JP6047952B2 - Speech synthesis apparatus and speech synthesis method - Google Patents

Description

  The present invention relates to a technique for synthesizing sounds such as speech sounds and singing sounds using speech segments.

  2. Description of the Related Art A unit connection type speech synthesizer that synthesizes a desired speech by connecting a plurality of speech units to each other has been proposed. Patent Document 1 discloses a technique for controlling the intelligibility of a synthesized sound (a degree of opening of a virtual speaker's mouth) by partially using speech segments.

  For example, as shown in FIG. 16, a speech unit V [w-a] including a phoneme section S1 of consonant (semi-vowel) phoneme / w / and a phoneme section S2 of vowel phoneme / a / is assumed. The phoneme section S2 is divided into a steady section EB in which the phoneme / a / waveform is maintained constantly, and a transition section EA in which the phoneme / w / in the immediately preceding phoneme section S1 transitions to the phoneme / a / waveform. . In the technique of Patent Document 1, the boundary time point (phoneme segmentation boundary) TB is variably set in the transition section EA, and the last frame of the section is repetitively connected to the section of the speech unit V before the boundary time TB. Thus, an audio signal having a desired length of time is generated. The position of the boundary time TB is variably set according to a variable instructed by the user (hereinafter referred to as “intelligibility variable”). According to the above configuration, it is possible to synthesize a voice having a low clarity (that is, a voice uttered without sufficiently opening the mouth) as compared with the case where the entire transition section EA is applied.

Japanese Patent No. 4265501

  In the technique of Patent Document 1, a configuration is assumed in which the boundary time point TB in the transition section EA is linearly moved on the time axis according to the clarity variable specified by the user. In other words, the time length from the start point of the transition section EA to the boundary time TB is proportional to the clarity variable specified by the user.

  On the other hand, a graph g in FIG. 16 is a graph schematically showing a temporal transition of a timbre (a degree of opening of a speaker's mouth) perceived by a listener of the synthesized sound. As shown in FIG. 16, the timbre of the speech segment V transitions from the timbre of the phoneme segment S1 to the timbre of the phoneme segment S2 from the timbre of the phoneme segment S1 from the start point to the end point of the transition segment EA. As understood from the graph g, the timbre perceived by the listener changes nonlinearly with respect to the elapsed time. That is, the timbre changes remarkably just after a short time in the vicinity of the start point of the transition section EA, but the timbre hardly changes even in the vicinity of the end point of the transition section EA.

  Therefore, when the boundary time TB is linearly moved according to the clarity variable specified by the user as described above, the clarity variable is slightly changed in a state where the boundary time TB is located in the vicinity of the start point of the transition section EA. The timbre of the synthesized sound changes noticeably, but the timbre of the synthesized sound hardly changes even if the intelligibility variable is similarly changed in a state where the boundary time TB is located near the end point of the transition section EA. As described above, since the change of the clarity variable and the timbre change of the synthesized sound do not match sensuously, it is difficult for the user to set the clarity variable so that the synthesized sound becomes a desired timbre. There is. In view of the above circumstances, an object of the present invention is to match the change in the clarity variable according to the instruction from the user with the timbre change in the synthesized sound.

  Means employed by the present invention to solve the above problems will be described. In order to facilitate the understanding of the present invention, in the following description, the correspondence between the elements of the present invention and the elements of the embodiments described later will be indicated in parentheses, but the scope of the present invention will be exemplified in the embodiments. It is not intended to be limited.

  The speech synthesizer of the present invention has a unit selection unit (for example, unit selection) for sequentially selecting speech units (for example, speech unit V) including a transition section (for example, transition section EA) in which the speech waveform changes over time. Section 34), variable setting means (for example, variable setting section 33) for variably setting an intelligibility variable (for example, intelligibility variable α) in accordance with an instruction from a user, and a boundary time point (for example, a boundary time point) in the transition section TB) is a means for setting, and the amount of movement at the boundary time (for example, the amount of change Δβ1) when the intelligibility variable changes by a predetermined amount from the first value (for example, numerical value α1), and the intelligibility variable has the first value According to the articulation variable set by the variable setting means so that the movement amount (for example, the change amount Δβ2) at the boundary time when it changes by a predetermined amount from a different second value (for example, the numerical value α2). Boundary setting means for variably setting the position of the boundary point; ; And a synthesis processing means for generating an audio signal (e.g., the synthesis processing section 46) segment selection means by utilizing the forward section of the boundary point of the speech unit selection (e.g. applying section W). In the above configuration, the position of the boundary time changes non-linearly with respect to the articulation variable according to the instruction from the user. Therefore, even when the timbre perceived by the listener from the speech segment changes nonlinearly with respect to the elapsed time, it is possible to match the change in the clarity variable according to the instruction from the user with the timbre change in the synthesized sound. Is possible.

  In a preferred aspect of the present invention, the boundary setting means sets the position of the boundary time according to the clarity variable so that the relationship between the clarity variable and the position of the boundary time is common to a plurality of speech segments. In the above aspect, since the relationship between the clarity variable and the position of the boundary time is common to a plurality of speech segments, there is an advantage that the setting of the boundary time by the boundary setting means is simplified. In addition, the specific example of the above aspect is later mentioned, for example as 1st Embodiment.

  In a preferred aspect of the present invention, each of the plurality of speech segments includes a first phoneme section (for example, phoneme section S1) and a second phoneme section (for example, phoneme section S2) corresponding to different phonemes, The phoneme section includes a transition section, and the boundary setting means causes the relationship between the clarity variable and the position at the boundary time to differ depending on the type of phoneme in the first phoneme section (for example, type C) in the speech segment. Next, the position of the boundary time is set according to the clarity variable. In the above aspect, since the relationship between the clarity variable and the position of the boundary time is individually set according to the phoneme type of the first phoneme section, the temporal transition of the timbre of the speech segment is the first phoneme section. Even when different depending on the type of phoneme, an effect that the change of the clarity variable and the timbre change of the synthesized sound can be matched is realized. In addition, the specific example of the above aspect is later mentioned as 2nd Embodiment, for example.

  The speech synthesizer according to a preferred aspect of the present invention is an index for calculating a timbre index value (for example, timbre index value Y [m]) indicating a difference in timbre of each frame with respect to a reference frame for each of a plurality of frames in the transition section. Computation means (for example, index computation unit 60) is provided, and the boundary setting means sets a time point when the timbre index value becomes a numerical value corresponding to the articulation variable in the temporal transition of the timbre index value. In the above aspect, since the time point at which the timbre index value becomes a numerical value corresponding to the intelligibility variable is set as the boundary time point, regardless of the characteristics of each speech unit, the change in the intelligibility variable and the timbre change in the synthesized sound Can be matched. In the above aspect, when the timbre index value becomes a numerical value corresponding to the articulation variable at a plurality of time points in the temporal transition of the timbre index value, the boundary setting means sets the boundary at the rearmost time point among the plurality of time points. If it is set at the time point, since a long section of the speech segment is used for synthesis of the speech signal, there is an advantage that a natural synthesized sound can be generated audibly. In addition, the specific example of the above aspect is later mentioned, for example as 6th Embodiment.

