WO2013014876A1 - Fragment processing device, fragment processing method, and fragment processing program - Google Patents

Abstract

A fragment processing device is provided which can reduce significant local spectral deformation while improving the continuity of audible sound in a continuous portion between fragments. A spectral envelope extraction means (71) extracts the spectral envelope from two continuous fragments in the period including the boundary time of the two fragments. An orthogonal transformation means (72) performs an orthogonal transformation on the spectral envelope. A smoothing means (73) smoothes the orthogonal transformation coefficient, which is the expansion coefficient resulting from the orthogonal transformation of the spectral envelope. An inverse transform means (74) performs an inverse transform of the orthogonal transformation on the corrected orthogonal transformation coefficient. A time-domain waveform generating means (75) generates a time-domain waveform from the spectral envelope.

Description

Segment processing apparatus, segment processing method, and segment processing program

The present invention relates to an element processing apparatus, an element processing method, and an element processing program for processing an element so that the elements are connected well.

As a speech synthesis method, a method of generating synthesized speech by connecting segments that are speech waveforms cut out from recorded speech is known. In this method, for example, in accordance with the reading of the input text, each segment corresponding to the reading is selected and the segment is connected. Speech synthesis that employs this method is called segment selection speech synthesis. Note that the segment is generated in advance based on the recorded voice. Each segment is cut out, for example, for each semi-syllable. In general, a plurality of types of segments are generated from various recorded voices for one voice (here, a voice of about half syllable).

Note that the segment is sometimes referred to as a speech segment.

Patent Document 1 describes a technique for calculating a frequency characteristic of a connection portion between connected pieces, calculating a spectrum envelope thereof, and correcting the frequency characteristic of the connection portion by the spectrum envelope.

JP 2009-237015 A

In the unit selection type speech synthesis, when a person listens to the synthesized speech, a sense of discontinuity may be felt. This is because a spectral gap is generated between the connected pieces. When selecting the pieces, a combination of the pieces is selected so that the gap becomes as small as possible, and the pieces are connected to each other. However, it has been difficult to eliminate the spectral gap.

In the technique described in Patent Document 1, the audibility continuity at the connecting portion of the segment can be improved by correcting the frequency characteristics by the spectrum envelope. However, the spectrum may be greatly deformed locally.

Therefore, the present invention provides a fragment processing apparatus, a fragment processing method, and a fragment processing method capable of improving the audible continuity of sound at a connected portion of a fragment while reducing local deformation of the spectrum. An object is to provide a fragment processing program.

The segment processing apparatus according to the present invention includes a spectrum envelope extraction unit that extracts a spectrum envelope in a period including a boundary time between two segments from two consecutive segments, and an orthogonal that performs orthogonal transform on the spectrum envelope. A transforming means, a smoothing means for smoothing an orthogonal transform coefficient, which is an expansion coefficient obtained as a result of performing an orthogonal transform on the spectrum envelope, and an inverse for performing an inverse transform of the orthogonal transform on the corrected orthogonal transform coefficient. Transform means, and time domain waveform generation means for generating a time domain waveform from a spectrum envelope obtained as a result of performing inverse transform of orthogonal transform on the corrected orthogonal transform coefficient.

Further, the segment processing method according to the present invention extracts a spectrum envelope in a period including a boundary time between the two segments from two consecutive segments, performs orthogonal transform on the spectrum envelope, and converts the spectrum envelope into a spectrum envelope. The orthogonal transformation coefficient, which is the expansion coefficient obtained as a result of performing the orthogonal transformation, is smoothed, the orthogonal transformation coefficient is inversely transformed to the corrected orthogonal transformation coefficient, and the orthogonal transformation is performed on the corrected orthogonal transformation coefficient. A time domain waveform is generated from a spectrum envelope obtained as a result of performing the inverse transformation.

In addition, the segment processing program according to the present invention allows a computer to perform spectrum envelope extraction processing for extracting a spectrum envelope in a period including a boundary time between two segments from two consecutive segments, orthogonal to the spectrum envelope. Orthogonal transformation processing that performs transformation, smoothing processing that smoothes the orthogonal transformation coefficient that is the expansion coefficient obtained as a result of performing orthogonal transformation on the spectrum envelope, and inverse transformation of orthogonal transformation to the corrected orthogonal transformation coefficient And a time domain waveform generation process for generating a time domain waveform from a spectrum envelope obtained as a result of performing an inverse transform of the orthogonal transform on the corrected orthogonal transform coefficient.

According to the present invention, it is possible to improve the continuity of the audible sound at the connecting portion of the segments while reducing local large deformation of the spectrum.

It is a block diagram which shows the example of embodiment of the unit processing apparatus of this invention. It is explanatory drawing which shows the example of the correction | amendment object period of the orthogonal transformation coefficient in two continuous elements. It is explanatory drawing which shows typically the example of the spectrum envelope which the spectrum envelope extraction part 2 extracted. It is explanatory drawing which shows typically the result of the orthogonal transformation with respect to a spectrum envelope. 3 is a block diagram illustrating an example of a smoothing processing unit 4. FIG. It is explanatory drawing for showing the process of the reference value calculation part. It is explanatory drawing which shows the correct | amended orthogonal transformation coefficient typically. It is a flowchart which shows the example of the process progress of the segment processing apparatus of this invention. It is explanatory drawing which shows typically the correction _| amendment in the period before boundary time _tbnd in a correction _| amendment object period. It is explanatory drawing which shows typically the correction _| amendment in the period after boundary time _tbnd in a correction _| amendment object period. It is a block diagram which shows the modification of embodiment of this invention. It is a block diagram which shows the example of the minimum structure of the fragment processing apparatus of this invention.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram showing an example of an embodiment of the fragment processing apparatus of the present invention. The segment processing apparatus of this embodiment includes a correction range determination unit 1, a spectrum envelope extraction unit 2, an orthogonal transformation unit 3, a smoothing processing unit 4, an inverse transformation unit 5, a pitch waveform generation unit 6, And a superposition addition unit 7.

The individual segment selected in accordance with the reading of the text to be synthesized is input to the correction range determination unit 1.

A fragment is expressed as a change in amplitude within a certain time. For example, the segment is represented by an individual time and an amplitude at each time. Then, by representing the time on the horizontal axis and the amplitude on the vertical axis, the segment can be represented as a waveform.

