JP2016118712A - Voice synthesis library generation device and voice synthesis device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To generate a synthesis voice which has high degree of being recognized as vague to a listener, while reducing a load applied to a user who generates a library of the voice synthesis.SOLUTION: A generation part 32 is an element for generating a second library L2 from a first library L1 for voice synthesis containing stock piece data Q for every voice stock piece P, and generates the second library L2 containing stock piece data Q of a voice stock piece P in which a vowel is substituted to a specific vowel in the first library L1, about the voice stock piece P containing the vowel other than the specific vowel, out of plural voice stock pieces P.SELECTED DRAWING: Figure 4

Description

  The present invention relates to a technology for synthesizing speech by connecting speech segments.

  2. Description of the Related Art A unit connection type speech synthesis technique for synthesizing synthesized speech of a desired tone color by connecting a plurality of speech units collected from recorded speech to each other is known. For example, in Patent Document 1, a plurality of stationary sound data is recorded in advance for each speech unit, and a plurality of stationary sound data (speech units) are selected according to a target pitch and interpolated with each other. A configuration for generating pitch steady sound data is disclosed.

JP2013-11863A

  By the way, in order to synthesize speech of various voice qualities, it is necessary to prepare a library of speech segments for each voice quality. For example, assuming the synthesis of two types of speech that have different phoneme (pronunciation content) intelligibility, a library of ambiguous speech is created by recording speech that is pronounced vaguely by the speaker, while the speaker is clearly There is a problem that it is necessary to create a library of clear voices by recording pronunciation sounds, and the burden of creating the library is large. In particular, since it is not easy to continuously produce a sound while maintaining a constant degree of ambiguity or clarity, this results in an excessive burden on the speaker. In view of the above circumstances, an object of the present invention is to generate synthesized speech that is highly perceived by the listener as being ambiguous while reducing the burden of creating a speech synthesis library.

  In order to solve the above problems, a speech synthesis library generating apparatus according to the first aspect of the present invention generates a second library from a first library for speech synthesis including segment data for each speech segment. A second library containing speech unit data of speech units obtained by replacing the vowels with the specific vowels in the first library for speech units containing vowels other than the specific vowels among a plurality of speech units. A generating unit for generating According to the above configuration, since the second library is generated from the existing first library, compared to the case where the first library and the second library are separately prepared from different recorded sounds, the second library is used for speech synthesis. It is possible to reduce the burden of creating a library. In addition, for speech units containing vowels other than specific vowels, the second library includes speech unit data in which the vowels are replaced with specific vowels in the first library, so that it is perceived as ambiguous by the listener. Can be generated from the segment data of the second library. For example, the voice of “U” tends to be perceived by listeners when the degree of mouth opening is small compared to other vowels (“A”, “I”, “E”, “O”) and the phoneme is ambiguous. There is. Considering the above tendency, a configuration in which “u” is a specific vowel is preferable.

  In a preferred example of the speech synthesis library generating device according to the first aspect, the generating unit is configured to generate a speech element including a specific consonant and a vowel whose acoustic characteristics approximate to the specific vowel in the first library. A second library including segment data indicating the segment of the specific consonant among the segment data is generated. In the above aspect, for the speech unit including the specific consonant and the vowel, the second library including the segment data indicating the segment of the specific consonant among the unit data of the speech unit in the first library is generated. Even if no segment data corresponding to a speech segment including a specific vowel exists in the first library, a synthesized speech that is perceived as ambiguous by the listener is generated using the segment data indicating the segment of the specific consonant. It is possible.

  In a preferred example of the speech synthesis library generating apparatus according to the first aspect, the generating unit is different from the first consonant in the first library with respect to a speech element including a first consonant and a vowel. A second library including segment data of speech segments including consonants and the specific vowel is generated. In the above aspect, for the speech unit including the first consonant and the vowel, the second including the segment data of the speech unit including the second consonant different from the first consonant and the specific vowel in the first library. Since the library is generated, even if no segment data including the first consonant and the specific vowel exists in the first library, the listener uses the segment data of the speech segment including the second consonant and the specific vowel. It is possible to generate synthesized speech that is perceived as ambiguous.

  In order to solve the above problems, a speech synthesizer according to a second aspect of the present invention is an element that generates a second library from a first library for speech synthesis that includes segment data for each speech segment. Generating a second library including speech unit data of speech units obtained by replacing the vowels with the specific vowels in the first library for speech units including vowels other than the specific vowels among the plurality of speech units. An element for selecting a first unit data of a speech unit corresponding to a phonetic character to be synthesized in the first library and a second unit data of the speech unit in the second library; A synthesis that generates a synthesized speech using a segment selection unit, a segment mixing unit that mixes the first segment data and the second segment data, and segment data after mixing by the segment mixing unit And a processing unit. In the second mode, for the speech unit containing vowels other than the specific vowel, the unit data of the speech unit obtained by replacing the vowel with the specific vowel in the first library is included in the second library. Similarly, it is possible to reduce the burden of creating a speech synthesis library. In addition, since synthesized speech is generated using the segment data after mixing the first segment data and the second segment data, it is perceived as ambiguous (or clear) by the listener according to the degree of mixing. It is possible to generate synthesized speech.

