k) is calculated for each subframe based on the pitch cycle obtained by the pitch cycle analysis unit 203. Any of the methods described with reference to FIGS. 6, 9, 12, and 14 can be applied to the method of calculating the subframe pitch period prediction value STP (k). Moreover, the effect by it is as above-mentioned. However, in the method described in FIG. 6, the center of the pitch period analysis in the pitch period analysis unit 203 is on the current frame f (0), and
In the methods described with reference to FIGS. 12 and 14, there is a difference that the center of the pitch period analysis is on the prefetching unit fr. In this embodiment, an example based on the method described in FIG. 6 will be described.

Subframe pitch period prediction value calculation unit 20
4, the pitch period T obtained by the pitch period analysis unit 203
A sub-frame pitch period predicted value STP (k) is obtained by interpolation from (0) and the pitch period T (-1) of the f (-1) frame.

Search range determination section 205 determines the search range for obtaining subframe pitch period ST (0), that is, the search range of the adaptive vector for adaptive codebook 206. Specifically, the subframe pitch cycle prediction value STP
The search range is ± NR / 2 centered on (0). That is, STP (0) -NR / 2≤t≤STP (0) + NR / 2
Only the cycle included in the range of -1 is searched. The search is performed by the following method.

The adaptive vector q corresponding to the pitch period t included in the search range determined by the search range determining unit 205.
(t, n) is taken out from the adaptive codebook 206 and is passed through the perceptual weighting synthesis filter 207 to generate a synthesis signal p (t, n). Then, the contribution degree cntb (t) between this synthesized signal p (t, n) and the target signal r (n) obtained by passing the input voice signal through the perceptual weighting filter 208 is passed through the subtractor 209 to the distortion calculation unit. At 210, it is calculated according to the following equation.

[0082]

[Equation 5]

Of all the candidates of pitch period t included in the search range, the pitch period t when the contribution cntb (t) becomes maximum is set as the subframe pitch period ST (0), and the subframe pitch period ST ( The value obtained by subtracting the subframe pitch cycle prediction value STP (0) from (0) is the relative pitch cycle ΔT.
(0) The process of obtaining this relative pitch period ΔT (0) is performed for all subframes sf (k) to obtain the relative pitch period ΔT (k) of all subframes.

Finally, the information of the pitch period T (0) obtained by the pitch period analysis unit 203 is output from the output terminal 212,
The contribution degree cntb from the output terminal 211 in the distortion calculation unit 210
Information on the relative pitch period ΔT (k) when (t) becomes maximum is output.

Next, the processing procedure of this embodiment will be described with reference to FIGS.
This will be described using the flowchart shown in FIG. 17 to 1
8 in steps S101, S102, S103, S
Since the processing of 104, S114 and S115 is the same as the processing of steps S11, S12, S13, S14, S16 and S17 in FIG. 8, description thereof will be omitted here. The processing unique to this embodiment will be described below.

In step S105, the subframe sf
Determine the search range of (k). The search range is determined using the subframe pitch period prediction value STP (k) as described above. In step S106, initial values of the variable cntbmax and the subframe pitch period ST (k) are given. c
ntbmax is given a sufficiently large value that can be an initial value, and ST (k) is given a minimum pitch period. Step S
In step 107, the adaptive vector q (t, n) corresponding to the cycle t included in the search range is selected from the adaptive codebook, and step S
At 108, the adaptive vector q (t, n) is passed through a perceptual weighting synthesis filter to generate a synthesis signal p (t, n). Step S
In step 109, the contribution cntb (t) between the target signal r (n) and the combined signal p (t, n) which has been obtained in advance is calculated. In step S110, the contribution cntb (t) in the pitch cycle t is calculated.
And cntbmax are compared.

Here, if cntb (t)> cntbmax holds, then in step S111, cntbma
cntb (t) is given to x, t is given to ST (k), and the process proceeds to step S112. If cntb (t)> cntbmax is not satisfied, the process proceeds to step S112.

In step S112, it is determined whether or not all the candidates of the pitch cycle t included in the search range have been searched,
If so, the process proceeds to step S113, and if not, the process returns to step S107 to continue the process using the pitch cycle t which has not been searched yet. Step S113
Then, the relative pitch period ΔT (k) of the subframe sf (k)
Is calculated using the subframe pitch period ST (k) and the subframe pitch period prediction value STP (k), and step S1
Proceed to 14.

(Seventh Embodiment) The seventh embodiment of the present invention will be described with reference to the flowcharts of FIGS.
The difference between the present embodiment and the sixth embodiment is that the method described in FIG. 9 is applied to the calculation of the subframe pitch period prediction value.

According to the present embodiment, as described above, the accuracy of the subframe pitch period prediction value STP (k) with respect to the actual pitch period is improved, so that the subframe sf
It is possible to reduce the search range of the subframe pitch period ST (k) of (k), and consequently reduce the number of bits and the amount of calculation required to express the relative pitch period ΔT (k). is there. 19 to 20, the processes of steps S201 and S204 to S215 are the same as steps S101 and S10 of FIGS.
Since it is the same as the processing of 4 to S115, the description thereof is omitted here.