The correction range determination unit 1 determines a range (period) to be corrected for the orthogonal transformation coefficient between two consecutive segments in order to increase the continuity of the audible sound at the connected portion of the segments. . Hereinafter, the range to be corrected is referred to as a correction target period. The orthogonal transform coefficient will be described later. FIG. 2 is an explanatory diagram illustrating an example of a correction target period of orthogonal transform coefficients in two continuous segments. Of the two consecutive pieces, the first piece is referred to as a preceding piece, and the subsequent piece is referred to as a subsequent piece. The correction range determination unit 1 may determine a period that satisfies the following conditions as the correction target period. The first condition is to include the boundary time of the segment (hereinafter referred to as t _bnd ). The second condition is that the period before the center time t _c1 of the preceding segment period and the period after the center time t _c2 of the subsequent segment period are not included. Accordingly, the correction period is the longest, from the central time t _c1 period preceding segment, a case where the distance from the center time t _c2 period subsequent segment was corrected target period (see FIG. 2). However, a period shorter than the period from t _c1 to t _c2 may be determined as the correction target period.

The correction range determination unit 1 may determine a fixed length period as a correction target period. Alternatively, a period of a predetermined ratio with respect to the segment length (for example, a period of 30% with respect to the total length of the preceding segment and the subsequent segment) may be determined as the correction target period. However, the above conditions shall be satisfied.

The spectrum envelope extraction unit 2 extracts the spectrum envelope corresponding to the portion from the portion overlapping the correction target period in the preceding segment. Similarly, the spectrum envelope extraction unit 2 extracts the spectrum envelope corresponding to the portion from the portion overlapping the correction target period in the subsequent segment.

The aspect in which the spectrum envelope extraction unit 2 extracts the spectrum envelope from the segment is not particularly limited. For example, the spectrum envelope extraction unit 2 may extract the spectrum envelope from the segment by performing STRIGHT analysis. STRAIGHT analysis is an example of a method for extracting a spectral envelope. Instead of the STRAIGHT analysis, the spectral envelope may be extracted from the segment by performing, for example, linear prediction analysis, LSP (Line Spectral Pair) analysis, or cepstrum analysis.

FIG. 3 is an explanatory diagram schematically showing an example of the spectrum envelope extracted by the spectrum envelope extraction unit 2. The horizontal axis shown in FIG. 3 represents time, and the vertical axis represents spectral density. Also, the start time and end time of the correction target period are denoted as t _str and t _fin , respectively. Further, the spectral density is obtained discretely. In this example, the case where the spectral density is calculated in accordance with the pitch period of the segment is taken as an example. Then, describing the time when the spectral density is computed pitch synchronously position time, represented by t _p. The first _{t p} in the spectrum envelope extracted from the preceding segment (the time at which the spectral density is _calculated), and _{t Beg1,} the last _{t p} in the spectral envelope and _{t end1.} Further, the first _{t p} in the spectrum envelope extracted from the subsequent _segment, and _{t Beg2,} the last _{t p} in the spectral envelope and _{t end2.} At this time, the following relationship is established.

t _str <t _beg1 <t _end1 <t _bnd <t _beg2 <t _end2 <t _fin

In the example shown in FIG. 3, the axis in the depth direction (in other words, the axis perpendicular to the paper surface) is the frequency axis. That is, the spectrum envelope illustrated in FIG. 3 is obtained for each subband.

The orthogonal transform unit 3 performs orthogonal transform on the spectrum envelope extracted by the spectrum envelope extraction unit 2. The orthogonal transformation is to obtain a expansion coefficient by an orthogonal function. Further, the orthogonal function is a function having orthonormality. Hereinafter, the expansion coefficient obtained by the orthogonal transformation is referred to as an orthogonal transformation coefficient. Specifically, a vector having an orthogonal transform coefficient as a component (hereinafter referred to as an orthogonal transform coefficient vector) is obtained by orthogonal transform on the spectrum envelope. Orthogonal transformation coefficient vector is obtained for each pitch synchronization position time t _p.

Examples of orthogonal transforms include discrete Fourier transform (DFT), fast Fourier transform (FFT), discrete cosine transform (DCT), wavelet transform, and the like. Here, a case where the orthogonal transform unit 3 performs fast Fourier transform is taken as an example.

FIG. 4 is an explanatory diagram schematically showing the result of orthogonal transformation with respect to the spectrum envelope. The horizontal axis shown in FIG. 4 represents time, and the vertical axis represents the orthogonal transform coefficient. Note that the orthogonal transform coefficient vector has n dimensions, and the results illustrated in FIG. 4 are obtained for each dimension. That is, in the example shown in FIG. 4, the axis in the depth direction (in other words, the axis perpendicular to the paper surface) is the axis of the dimension.

As illustrated in FIG. 4, orthogonality is obtained at a connection portion (near time t _bnd ) between the orthogonal transformation result with respect to the spectral envelope extracted from the preceding element and the orthogonal transformation result with respect to the spectral envelope extracted from the subsequent element. There is a gap in conversion coefficients. The smoothing processing unit 4 corrects the orthogonal transformation coefficient obtained by the orthogonal transformation to smooth the change from the orthogonal transformation coefficient on the preceding element side to the orthogonal transformation coefficient on the subsequent element side. The smoothing processing unit 4 eliminates the gap between orthogonal transform coefficients in the vicinity of the time t _bnd by this smoothing processing. The smoothing processing unit 4 performs this process for each dimension.

FIG. 5 is a block diagram illustrating an example of the smoothing processing unit 4. The smoothing processing unit 4 includes a reference value calculation unit 41 and a correction unit 42.

FIG. 6 is an explanatory diagram for illustrating processing of the reference value calculation unit 41. The orthogonal transformation coefficient before correction is represented as X (ω), and the orthogonal transformation coefficient after correction is represented as Y (ω). Note that ω is a frequency bin, and ωε {0, 1, 2,..., FFT_LEN−1}. In addition, when an orthogonal transform coefficient at a certain time is expressed, the time is added as a subscript. For example, the orthogonal transformation coefficient before correction at time t _end1 is _denoted as X _tend1 (ω). Further, the orthogonal transform coefficient before correction at time t _beg2 is _denoted as X _tbeg2 (ω). Similarly, the corrected orthogonal transformation coefficient Y (ω) is represented by adding the corresponding time as a subscript. Note that the orthogonal transformation coefficient before correction in an arbitrary pitch synchronization position time _{t p,} and the orthogonal transform coefficients after the correction, _{respectively, X tp (omega),} referred to as _{Y tp (ω).}