  The speech synthesizer according to a preferred aspect of the present invention comprises limit setting means (for example, limit setting unit 42) for setting a limit time point in the transition segment of the speech unit selected by the unit selection means, and the boundary setting means includes: Then, the boundary time is set in the fluctuation section (for example, the fluctuation section EC) from the limit time set by the limit setting means to the end point of the transition section. In the above aspect, the boundary time is set behind the limit time set in the transition section. That is, the time point at which the influence of the phonemes in the first phoneme segment in the transition segment remains excessively is not set as the boundary time point. Accordingly, there is an advantage that a synthesized sound with an audibly natural impression can be generated regardless of the configuration in which the audio signal is generated by repeating the frame corresponding to the boundary time point.

  In a preferred example of the configuration including the limit setting means, each of the plurality of speech segments includes a first phoneme section and a second phoneme section corresponding to different phonemes, and the second phoneme section includes a transition section. The limit setting means sets a time point according to the phoneme type of the first phoneme section among the transition sections of the second phoneme section as the limit time point. Of the transition sections, sections in which the influence of phonemes in the first phoneme section remains excessively tend to differ depending on the type of phonemes in the first phoneme section. In the above aspect, since the position of the limit time point in the transition section is variably set according to the phoneme type of the first phoneme section, an appropriate position according to the phoneme type of the first phoneme section is set as the limit time point. There is an advantage that it can be set.

  A speech synthesizer according to a preferred example of a configuration including a limit setting unit includes a second index calculation unit that calculates a speech naturalness index value when each frame is repeated for each of a plurality of frames in a transition section. For example, an index calculation unit 48) is provided, and the limit setting means sets the limit time according to the index value of each frame. In the above aspect, since the limit time is set according to the index value of each frame in the transition section, there is an advantage that an appropriate limit time can be set according to the characteristics of the speech unit. Further, the first index value corresponding to the volume of each frame in the transition section and the second index value corresponding to the intensity of the anharmonic component of each frame in the transition section are used as second index calculation means (for example, an index calculation unit). 48) is calculated as an index value, and, of the transition sections, the time when the volume indicated by the first index value exceeds a predetermined value and the intensity of the anharmonic component indicated by the second index value falls below the predetermined value is set as a limit setting means. Is set as the limit time, for example, when the phoneme in the first phoneme section is an unvoiced consonant (for example, a plosive sound, a smashing sound, or a friction sound), the limit time point can be set at an appropriate position in the transition section. There is an advantage.

  The speech synthesizer according to each aspect described above is realized by hardware (electronic circuit) such as a DSP (Digital Signal Processor) dedicated to speech synthesis, and a general-purpose arithmetic processing device such as a CPU (Central Processing Unit). And collaboration with the program. The program of the present invention (for example, the program PGM) has a segment selection process for sequentially selecting speech segments including a transition section in which a speech waveform changes over time, and a variable intelligibility according to an instruction from a user. Variable setting processing for setting the boundary time in the transition section, the movement amount at the boundary time when the clarity variable changes by a predetermined amount from the first value, and the clarity variable is the first Boundary where the position of the boundary time point is variably set according to the articulation variable set in the variable setting process so that the amount of movement at the boundary time point when the predetermined value changes from the second value different from the value The computer is caused to execute a setting process and a synthesizing process for generating an audio signal using a section in front of the boundary time point among the speech elements selected in the element selection process. According to the above program, the same operation and effect as the speech synthesizer of the present invention are realized. The program of the present invention is provided to a user in a form stored in a computer-readable recording medium and installed in the computer, or provided from a server device in a form of distribution via a communication network and installed in the computer. Is done.

1 is a block diagram of a speech synthesizer according to a first embodiment of the present invention. It is a schematic diagram of the segment group stored in the storage device. It is explanatory drawing of the relationship between the waveform of a speech unit, and unit data. It is a schematic diagram of an edit screen. It is a block diagram of a speech synthesizer. It is a graph which shows the relationship between a clarity variable and a boundary variable. It is a schematic diagram of the segment data in 2nd Embodiment. It is a block diagram of the speech synthesizer in a 3rd embodiment. It is explanatory drawing of the relationship between the waveform of the speech unit and unit data in 3rd Embodiment. It is explanatory drawing of the calculation method of the boundary variable in 3rd Embodiment. It is a block diagram of the speech synthesizer in the fifth embodiment. It is explanatory drawing of an index value. It is a block diagram of the speech synthesizer in a 6th embodiment. It is explanatory drawing of operation | movement of 6th Embodiment. It is explanatory drawing of other operation | movement of 6th Embodiment. It is explanatory drawing of background art.

<First Embodiment>
FIG. 1 is a block diagram of a speech synthesizer 100 according to the first embodiment of the present invention. The speech synthesizer 100 is a signal processing device that generates speech such as speech sounds and singing sounds through segment-connected speech synthesis processing. As shown in FIG. 1, the arithmetic processing device 12, the storage device 14, and the display device are used. 22, an input device 24, and a sound emitting device 26.

  The arithmetic processing unit 12 (CPU) executes a program PGM stored in the storage device 14 to thereby generate a plurality of functions (display control unit 32, variable setting unit) for generating an audio signal VOUT representing the waveform of the synthesized sound. 33, a segment selection unit 34, and a speech synthesis unit 36) are realized. A configuration in which each function of the arithmetic processing unit 12 is distributed over a plurality of integrated circuits, or a configuration in which a dedicated electronic circuit (DSP) realizes a part of the functions may be employed.

  The display device 22 (for example, a liquid crystal display device) displays an image instructed from the arithmetic processing device 12. The input device 24 is a device (for example, a mouse or a keyboard) that receives an instruction from a user. The sound emitting device 26 (for example, a headphone or a speaker) emits a sound wave corresponding to the sound signal VOUT generated by the arithmetic processing device 12.

  The storage device 14 stores a program PGM executed by the arithmetic processing device 12 and various data (segment group QA, synthesis information QB) used by the arithmetic processing device 12. A known recording medium such as a semiconductor recording medium or a magnetic recording medium or a combination of a plurality of types of recording media is employed as the storage device 14.

  The unit group QA stored in the storage device 14 is a set (speech synthesis library) of a plurality of unit data D corresponding to different speech units V as shown in FIG. In the first embodiment, a diphone (phoneme chain) in which two phoneme sections S (S1, S2) corresponding to different phonemes are connected is assumed as a speech unit V. The phoneme segment S2 is located behind the phoneme segment S1. In the following, silence is described as a consonant phoneme for convenience.

  FIG. 3 is a waveform diagram of one speech unit V. FIG. 3 illustrates a waveform of a speech unit V [w−a] in which a phoneme segment S1 of a vowel / a / is followed by a phoneme segment S2 of a vowel / a / phoneme / w /. The phoneme boundary GA in FIG. 3 means the boundary between the phoneme segment S1 and the phoneme segment S2. The vowel phoneme section S2 is divided into a transition section EA and a steady section EB across the state boundary GB.

  The steady section EB in FIG. 3 is a section in which the waveform of the phoneme / a / in the phoneme section S2 is maintained in a steady state. On the other hand, the transition section EA is a section in which the waveform (tone color) of the speech segment V transitions from the phoneme / w / waveform of the previous phoneme section S1 to the phoneme / a / waveform of the phoneme section S2. That is, the shape of the speaker's mouth of the speech segment V starts to change at the phoneme boundary GA from the shape corresponding to the phoneme / w / in the phoneme segment S1, and the phoneme in the phoneme segment S2 from the phoneme boundary GA to the state boundary GB. It changes over time to a shape corresponding to / a /, and after reaching the shape corresponding to phoneme / a / at the state boundary GB, it is kept steady thereafter.