The processing characteristic of this embodiment is step S
202 and S203. In step S202, the pitch period is analyzed with the center of the pitch period analysis placed in the look-ahead part. In step S203, the calculation of the subframe pitch period prediction value STP (k) is determined by using the pitch period Tr of the prefetching unit fr and the pitch period T (0) of the frame f (0), or the frame f (0). Pitch period of T (0)
And the pitch period T (-1) of the frame f (-1) is used.

(Eighth Embodiment) The eighth embodiment will be described with reference to the flowcharts of FIGS. The difference between the present embodiment and the seventh embodiment is that the method described in FIG. 12 is applied to the subframe pitch period prediction value calculation unit.

According to this embodiment, the size of the read-ahead portion can be reduced as described above, so that the delay can be shortened. 21 to 22, the processes of steps S301, S302 and S304 to S315 are the same as steps S201 and S2 of FIGS.
02 and S204 to S215, the description is omitted here.

The processing characteristic of this embodiment is step S3.
In step S303, the sub-frame pitch cycle predicted value STP (k) is calculated using the pitch cycle Tr of the prefetching unit fr and the pitch cycle T (0) of the frame f (0).

(Ninth Embodiment) The ninth embodiment of the present invention will be described with reference to the flowcharts of FIGS.
The difference between this embodiment and the eighth embodiment is that the method described with reference to FIG. 14 is applied to the subframe pitch period prediction value calculation unit.

According to the method of this embodiment, even if the analysis of the pitch period T (0) in the frame f (0) fails as described above and the pitch period T (0) largely deviates from the actual pitch period, the prefetch section fr is used. Pitch cycle Tr and frame f calculated in
By interpolating with the pitch period T (-1) of (-1), the subframe pitch period prediction value STP (k) can be accurately obtained, and the pitch period T (0) of the frame f (0) is The pitch period can be efficiently expressed as in the case where the pitch period is accurately obtained.

23 to 24, step S40
2, the processing of S402 and S407 to S418 is as shown in FIG.
1 to 22 steps S301, S302 and S
The processing is the same as the processing of 304 to S315, and also shown in FIGS.
The processing of steps S403 to S406 in FIG.
Since it is the same as the processing of steps S43 to S46 in 5, the description thereof will be omitted here.

(Tenth Embodiment) The tenth embodiment of the present invention will be described with reference to FIG. In FIG. 25, an input terminal 301, a frame / subframe configuration unit 302, a pitch period analysis unit 303, a subframe pitch period calculation unit 304, a search range determination unit 305, an adaptive codebook 306, a perceptual weighting synthesis filter 307, and a perceptual weight filter. Thirty
8, subtractor 309, distortion calculator 310, output terminal 311,
The reference numeral 312 is the same as the elements denoted by reference numerals 201 to 212 in FIG. 16, and thus the description thereof is omitted here. The feature of this embodiment lies in that the pitch period analysis position is determined by the pitch period analysis position determination unit 313 in the pitch period analysis unit 303.

The effect of this embodiment will be described with reference to FIG. When the characteristics of the input speech signal change within the frame as shown in the figure, an accurate pitch cycle may not be obtained if the pitch cycle analysis position is fixed. In the case of FIG. 26, the fact that the signal to be pitch-analyzed when the pitch-period analysis position is fixed does not include a periodic component is the reason why an accurate pitch period cannot be obtained. However, if the pitch analysis position can be made variable, the pitch period analysis position can be set at a position where a waveform having a pitch period component exists as shown in FIG. 26, so that an accurate pitch period can be obtained. . However, in this case, additional information is required to represent the pitch period analysis position.

In FIG. 25, the pitch period analysis unit 313
Determines the pitch period analysis position. In the present embodiment, the short-term power of the input audio signal or the power of the prediction residual signal after passing the low-pass filter is used as an index,
A method of setting a pitch period analysis position in a portion where the power becomes large is used. The information of the pitch cycle analysis position is sent to the pitch cycle analysis unit 303 and simultaneously output from the output terminal 314. The pitch period analysis unit 303 analyzes the pitch period according to the information of the pitch period analysis position sent from the pitch period analysis position determination unit 313.

(Eleventh Embodiment) The eleventh embodiment of the present invention will be described with reference to FIG. In FIG. 27, an input terminal 401, a frame / subframe configuration unit 402,
Pitch period analysis unit 403, subframe pitch period calculation unit 404, search range determination unit 405, adaptive codebook 406,
Perceptual weighting synthesis filter 407, perceptual weighting filter 4
08, subtractor 409, distortion calculator 410, output terminal 41
Since 1 and 412 are the same as the elements denoted by reference numerals 201 to 212 in FIG. 16, description thereof will be omitted here.