The reference value calculation unit 41 calculates a reference value that serves as a reference when the change of each orthogonal transform coefficient in the correction target period t _str to t _fin is smoothed. At the boundary time t _bnd , there is no orthogonal transformation coefficient before correction, but for convenience, an orthogonal transformation coefficient Y _tbnd (ω) after correction at the boundary time t _bnd is assumed and the orthogonal transformation coefficient Y _tbnd (ω). Is the reference value. _If the orthogonal transformation coefficient X _tend1 (ω) at the pitch synchronization position time t _{end1 at} the end of the preceding element side is less than the reference value, each orthogonal transformation coefficient on the preceding element side is increased, and the orthogonal transformation coefficient X _tend1 (ω) becomes If the reference value is exceeded, each orthogonal transform coefficient on the leading element side is reduced. Similarly, _if the orthogonal transformation coefficient X _tbeg2 (ω) at the pitch synchronization position time t _{beg2 at} the starting end on the subsequent element side is less than the reference value, each orthogonal transformation coefficient on the subsequent element side is increased, and the orthogonal transformation coefficient X _tbeg2 ( If ω) exceeds the reference value, each orthogonal transform coefficient on the subsequent segment side is decreased. The reference value calculation unit 41 uses, as the reference value Y _tbnd (ω), the orthogonal transformation coefficient X _tend1 (ω) at the pitch synchronization position time t _{end1 at} the end of the preceding element side and the pitch synchronization position time t at the _{start edge} of the subsequent element side. obtaining a value between the orthogonal transform coefficient _{X tbeg2 (ω)} in _{Beg2. That, X tend1 (ω) ≦ Y} tbnd (ω) ≦ X tbeg2 (ω), _or is asked to _{X tbeg2 (ω) ≦ Y tbnd} (ω) ≦ X tend1 reference value _{Y Tbnd} satisfying (ω) (ω) That's fine.

In this example, the reference value calculation unit 41 sets the average value of X _tend1 (ω) and X _tbeg2 (ω) as Y _tbnd (ω).

Depending on how the reference value is defined, Y _tbnd (ω) = X _tend1 (ω) or Y _tbnd (ω) = X _tbeg2 (ω) may be satisfied. When Y _tbnd (ω) = X _tend1 (ω), it is not necessary to change each orthogonal transform coefficient on the leading element side. Similarly, when Y _tbnd (ω) = X _tbeg2 (ω), it is not necessary to change each orthogonal transform coefficient on the subsequent _segment side.

Correcting unit 42, based on the reference value _{Y tbnd (ω),} the more the pitch synchronization position time _{t p} near the boundary time _{t bnd,} the amount of change from the orthogonal transform coefficient before correction is increased, far from the boundary time _{t bnd} as the pitch synchronization position time t _p, as the amount of change from the orthogonal transform coefficient before correction is reduced, it corrects the orthogonal transform coefficients of each pitch synchronization position time t _p belonging to the correction period t _{str ~} t _fin.

FIG. 7 is an explanatory diagram schematically showing the orthogonal transform coefficient corrected by the correction unit 42. As illustrated in FIG. 7, the correction of the orthogonal transform coefficient by the correction unit 42 eliminates the gap in the vicinity of the time t _bnd , and the change in the orthogonal transform coefficient with the passage of time is smoothed.

The inverse transform unit 5 performs the inverse transform of the orthogonal transform performed by the orthogonal transform unit 3 on the orthogonal transform coefficient corrected by the correction unit 42.

The inverse transform performed by the inverse transform unit 5 may be determined in advance according to the type of orthogonal transform performed by the orthogonal transform unit 3. For example, in the configuration in which the orthogonal transform unit 3 performs discrete Fourier transform, the inverse transform unit 5 performs inverse discrete Fourier transform (Inverse DFT). In the configuration in which the orthogonal transform unit 3 performs the fast Fourier transform, the inverse transform unit 5 performs the inverse fast Fourier transform (Inverse FFT). In the configuration in which the orthogonal transform unit 3 performs discrete cosine transform, the inverse transform unit 5 performs inverse discrete cosine transform (Inverse DCT). In the configuration in which the orthogonal transform unit 3 performs wavelet transform, the inverse transform unit 5 performs inverse wavelet transform.

The corrected spectral envelope is obtained by the inverse transformation by the inverse transformation unit 5.

The pitch waveform generation unit 6 generates a pitch waveform from the corrected spectral envelope (result of inverse transformation by the inverse transformation unit 5).

The superposition adding unit 7 generates a synthesized speech waveform by superposing and adding the pitch waveforms generated by the pitch waveform generating unit 6. A pitch pattern is input to the overlay adding unit 7. The pitch pattern is a time series of pitch frequencies. It is only necessary to sequentially input the pitch frequency corresponding to the pitch waveform generated in the pitch waveform generation unit 6 to the overlay addition unit 7. The superposition adding unit 7 determines the pitch waveform arrangement interval according to the input pitch frequency, and superimposes and adds the pitch waveforms.

Correction range determination unit 1, spectrum envelope extraction unit 2, orthogonal transformation unit 3, smoothing processing unit 4 (reference value calculation unit 41 and correction unit 42), inverse transformation unit 5, pitch waveform generation unit 6, and overlay addition unit 7 Is realized, for example, by a CPU of a computer that operates in accordance with a segment processing program. In this case, for example, a computer program storage device (not shown) stores the fragment processing program, and the CPU reads the program, and according to the program, the correction range determination unit 1, the spectrum envelope extraction unit 2, and the orthogonal transform unit 3. The smoothing processing unit 4 (the reference value calculation unit 41 and the correction unit 42), the inverse conversion unit 5, and the pitch waveform generation unit 6 may be operated. The correction range determination unit 1, the spectrum envelope extraction unit 2, the orthogonal transformation unit 3, the reference value calculation unit 41, the correction unit 42, the inverse transformation unit 5, the pitch waveform generation unit 6, and the overlay addition unit 7 are separate units. It may be realized with.

Next, the operation will be described.
FIG. 8 is a flowchart showing an example of processing progress of the fragment processing apparatus of the present invention. The individual segments selected in accordance with the reading of the text to be subjected to speech synthesis are sequentially input to the correction range determination unit 1 according to the reading order. The correction range determination unit 1 determines the correction target period of the orthogonal transform coefficient for two consecutive pieces (step S1). As already described, the correction range determination unit 1 determines that the correction target period includes the segment boundary time t _bnd (first condition) and the center time t _c1 of the preceding segment period (see FIG. 2). The correction target period so as to satisfy the condition (second condition) that the correction target period does not include the period before the center time t _c2 (see FIG. 2) of the period before) and the subsequent segment period (see FIG. 2). Can be determined.