  In order to solve the above problems, the speech synthesizer according to the third aspect of the present invention provides a speech synthesis unit corresponding to a phonetic character to be synthesized from a speech synthesis library including segment data for each speech unit. An element for selecting element data, and for a speech element including a vowel other than a specific vowel, the element selection unit for selecting element data of a speech element obtained by replacing the vowel with the specific vowel, and And a synthesis processing unit that generates synthesized speech using the speech unit selected by the unit selection unit. In the third aspect, for speech units containing vowels other than specific vowels, the unit data of speech units obtained by replacing the vowels with specific vowels are selected from the existing library for speech synthesis. Compared with a configuration in which a plurality of libraries corresponding to the above are prepared in advance, it is possible to generate synthesized speech that is perceived as ambiguous by the listener while reducing the burden of creating a speech synthesis library. Further, compared with the first mode in which the second library generated from the first library needs to be stored, the third mode also has an advantage that the storage of the second library is unnecessary.

  In a preferred example of the speech synthesizer according to the third aspect, the unit selection unit is configured to generate, for a speech unit including a specific consonant and a vowel whose acoustic characteristics approximate to the specific vowel, Of these, segment data indicating the segment of the specific consonant is selected. In the above aspect, for a speech unit including a specific consonant and a vowel whose acoustic characteristics approximate to the specific vowel, the segment data indicating the segment of the specific consonant is selected from the segment data of the speech unit. Generates synthesized speech that is perceived as ambiguous by the listener by using segment data indicating a specific consonant segment even if the segment data corresponding to the speech segment containing the specific vowel does not exist in the speech synthesis library The effect of being able to do is realized.

  In a preferred example of the speech synthesizer according to the third aspect, the unit selection unit obtains a second consonant different from the first consonant and the specific vowel for a speech unit including a first consonant and a vowel. Select the speech segment data to include. In the above aspect, since the speech unit including the second consonant and the specific vowel different from the first consonant is selected for the speech unit including the first consonant and the vowel, the first consonant and the specific are selected. Generates synthesized speech that is perceived as ambiguous by listeners using segment data of speech units including second consonants and specific vowels, even if segment data including vowels does not exist in the speech synthesis library The effect of being able to do is realized.

  In a preferred example of the speech synthesizer according to the third aspect, the unit selection unit specifies the first unit data of the speech unit and the specific vowel for the speech unit including a vowel other than the specific vowel. Selecting a second unit data of a speech unit replaced with a vowel, and comprising a unit mixing unit that mixes the first unit data and the second unit data, and the synthesis processing unit includes: A synthesized speech is generated using the segment data after mixing by the segment mixing unit. In the above aspect, since the first unit data and the second unit data are mixed, the voice of various voice qualities is compared with the configuration in which the synthesized speech is generated using only the second unit data. There is an advantage that it can be generated.

  In a preferred example of the speech synthesizer according to the second aspect and the third aspect, the speech synthesizer includes a variable setting unit that sets a mixing ratio according to the feature quantity to be synthesized, and the variable setting unit sets the segment mixing unit The first segment data and the second segment data are mixed at the mixing ratio. In the above configuration, since the mixing ratio between the first segment data and the second segment data is set according to the feature quantity to be combined, for example, the user is compared with the configuration in which the user instructs the mixing ratio. It is possible to reduce the burden. A good example of the feature quantity is pitch. For example, on the premise of the tendency that the higher the sound range, the more likely the phoneme is perceived as ambiguous, the variable setting unit sets the mixing ratio so that the higher the pitch of the synthesis target, the higher the ratio of the second segment data. A configuration for setting is preferable.

  The library generation device and the speech synthesizer according to each of the above aspects are realized by dedicated hardware (electronic circuit), or by cooperation of a general-purpose arithmetic processing device such as a CPU (Central Processing Unit) and a program. Realized. The program of the present invention can be provided in a form stored in a computer-readable recording medium and installed in the computer. The recording medium is, for example, a non-transitory recording medium, and an optical recording medium (optical disk) such as a CD-ROM is a good example, but a known arbitrary one such as a semiconductor recording medium or a magnetic recording medium This type of recording medium can be included. For example, the program of the present invention can be provided in the form of distribution via a communication network and installed in a computer. The present invention is also specified as the operation method (library generation method) of the speech synthesis library generation device and the operation method (speech synthesis method) of the speech synthesis device according to each aspect described above.