The feature of this embodiment is that the pitch period analysis unit 4
The continuity determination unit 413 determines whether or not the change in pitch period obtained in 03 is continuous, and determines the search range of the adaptive codebook 406 according to the determination result. That is, in the present embodiment, when it is determined that the pitch period changes continuously, the subframe pitch period predicted value is obtained as described above, and the search range determination unit 405 determines the adaptive codebook 406 based on the value. Determine the search range. If it is determined that the adaptive codebook 406 is not continuous.
The search is performed for all the candidates in.

This embodiment has the following effects.
When the pitch period calculated by the pitch period analysis unit 403 is continuously changing, the method of interpolating the pitch period and searching only the periphery thereof as described above works effectively. However, the pitch period of the voice signal does not always change stably, the pitch period may change greatly, and the pitch component may not exist like the unvoiced part. In such a case, if the search range is limited by forcibly interpolating based on the assumption that the pitch period is continuously changing, serious quality deterioration will be caused. When the pitch period is not continuous in this way, in the present embodiment, the above problem can be avoided by switching to search for all candidates in the adaptive codebook 406.

The same effect can be obtained by switching whether or not to search all candidates according to the strength of periodicity at the pitch period T obtained by the pitch period analysis unit 403.
That is, when the variable representing the strength of the pitch periodicity, which is represented by ρ (T) in the equation (6), is equal to or less than a predetermined threshold value, the full search unit 414 searches all candidates. By switching the switch 415 to, it is possible to avoid deterioration in the region where the pitch period does not exist.

The continuity judging section 413 judges the continuity of the pitch cycle using the pitch cycle obtained by the pitch cycle analyzing section 403. In the present embodiment, the case where the pitch period T (0) obtained in the frame f (0) and the pitch period T (-1) obtained in the frame f (-1) are used will be described.
pitch period Tr of r and pitch period T of frame f (0)
When (0) is used, or when the continuity of the pitch period Tr of the prefetch unit fr and the pitch period T (-1) of the frame f (-1) is used, the pitch period Tr of the prefetch unit fr and the frame f
It is also applicable when using the continuity of the pitch period T (0) of (0) and the pitch period T (-1) of the frame f (-1). The continuity of the pitch period is determined according to the following equation (9).

[0106]

[Equation 6]

When the equation (9) is satisfied, it is determined that the pitch cycle is continuously changing, and the switch 415 selects the output of the search range determining unit 405 to perform the limited search as described above. Adaptive codebook 406 only for range
Search for. When the expression (9) is not satisfied, it is determined that the pitch cycle does not change continuously, and the switch 415
Selects all search section 414 and searches for all candidates in adaptive codebook 406. The determination result of the continuity determination unit 414 is output from the output terminal 416. Furthermore, if the output from the continuity determination unit 414 is such that the pitch cycle does not change continuously, it is not necessary to output the pitch cycle obtained by the pitch cycle analysis unit 403 from the output terminal 412. .

(Twelfth Embodiment) The twelfth embodiment of the present invention will be described with reference to FIG. In FIG. 28, an input terminal 501, a frame / subframe configuration unit 502,
The pitch cycle analysis unit 503, the subframe pitch cycle calculation unit 504, the adaptive codebook 506, the perceptual weighting synthesis filter 507, the perceptual weighting filter 508, the subtractor 509, the distortion calculation unit 510, and the output terminals 511 and 512 are shown in FIG. Since they are the same as the elements denoted by reference numerals 201 to 204 and 206 to 212, the description thereof will be omitted here.

The feature of this embodiment is that the relative pitch pattern codebook 505 is composed of a plurality of sets of patterns of relative pitch periods (referred to as relative pitch patterns) that represent variations in subframe pitch periods of a plurality of subframes. That is, the relative pitch periods of a plurality of subframes are regarded as a vector, and the relative pitch pattern in the relative pitch pattern codebook 505 that is most similar to this vector is selected.

According to this embodiment, the following effects can be obtained. If the relative pitch period ΔT (k) is scalar-quantized for each subframe, for example, the search range NR = 8 as described above.
, Then 3 bits are required to represent it.
That is, to represent the information of the relative pitch period ΔT (k),
Only 3 bits x 4 subframes = 12 bits are required per frame.

On the other hand, in the present embodiment, the relative pitch pattern codebook 505 having the pattern of the relative pitch period having a high appearance frequency as the relative pitch pattern is provided.
Here, it is assumed that four subframes (sf (0) to sf (3)
) Relative pitch period ΔT (0) to ΔT (3) represents one vector (4 dimensions), and if the relative pitch pattern codebook 505 is composed of 128 (7 bits) types of relative pitch patterns, there are originally four 12 bits were required to represent the relative pitch period ΔT (k) of the subframe, but 7
Since it is expressed in bits, there is an advantage that a significant bit reduction is achieved.