Next, the spectrum envelope extracting unit 2 extracts the spectrum envelope corresponding to the correction target period from the portion overlapping the correction target period in the preceding element, and similarly, from the part overlapping the correction target period in the subsequent element, A spectrum envelope corresponding to the portion is extracted (step S2). In step S2, the spectrum envelope extraction unit 2 may extract the spectrum envelope from the segment by performing, for example, STRIGHT analysis. Alternatively, the spectral envelope may be extracted from the segment by other methods such as linear prediction analysis, LSP analysis, and cepstrum analysis. Through the processing in step S2, the spectrum envelope illustrated in FIG. 3 is obtained for each subband.

Next, the orthogonal transform unit 3 performs orthogonal transform on the spectrum envelope extracted in step S2 (step S3). As a result, an orthogonal transform coefficient vector is obtained for each pitch synchronization position time. In other words, the relationship between the time and the orthogonal transformation coefficient as illustrated in FIG. 4 is obtained for each dimension. As already explained, in the example shown in FIG. 4, the axis in the depth direction is the axis of the dimension.

In step S3, the orthogonal transform unit 3 may perform, for example, a fast Fourier transform as the orthogonal transform. Here, a case where fast Fourier transform is performed as the orthogonal transform is illustrated, but the orthogonal transform unit 3 may perform other orthogonal transform such as discrete Fourier transform, discrete cosine transform, and wavelet transform.

Next, the reference value calculation unit 41 corrects the orthogonal transformation coefficient X _tend1 (ω) before correction at the end pitch synchronization position time t _end1 on the preceding unit side and the correction at the pitch synchronization position time t _beg2 on the start side of the subsequent unit side. A reference value Y _tbnd (ω) is calculated based on the previous orthogonal transform coefficient X _tbeg2 (ω) (step S4). In this example, the reference value calculation unit 41 uses an average value of X _tend1 (ω) and X _tbeg2 (ω) as a reference value. That is, in this example, the reference value calculation unit 41 may calculate the reference value Y _tbnd (ω) by calculating the following equation (1).

Y _tbnd (ω) = {X _tend1 (ω) + X _tbeg2 (ω)} / 2 Formula (1)

Equation (1) is an example of a method for calculating the reference value Y _tbnd (ω). If X _tend1 (ω) ≦ Y _tbnd (ω) ≦ X _tbeg2 (ω), or X _tbeg2 (ω) ≦ Y _tbnd (ω) ≦ X _tend1 (ω), the reference value Y is satisfied by another method. _tbnd (ω) may be calculated.

After step S4, the correction unit 42 corrects the orthogonal transform coefficients in the pitch synchronization position time t _p in the correction target period (step S5). By this correction, the correction unit 42 smoothes the change from the orthogonal transformation coefficient on the preceding element side to the orthogonal transformation coefficient on the subsequent element side.

Specifically, the correction unit 42 uses the reference value _Y tbnd _(ω), in each pitch synchronization position time t _p of the period of the preceding element side in the correction target period (the period before the boundary time t _bnd) The orthogonal transformation coefficient of is corrected. Similarly, the correction unit 42 uses the reference value _Y tbnd _(ω), orthogonal at each pitch synchronization position time t _p of the period of the subsequent element side in the correction target period (period after the boundary time t _bnd) Correct the conversion factor.

First, correction in a period before the segment boundary time t _bnd will be described. FIG. 9 is an explanatory diagram schematically illustrating the correction in the period before the boundary time t _bnd in the correction target period. FIG. 9 illustrates a case where the orthogonal transformation coefficient is increased in a period before the reference time Y _tbnd (ω) is larger than X _tend1 (ω) and before the boundary time t _bnd .

In this correction _process, is regarded as a _{t end1} ≒ _{t bnd,} obtains the _{Y tp} (omega) so as to satisfy the following equation (2).

The left-hand side of equation (2), the correction period of the starting time _{t str} time to the boundary time _{t bnd} the segment from for (see FIG. 9), the time _{t str} arbitrary pitch synchronization position time _{t p} to the time from ( (See FIG. 9). The proportion of the right side of the equation (2) _is for _{"Y tbnd (ω) -X tend1} (ω)", the change amount of the correction at the pitch synchronization position time _{t p "Y tp (ω)} -X tp (ω)" It is. That is, Equation (2) _is regarded as _{t end1} ≒ _{t bnd,} represents that the two above-mentioned ratio is the same.

Then, by obtaining the Y _{tp (omega)} to satisfy equation (2), as the pitch synchronizing position time t _p near the boundary time t _bnd, the amount of change from the orthogonal transform coefficient before correction is increased, boundary time t farther pitch synchronization position time _{t p} from _bnd, as the amount of change from the orthogonal transform coefficient before correction is reduced, it is possible to correct the _{X tp} (omega) in the period _t str _{~ t bnd.}

Solving equation (2) with respect to Y _tp (ω) yields equation (3) below.

That is, the correction unit 42, by performing the calculation of the above formula (3) with respect to each pitch synchronization position time t _p for the period before the boundary time t _bnd, orthogonal corrected at each pitch synchronization position time t _p A conversion coefficient Ytp (ω) is calculated. As a result, as the pitch synchronization position time t _p near the boundary time t _bnd, correction change from the orthogonal transformation coefficient before increases, the farther the pitch synchronization position time t _p from the boundary time t _bnd, before correction orthogonal transform The amount of change from the coefficient is small.

Even when the reference value Y _tbnd (ω) is smaller than X _tend1 (ω) and the orthogonal transform coefficient is decreased in the period before the boundary time t _bnd , the correction unit 42 uses the equation (3). The corrected orthogonal transformation coefficient Y _tp (ω) may be calculated by the above calculation.

Next, correction in a period after the segment boundary time t _bnd will be described. FIG. 10 is an explanatory diagram schematically illustrating correction in a period after the boundary time t _bnd in the correction target period. FIG. 10 illustrates a case where the orthogonal transformation coefficient is increased in a period after the reference value Y _tbnd (ω) is larger than X _tbeg2 (ω) and after the boundary time t _bnd .

In this correction _process, is regarded as a _{t beg2} ≒ _{t bnd,} obtains the _{Y tp} (omega) so as to satisfy the following equation (4).