  The graph g in FIG. 3 shows that the timbre perceived by the listener of the synthesized sound ranges from the phoneme / w / of the phoneme section S1 to the phoneme of the phoneme section S2 from the start point (phoneme boundary GA) to the end point (state boundary GB) of the transition section EA. It is the graph which illustrated a mode that it changed temporally to / a /. As can be understood from the graph g, the timbre perceived by the listener (the opening degree of the speaker's mouth) changes nonlinearly with respect to the elapsed time. That is, the timbre change amount with respect to a predetermined unit time is different at each time point in the transition section EA. Specifically, the timbre changes remarkably just after a short time in the vicinity of the start point (phoneme boundary GA) of the transition section EA, but the time passes in the vicinity of the end point (state boundary GB) of the transition section EA. However, the timbre hardly changes.

  As shown in FIG. 2, the unit data D of each speech unit V is configured to include section information DB and a plurality of unit data U. The section information DB specifies the phoneme boundary GA and the state boundary GB in the speech unit V. Each of the plurality of unit data U designates the frequency spectrum of the speech of each frame obtained by dividing the speech segment V (phoneme segment S1 and phoneme segment S2) on the time axis.

  As shown in FIG. 1, the synthesized information (score data) QB stored in the storage device 14 is composed of a synthesized sound pronunciation character X1, a pronunciation period X2, a pitch (pitch) X3, and an articulation variable α. Specify in chronological order for each note. The pronunciation character X1 is a character string of lyrics when, for example, a singing sound is synthesized. The articulation variable α is a variable indicating the degree to which the synthesized sound is perceived as perceptually clear. The voice is perceptually perceived more clearly as the speaker opens his / her mouth when speaking. Therefore, the articulation variable α can also be expressed as a variable indicating the degree of mouth opening of the virtual speaker of the synthesized sound.

  The display control unit 32 of the arithmetic processing device 12 in FIG. 1 causes the display device 22 to display the editing screen 50 in FIG. 4 that is visually recognized by the user for generating and editing the composite information QB. The edit screen 50 is divided into a first area 51 and a second area 52. In the first area 51, a time axis (horizontal axis) and a pitch axis (vertical axis) are set, and a sound indicator 54 that represents each note of the synthesized sound corresponds to an instruction from the user to the input device 24. Arranged. The pitch X3 of the synthesis information QB is set according to the position on the pitch axis of each sound indicator 54, and the sound generation period X2 is set according to the position and size on the time axis. Further, the character designated by the user for each sound indicator 54 is set as the pronunciation character X1 of the composite information QB.

  In the second area 52, variable indicators 56 corresponding to the sound indicators 54 are arranged on the same time axis as the first area 51. Each variable indicator 56 is an image (bar graph) expressing the magnitude of the articulation variable α with a length d in the vertical direction. The user can change the length d of each variable indicator 56 within the range of 0 or more and 127 or less by appropriately operating the input device 24.

  The variable setting unit 33 in FIG. 1 variably sets the articulation variable α in the composite information QB in accordance with an instruction from the user to the input device 24. Specifically, the variable setting unit 33 sets the clarity variable α within a range of 0 or more and 1 or less so as to be proportional to the length d of the variable indicator 56 designated by the user. In other words, the articulation variable α is calculated by normalizing (for example, dividing by 127) the numerical value set by the user by the operation on the variable indicator 56 so that the maximum value becomes 1.

  The unit selection unit 34 in FIG. 1 sequentially selects the speech unit V corresponding to each phonetic character X1 specified in time series by the synthesis information QB from the unit group QA. The speech synthesizer 36 generates a speech signal VOUT using the segment data D of the speech segments V that are sequentially selected by the segment selector 34. Schematically, the speech synthesizer 36 expands / contracts the segment data D on the time axis according to the sound generation period X2 specified by the synthesis information QB, and converts the frequency spectrum indicated by each unit data U after expansion / contraction into a time waveform. After conversion, the audio signal VOUT is generated by adjusting the pitch X3 of the synthesis information QB and connecting them to each other.

  FIG. 5 is a block diagram of the speech synthesizer 36. As shown in FIG. 5, the speech synthesis unit 36 according to the first embodiment includes a boundary setting unit 44 and a synthesis processing unit 46. As shown in FIG. 3, the boundary setting unit 44, when the phoneme segment S2 of the speech unit V selected by the unit selection unit 34 corresponds to temporally sustainable phonemes such as vowels, friction sounds, and nasal sounds, A boundary time TB is set in the transition section EA of the phoneme section S2. Specifically, the boundary setting unit 44 sets a variable (hereinafter referred to as “boundary variable”) β that specifies the position on the time axis of the boundary time TB in the transition section EA. The boundary variable β ranges from the minimum value 0 that specifies the start point (phoneme boundary GA) of the transition section EA as the boundary time TB to the maximum value 1 that specifies the end point (state boundary GB) of the transition section EA as the boundary time TB. Within the range (0 ≦ β ≦ 1), it is variably set according to the articulation variable α designated by the synthesis information QB.

FIG. 6 is a graph showing the relationship between the articulation variable α and the boundary variable β. The boundary setting unit 44 of the first embodiment calculates the square of the clarity variable α as the boundary variable β (β = α ² ). Accordingly, as shown in FIG. 6, the change amount of the boundary variable β (that is, the movement amount at the boundary time TB) Δβ1 when the clarity variable α changes from the numerical value α1 to the positive side by a predetermined amount δ, and the clarity variable α Is different from the change amount Δβ2 of the boundary variable β when the value changes from the numerical value α2 (α2> α1) to the positive side by the same predetermined amount δ. Specifically, the change amount Δβ2 exceeds the change amount Δβ1. That is, as the intelligibility variable α approaches the maximum value 1, the movement amount of the boundary time TB with respect to the change in the intelligibility variable α increases. As understood from the above description, the boundary variable β (the position of the boundary time TB) changes nonlinearly in accordance with the clarity variable α designated by the user. In the first embodiment, the relationship between the clarity variable α and the boundary variable β is common to all types of speech segments V. As shown in FIG. 3, a section (a section from the start point of the phoneme section S1 to the boundary time TB) W before the boundary time TB set by the boundary setting unit 44 in the speech unit V is referred to as an “applied section” below. write.

  The synthesis processing unit 46 in FIG. 5 generates the audio signal VOUT using the application section W of the audio unit V selected by the unit selection unit 34. Specifically, as shown in FIG. 3, the composition processing unit 46 positions the unit data group Z1 composed of the unit data U in the application section W in the segment data D at the end in the application section W. A unit data group Z2 in which one unit data U (shaded portion in FIG. 3) is repeatedly arranged is connected. The number of unit data U constituting the unit data group Z2 is variably set so that the total length of the unit data group Z1 and the unit data group Z2 becomes a target length corresponding to the sound generation period X2.

  The synthesis processing unit 46 converts the frequency spectrum indicated by the unit data U of the unit data group Z1 and the unit data group Z2 into a time waveform and adjusts it to the pitch X3 designated by the synthesis information QB, and mutually exchanges frames in succession. To generate an audio signal VOUT. When the articulation variable α is set to the maximum value 1 and the target length corresponding to the pronunciation period X2 is less than a predetermined value (for example, the time length of the speech segment V), the synthesis processing unit 46 determines the steady interval EB. A plurality of unit data U of the speech unit V including is removed from the rear and adjusted to the target length to generate the speech signal VOUT (that is, the addition of the unit data group Z2 is not executed).