The relative pitch pattern codebook 505 has J types of typical relative pitch patterns pv (j) which are obtained in advance as shown in the following equation. pv (j) = {v (j, k); k = 0 to K-1} (10) (0≤j≤J-1) In the adder 513, the subframe pitch period prediction is performed as shown in the following equation. The value STP (k) and the j-th relative pitch pattern pv (j) are added to obtain a set Tx (j) of pitch periods of K subframes. Tx (j) = STP (k) + v (j, k) (11) Based on the set of pitch periods Tx (j) thus obtained, when the contribution over K subframes is maximized The relative pitch pattern of is searched in a closed loop. At this time, the internal state of adaptive codebook 506 is updated for each subframe. The number of subframes at this time is not limited to K, and the relative pitch period can be obtained for the number of subframes of 2 or more and K or less. Moreover, in order to reduce the amount of calculation, first, the candidates of the relative pitch pattern are limited by a simple index with a small amount of calculation (for example, the numerator term of the contribution), and the accurate contribution is calculated for the remaining candidates. Finally, one relative pitch pattern may be determined.

(Thirteenth Embodiment) The thirteenth embodiment of the present invention will be described with reference to FIG. In FIG. 29, an input terminal 601, a frame / subframe configuration unit 602,
Pitch cycle analysis section 603, subframe pitch cycle calculation section 604, relative pitch pattern codebook 605, adaptive codebook 606, perceptual weighting synthesis filter 607, perceptual weighting filter 608, subtracter 609, distortion calculation section 610, output terminal 611. , 612 and the adder 613 are the same as the elements denoted by reference numerals 501 to 513 in FIG. 20, and therefore the description thereof is omitted here.

The feature of this embodiment is that the relative pitch pattern is determined in consideration of the influence of the noise codebook 614 together with the adaptive codebook 606. With such a configuration, it becomes possible to determine a more accurate relative pitch pattern.

At this time, the gain by which the adaptive vector is multiplied by the multiplier 615 is given from the terminal 616. This gain is the ideal gain gopt expressed by the following equation, or
Here, the element closest to the ideal gain gopt in the adaptive vector gain codebook (not shown) is used.

[0116]

[Equation 7]

Where r (n) is the target vector and p (t, n)
Represents a signal obtained by passing the adaptive vector in the pitch period t through the perceptual weighting synthesis filter 607.

Similarly, the gain by which the noise vector is multiplied by the multiplier 617 is given from the terminal 618. As this gain, an ideal gain bopt represented by the following expression or an element closest to the ideal gain bopt in a noise vector gain codebook not shown here is used.

[0119]

[Equation 8]

Where t (n) is the target vector and e (i, n)
Represents a signal obtained by passing the noise vector at index i through the perceptually weighted synthesis filter 607.

When the relative pitch pattern when the degree of contribution over a plurality of subframes is maximum is determined,
The index of the relative pitch pattern at that time is output from the output terminal 611. Further, the output terminal 612 outputs the pitch cycle information obtained by the pitch cycle analysis unit 603.

At this time, the internal state of adaptive codebook 606 is updated for each subframe. The number of subframes at this time is not limited to K, and the relative pitch period can be obtained for the number of subframes of 2 or more and K or less. Also, in order to reduce the amount of calculation, first, the relative pitch pattern candidates are limited by a simple index with a small amount of calculation (for example, the numerator term of the contribution degree), and the accurate contribution degree is calculated for the remaining candidates. Finally, one relative pitch pattern may be determined.

(Fourteenth Embodiment) The fourteenth embodiment of the present invention will be described with reference to FIG. In this embodiment, the CE
This is an example in which the present invention is applied to an LP encoder (speech encoder). Further, although the present embodiment has one representative configuration in which some of the embodiments described above are combined, the present invention is not limited to this.
Various combinations can be applied to the CELP method.

In FIG. 30, a digitized audio signal is input from an input terminal 501, and a frame / subframe forming section 502 forms a frame and a subframe. After that, the audio signal divided into the frames and subframes is given to the LPC coefficient analysis unit 521,
The LPC coefficient is calculated by the LPC analysis. This LP
The C coefficient is used when determining the transfer functions of the perceptual weighting filter 508 and the perceptual weighting synthesis filter 507.

LP determined by the LPC coefficient analysis unit 521
The C coefficient is quantized in the LPC coefficient quantization unit 522, and the index obtained by the quantization is given to the multiplexer 523, and is multiplexed with other information described later. The LPC coefficient decoded after quantization is used when determining the transfer function of the perceptual weighting synthesis filter 507.

The input voice signal is also given to the pitch period analysis position determining unit 525. Pitch cycle analysis position determination unit 5
In 25, a prescribed number of powers of a predetermined short section of the audio signal are obtained, and an index representing a region in which the short section power is the largest is given to the pitch cycle analysis unit 503 and the multiplexer 523. Pitch cycle analysis unit 503
Then, the pitch period is analyzed using the voice signal, the prediction residual signal, the prediction residual signal after passing through the low pass filter, and the like, centering on the analysis position determined by the pitch period analysis position determining unit 525.