Equation (4) left-hand side of the time from the segment of boundary time _{t bnd} to the end time _{t fin} of the correction target period for (see FIG. 10), the time from an arbitrary pitch synchronization position time _{t p} to time _{t fin} ( (See FIG. 10). The proportion of the right side of the equation (4) _is for the _{"Y tbnd (ω) -X tbeg2} (ω)", the change amount of the correction at the pitch synchronization position time _{t p "Y tp (ω)} -X tp (ω)" It is. That is, equation (4) _is regarded as _{t beg2} ≒ _{t bnd,} represents that the two above-mentioned ratio is the same.

Then, by obtaining the Y _{tp (omega)} to satisfy equation (4), as the pitch synchronizing position time t _p near the boundary time t _bnd, the amount of change from the orthogonal transform coefficient before correction is increased, boundary time t farther pitch synchronization position time _{t p} from _bnd, as the amount of change from the orthogonal transform coefficient before correction is reduced, it is possible to correct the _{X tp} (omega) in the period _t bnd _{~ t fin.}

When Formula (4) is solved with respect to Y _tp (ω), Formula (5) shown below is obtained.

That is, the correction unit 42, by performing the calculation of the above equation (5) with respect to each pitch synchronization position time t _p in the period after the boundary time t _bnd, orthogonal corrected at each pitch synchronization position time t _p A conversion coefficient Y _tp (ω) is calculated. As a result, as the pitch synchronization position time t _p near the boundary time t _bnd, correction change from the orthogonal transformation coefficient before increases, the farther the pitch synchronization position time t _p from the boundary time t _bnd, before correction orthogonal transform The amount of change from the coefficient is small.

Even when the reference value Y _tbnd (ω) is smaller than X _tbeg2 (ω) and the orthogonal transform coefficient is decreased in the period after the boundary time t _bnd , the correction unit 42 is configured to use the equation (5). The corrected orthogonal transformation coefficient Y _tp (ω) may be calculated by the above calculation.

The reference value calculation unit 41 and the correction unit 42 perform the processes of steps S4 and S5 for each dimension of the orthogonal transform coefficient.

After step S5, the inverse transform unit 5 performs the inverse transform of the orthogonal transform in step S3 on the corrected orthogonal transform coefficient obtained in step S5 (step S6). If the fast Fourier transform is performed in step S3, the inverse transform unit 5 may perform the inverse fast Fourier transform. As a result of step S6, a spectrum envelope is obtained. This spectral envelope can be said to be a corrected spectral envelope.

The pitch waveform generation unit 6 converts the corrected spectral envelope (the spectral envelope obtained as a result of step S6) into a time domain waveform (that is, a pitch waveform) using inverse Fourier transform (step S7). In this conversion, an appropriate phase spectrum is used. The pitch waveform generation unit 6 may use, for example, a zero phase or a fixed phase as the phase spectrum. Further, if the phase before extraction of the spectral envelope in step S2 can be used, that phase may be used.

The spectral envelope that is corrected in the pitch synchronization position time t _p and Y _{tp (ω).} Also, the phase used when generating the pitch waveform is θ (ω). At this time, the pitch waveform generation unit 6 generates the pitch waveform y (t) by calculating the following equation (6).

y _tp (t) = F ⁻¹ {Y _tp (ω) e ^{jθ (ω)} } Equation (6)

In Formula (6), F ⁻¹ {·} represents inverse Fourier transformation. If the phase used when generating the pitch waveform is, for example, a zero phase, θ (ω) = 0 in equation (6).

Time domain waveform (pitch waveform) is represented by individual time and amplitude at each time.

The superposition adding unit 7 generates a synthesized speech waveform by superposing and adding the pitch waveforms generated by the pitch waveform generating unit 6 (step S8). The superposition adding unit 7 may determine the pitch waveform arrangement interval in accordance with the input pitch frequency and superimpose and add the pitch waveforms.

According to the present embodiment, the orthogonal transform coefficient obtained by orthogonal transform of the spectral envelope is smoothed, and the gap of the orthogonal transform coefficient in the vicinity of the time t _bnd is eliminated. Then, an inverse transform of the orthogonal transform is performed on the smoothed (in other words, corrected) orthogonal transform coefficient to generate a spectrum envelope, and a pitch waveform is generated from the spectrum envelope. Therefore, since the external shape of the spectrum is adjusted, it is possible to reduce the large local deformation of the spectrum. In addition, since the pitch waveforms thus obtained are superimposed and added, the continuity of the audible sound at the connecting portion of the segments can be improved.

Also, as described above, the spectral envelope is extracted, and the orthogonal transformation result (orthogonal transformation coefficient) for the spectral envelope is smoothed. Therefore, the influence of pitch can be eliminated during smoothing.

Next, a modification of the above embodiment will be described. In the above embodiment, the smoothing processing unit 4 includes the reference value calculation unit 41 and the correction unit 42, the reference value calculation unit 41 calculates the reference value, and the correction unit 42 uses the reference value to perform the orthogonal transform coefficient. The case of correcting the above has been described. In each modification shown below, the orthogonal transform coefficient is corrected by another method.

FIG. 11 is a block diagram showing a modification of the embodiment of the present invention. Elements other than the smoothing processing unit 14 are the same as those described in the above embodiment. Elements other than the smoothing processing unit 14 are denoted by the same reference numerals as those in FIG. In addition, the smoothing process part 14 is implement | achieved by CPU of the computer which operate | moves according to a segment processing program, for example.

In the above embodiment, in accordance with the pitch period of the segment, although the case where the spectral density and the orthogonal transform coefficient vector is calculated for each pitch synchronization position time t _p example, in the modification described below, An example will be described in which the spectral density and the orthogonal transform coefficient vector are calculated at a fixed period from the start time _tstr of the correction target period. This period is _{t s.} In addition, the correction period and _{t str ~ t str + (N} -1) · t s. In this case, referred to N the number of frames correction period, the t _s can also be referred to as a length of a frame-shift correction period.

At this time, the orthogonal transformation coefficient before correction at time t is assumed to be X _t (ω). The corrected orthogonal transform coefficient at time t is Y _t (ω). However, the time t is expressed as follows.

Time tε {t _str , t _str + t _s , t _str + 2t _s ,..., T _str + (N−1) · t _s }

Ω is a frequency bin, and ω∈ {0, 1, 2,..., FFT_LEN−1}.