  As described above, the boundary time point TB is set before the arrival of the state boundary GB in which the speech waveform reaches the steady state in the phoneme section S2 (that is, before the speaker's mouth is fully opened). The applicable section W in front of the boundary time TB is used for generating the audio signal VOUT. Therefore, it is possible to generate a synthesized sound that the speaker speaks without fully opening his / her mouth.

  In the first embodiment, the boundary variable β (the position of the boundary time TB on the time axis) changes nonlinearly with respect to the articulation variable α in accordance with an instruction from the user. Specifically, the amount of change in the boundary variable β increases as the clarity variable α approaches the maximum value 1. On the other hand, the timbre perceived by the listener for the sound in the transition section EA changes nonlinearly according to the position in the transition section EA. Specifically, the timbre change amount decreases as the position is closer to the end point of the transition section EA. Therefore, according to the first embodiment, the change in the clarity variable α designated by the user and the synthesized sound are compared with the configuration in which the boundary time TB is linearly moved according to the clarity variable α designated by the user. It is possible to reduce inconsistency with the timbre change. For example, when the user changes the articulation variable α from the maximum value 1 to half, the timbre of the synthesized sound changes as if the opening degree of the mouth is changed to half.

Second Embodiment
A second embodiment of the present invention will be described below. In the first embodiment, the relationship between the clarity variable α and the boundary variable β is made common to all types of speech segments V. However, the time change of the timbre (mouth openness) in the transition section EA changes according to the type of the speech segment V (particularly the type of phoneme in the phoneme section S1). In consideration of the above tendency, in the second embodiment, the relationship between the articulation variable α and the boundary variable β is changed according to the type of phoneme in the phoneme section S1 of the speech unit V. In addition, about each element which an effect | action and a function are equivalent to 1st Embodiment in each form illustrated below, each detailed description which diverted the code | symbol referred by the above description is abbreviate | omitted suitably.

  FIG. 7 is a schematic diagram of the segment data D in the second embodiment. The segment data D of each speech segment V in the second embodiment is configured to include classification information DA in addition to the section information DB and the plurality of unit data U similar to those in the first embodiment. The classification information DA specifies the classification of each phoneme constituting the speech segment V. For example, as shown in FIG. 7, vowels (/ a /, / i /, / u /), plosives (/ t /, / k /, / p /), rubbing sounds (/ ts /), nasal sounds ( / m /, / n /), stream sound (/ r /), friction sound (/ s /, / f /), semi-vowel (/ w /, / y /), silence (/ Sil /), etc. Each of the phoneme segment S1 and the phoneme segment S2 of the speech segment V is specified by the classification information DA.

  As shown in FIG. 7, each phoneme is divided into a plurality of types C (C1 to C3). Specifically, consonant phonemes are classified into various types C according to the level of voicedness. For example, in the case of Japanese phonemes, phonemes rich in harmonic components such as semi-vowels (/ w /, / y /), nasal sounds (/ m /, / n /) and stream sounds (/ r /), or voiced friction sounds Consonants with high voicedness (for example, voiced consonants), such as phonemes rich in anharmonic components such as (/ z /) and voiced plosives (/ d /), are classified into type C1, and plosives (/ t / , / K /, / p /), deaf sounds (/ ts /), and unvoiced consonants such as friction sounds (/ s /, / f /) are classified into type C2. Silence (/ Sil /) is classified as type C3. Vowels (/ a /, / i /, / u /) are classified into type C2.

  In the boundary setting unit 44 of the second embodiment, the relationship between the clarity variable α instructed by the user and the boundary variable β (the position of the boundary time TB) is the phoneme type of the phoneme segment S1 in the phoneme segment V. The boundary variable β corresponding to the articulation variable α is calculated so as to be different according to C. Specifically, the function for calculating the boundary variable β from the clarity variable α is different between the case where the phoneme classification of the phoneme section S1 specified by the classification information DA belongs to the type C1 and the type C2. To do.

  In the first embodiment, the same effect as in the second embodiment is realized. In the second embodiment, the relationship between the articulation variable α and the boundary variable β changes according to the phoneme type C (classification) of the phoneme segment S1, so that the timbre in the transition segment EA (how the mouth opens) Even if the temporal transition of the voice segment differs depending on the phoneme type C in the phoneme segment S1, the inconsistency between the change in the clarity variable α designated by the user and the timbre change in the synthesized sound is represented by each speech unit. It is possible to reduce V.

<Third Embodiment>
In the transition segment EA of the phoneme segment V, the phoneme waveform of the phoneme segment S1 transitions to the phoneme waveform of the phoneme segment S2. That is, the influence of the phoneme of the immediately preceding phoneme section S1 remains in the transition section EA near the phoneme boundary GA. Therefore, when the boundary time TB is set at a position close to the phoneme boundary GA in the transition section EA, the phoneme of the phoneme section S1 is originally included in the unit data group Z2 in which only the phoneme of the phoneme section S2 should be reflected. There is a possibility that the unit data U including the influence of the above will be repeated and the synthesized sound becomes an unnatural sound. Against the background described above, in the third embodiment, the immediately preceding phoneme section in the unit data U that is repeated in the unit data group Z2 (that is, the last unit data U corresponding to the boundary time TB in the application section W). The position of the boundary time TB set in the transition section EA is limited so that the influence of the phoneme of S1 is sufficiently reduced.

  FIG. 8 is a block diagram of the speech synthesizer 36 in the third embodiment. As shown in FIG. 8, the speech synthesizer 36 of the third embodiment has a configuration in which a limit setting unit 42 is added to the speech synthesizer 36 of the first embodiment. As shown in FIG. 9, the limit setting unit 42 sets a limit time TA within the transition segment EA of the phoneme segment S2 corresponding to the phoneme of the vowel in the speech segment V selected by the segment selection unit 34. The limit time point TA is a time point ahead of the end point (state boundary GB) of the transition section EA by a time (R × EA) corresponding to the ratio R of the transition section EA. Limit information QC specifying the ratio R is stored in advance in the storage device 14, and the limit setting unit 42 sets the limit time TA according to the limit information QC acquired from the storage device 14. The ratio R is selected within a range of 0 or more and 1 or less so that the influence of the phoneme in the phoneme section S1 immediately before the transition section EA is sufficiently reduced at the limit time TA.

  The boundary setting unit 44 in FIG. 8 variably sets the boundary variable γ that specifies the position of the boundary time TB in the transition section EA according to the articulation variable α set by the variable setting unit 33. Similarly to the boundary variable β of the first embodiment, the boundary variable γ is a variable that designates the position on the time axis of the boundary time TB with the start point (phoneme boundary GA) of the transition section EA as the minimum value 0. As shown in FIG. 9, the boundary setting unit 44 is within a section (hereinafter referred to as “variation range”) EC from the limit time TA set by the limit setting unit 42 to the state boundary GB in the transition section EA in the phoneme section S2. The boundary variable γ is calculated so that the boundary time point TB is located in

  FIG. 10 is an explanatory diagram of the boundary variable γ. Part (A) of FIG. 10 illustrates a boundary time point TB specified in the transition section EA by the boundary variable β of the first embodiment that changes nonlinearly with the articulation variable α. On the other hand, part (B) of FIG. 10 shows a fluctuation section EC defined by the limit time TA of the third embodiment. Assuming that the total time length of the transition section EA is 1, as shown in part (B) of FIG. 10, the transition section EA includes the section over the time (1-R) ahead of the limit time TA and the limit time TA. Is divided into a fluctuation section EC over time R.