It is a block diagram of the speech synthesizer concerning a 1st embodiment. 6 is an explanatory diagram of an edit screen 40. FIG. It is explanatory drawing of the replacement process of the segment data Q by the production | generation part 32. FIG. It is explanatory drawing of the segment data Q corresponding to the speech segment P. It is a flowchart of operation | movement of the library production | generation process of 1st Embodiment. It is a block diagram of the speech synthesizer concerning a 2nd embodiment. It is a flowchart of the operation | movement of the process of the segment selection of 2nd Embodiment. It is a block diagram of the speech synthesizer concerning a 3rd embodiment. It is explanatory drawing of the setting of the mixture ratio K in 3rd Embodiment. It is a block diagram of the speech synthesizer in 4th Embodiment. It is explanatory drawing about the interpolation process in the speech synthesizer which concerns on a modification. It is explanatory drawing about the interpolation process in the speech synthesizer which concerns on a modification.

<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 according to the first embodiment generates a speech signal V of a singing voice of an arbitrary piece of music (hereinafter referred to as “synthetic music”) by a unit connection type speech synthesis that connects a plurality of speech units. Device.

  As illustrated in FIG. 1, the speech synthesizer 100 includes a computer system (for example, a mobile phone, a personal computer, or the like) that includes an arithmetic processing device 10, a storage device 12, a display device 14, an input device 16, and a sound emitting device 18. Information processing device). The display device 14 (for example, a liquid crystal display panel) displays an image instructed from the arithmetic processing device 10. The input device 16 is an operation device (for example, a pointing device such as a mouse or a keyboard) operated by the user for various instructions to the speech synthesizer 100, and includes a plurality of operators operated by the user, for example. Is done. Note that a touch panel configured integrally with the display device 14 may be employed as the input device 16. The sound emitting device 18 (for example, a speaker or headphones) reproduces sound according to the audio signal V.

  The storage device 12 stores a program PGM executed by the arithmetic processing device 10 and various data used by the arithmetic processing device 10. 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 arbitrarily employed as the storage device 12. The storage device 12 of the first embodiment stores the first library L1 and the synthesis information S as exemplified below.

  The first library L1 is a set (speech synthesis library) of a plurality of segment data Q corresponding to different speech segments P. One speech element P in the first embodiment is a phoneme chain (diphone) in which two phonemes are connected. The unit data Q of one speech unit P in the first library L1 is data representing the speech waveform of the section corresponding to the speech unit P in the recorded speech of a specific speaker. Specifically, each piece of data Q can be expressed as a time series of a time domain speech waveform sample or a frequency domain spectrum time series calculated for each frame of the speech waveform. Each piece data Q of the first library L1 of the present embodiment represents a voice that allows the listener to clearly perceive the phoneme.

  The synthesis information S is time-series data for specifying the singing voice of the synthesized music, and for each note constituting the synthesized music, the pronunciation character X1, the pronunciation period X2, and the pitch (for example, note number) X3 are designated in time series. . The pronunciation character X1 designates the pronunciation content of the speech to be synthesized (that is, the lyrics of the synthesized music). The sound generation period X2 is defined by, for example, a sound generation start time and duration (or end time).

  The arithmetic processing unit 10 (CPU) of FIG. 1 executes a program PGM stored in the storage unit 12 to thereby edit a plurality of functions (generation unit 32, display for editing the synthesis information S and generating the audio signal V). A control unit 33, an element selection unit 34, an instruction reception unit 35, and a composition processing unit 36) are realized. A configuration in which each function of the arithmetic processing device 10 is distributed to a plurality of devices, or a configuration in which a dedicated electronic circuit (for example, DSP) realizes a part of the functions of the arithmetic processing device 10 may be employed.

  The instruction receiving unit 35 receives an instruction from a user according to an operation on the input device 16. The display control unit 33 displays various images on the display device 14. Specifically, the display control unit 33 of the first embodiment causes the display device 14 to display the editing screen 40 of FIG. 2 for the user to confirm the content of the composite music specified by the composite information S. As illustrated in FIG. 2, the note string image 42 of the editing screen 40 is a piano roll type coordinate plane in which a time axis (horizontal axis) and a pitch axis (vertical axis) intersecting each other are set.

  The display control unit 33 arranges the note image 54 and the pronunciation character X1 on the editing screen 40 for each note designated by the synthesis information S. The musical note iconic image 54 is an image representing each musical note of the synthesized music. Specifically, the position of the musical note iconic image 54 in the direction of the pitch axis is set according to the pitch X3 specified by the synthesis information S, and the position and display length of the musical note iconic image 54 in the direction of the time axis are determined by the synthesis information. S is set according to the sound generation period X2 designated. The phonetic character X1 is arranged in the vicinity of the head of the musical note iconic image 54.

