JPH09146596A - Sound signal synthesizing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a sound signal synthesizing method which can reproduce spectrum envelope in a frame and the change of spectral fine structure with time by interpolating and synthesizing a sound signal in synchronization with the information on the pitch period to which smoothing has been applied. SOLUTION: When an information obtained by an analysis based on the unit of the fundamental frame of a specified length is received from the analyser section 20 of a first vocoder through a transmission line 10, a sub-frame is set inside an extend frame corresponding to the present fundamental frame in the interpolation processing section 33 of synthesizer section 30 of a second vocoder so that the sub-frame length showing the pitch period related to each sub-frame is changed smoothly. A line spectrum pair coefficient and/or frame power information are generated for every sub-frame based on the information related to the present frame. A synthesis processing section 34 synthesizes a sound signal based on the information related to the present fundamental frame, line spectrum pair coefficient and/or frame power information.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice signal synthesizing method used in a short wave vocoder or the like.

[0002]

2. Description of the Related Art Since the Radio Law imposes a restriction that an occupied band is ≤3 kHz, it is necessary to reduce the bit rate when developing a vocoder suitable for a short wave band transmission line. On the other hand, since the short wave band transmission line has the characteristic that fading and interference easily occur, it is necessary to consider error resilience when lowering the bit rate.

FIG. 1 shows a state in which first and second low-bit rate vocoders for short waves are connected via a transmission line 10. However, for simplicity of illustration, only the analysis unit 10 is shown for the first vocoder, and only the combining unit 30 is shown for the second vocoder. The analysis unit 10 of the first vocoder includes an analysis processing unit 21 that analyzes an input voice signal and an encoder 2 that encodes the output of the analysis processing unit 21.
3-1 to 23-4 and a multiplexer 22 that multiplexes the outputs of the respective encoders, and the synthesizing unit 30 of the second vocoder converts the signal received from the first vocoder via the transmission path. Demultiplexer 31 for demultiplexing, decoders 32-1 to 32-4 for decoding the demultiplexed signal, interpolation processing unit 33 for performing interpolation processing based on the outputs of decoders 32-1 to 32-4, and interpolation It has a synthesis processing unit 34 that synthesizes an audio signal from the output of the processing unit 33.

In the analysis processing unit 21, its length is long (for example, 30 mse) in order to meet the request for a lower bit rate.
For each analysis section (frame) set in c), the input voice signal is analyzed using a method such as vector quantization, and the information obtained as a result is output. The information generated at this time includes pitch period, line spectrum pair coefficient, frame power, voiced / unvoiced information, and the like. These pieces of information are supplied to the interpolation processing unit 33 through encoding / multiplexing / transmission path transmission / demultiplexing / decoding. The interpolation processing unit 33 uses one frame (hereinafter referred to as a basic frame) to a plurality of frames (hereinafter referred to as subframes) in order to recover information that is lost due to the frame length being set to be long as described above. ), The pitch period,
Linearly interpolate line spectrum pair coefficients, frame power, etc. The synthesizing unit 34 determines the pitch period obtained as a result,
The audio signal is synthesized based on the line spectrum pair coefficient, the frame power, and the like.

[0005]

However, in such a method, various kinds of information necessary for faithfully reproducing an audio signal, such as a spectral envelope in a frame and a temporal transition of a fine spectral structure, are reproduced. Can not do it. An object of the present invention is to realize an audio signal synthesizing method capable of reproducing such information by a process relating to a pitch cycle.