A time point behind the limit time point TA by a time tx is defined as a boundary time point TB indicated by the boundary variable γ of the third embodiment. Now, the position (part (A)) of the boundary time TB indicated by the boundary variable β with respect to the entire transition section EA and the position of the boundary time TB indicated by the boundary variable γ with respect to the fluctuation section EC of the transition section EA. Assuming that (part (B)) is equivalent (1: β = R: tx), the time tx is expressed as a multiplication value (βR) of the boundary variable β and the ratio R. Further, considering that the section ahead of the limit time TA in the transition section EA is the time (1-R) as described above, the boundary for designating the boundary time TB with the start point GA of the transition section EA as the minimum value 0. The variable γ is expressed by the following formula (1).
γ = (1−R) + βR
= 1- (1-β) R (1)

The boundary setting unit 44 of the third embodiment calculates a boundary variable β corresponding to the articulation variable α in the same manner as in the first embodiment (β = α ² ), and the ratio indicated by the boundary variable β and the limit information QC The boundary variable γ is calculated by executing the calculation of the mathematical formula (1) for R and R. Therefore, the boundary variable γ changes nonlinearly with the articulation variable α, similarly to the boundary variable β of the first embodiment. Further, since the boundary variable β is a numerical value of 0 or more and 1 or less, the boundary variable γ is a numerical value within the range of (1−R) or more and 1 or less. That is, the boundary variable γ is in accordance with the articulation variable α in the fluctuation section EC from the limit time TA after the time (1-R) to the end point GB of the transition section EA by the time (1-R) with respect to the start point GA of the transition section EA. This is a variable for designating a non-linearly changing boundary time TB.

  In the third embodiment, the same effect as in the first embodiment is realized. In the third embodiment, the position of the boundary time point TB is limited to the limit time point TA after the phoneme boundary GA, and is thus repeated in the unit data group Z2 used for generating the audio signal VOUT. Unit data U (ie, unit data U corresponding to the boundary time TB), the influence of the phoneme in the immediately preceding phoneme section S1 is sufficiently reduced. Therefore, there is an advantage that an acoustically natural voice can be synthesized.

<Fourth embodiment>
The position of the first unit data U in which the influence of the phoneme in the phoneme section S1 is sufficiently reduced within the transition section EA of the phoneme section S2 (the synthesized sound is generated even when it is repeated among the plurality of unit data U in the phoneme section S2). There is a tendency that the position of the earliest unit data U that does not result in an unnatural sound is different depending on the type C of the phoneme in the immediately preceding phoneme section S1. For example, if the phoneme in the phoneme segment S1 belongs to a phoneme type C1 such as a semi-vowel with high voicedness, even if the unit data U near the phoneme boundary GA in the phoneme segment S2 is repeated, the synthesized sound is not so unnatural. It does not become a sound. On the other hand, when the phoneme in the phoneme segment S1 belongs to the phoneme type C2 such as a plosive having a rich anharmonic component (noise component) and a small amplitude, the unit data U near the phoneme boundary GA in the phoneme segment S2 is repeated. Then, the unnaturalness of the synthesized sound derived from the phoneme in the phoneme section S1 is noticeably perceived. Considering the above tendency, in the fourth embodiment, the position of the limit time TA with respect to the transition section EA in the phoneme section S2 is changed according to the phoneme type C of the immediately preceding phoneme section S1. The configuration of the fourth embodiment is the same as that of the third embodiment.

  The storage device 14 of the fourth embodiment stores limit information QC that defines the position of the limit time TA. The limit information QC specifies the ratio R (R1 to R3) of the fluctuation section EC to the transition section EA for each phoneme type C (C1 to C3). The limit setting unit 42 is included in the transition segment EA of the speech unit V according to the ratio R indicated by the limit information QC for the type C of the phoneme segment S1 in the speech unit V selected by the segment selection unit 34. Set the limit time TA. Each segment data D in the storage device 14 is configured to include classification information DA as in the second embodiment, and the type C of the phoneme section S1 is specified from the classification information DA of the segment data D.

  Specifically, when the phoneme in the phoneme section S1 belongs to the type C1 (consonant with high voicedness), the time point ahead by the time (R1 × EA) corresponding to the ratio R1 from the end point GB of the transition section EA is limited. Set as time TA. Similarly, when the phoneme in the phoneme segment S1 corresponds to the type C2 (unvoiced consonant or vowel), the time point ahead by the time (R2 × EA) corresponding to the ratio R2 from the end point GB of the transition zone EA is the limit time point TA. When the phoneme in the phoneme section S1 corresponds to the type C3 (silence), a time point ahead by the time (R3 × EA) from the end point GB of the transition section EA is set as the limit time point TA. As in the third embodiment, the boundary time TB corresponding to the articulation variable α is set in the transition section EC from the limit time TA to the end point GB in the transition section EA.

  Among the plurality of unit data U in the phoneme section S2, the position of the earliest unit data U that does not result in an unnatural speech affected by the phoneme in the phoneme section S1 is specified for the synthesized sound generated by the repetition. The ratio R for each phoneme type C is selected experimentally or statistically. For example, when the phoneme in the phoneme segment S1 belongs to the type C1, the synthesized sound does not become so unnatural even if the unit data U near the phoneme boundary GA in the phoneme segment S2 is repeated, but the phoneme in the phoneme segment S1 Considering the tendency that if the unit data U in the vicinity of the phoneme boundary GA of the phoneme section S2 is repeated when belonging to the type C2, the unnaturalness of the synthesized sound derived from the phoneme of the phoneme section S1 becomes obvious, and the ratio R1 is It is set to a numerical value that exceeds the ratio R2 and ratio R3. Therefore, assuming that the time lengths of the transition sections EA are common, the limit time TA when the phoneme of the phoneme section S1 belongs to the type C2 or the type C3 is greater than the limit time TA when the phoneme of the phoneme section S1 belongs to the type C1. Will be later in time. Specifically, the ratio R1 is set to about 0.8 (80%), the ratio R2 is set to about 0.61 (61%), and the ratio R3 is set to about 0.5 (50%). .

  In the fourth embodiment, the same effect as in the third embodiment is realized. In the third embodiment in which the limit time TA is set according to the common ratio R regardless of the phonemes in the phoneme section S1, the unnaturalness of the synthesized sound derived from the phonemes in the phoneme section S1 is suppressed for all phonemes. In order to do this, it is necessary to set the limit time TA to the rear side of the transition section EA. Therefore, the fluctuation section EC is limited to a short time, and there is a possibility that the intelligibility (openness) of the synthesized sound cannot be sufficiently reduced. On the other hand, when the time near the phoneme boundary GA is selected as the limit time TA in order to sufficiently reduce the clarity of the synthesized sound in the third embodiment, the synthesized sound is unnatural due to the phoneme in the phoneme section S1. Sound. In the fourth embodiment, since the position (ratio R) of the limit time TA with respect to the transition section EA is set according to the phoneme type C of the immediately preceding phoneme section S1, a sufficient range of change in the clarity of the synthesized sound is ensured. There is an advantage that it is possible to achieve both (to sufficiently reduce the intelligibility) and to reduce the unnaturalness of the synthesized sound caused by the phonemes in the phoneme section S1.