The continuity of the pitch cycle obtained here and the pitch cycle obtained in the previous frame processing is judged by the continuity judging section 526, and the judgment result is given to the multiplexer 523. If the determination result is continuous, the pitch period obtained by the pitch period analysis unit 503 is set to the multiplexer 52.
3, the subframe pitch period prediction value calculation unit 504 calculates the pitch period prediction value for each subframe. Search range determination section 530 determines a pitch period search range for each subframe based on the subframe pitch period prediction value. In this case, the adaptive vector of the pitch period included in the search range determined by the search range determining unit 530 is selected as a candidate.

When the determination result of the continuity determination unit 526 is not continuous, the switch 531 is switched so that the full search unit 527 is selected, and all adaptive vectors included in the adaptive codebook 528 are selected as candidates. In this case, the pitch cycle calculated by the pitch cycle analysis unit 503 does not need to be given to the multiplexer 523.

In the multiplier 532, the adaptive vector selected from the adaptive codebook 528 and the adaptive vector gain codebook 5
The product is multiplied by the adaptive vector gain selected from 33. Similarly, in the multiplier 534, the noise vector selected from the noise codebook 529 and the noise vector gain codebook 53
The product with the noise vector gain selected from 5 is taken. The adder 536 takes the sum of the signals obtained from the multipliers 532 and 534 to generate a drive vector.

The drive vector thus generated is passed through the perceptual weighting synthesis filter 507 to generate a synthesis vector. The subtractor 509 calculates the difference between the target vector obtained by passing the voice signal through the perceptual weighting filter 508 and the combined vector, and based on this difference signal, the distortion calculation unit 511.
To find the distortion. The combination of the adaptive vector, the adaptive vector gain, the noise vector, and the noise vector gain when this distortion is minimized is efficiently searched.

For example, as shown in the flowchart of FIG. 31, there is a method of serially obtaining an adaptive vector, an adaptive vector gain, a noise vector, and a noise vector gain for each subframe. Further, as shown in the flowchart of FIG. 32, in the case of a configuration in which the adaptive vector gain and the noise vector gain are simultaneously optimized by vector quantization for each subframe, the adaptive vector, the noise vector, and the gain vector are obtained in this order. There is also a method.

That is, in FIG. 31, first, step S60.
After setting the counter k to 0 in step 1, step S602
An adaptive vector, an adaptive vector gain, a noise vector, and a noise vector gain are sequentially obtained in S603, S604, and S605, and the counter k is incremented by 1 in step S606. In step S607, it is determined whether the counter k matches K. If they do not match, the process returns to step S602 to continue the process, and if they match, the process ends.

Further, in FIG. 32, the counter k is first set to 0 in step S701, and then steps S702 and S7 are performed.
03, S704: Adaptive vector, noise vector, gain vector (adaptive vector gain and noise vector gain)
Are sequentially obtained, and the counter k is incremented by 1 in step S705. In step S706, the counter k is K.
It is determined whether or not If they do not match, the process returns to step S702 to continue the process, and if they match, the process ends.

In this way, the multiplexer 52 is provided with an index representing the adaptive vector, the adaptive vector gain, the noise vector, and the noise vector gain when the distortion is minimized.
Give to 3. When the continuity determination unit 526 determines that the adaptive vector is continuous, the index of the adaptive vector is represented as a relative value with respect to the subframe pitch cycle predicted value.

In the multiplexer 523, the object to be multiplexed differs depending on whether the determination result of the continuity determining unit 526 is continuous or not. That is, the continuity determination unit 5
26 is continuous, the LPC coefficient index obtained by the LPC coefficient quantizing unit 522, the analysis position index obtained by the pitch period analysis position determining unit 525, the determination information obtained by the continuity determining unit 526, the pitch period The pitch period information obtained by the analysis unit 503, the relative value of the adaptive vector to the predicted value of the subframe pitch period, the index of the adaptive vector gain, the noise vector index, and the index of the noise vector gain are multiplexed and encoded from the encoded data output terminal 524. Output as the converted data. On the other hand, when the determination result of the continuity determining unit 526 is not continuous, the LPC coefficient index obtained by the LPC coefficient quantizing unit 522, the determination information obtained by the continuity determining unit 526, the adaptive vector index, the adaptive vector gain index, The noise vector index and the noise vector gain index are multiplexed and output terminal 524
Output from.

(Fifteenth Embodiment) The fifteenth embodiment of the present invention will be described with reference to FIG. The present embodiment shows an example in which the present invention is applied to a CELP type decoder (speech decoding device). Although the present embodiment shows a decoder unit corresponding to a CELP encoder unit configured by combining some of the embodiments described above, the present invention is not limited to this, and CELP configured by various combinations. It can be applied to the decoder unit corresponding to the encoder unit.