[0006]

Means for Solving the Problems and Effects of the Invention In order to achieve such an object, a first structure of the present invention is a pitch period obtained by an analysis _using a basic frame of a predetermined length F _anl as a unit. Information P _{ref and} line spectrum pair coefficient L _ref
And / or the frame power information G _ref is received from the analysis unit of the first vocoder via the transmission line, in the audio signal synthesizing method executed by the synthesizing unit of the second vocoder, It is characterized in that it is repeatedly executed every time: (1) In the extension frame corresponding to the current basic frame, the length F _{pre of} the extension frame, the pitch period information P _ref related to the current basic frame, and the previous extension. Based on the length P _pst of the last sub-frame in the frame, the sub-frame length P indicating the pitch period of each sub-frame
The first step of setting N subframes so that _smz [n] changes smoothly with respect to n (N is a natural number of 2 or more, and n is an integer satisfying 0 ≦ n ≦ N−1); (2 ) Line spectrum pair coefficient L for the current basic frame
_The line spectrum pair coefficient L based on the _ref and / or the frame power information G _ref and for each subframe.
_smz [n] and / or frame power information G _smz [n]
(3) Pitch period information P _smz [n] and line spectrum pair coefficient L _smz [n] and / or frame power information G
Third step of synthesizing voice signal based on _smz [n]; and (4) Surplus amount ΔF of the extended frame length F _pre with respect to sum F _syn of n of subframe length P _smz [n].
_Is added to the basic frame length F _anl , and the resulting value F _nxt is set to the length F _pre of the extension frame corresponding to the next incoming basic frame.

As described above, in the present configuration, the line spectrum pair coefficient L _smz [n] is synchronized in the second step, so to speak, in synchronism with the pitch period information P _smz [n] smoothed in the first step prior to the synthesis. ] And / or frame power information G _smz [n] is generated (interpolation), and the audio signal is synthesized in the third step based on the result. Therefore, it becomes possible to reproduce the characteristics such as the temporal change of the spectral envelope and the spectral fine structure, and the quality of the speech signal obtained by the synthesis can be improved.

The second configuration of the present invention is that the pitch period information P _ref , the line spectrum pair coefficient L _ref and / or the frame power information G _ref , which are obtained by the analysis _using a basic frame of a predetermined length F _anl as a unit, and When voiced / unvoiced information indicating whether the current basic frame is a voiced frame or an unvoiced frame is received from the analysis unit of the first vocoder via the transmission path, the synthesis unit of the second vocoder The voice signal synthesizing method to be performed is characterized in that the following steps are repeatedly performed for each of the basic frames: (0) is the current basic frame a voiced frame based on the received voiced / unvoiced information? 0th step of determining whether it is an unvoiced sound frame; (6) If it is determined in step 0 that the frame is a voiced sound, The sixth step of transitioning to the voice signal synthesizing method according to the configuration; A seventh step in which a frame having a length obtained by adding ΔF is regarded as the extended frame, and then the second to fourth steps are performed. According to this configuration, it is possible to deal with unvoiced sound frames.

According to a third configuration of the present invention, in the second configuration, in the seventh step, when the previous basic frame is a voiced sound frame and the current basic frame is an unvoiced sound frame, the previous expansion is performed. Surplus in frame ΔF
Is added to the subframe length related to the equal division, and the step is set to the length of the beginning subframe of the extension frame corresponding to the current basic frame. According to this configuration, it is possible to make adjustment so that no particular contradiction occurs in the transition from voiced sound to unvoiced sound.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of the present invention will be described below with reference to the drawings. Since the present invention can be implemented by the device shown in FIG. 1, the following description is premised on the device shown in FIG. 1, but the present invention does not require the details of the functional configuration of the device to be limited. Similarly, the details of the procedure described below are not limited.

FIG. 2 shows a procedure of processing executed by the interpolation processing unit 33 in the apparatus according to the embodiment of the present invention. In this procedure, the interpolation processing unit 33
First, the basic frame length F _anl and the extension frame length F
_pre, pitch period preset value P _pst, line spectral pair coefficients preset value L _pst, frame power (Synthesis side gain) preset value G _pst, it can accommodate pitch period number in the extended frame N, evenly divided number N _UV during unvoiced Variables such as are initialized according to the system design (10
0). The interpolation processing unit 33 then receives the demultiplexer 31.
The voiced sound / unvoiced sound information decoded by the decoder 32-1 is input from the information demultiplexed by the above, and based on this, the synthesizing unit 30 of the second vocoder transmits the information of the first vocoder via the line 10 or the like. It is determined whether the frame received from the analysis unit 20 is a voiced sound frame (102).