The operations in steps S1 to S3 and steps S6 to S8 (see FIG. 8) are the same as those in the above embodiment. Then, instead of the above-described steps S4 and S5 (see FIG. 8), the smoothing processing unit 14 may perform the following processing to calculate the corrected orthogonal transformation coefficient Y _t (ω). That is, the smoothing processing unit 14 calculates the following orthogonal transformation coefficient Y _t (ω) by performing the calculation of Expression (7) below for all ω and t.

In the equation _{(7), X tstr + k} · ts (ω) is an orthogonal transform coefficient before correction at time _{t str} + k · _{t s.}

M in Equation (7) is a moving average window width. Further, X _t (ω) = 0 is set for t in the range of t ≦ 0 or t _str + N · t _s ≦ t.

The smoothed orthogonal transform coefficient Y _t (ω) can be obtained by the calculation of Expression (7). In this method, each orthogonal transform coefficient is corrected by a moving average.

Further, the smoothing processing unit 14 may calculate the corrected orthogonal transformation coefficient Y _t (ω) by performing the calculation of the following equation (8) for all ω and t.

Y _t (ω) = αX _t (ω) + (1−α) Y _t−ts (ω) Equation (8)

In the formula _{(8), Y t-ts} (ω) is an orthogonal transform coefficients after the correction at time _{t-t s.}

Α in equation (8) is a time constant. However, Y _tstr−ts (ω) = 0. That is, the time _t str orthogonal transform coefficients after the correction in _{-t s Y tstr-ts (ω} ) is assumed to be 0.

Equation (8) is obtained by multiplying the orthogonal transformation coefficient X _t (ω) before correction at time t by the time constant and the orthogonal transformation coefficient Y _t-ts (ω) after correction at the time before that fixed period. 1 is obtained by calculating the sum of the result obtained by multiplying the value obtained by subtracting the time constant from 1 and obtaining the corrected orthogonal transform coefficient Y _t (ω) at time t. The smoothed orthogonal transform coefficient Y _t (ω) can be obtained by the calculation of Expression (8). This method can be said to be smoothing using first-order leak integration.

Further, the smoothing processing unit 14 may calculate the corrected orthogonal transformation coefficient Y _t (ω) as follows.

The smoothing processing unit 14 selects one ω and performs the following processing on the ω. First, the smoothing processing unit 14 calculates the following equation (9) in the order of t = t _str , t _str + t _s , t _str + 2t _{s ,.}

In addition, in Formula (9), (a) described as a subscript is a code | symbol for distinguishing from the calculation result of Formula (11) mentioned later.

Α ₁ in equation (9) is a time constant. Further, it is assumed that the following expression (10) is established. It should be noted that, the left-hand side of the subscript of the formula (10), refers to the time _t str -t _s.

It can be said that the value calculated by Equation (9) is the first provisional corrected orthogonal transform coefficient. Expression (9) is obtained by multiplying the orthogonal transformation coefficient X _t (ω) before correction at time t by the first time constant α ₁ and the first provisional correction at the time before that fixed period. the orthogonal transform coefficients after, by calculating the sum of the result of multiplying the value of the constant alpha ₁ is subtracted when the 1 of the first, obtaining the orthogonal transform coefficients of the first tentative corrected at time t Means that.

Next, the smoothing processing unit 14 performs t = t _str + (N−1) · t _s , t _str + (N−2) · t _s , t _str + (N−3) · t _s ,. The following formula (11) is calculated in the order of.

In addition, (b) described as a subscript in Formula (11) is a code | symbol for distinguishing from the calculation result of the above-mentioned Formula (9).

Α ₂ in equation (11) is a time constant. Further, it is assumed that the following expression (12) is established. It should be noted that, the left-hand side of the subscript of the formula (12), refers to the time _t str + N · _{t s.}

It can be said that the value calculated by Equation (11) is the second provisional corrected orthogonal transform coefficient. Expression (11) is obtained by multiplying the orthogonal transformation coefficient X _t (ω) before correction at time t by the second time constant α ₂ and the second provisional correction at the time after a certain period. the orthogonal transform coefficients after, by calculating the sum of the result of multiplying the 1 second time value constant alpha ₂ is subtracted to obtain the orthogonal transform coefficients of the second provisional corrected at time t Means that.

Subsequently, the smoothing processing unit 14 calculates the average value of the calculation result of Expression (9) and the calculation result of Expression (11) for all t, thereby correcting the orthogonal transform coefficient Y _t (ω after correction). ) Is calculated. That is, the smoothing processing unit 14 calculates Y _t (ω) by performing the calculation of Expression (13) shown below for all t.

After calculating Y _t (ω) for all t, the smoothing processing unit 14 selects the next ω and repeats the same processing. The smoothing processing unit 14 ends the process when there is no unselected ω.

The smoothed orthogonal transform coefficient Y _t (ω) can also be obtained by the above processing. This method can also be said to be smoothing using first-order leak integration.

In each of the above modifications, the same effect as that of the above-described embodiment can be obtained.

Next, the minimum configuration of the present invention will be described. FIG. 12 is a block diagram showing an example of the minimum configuration of the fragment processing apparatus of the present invention. The segment processing apparatus of the present invention includes a spectrum envelope extracting unit 71, an orthogonal transform unit 72, a smoothing unit 73, an inverse transform unit 74, and a time domain waveform generating unit 75.

The spectrum envelope extracting means 71 (for example, the spectrum envelope extracting unit 2) extracts a spectrum envelope in a period (for example, a correction target period) including a boundary time between the two elements from two consecutive elements.

Orthogonal transformation means 72 (for example, orthogonal transformation unit 3) performs orthogonal transformation on the spectrum envelope.

The smoothing means 73 (for example, the smoothing processing unit 4) smoothes the orthogonal transform coefficient that is the expansion coefficient obtained as a result of performing the orthogonal transform on the spectrum envelope.

The inverse transform means 74 (for example, the inverse transform unit 5) performs inverse transform of orthogonal transform on the corrected orthogonal transform coefficient.

The time domain waveform generation means 75 (for example, the pitch waveform generation unit 6) generates a time domain waveform from a spectrum envelope obtained as a result of performing inverse transformation of orthogonal transformation on the corrected orthogonal transformation coefficient.

Such a configuration makes it possible to match the outer shape of the spectrum, so that the local deformation of the spectrum can be reduced. Moreover, the continuity of the audible sound at the connecting portion of the segment can be improved.

Some or all of the above-described embodiment and its modifications may be described as in the following supplementary notes, but are not limited to the following.