The multiplexed coded data is input from the input terminal 801 and converted into indexes of various parameters in the demultiplexer 802. The LPC coefficient decoding unit 811, based on the LPC coefficient index, sets the LP
The C coefficient is decoded and given to the synthesis filter 812. The sub-frame pitch cycle generation unit 803 is given information related to the pitch cycle, and generates the pitch cycle used in the adaptive codebook 804. Details of generation of this pitch period will be described later with reference to FIG. 34.

An adaptive vector is obtained from adaptive codebook 804 using the pitch period generated here as an index. A multiplier 806 multiplies this adaptive vector by the adaptive vector gain obtained from the adaptive vector gain codebook 805 using the adaptive vector gain index as an index. Similarly, a noise vector obtained from the noise codebook 807 using the noise vector index as an index, and a noise vector gain codebook 8 using the noise vector gain index as an index.
The noise vector gain obtained from 08 is multiplied by a multiplier 809.
Ride with.

The two signals after the multiplication are added to the adder 810.
Is added to generate a drive signal, and the drive signal is given to the synthesis filter 812 to generate a synthesis signal. This combined signal is input to the post filter 813, and after being subjected to perceptual improvement such as formman emphasis, pitch emphasis, and gain adjustment, it is output from the output terminal 815.

Next, the subframe pitch period generator 803 will be described with reference to FIG. Pitch cycle continuity determination information is input from the input terminal 901, and the switch 904 and the switch 911 are switched according to the information. When the pitch cycle continuity determination information indicates “discontinuity”, the pitch cycle index for each subframe is input from the input terminal 902, and the subframe pitch cycle decoding unit 910 inputs the subframe pitch via the switch 904. The cycle is decoded, and the output terminal 912 is passed through the switch 911.
Is output from.

When the pitch cycle continuity determination information indicates "continuous", a pitch cycle analysis position index, a pitch cycle index, and a relative pitch cycle index are given from the input terminal 902, and the pitch is analyzed via the switch 904. It is given to the period analysis position decoding unit 905, the pitch period decoding unit 906, and the relative pitch period decoding unit 907, respectively. The subframe pitch period prediction value calculation unit 908 obtains the subframe pitch period prediction value using the pitch period analysis position and the pitch period obtained as a result. The subframe pitch cycle calculation unit 909 calculates the subframe pitch cycle using the predicted value of the subframe pitch cycle and the relative pitch cycle, and outputs the subframe pitch cycle from the output terminal 912 via the switch 911.

FIG. 35 shows the configuration of a computer system which realizes the voice encoding / decoding method according to the present invention.
This computer system is, for example, a personal computer, and has a CPU 1001 that controls arithmetic processing, an input unit 1002 such as a keyboard, a pointing device, and a microphone, an output unit 1003 such as a display and a speaker, and a ROM 1004 as main memory.
And a RAM 1005, a hard disk device 1006 as an external storage device, and a floppy disk device 1007.
And an optical disk device 1008 connected by a bus 1009.

Here, the hard disk drive 1006, floppy disk drive 1007 and optical disk drive 10
Any one of the recording media 08 stores a program and encoded data for executing the audio encoding process described in the above embodiment. Then, according to this program, the CPU 1001 performs the audio encoding process on the input audio signal input from the input unit 1001, and the CPU 1 executes the encoded data read from the recording medium.
The audio decoding processing is performed by 001, and the output unit 1003
Is output from. By doing so, it is possible to carry out the voice encoding / decoding processing of the present invention using an ordinary personal computer.

The present invention can be implemented with various modifications as follows. (1) In the above embodiments, interpolation or extrapolation is used as a method of obtaining the subframe pitch period prediction value.
The present invention is not limited to this. For example, the number of pitch periods used for prediction may be increased to improve the accuracy of prediction, or higher-level prediction such as a higher-order function or spline function may be performed. By doing so, the error between the actually measured value and the predicted value of the subframe pitch is reduced, and the subframe pitch period can be expressed with a smaller amount of information.

(2) When quantizing the pitch period obtained by the pitch period analysis unit, prediction is performed with the pitch period obtained in the past, and the difference from the predicted value is quantized to improve the quantization efficiency. You can also let it.

(3) When the search range is determined from the subframe pitch cycle predicted value, the search range is not set only in the vicinity of the subframe pitch cycle predicted value, but in the vicinity of an integer multiple of the subframe pitch cycle predicted value or an integer. It can also be set in the vicinity of one-half.

(4) The embodiments described above may be applied for the purpose of calculating the pitch period of the pitch filter which is one component of the perceptual weighting filter. Similarly, the embodiments described so far may be applied for the purpose of calculating the pitch period of the pitch filter, which is one component of the post filter not explicitly shown here.
In that case, no additional information is required, so that it can be easily applied to a voice encoding device.

(5) The speech coding method according to the above-described embodiments may be switched and used in frame units. In that case, additional information is newly required to indicate which method of the embodiment is used.

[0149]

As described above, according to the speech coding method of the present invention, it is possible to reduce the amount of calculation when searching the subframe pitch period while maintaining the quality of synthesized speech. It is possible to effectively reduce the number of bits for expressing the information of the subframe pitch period.