When it is determined that the frame is a voiced sound frame, the interpolation processing unit 33 outputs the pitch period P _ref and the line spectrum pair coefficient L _ref from the decoders 32-2 to 32-4, respectively.
And the frame power G _ref are input (104). The interpolation processing unit 33 uses the following equation

(Equation 1) According to the above, the number N of pitch periods that can be accommodated in the extension frame (described later) is calculated (106). That is, one extension frame is divided into N subframes. However, although the above equation is used a simple average value of the pitch period preset value P _{ps t} that pitch period P _ref and the initial settings entered in the denominator, may be used a weighted average value.

The interpolation processing unit 33 further uses the following equation

(Equation 2) By performing the linear interpolation calculation according to the following, the pitch period P is calculated for each of the N subframes in the extension frame.
_{The smz} [n] is _calculated (108). The interpolation processing unit 33 confirms that all subframes related to the obtained pitch period P _smz [n] are included in the extension frame by the following equation.

(Equation 3) The synthetic frame length F _syn is _{calculated according} to (110) and compared with the extended frame length F _pre (112). If it does not fit (that is, if F _syn > F _pre ), the interpolation processing unit 33
Decrements the number N of subframes by 1 (114) and returns to step 108. Conversely, if it fits (ie F _syn
≦ F _pre ), the interpolation processing unit 33 uses the same formula as the formula (2).

(Equation 4) , That is, by the calculation synchronized with the linear interpolation of the pitch period, the line spectrum pair coefficient L is calculated for each subframe.
_{The smz} [n] and the frame power G _smz [n] are obtained (116).

After the execution of step 116, the interpolation processing unit 33
Is the composite frame length F with respect to the extended frame length F _pre
syn calculated residual ΔF = F _pre -F _syn of (118), calculated residual frame length [Delta] F the basic frame length, i.e. the next basic frame by adding a frame length F _anl using in the analysis unit 20 The extended frame length F _pre is calculated (120) and the preset values P _pst and L are set.
The pitch period P _smz of the last sub-frame in _pst and G _pst
[N-1], line spectrum pair coefficient L _smz [N-1] and frame power G _smz [N-1] are set (1
22) and returns to step 102. That is, the extension frame used in the above description is a frame obtained by extending the basic frame with the residual ΔF that does not belong to any subframe of the length of the previous extension frame.

When it is determined in step 102 that the frame is unvoiced, the interpolation processing unit 33 sets the subframe length to the length _obtained by dividing the basic frame length F _anl by the equal division number N _UV . If the immediately preceding frame is a voiced sound frame, a non-zero residual ΔF is generally generated in the voiced sound frame, so the interpolation processing unit 33 expands the first sub-frame with the residual ΔF. . The interpolation processing unit 33 executes the same processing as steps 116 to 122 (128 to 134) and returns to step 102. However, in steps 128 to 134, N is replaced with N _UV and ΔF is treated as 0. In addition, when the voice is unvoiced, the pitch cycle calculation is not executed. As P _ref , the pitch tracking value calculated by the analysis unit 20 is updated every frame until voiced sound is obtained.

FIG. 3 schematically shows the interpolation state in this embodiment. Particularly, FIG. 3A shows a case of a voiced sound frame, and one extension frame is divided into a plurality of subframes. Further, in this case, the length of each subframe is determined by linear interpolation. Next, FIG. 3B shows a case of an unvoiced sound frame, in which one basic frame is divided into a plurality of subframes. In this case, the lengths of the subframes are equal to each other because they are determined by the equal division. 3C is an example immediately after the transition from the voiced sound frame to the unvoiced sound frame, which is different from FIG. 3B in that the residual ΔF is added to the first subframe.