(Supplementary note 1) Spectral envelope extraction means for extracting a spectrum envelope in a period including a boundary time between the two pieces from two consecutive pieces, and orthogonal transform means for performing orthogonal transformation on the spectrum envelope, Smoothing means for smoothing orthogonal transform coefficients, which are expansion coefficients obtained as a result of performing orthogonal transform on the spectrum envelope, and inverse transform means for performing inverse transform of the orthogonal transform on the corrected orthogonal transform coefficients And a time domain waveform generating means for generating a time domain waveform from a spectrum envelope obtained as a result of performing the inverse transformation of the orthogonal transformation on the corrected orthogonal transformation coefficient.

(Supplementary Note 2) The smoothing means includes an orthogonal transform coefficient at a time closest to a boundary time between the two segments among the orthogonal transform coefficients on the previous segment side in two consecutive segments, and the two segments. Among the orthogonal transform coefficients on the element side after the reference time, the reference determining means for determining the value between the orthogonal transform coefficient at the time closest to the boundary time as the orthogonal transform coefficient at the boundary time, and the orthogonal transform at the boundary time The segment processing apparatus according to appendix 1, including correction means for correcting the orthogonal transform coefficient on the previous segment side and the orthogonal transform coefficient on the subsequent segment side based on the coefficient.

(Additional remark 3) As for the correction | amendment means, the variation | change_quantity from the orthogonal transformation coefficient before correction | amendment becomes large as the time near the boundary time of two segments, and the variation | change_quantity from the orthogonal transformation coefficient before correction | amendment becomes the time far from the said boundary time. The segment processing apparatus according to appendix 2, wherein the orthogonal transform coefficient on the previous segment side and the orthogonal transform coefficient on the subsequent segment side are corrected so that becomes smaller.

(Supplementary note 4) The segment processing apparatus according to supplementary note 1, wherein the smoothing unit corrects the orthogonal transform coefficient on the previous segment side and the orthogonal transform coefficient on the subsequent segment side by moving average.

(Supplementary Note 5) The smoothing means multiplies the result obtained by multiplying the orthogonal transform coefficient before correction by a constant, and the result obtained by multiplying the orthogonal transform coefficient after correction at a time before a certain period by a value obtained by subtracting the constant from 1. 2. The element processing device according to appendix 1, wherein the orthogonal transformation coefficient on the previous element side and the orthogonal transformation coefficient on the subsequent element side are corrected by calculating the sum of each time.

(Supplementary note 6) The smoothing means converts the first orthogonal transform coefficient from 1 to the result obtained by multiplying the orthogonal transform coefficient before correction by the first constant and the first provisional corrected orthogonal transform coefficient at a time before a predetermined period. By calculating the sum of the result of subtracting the constant of 1 and the result of multiplication for each time, the first provisional corrected orthogonal transform coefficient at each time is calculated in time order, and the orthogonal transform coefficient before correction And the result of multiplying the second constant by a value obtained by subtracting the second constant from the second provisional corrected orthogonal transform coefficient at a time after a fixed period. By calculating for each time, the second provisional corrected orthogonal transform coefficient at each time is calculated in the reverse order of the time order, and the first provisional corrected orthogonal transform coefficient at each time and the first The average value of the two orthogonal correction coefficients after tentative correction is calculated at each time. Fragment processing apparatus according to note 1, calculated as the orthogonal transform coefficients after the correction.

(Supplementary note 7) A spectrum envelope in a period including a boundary time between the two segments is extracted from two consecutive segments, orthogonal transform is performed on the spectrum envelope, and orthogonal transform is performed on the spectrum envelope. Smoothing the orthogonal transformation coefficient, which is the expansion coefficient obtained as a result of performing, performing the inverse transformation of the orthogonal transformation on the corrected orthogonal transformation coefficient, and performing the inverse transformation of the orthogonal transformation on the corrected orthogonal transformation coefficient A segment processing method, comprising: generating a time domain waveform from a spectrum envelope obtained as a result of performing.

(Supplementary note 8) Spectral envelope extraction processing for extracting a spectral envelope in a period including a boundary time between the two segments from two continuous segments, and orthogonal transformation processing for performing orthogonal transformation on the spectral envelope A smoothing process for smoothing an orthogonal transform coefficient, which is an expansion coefficient obtained as a result of performing an orthogonal transform on the spectrum envelope, and an inverse transform process for performing an inverse transform of the orthogonal transform on the corrected orthogonal transform coefficient And a segment processing program for executing time domain waveform generation processing for generating a time domain waveform from a spectrum envelope obtained as a result of performing inverse transformation of the orthogonal transformation on the corrected orthogonal transformation coefficient.

(Supplementary Note 9) A spectrum envelope extraction unit that extracts a spectrum envelope in a period including a boundary time between the two units from two consecutive units, an orthogonal transform unit that performs orthogonal transform on the spectrum envelope, A smoothing unit that smoothes orthogonal transform coefficients that are expansion coefficients obtained as a result of performing orthogonal transform on the spectrum envelope, and an inverse transform unit that performs inverse transform of the orthogonal transform on the corrected orthogonal transform coefficient And a time domain waveform generation unit that generates a time domain waveform from a spectrum envelope obtained as a result of performing inverse transformation of the orthogonal transformation on the corrected orthogonal transformation coefficient.

(Supplementary Note 10) The smoothing unit includes an orthogonal transform coefficient at a time closest to a boundary time of the two segments among the orthogonal transform coefficients on the previous segment side in two consecutive segments, and the two segments A reference determining unit that determines a value between the orthogonal transformation coefficient at the time closest to the boundary time as the orthogonal transformation coefficient at the boundary time, and the orthogonal transformation at the boundary time The segment processing apparatus according to appendix 9, further comprising: a correction unit that corrects the orthogonal transform coefficient on the previous segment side and the orthogonal transform coefficient on the subsequent segment side based on the coefficient.

(Supplementary Note 11) The correction unit increases the amount of change from the orthogonal transformation coefficient before correction as the time is closer to the boundary time between the two segments, and the amount of change from the orthogonal transformation coefficient before correction as the time is farther from the boundary time. The segment processing apparatus according to appendix 10, wherein the orthogonal transform coefficient on the previous segment side and the orthogonal transform coefficient on the subsequent segment side are corrected so as to be smaller.

(Supplementary note 12) The segment processing apparatus according to supplementary note 9, wherein the smoothing unit corrects the orthogonal transform coefficient on the previous segment side and the orthogonal transform coefficient on the subsequent segment side by moving average.