[0150]

[Brief description of drawings]

FIG. 1 is a diagram for explaining the basic principle of the present invention.

FIG. 2 is a block diagram for explaining the basic operation of pitch period analysis in the speech coding method according to the present invention.

FIG. 3 is a diagram showing a frame structure and a subframe structure.

FIG. 4 is a block diagram for explaining a speech encoding method according to a second embodiment of the present invention.

FIG. 5 is a block diagram showing a configuration of a subframe pitch extraction unit according to the second embodiment.

FIG. 6 is a diagram for explaining a method of calculating a subframe pitch period according to the second embodiment.

FIG. 7 is a diagram for explaining a method of calculating a subframe pitch period according to the second embodiment.

FIG. 8 is a flowchart for explaining a processing procedure of subframe pitch period calculation in the second embodiment.

FIG. 9 is a diagram for explaining a method of calculating a subframe pitch period according to the third embodiment of the present invention.

FIG. 10 is a flowchart for explaining a processing procedure of subframe pitch period calculation in the third embodiment.

FIG. 11 is a diagram for explaining the effect of the third embodiment.

FIG. 12 is a diagram for explaining a subframe pitch period calculation method according to the fourth embodiment of the present invention.

FIG. 13 is a flowchart for explaining a processing procedure of subframe pitch period calculation in the fourth embodiment.

FIG. 14 is a diagram for explaining a subframe pitch period calculation method according to the fifth embodiment of the present invention.

FIG. 15 is a flowchart for explaining a processing procedure of subframe pitch period calculation in the fifth embodiment.

FIG. 16 is a block diagram for explaining a speech coding method according to a sixth embodiment of the present invention.

FIG. 17 is a diagram showing a part of a flowchart for explaining a processing procedure of subframe pitch period calculation in the sixth embodiment.

FIG. 18 is a diagram showing another part of the flowchart for explaining the processing procedure of subframe pitch period calculation in the sixth embodiment.

FIG. 19 is a diagram showing a part of a flowchart for explaining a processing procedure of subframe pitch period calculation in the seventh embodiment of the present invention.

FIG. 20 is a diagram showing another part of the flowchart for explaining the processing procedure of subframe pitch period calculation in the seventh embodiment.

FIG. 21 is a diagram showing a part of a flowchart for explaining a processing procedure of subframe pitch period calculation according to an eighth embodiment of the present invention.

FIG. 22 is a diagram showing another part of the flowchart for explaining the processing procedure of subframe pitch period calculation in the eighth embodiment.

FIG. 23 is a diagram showing a part of a flowchart for explaining a processing procedure of subframe pitch period calculation in the ninth embodiment of the present invention.

FIG. 24 is a diagram showing another part of the flowchart for explaining the processing procedure of subframe pitch period calculation in the ninth embodiment.

FIG. 25 is a block diagram for explaining a speech encoding method according to a tenth embodiment of the present invention.

FIG. 26 is a diagram for explaining the effect of the tenth embodiment.

FIG. 27 is a block diagram for explaining a speech encoding method according to an eleventh embodiment of the present invention.

FIG. 28 is a block diagram for explaining a speech encoding method according to a twelfth embodiment of the present invention.

FIG. 29 is a block diagram for explaining a speech coding method according to a thirteenth embodiment of the present invention.

FIG. 30 is a block diagram showing the configuration of a speech encoding apparatus according to the fourteenth embodiment in which the speech encoding method of the present invention is applied to the CELP system.

FIG. 31 is an adaptive vector according to the fourteenth embodiment,
Flowchart for explaining the procedure for obtaining adaptive vector gain, noise vector, and noise vector gain

FIG. 32 is an adaptive vector according to the fourteenth embodiment,
Flowchart for explaining the procedure for obtaining the noise vector and the gain vector

FIG. 33 is a block diagram showing the configuration of a speech decoding apparatus according to the fifteenth embodiment in which the speech decoding method of the present invention is applied to the CELP method.

FIG. 34 is a block diagram showing the configuration of a subframe pitch period generation unit in the fifteenth embodiment.

FIG. 35 is a block diagram showing the configuration of a computer system that realizes a speech encoding / decoding method according to the present invention.

[Explanation of symbols]