  In the fourth embodiment, the ratio R of the fluctuation section EC to the transition section EA is individually set for each type C. However, the method for changing the limit time TA for each phoneme type C in the phoneme section S1 is changed as appropriate. Is done. For example, the limit information QC stored in the storage device 14 indicates the time length (number of frames) of the period from the start point GA to the limit point TA or the period from the limit point TA to the end point GB for each phoneme type C. A configuration designated in the above can also be adopted.

<Fifth Embodiment>
In the third and fourth embodiments, the limit setting unit 42 sets the limit time TA within the transition section EA using the limit information QC stored in advance in the storage device 14. In the fifth embodiment, the limit setting unit 42 sets the limit time TA using the analysis result of the speech segment V.

  FIG. 11 is a block diagram of the speech synthesizer 36 in the fifth embodiment. As shown in FIG. 11, the speech synthesizer 36 of the fifth embodiment has a configuration in which an index calculator 48 is added to the speech synthesizer 36 (FIG. 8) of the third embodiment. The index calculation unit 48 repeats one unit data U of each frame for each of a plurality of frames in the transition segment EA in the phoneme segment S2 of the speech segment V selected by the segment selection unit 34. An index value K that is a measure of auditory naturalness of the generated synthesized sound is calculated.

  A typical frame in which the synthesized sound becomes audibly unnatural sound when one unit data U is repeated is a frame whose volume is lower than that of a voiced sound, or a harmonic component (a fundamental component and each harmonic component). ) Is a frame in which the intensity of the anharmonic component is high. Specifically, when the unit data U of the front frame in the transition section EA located immediately after the phoneme section S1 of a phoneme such as a plosive or a smashing sound is repeated, the synthesized sound is audibly unnatural speech. It becomes. In consideration of the above tendency, the index calculation unit 48 selects the speech unit selected by the unit selection unit 34 from the index value K1 related to the volume of each frame and the index value K2 related to the intensity of the anharmonic component of each frame. The index value K is calculated for each frame in the V transition section EA.

  The index value K1 of each frame is calculated, for example, as the ratio of the volume A of the frame to the predetermined volume A0 (K1 = A / A0). The predetermined volume A0 is, for example, the volume of the last frame in the transition section EA (highly likely to be the maximum value in the transition section EA). Therefore, the index value K1 becomes a larger numerical value as the frame has a louder volume A within the transition section EA (that is, the frame having a higher possibility that the synthesized sound obtained by repeating the unit data U becomes a perceptually natural voice).

  The index value K2 of each frame is calculated as the ratio (K2 = P / PS) of the average power P of the frame to the average power PS when the anharmonic component is reduced or removed from the audio component of the frame. FIG. 12 shows the frequency spectrum SP1 designated by the unit data U of one frame in the transition section EA. The frequency spectrum SP1 includes an inharmonic component that exists between the harmonic frequencies in addition to a harmonic component having a peak intensity at each harmonic frequency Fn (basic frequency and each harmonic frequency).

  FIG. 12 also shows a frequency spectrum SP2 (shaded portion) obtained by removing the anharmonic component from the frequency spectrum SP1. The frequency spectrum SP2 is a spectrum in which a predetermined harmonic component H is arranged at each harmonic frequency Fn of the frequency spectrum SP1, and the intensity of each harmonic component H is adjusted so as to match the envelope ENV of the frequency spectrum SP1. . The index calculation unit 48 calculates, for each frame, the ratio of the average power P of the frequency spectrum SP1 to the average power PS of the frequency spectrum SP2 as an index value K2. Therefore, the index value K2 is a smaller numerical value for a frame in which the intensity of the inharmonic component relative to the harmonic component is low (that is, a frame in which the synthesized sound obtained by repeating the unit data U is more likely to be perceptually natural speech).

  The limit setting unit 42 of the fifth embodiment sets the limit time TA according to the index value K (K1, K2) of each frame in the transition section EA. That is, the limit setting unit 42 sets the time point of the earliest frame in which the naturalness of the synthesized sound indicated by the index value K among the plurality of frames in the transition section EA exceeds the target value as the limit time point TA.

  Specifically, the index calculation unit 48 sequentially selects frames from the beginning of the transition section EA and calculates the index value K1 and the index value K2 of the frame, and the limit setting unit 42 sets the index value K1 to a predetermined value. Whether or not the index value K2 is below a predetermined threshold Kth2 (that is, whether the strength of the anharmonic component relative to the harmonic component is below the target value) Or not). The limit setting unit 42 sets, as the limit time TA, the time of the earliest frame in which both the determination of the index value K1 and the determination of the index value K2 are affirmative. That is, a time point when the intensity of the non-harmonic component with respect to the harmonic component is sufficiently low and the sound volume is large (a time point when the synthesized sound generated by repetition of the unit data U becomes audibly natural sound) is set as the limit time point TA. . Accordingly, in the fifth embodiment, as a result, as in the fourth embodiment, the time point corresponding to the phoneme type C in the phoneme section S1 is set as the limit time point TA. For example, the limit time TA when the phoneme in the phoneme section S1 belongs to the type C2 is a time point later in time than the limit time TA when the phoneme in the phoneme section S1 belongs to the type C1. The operations of the boundary setting unit 44 and the composition processing unit 46 are the same as those in the third embodiment and the fourth embodiment.

  In the fifth embodiment, the same effect as in the third embodiment is realized. In the fifth embodiment, since the limit time TA is set according to the index value K (K1, K2) calculated for each frame in the transition section EA, the limit information QC that defines the limit time TA is obtained in advance. Compared with the first embodiment and the second embodiment prepared in the above, there is an advantage that an appropriate limit time TA can be set according to the characteristics of each speech unit V.

<Sixth Embodiment>
FIG. 13 is a block diagram of the speech synthesizer 36 in the sixth embodiment. As shown in FIG. 13, the speech synthesis unit 36 of the sixth embodiment has a configuration in which an index calculation unit 60 is added to the speech synthesis unit 36 of the first embodiment. As shown in FIG. 14, the index calculation unit 60 uses the timbre index value Y [m] (m) for each of a plurality of (M) frames in the transition section EA of the speech unit V selected by the unit selection unit 34. = 1 to M). The timbre index value Y [m] is a scale indicating a timbre difference between the m-th frame in the transition section EA and a specific frame (hereinafter referred to as “reference frame”) in the transition section EA. In the following description, a case where one frame located at the end (Mth) in the transition section EA is selected as the reference frame will be exemplified.

  In the sixth embodiment, one unit data U corresponding to a phoneme of a voiced sound includes envelope shape data E in addition to data indicating the frequency spectrum of each frame, as shown in FIG. The envelope shape data E is composed of a plurality of variables indicating the shape characteristics of the envelope of the frequency spectrum of speech. The envelope shape data E of the first embodiment is a vector having, for example, a known SMS plus an ExR (Excitation plus Resonance) parameter including an excitation waveform envelope e1, a chest resonance e2, a vocal tract resonance e3, and a difference spectrum e4. (Spectral Modeling Synthesis) Generated by analysis. EpR parameters and SMS analysis are also disclosed in, for example, Japanese Patent No. 3711880 and Japanese Patent Application Laid-Open No. 2007-226174.