(Supplementary note 13) The smoothing unit multiplies the result of multiplying the orthogonal transform coefficient before correction by a constant, and the result of multiplying the orthogonal transform coefficient after correction at a time before a certain period by a value obtained by subtracting the constant from 1. The segment processing apparatus according to appendix 9, wherein the orthogonal transform coefficient on the previous segment side and the orthogonal transform coefficient on the subsequent segment side are corrected by calculating the sum of the values at each time.

(Supplementary Note 14) The smoothing unit converts the result of multiplying the orthogonal transformation coefficient before correction by the first constant and the first provisional corrected orthogonal transformation coefficient at a time before a predetermined period from 1 to the first By calculating the sum of the result of subtracting the constant of 1 and the result of multiplication for each time, the first provisional corrected orthogonal transform coefficient at each time is calculated in time order, and the orthogonal transform coefficient before correction And the result of multiplying the second constant by a value obtained by subtracting the second constant from the second provisional corrected orthogonal transform coefficient at a time after a fixed period. By calculating for each time, the second provisional corrected orthogonal transform coefficient at each time is calculated in the reverse order of the time order, and the first provisional corrected orthogonal transform coefficient at each time and the first The average value of the two orthogonal correction coefficients after tentative correction is calculated at each time. Fragment processing apparatus according to note 9 to calculate the orthogonal transformation coefficient after correction.

This application claims priority based on Japanese Patent Application 2011-165707 filed on July 28, 2011, the entire disclosure of which is incorporated herein.

The present invention has been described above with reference to the embodiments, but the present invention is not limited to the above-described embodiments. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

Industrial applicability

The present invention is preferably applied to a segment processing apparatus that performs processing on segments so that the segments are connected well in segment selection type speech synthesis.

DESCRIPTION OF SYMBOLS 1 Correction range determination part 2 Spectral envelope extraction part 3 Orthogonal transformation part 4,14 Smoothing process part 5 Inverse transformation part 6 Pitch waveform generation part 7 Overlay addition part 41 Reference value calculation part 42 Correction part

Claims (8)

1. Spectrum envelope extraction means for extracting a spectrum envelope in a period including a boundary time between the two segments from two consecutive segments;
   Orthogonal transform means for performing orthogonal transform on the spectral envelope;
   Smoothing means for smoothing orthogonal transform coefficients, which are expansion coefficients obtained as a result of performing orthogonal transform on the spectral envelope;
   Inverse transform means for performing inverse transform of the orthogonal transform on the corrected orthogonal transform coefficient;
   A segment processing apparatus comprising: time domain waveform generation means for generating a time domain waveform from a spectrum envelope obtained as a result of performing inverse transformation of the orthogonal transformation on the corrected orthogonal transformation coefficient.
2. The smoothing means is
   Of the orthogonal transformation coefficients on the preceding element side in the two consecutive elements, the orthogonal transformation coefficient at the time closest to the boundary time of the two elements, and the orthogonal transformation coefficient on the subsequent element side in the two elements A reference determining means for determining a value between an orthogonal transform coefficient at a time closest to the boundary time as an orthogonal transform coefficient at the boundary time;
   The segment processing apparatus according to claim 1, further comprising: a correcting unit that corrects the orthogonal transform coefficient on the previous unit side and the orthogonal transform coefficient on the subsequent unit side based on the orthogonal transform coefficient at the boundary time.
3. The correcting means increases the amount of change from the orthogonal transformation coefficient before correction as the time is closer to the boundary time between the two segments, and decreases the amount of change from the orthogonal transformation coefficient before correction as the time is farther from the boundary time. The segment processing apparatus according to claim 2, wherein the orthogonal transform coefficient on the previous segment side and the orthogonal transform coefficient on the subsequent segment side are corrected.
4. The segment processing apparatus according to claim 1, wherein the smoothing unit corrects the orthogonal transform coefficient on the previous segment side and the orthogonal transform coefficient on the subsequent segment side by moving average.
5. The smoothing means calculates the sum of the result obtained by multiplying the orthogonal transform coefficient before correction by a constant and the result obtained by multiplying the orthogonal transform coefficient after correction at a time before a predetermined period by the value obtained by subtracting the constant from 1. The segment processing apparatus according to claim 1, wherein the orthogonal transform coefficient on the previous segment side and the orthogonal transform coefficient on the subsequent segment side are corrected by calculating each time.
6. The smoothing means is
   Multiply the result of multiplying the orthogonal transformation coefficient before correction by the first constant and the first provisional corrected orthogonal transformation coefficient at a time before a certain period by the value obtained by subtracting the first constant from 1 By calculating the sum with the result for each time, the first provisional corrected orthogonal transform coefficient at each time is calculated in order of time,
   Multiply the result of multiplying the orthogonal transformation coefficient before correction by the second constant and the value obtained by subtracting the second constant from 1 to the second provisional corrected orthogonal transformation coefficient at a time after a certain period. By calculating the sum with the result for each time, the second provisional corrected orthogonal transform coefficient at each time is calculated in the reverse order of the time order,
   The average value of the first provisional corrected orthogonal transform coefficient and the second provisional corrected orthogonal transform coefficient at each time is calculated as the corrected orthogonal transform coefficient at each time. The fragment processing apparatus described.
7. Extracting the spectral envelope in a period including the boundary time between the two segments from two consecutive segments,
   Performing orthogonal transform on the spectral envelope;
   Smoothing the orthogonal transform coefficient, which is the expansion coefficient obtained as a result of performing orthogonal transform on the spectral envelope,
   The inverse transform of the orthogonal transform is performed on the corrected orthogonal transform coefficient,
   A segment processing method, comprising: generating a time domain waveform from a spectrum envelope obtained as a result of performing inverse transformation of the orthogonal transformation on the corrected orthogonal transformation coefficient.
8. On the computer,
   A spectral envelope extraction process for extracting a spectral envelope in a period including a boundary time between the two segments from two consecutive segments;
   An orthogonal transformation process for performing orthogonal transformation on the spectrum envelope;
   A smoothing process for smoothing orthogonal transform coefficients, which are expansion coefficients obtained as a result of performing orthogonal transform on the spectral envelope;
   An inverse transform process for performing the inverse transform of the orthogonal transform on the corrected orthogonal transform coefficient, and
   A segment processing program for executing time domain waveform generation processing for generating a time domain waveform from a spectrum envelope obtained as a result of performing inverse transformation of the orthogonal transformation on the corrected orthogonal transformation coefficient.

Images