11 ... Voice signal input terminal 12 ... Frame configuration section 13 ... Pitch cycle analysis section 14 ... Pitch cycle information output terminal 21 ... Voice signal input terminal 22 ... Frame / subframe configuration section 23 ... Pitch cycle analysis section 24 ... Subframe pitch cycle Extraction unit 25 ... Subframe pitch period information output terminals 201, 301, 401, 501, 601 ... Audio signal input terminals 202, 302, 402, 502, 602 ... Frame
Subframe configuration units 203, 303, 403, 503, 603 ... Pitch period analysis units 204, 304, 404, 504, 604 ... Subframe pitch period prediction value calculation units 205, 305, 405, 530 ... Search range determination unit 206, 306, 406, 506, 528, 606 ... Adaptive codebook 207, 307, 407, 507, 607 ... Auditory weighted synthesis filter 208, 308, 408, 508, 608 ... Auditory weighted filter 209, 309, 409, 509, 609 ... Subtractors 210, 310, 410, 510, 610 ... Distortion calculation units 211, 311, 411, 511, 611 ... Relative pitch cycle information output terminals 212, 312, 412, 512, 612 when distortion is minimum ... Pitch cycle information output Terminals 313, 525 ... Pitch cycle analysis Position determination unit 314 ... Pitch cycle analysis Position information output terminals 413, 526 ... Continuity determination sections 414, 527 ... Full search section 415, 531 ... Switch 416 ... Continuity determination result output terminals 505, 605 ... Relative pitch pattern codebooks 513, 536, 613, 619 ... Addition 521 ... LPC coefficient analysis section 522 ... LPC coefficient quantization section 523 ... Multiplexer 524 ... Coded data output terminals 529, 614 ... Noise codebook 532, 534, 615, 617 ... Multiplier 533 ... Adaptive vector codebook 535 ... Adaptation Vector gain codebook 801 ... Encoded data input terminal 802 ... Demultiplexer 803 ... Subframe pitch period generation unit 804 ... Adaptive codebook 805 ... Adaptive vector gain codebook 806, 809 ... Multiplier 807 ... Noise codebook 808 ... Noise vector Gain codebook 810 ... Adder 811 ... LPC coefficient recovery Part 812 ... synthesis filter 813 ... postfilter 814 ... audio signal output terminal

─────────────────────────────────────────────────── ─── Continuation of the front page (56) Reference JP-A-9-34499 (JP, A) JP-A-8-202398 (JP, A) JP-A-10-97296 (JP, A) JP-A-8- 328588 (JP, A) JP 5-73098 (JP, A) JP 4-30200 (JP, A) Special table 9-512644 (JP, A) (58) Fields investigated (Int.Cl. ⁷ , DB name) G10L 19/04 G10L 19/12

Claims (8)

(57) [Claims] 1. A voice encoding method for performing an encoding process, comprising dividing an input voice signal into frames of a predetermined length and encoding a pitch period of the input voice signal, wherein the frame is divided into the frames. A speech coding method, characterized in that, at the time of coding the pitch cycle of the current frame of the input speech signal, the pitch cycle of the future frame is obtained with respect to the current frame, and the pitch cycle of the future frame is coded. . 2. A voice encoding method for performing an encoding process, which comprises dividing an input voice signal into frames of a predetermined length and encoding a pitch period of the input voice signal. A voice code, characterized in that, at the time of encoding the pitch period of the current frame of the input speech signal, the pitch period of the frame of the look-ahead part is obtained for the current frame, and the pitch period of the frame of the look-ahead part is encoded. Method. 3. The pitch period of the future frame is adaptively determined based on the magnitude of the short-term power of the input speech signal, the prediction residual signal, or the prediction difference signal after passing through the low pass filter. The speech coding method according to claim 1, wherein the speech coding method is obtained at the analyzed position. 4. The pitch period of the frame of the prefetching unit is:
The analysis position adaptively determined based on the magnitude of the short-term power of the input voice signal, the prediction difference signal, or the prediction difference signal after passing through the low-pass filter. 2. The voice encoding method according to 2. 5. A speech coder that divides an input speech signal into frames of a predetermined length and performs a coding process including a process of coding a pitch period of the input speech signal, wherein the speech coding device is divided into the frames. Means for obtaining a pitch period of a future frame with respect to the present frame when encoding the pitch period of the present frame of the input speech signal, and means for encoding the pitch period of the future frame obtained by this means A speech coding apparatus comprising: 6. A speech coder that divides an input speech signal into frames of a predetermined length and performs a coding process including a process of coding a pitch period of the input speech signal, wherein the speech coding device is divided into the frames. When the pitch period of the current frame of the input audio signal is encoded, a means for obtaining the pitch period of the frame of the look-ahead portion for the current frame and the pitch period of the frame of the look-ahead portion obtained by this means are encoded. A speech coding apparatus having means. 7. A computer readable program recorded with a program for performing a voice coding process including a process of dividing an input voice signal into frames of a predetermined length and coding a pitch period of the input voice signal. A storage medium, wherein when the pitch period of the current frame of the input audio signal that has been divided into the frames is encoded, a process of obtaining the pitch period of the future frame for the frame and the pitch period of the future frame are encoded. A storage medium in which a program for executing the processing to be converted is recorded. 8. A computer readable program recording a program for performing a voice coding process, which comprises a process of dividing an input voice signal into frames of a predetermined length and coding a pitch period of the input voice signal. A storage medium, wherein, when encoding the pitch period of the current frame of the frame-divided input audio signal, the process of obtaining the pitch period of the frame of the look-ahead unit for the frame, and the pitch period of the frame of the look-ahead unit A storage medium recording a program for executing the process of encoding.