  The excitation waveform envelope (Excitation Curve) e1 is a variable that approximates the envelope of the spectrum of vocal cord vibration. Chest resonance e2 designates the bandwidth, center frequency, and amplitude value of a predetermined number of resonances (bandpass filters) that approximate the chest resonance characteristics. Vocal Tract Resonance (e3) designates a bandwidth, a center frequency, and an amplitude value for each of a plurality of resonances that approximate the vocal tract resonance characteristics. The difference spectrum e4 means the difference (error) between the spectrum approximated by the excitation waveform envelope e1, the chest resonance e2 and the vocal tract resonance e3 and the spectrum of the voice.

The envelope shape data E can be used as a variable indicating the tone color of each frame. Therefore, the index calculation unit 60 calculates a timbre index value Y [m] indicating a difference between the two according to the envelope shape data E of each frame and the envelope shape data E of the reference frame. Specifically, the index calculation unit 60 calculates the timbre index value Y [m] by the calculation of the following formula (2).
Y [m] = D {E (M), E (m)} / D {E (M), E (1)} (2)
The operator D {E (m1), E (m2)} in Equation (2) is obtained by calculating the envelope shape data E (m1) of the m1st frame and the envelope shape data E of the m2th frame in the transition section EA. This means a distance from (m2) (for example, a Euclidean distance between vectors indicated by each envelope shape data E). That is, the distance D {E (M), E (m)} of the numerator in Expression (2) is determined by the envelope shape data E (M) of the reference frame and the envelope shape data E (m) of the mth frame in the transition section EA. m), and the denominator distance D {E (M), E (1)} in equation (2) is the envelope shape data E (M) of the reference frame and the envelope of the first frame in the transition section EA This is the distance from the shape data E (1).

  Since the timbre of the speech segment V changes over time from the timbre of the phoneme segment S1 to the timbre of the phoneme segment S2 from the start point GA to the end point GB of the transition segment EA, the distance D {E (M), E (m)} is the maximum value D {E (M), E (1)} in the first frame of the transition section EA, and generally decreases with the passage of time and is the minimum in the last frame (reference frame). The value is 0. The division by the distance D {E (M), E (1)} in Equation (2) means an operation for normalizing the timbre index value Y [m] to a numerical value in the range of 0 to 1 inclusive. That is, as shown in FIG. 14, the timbre index values Y [1] to Y [M] have a maximum value of 1 in the first frame of the transition section EA, and decrease as time passes as a general trend. The minimum value becomes 0 in the frame (reference frame).

  As understood from the above description, the timbre index value Y [m] functions as a variable that changes linearly with respect to a change in the timbre perceived by the listener (change in the degree of opening of the speaker's mouth). For example, in the transition zone EA, the tone color index value Y [m in a frame having a tone color intermediate between the phoneme in the immediately preceding phoneme segment S1 and the phoneme in the phoneme segment S2 (for example, a frame in which the speaker has half his mouth open). ] Is 0.5, and a timbre frame in which the timbre of the phoneme segment S1 and the timbre of the phoneme segment S2 are mixed at a ratio of 2: 8 (for example, the speaker has opened his / her mouth by 20%) ), The timbre index value Y [m] is 0.2.

  In the boundary setting unit 44 of the sixth embodiment, in the temporal transition of the timbre index value Y [m] in the transition section EA, the timbre index value Y [m] is a numerical value corresponding linearly to the articulation variable α. Is selected as the boundary time TB. Specifically, the boundary setting unit 44 calculates a numerical value (1-α) that changes linearly within a range from 0 to 1 according to the clarity variable α, and as shown in FIG. The time point ty at which the timbre index value Y [m] matches the numerical value (1-α) is set as the boundary time point TB. For example, when the intelligibility variable α is 0.5, the time point (frame) in which the timbre index value Y [m] becomes 0.5 in the transition section EA is set as the boundary time point TB, and the intelligibility variable α is In the case of 0.2, the time point at which the timbre index value Y [m] becomes 0.8 in the transition section EA is set as the boundary time point TB.

  FIG. 14 illustrates the case where the timbre index value Y [m] monotonously decreases with time. However, as shown in FIG. 15, the timbre index value Y [m] may not monotonously decrease depending on the speech segment V. There is sex. When the speech segment V increases and decreases in the transition section EA, the timbre index value Y [m] is a numerical value (1-α corresponding to the articulation variable α at a plurality of time points (frames) ty in the transition section EA. ). As described above, when the timbre index value Y [m] is a numerical value (1-α) corresponding to the articulation variable α at a plurality of time points ty, the boundary setting unit 44 is the rearmost among the plurality of time points ty. The time point is selected as the boundary time point TB.

  Also in the sixth embodiment, since the applicable section W in front of the boundary time TB in the speech unit V is used for generating the speech signal VOUT, the speaker sufficiently opens his mouth as in the first embodiment. It is possible to generate a synthesized sound as if it were uttered. In the sixth embodiment, the timbre index value Y [m] that changes linearly with respect to the timbre change in the transition section EA (change in the mouth openness of the speaker) is calculated and instructed by the user. The time point at which the timbre index value Y [m] matches the numerical value (1-α) corresponding to the articulation variable α is set as the boundary time point TB. Therefore, as in the first embodiment, it is possible to reduce the inconsistency between the change in the clarity variable α designated by the user and the timbre change in the synthesized sound. Specifically, when the user changes the clarity variable α from the maximum value 1 to half, the timbre of the synthesized sound changes as if the opening degree of the mouth is changed to half.

  In addition, in the sixth embodiment, the boundary time point TB is set according to the time transition of the timbre index value Y [m] calculated for each speech unit V, so the clarity variable α and the position of the boundary time point TB are Compared with the first to fifth embodiments in which the relationship is determined in advance, the inconsistency between the change in the clarity variable α designated by the user and the timbre change in the synthesized sound is reduced. There is an advantage that an appropriate boundary time TB can be set according to the characteristics of each speech unit V. In the sixth embodiment, when the timbre index value Y [m] becomes a numerical value (1-α) corresponding to the articulation variable α at a plurality of time points ty on the time axis, The most backward time is selected as the boundary time TB. In the above configuration, for example, a larger number of unit data U in the speech segment V is used as the unit data group Z1 of the audio signal VOUT as compared with the configuration in which the most forward time among the plurality of times ty is selected as the boundary time TB. The time length of the unit data group Z2 that is used for synthesis and repeats one unit data U is relatively shortened. Therefore, there is an advantage that a synthetic sound that is audibly natural can be generated.

  In addition, it is also possible to add the limit setting part 42 of 3rd Embodiment to 5th Embodiment to 6th Embodiment. That is, the timbre index value Y [m] is calculated for each frame in the variation section EC from the limit time TA set by the limit setting section 42 to the end point GB of the transition section EA, and the timbre index value Y [m] is the clarity. A time point ty in the fluctuation section EC that is a numerical value (1-α) corresponding to the variable α is selected as the boundary time point TB.

<Modification>
Each of the above forms can be variously modified. Specific modifications are exemplified below. Two or more modes arbitrarily selected from the following examples can be appropriately combined.

(1) By specifying the relationship between the articulation variable α and the boundary time TB (boundary variables β, γ) in each segment data D, it is also possible to set each speech segment V individually. It is also possible to individually specify the limit time TA for each speech unit V by including the limit information QC specifying the limit time TA in each unit data D.