As described above, in this embodiment, step 1
In step 116, the line spectrum pair coefficient L is synchronized with the smoothed pitch period P _smz [n].
_smz [n] and frame power G _smz [n] are generated, and the synthesis processing unit 34 synthesizes the audio signal based on the result. Therefore, it becomes possible to reproduce the characteristics such as the temporal change of the spectral envelope and the spectral fine structure, and the quality of the speech signal obtained by the synthesis can be improved. Moreover, since even division is adopted, unvoiced frames can be dealt with. Furthermore, since the residual ΔF is added to the length of the first subframe when the voiced sound frame is changed to the unvoiced sound frame, no particular contradiction occurs when the voiced sound is changed to the unvoiced sound.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an example configuration of a vocoder.

FIG. 2 is a flowchart showing a flow of operations of an interpolation processing unit according to the embodiment of the present invention.

3A and 3B are diagrams showing an interpolation state in this embodiment, where FIG. 3A is a voiced sound, FIG. 3B is an unvoiced sound, and FIG.
Indicates the transition from voiced sound to unvoiced sound.

[Explanation of symbols]

10 transmission lines, 20 analysis unit, 30 synthesis unit, 33 interpolation processing unit, 34 synthesis processing unit.

Claims (3)

[Claims] 1. Pitch period information P _{ref and} line spectrum pair coefficient L _ref and / or frame power information G obtained by an analysis _using a basic frame of a predetermined length F _anl as a unit.
_{When ref} is received from the analysis unit of the first vocoder via the transmission line, in the audio signal synthesizing method executed by the synthesizing unit of the second vocoder, the following steps are repeatedly performed for each basic frame. Characteristic audio signal synthesizing method: Length of the extension frame F _pre , pitch period information P _ref related to the current base frame, and last subframe in the previous extension frame in the extension frame corresponding to the current base frame. Sub-frame length P _smz indicating the pitch period of each sub-frame based on the length P _pst of
The first step of setting N subframes so that [n] changes smoothly with respect to n (N is a natural number of 2 or more, n is an integer satisfying 0 ≦ n ≦ N−1); current basic A line spectrum pair coefficient L _smz [n] and / or frame power information G _smz [n] is generated based on the line spectrum pair coefficient L _ref and / or the frame power information G _{ref related} to the frame and for each subframe. Second step; pitch period information P _smz [n] and line spectrum pair coefficient L
_smz [n] and / or frame power information G _smz [n]
A third step of synthesizing a speech signal based on the above, and adding a surplus ΔF of the extended frame length F _pre to the sum F _syn of n of the sub-frame length P _smz [n] to the basic frame length F _anl , and The resulting value F _nxt is
The fourth step of setting the length F _pre of the extension frame corresponding to the next basic frame. 2. Pitch period information P _ref , line spectrum pair coefficient L _ref and / or frame power information G _ref , which are obtained by analysis in units of a basic frame of a predetermined length F _anl .
Also, when voiced / unvoiced information indicating whether the current basic frame is a voiced frame or an unvoiced frame is received from the analysis unit of the first vocoder via the transmission path, the synthesis unit of the second vocoder is In the method of synthesizing a voice signal according to claim 1, the following steps are repeatedly performed for each of the above basic frames: a voice signal synthesizing method, in which the current basic frame is a voiced frame based on the received voiced / unvoiced information. 0th step of judging whether there is an unvoiced sound frame or not; 0th
When it is determined that the frame is a voiced sound in step,
The sixth step of transitioning to the voice signal synthesizing method according to claim 1; and, when it is determined that the frame is an unvoiced sound frame in the 0th step, the surplus ΔF related to the current basic frame or the immediately preceding voiced sound frame is calculated. A seventh step in which the frame having the added length is regarded as the above-mentioned extended frame, and then the second to fourth steps are performed. 3. The speech signal synthesizing method according to claim 2, wherein in the seventh step, when the previous basic frame is a voiced sound frame and the current basic frame is an unvoiced sound frame, Surplus ΔF
A method of synthesizing a voice signal, comprising the step of setting a value obtained by adding to the subframe length according to the equal division to the length of the beginning subframe of the extension frame corresponding to the current basic frame.