(2) In the third embodiment and the fourth embodiment, the ratio R of the time from the limit time TA to the state boundary GB in the transition section EA is specified, but from the phoneme boundary GA to the limit time TA in the transition section EA. It is also possible to specify the time ratio R.

(3) In each of the above-described forms, an extended sound (steady vowel sound) is generated by repeating the unit data U. Therefore, the stationary section EB in the phoneme section S2 of each speech unit V may be omitted. Is possible.
According to the configuration in which the stationary section EB is omitted, there is an advantage that the data amount of the element group QA can be reduced. However, since the decompressed sound generated by repeating the unit data U may be unnatural speech compared to the actually recorded decompressed sound, the segment of the speech unit V so as to include the stationary section EB. In the case where the data D is generated and the sound generation period X2 is short, the configurations of the above-described embodiments in which the speech segment V including the stationary section EB is used as it is for the generation of the synthesized sound are suitable. As understood from the above examples, the transition section EA means a part of the phoneme section S2 (except for the steady section EB) or the entire section of the speech unit V.

(4) The type C of each phoneme is changed as appropriate. For example, when a speech segment V is generated (recorded) so that a silent segment is interposed between phonemes of successive vowels, if the vowel is classified into type C1, the silent segment is excessive in the fourth embodiment. A configuration in which vowels are classified into type C2 as shown in the above example is preferable because they can be expanded and become unnatural synthesized sounds. However, if there is no expansion of the silent section (for example, if there is no silent section between adjacent vowel phonemes) or if there is no particular problem, the vowel phonemes can be classified as type C1. It is.

(5) In each of the above embodiments, the clarity variable α is controlled in accordance with an instruction from the user, but the method of setting the clarity variable α is arbitrary. For example, it is possible to control the articulation variable α in accordance with the tempo specified for the synthesized sound. For example, when the tempo specified for the synthesized sound is set to a predetermined reference value (for example, a general numerical value such as 120 BPM (Beat Per Minute)), the articulation variable α is set to the maximum value 127, and from the reference value A configuration in which the articulation variable α decreases as the tempo increases (for example, increases beyond the reference value) is preferable. Further, for example, the actual duration of each synthesized sound is calculated according to the tempo and the sound generation period X2, and the duration is a predetermined threshold (for example, a time that is a predetermined multiple of the duration of the phoneme recorded in the speech segment V). It is also possible to reduce the articulation variable α when below. As described above, according to the configuration in which the intelligibility variable α is controlled according to the tempo and the duration, it is possible to reproduce the actual pronunciation tendency that the intelligibility decreases as the sound is pronounced quickly with the synthesized sound. is there.

(6) There is a possibility that the synthesized sound generated only by repeating the unit data U as in each of the above-described forms is perceived as an artificial and unnatural sound. Therefore, it is also preferable to add a fluctuation component extracted from an actual utterance sound (a fluctuation component that fluctuates in time among the extended sounds) to the voice generated from the time series of the unit data U.

(7) The index value K in the fifth embodiment is changed as appropriate. For example, the limit time TA is set using only one of the volume index value K1 and the anharmonic component index value K2, or the limit time TA is set using an index value K other than the index value K1 and the index value K2. A configuration for setting the value can also be adopted. In addition, the calculation method of the index value K1 and the index value K2 is appropriately changed. For example, in the above-described example, the index value K1 has a larger numerical value as the volume increases, and the index value K2 decreases as the inharmonic component intensity decreases. However, the index value K1 decreases as the volume increases. It is also possible to calculate the index value K1 and the index value K2 so that the index value K2 becomes larger as the intensity of the anharmonic component is lower.

(8) In the sixth embodiment, the timbre index value Y [m] is calculated using the EpR parameter specified by the unit data U as timbre information. However, the information indicating the timbre of the speech segment V is used as the EpR parameter. It is not limited. For example, the timbre index value Y [m] can be calculated using a cepstrum calculated from the frequency spectrum indicated by the unit data U as timbre information instead of EpR.

(9) In each of the above-described embodiments, the configuration in which the storage device 14 that stores the unit group QA is mounted on the speech synthesizer 100 is exemplified. However, an external device (for example, a server device) independent of the speech synthesizer 100 is provided. A configuration for holding the element group QA is also employed. The speech synthesizer 100 (unit selection unit 34) obtains a speech unit V (unit data D) from an external device via, for example, a communication network, and generates a speech signal VOUT. Similarly, the synthesis information QB can be held in an external device independent of the speech synthesizer 100. As can be understood from the above description, the element for storing the segment data D and the synthesis information QB (the storage device 14 in each of the above embodiments) is not an essential element of the speech synthesizer 100.

DESCRIPTION OF SYMBOLS 100 ... Speech synthesizer, 12 ... Arithmetic processing unit, 14 ... Memory | storage device, 22 ... Display device, 24 ... Input device, 26 ... Sound emission device, 32 ... Display control part, 33 ... Variable Setting unit 34... Segment selection unit 36... Speech synthesis unit 42. Limit setting unit 44. Boundary setting unit 46. Synthesis processing unit 48 and 60.

Claims (6)

Unit selection means for sequentially selecting speech units including a transition section in which a speech waveform changes over time;
Variable setting means for variably setting a clarity variable in accordance with an instruction from a user;
A means for setting a boundary time point in the transition section, wherein the larger the intelligibility variable is , the closer the boundary time point is to the end point of the transition section, and the larger the intelligibility variable is, Boundary setting means for variably setting the position of the boundary time according to the clarity variable set by the variable setting means so that the movement amount at the boundary time increases .
A speech synthesizer comprising: a synthesis processing unit that generates a speech signal using a section in front of the boundary time point among speech units selected by the unit selection unit. The voice according to claim 1, wherein the boundary setting means sets the position of the boundary time according to the clarity variable so that the relationship between the clarity variable and the position of the boundary time is common to a plurality of speech segments. Synthesizer. Each of the plurality of speech segments includes a first phoneme section and a second phoneme section corresponding to different phonemes, and the second phoneme section includes the transition section,
The boundary setting means responds to the clarity variable so that a relationship between the clarity variable and the position at the boundary time differs according to the phoneme type of the first phoneme section of the speech segment. The speech synthesizer according to claim 1, wherein the position at the boundary time is set. Comprising index calculation means for calculating a timbre index value indicating a difference in timbre of each frame with respect to a reference frame for each of the plurality of frames in the transition section;
The speech synthesizer according to claim 1, wherein the boundary setting means sets a time point at which the timbre index value becomes a numerical value corresponding to the clarity variable in the temporal transition of the timbre index value. When the timbre index value becomes a numerical value corresponding to the clarity variable at a plurality of time points in the temporal transition of the timbre index value, the boundary setting means determines the most backward time point among the plurality of time points. The speech synthesizer according to claim 4, wherein the speech synthesizer is set at a boundary time.   Computer system
  Sequentially select speech segments that contain transition intervals where the speech waveform changes over time,
  Set the articulation variable variably according to instructions from the user,
  Set a boundary time point in the transition section,
  A speech signal is generated using a section in front of the boundary time among the selected speech segments.
  A speech synthesis method,
  In the setting of the boundary time point, the larger the articulation variable, the closer the boundary time point is to the end point of the transition section, and the larger the articulation variable, the more the movement amount of the boundary time point with respect to the change of the articulation variable. The position of the boundary time point is variably set according to the set clarity variable so as to increase
  Speech synthesis method.

Images