  The user may instruct addition or movement of the note image 54 or addition or change of the pronunciation character X1 by appropriately operating the input device 16 while confirming the note string image 42 on the editing screen 40 of FIG. Is possible. The display control unit 33 updates the editing screen 40 (the musical note iconic image 54 and the pronunciation character X1) according to the instruction received from the user by the instruction receiving unit 35.

  The segment selection unit 34 in FIG. 1 sequentially selects the segment data Q of the speech segment P corresponding to the phonetic character X1 for each note designated by the synthesis information S from the first library L1. The synthesis processing unit 36 generates the audio signal V using the segment data Q selected from the first library L1 by the segment selection unit 34 and the synthesis information S. Specifically, the synthesis processing unit 36 adjusts the speech waveform indicated by the segment data Q selected by the segment selection unit 34 to the pitch X3 and the sound generation period X2 specified by the synthesis information S, and connects them to each other. As a result, the audio signal V is generated. The voice signal V generated by the synthesis processing unit 36 is supplied to the sound emitting device 18 so that the singing voice of the synthesized music is reproduced. As described above, since each piece data Q represents sound with clear pronunciation, the reproduced sound of the sound signal V generated using the first library L1 is perceived as clear by the listener.

  1 generates a second library L2 for generating a synthesized speech in which the phoneme is perceived as ambiguous compared to the speech signal V generated using the first library L1. Generate from L1. Similar to the first library L1, the second library L2 is a set of a plurality of unit data Q corresponding to different speech units P. As illustrated in FIG. 1, the second library L2 generated by the generation unit 32 is stored in the storage device 12 together with the first library L1 and used for generation of the audio signal V by the synthesis processing unit 36.

  Specifically, the generation unit 32 uses the vowel in the first library L1 for a speech unit P including a vowel other than a specific vowel that the listener tends to perceive as ambiguous (hereinafter referred to as “specific vowel”). The second library L2 including the segment data Q of the speech segment P in which is replaced with a specific vowel is generated. The vowel [M] (U) may be perceived as ambiguous compared to other vowels ([a], [i], [e], [o]), because the degree of mouth opening during pronunciation may be small. There is a tendency to be easy. Considering the above tendency, in the first embodiment, the vowel [M] is exemplified as the specific vowel. In addition, the notation of each phoneme is based on X-SAMPA (Extended Speech Assessment Method Phonetic Alphabet). The symbol “Sil” means silence. In the following description, the symbol “V” means a vowel, and the symbol “C” means a consonant.

  FIG. 3 is a diagram for specifically explaining the replacement processing by the generation unit 32. FIG. 3 shows the content of the segment data Q corresponding to the speech segment P common to the first library L1 and the second library L2. Since one speech element P in the first library L1 is a phoneme chain obtained by connecting two phonemes, the speech element P including two vowels is used as the speech element P including vowels other than the specific vowel. And two types of speech units including vowels and consonants are assumed.

(1) Vowel + Vowel: [VV]
FIG. 3A is an explanatory diagram of the unit data Q of the speech unit P [VV] in which two vowels are arranged in the second library L2. As illustrated in FIG. 3 (a), the first library L1 includes speech segments P ([aa], [ii], [[] corresponding to five types of vowels (A, I, U, E, O)). As for MM], [ee], [oo]), segment data Q ([aa], [ii], [MM], [ee], [oo]) representing the speech waveform of the speech segment P is stored. Is done. The generating unit 32 generates a vowel other than the specific vowel ([a], [i], [oo]) from each speech unit P [VV] ([aa], [ii], [MM], [ee], [oo]). The unit data Q [MM] stored in the first library L1 is duplicated for the speech unit P [MM] in which [e], [o]) is replaced with the specific vowel [M], and the corresponding unit in the second library L2 Stored as unit data Q of speech unit P [VV]. For example, for the speech unit [aa] in the second library L2, the segment data Q [MM] stored in the first library L1 for the speech unit [MM] obtained by replacing the phoneme [a] with the specific vowel [M]. ] Is duplicated. In the above description, the speech unit P composed of combinations of the same vowels is illustrated, but the speech unit P composed of combinations of different vowels ([ai], [aM], [ae ], [ao], [ia], [iM], [ie], [io], [Ma], [Mi], [Me], [Mo], [oa], [oi], [oM], Similarly, for [oe]), the first speech unit P [MM] in which vowels other than the specific vowel ([a], [i], [e], [o]) are replaced with the specific vowel [M] is the first. The segment data Q [MM] stored in the library L1 is duplicated and stored as the segment data Q of the speech segment P [VV] in the second library L2.
Note that for the speech unit P [MM] composed only of specific vowels, the segment data Q [MM] of the speech unit [MM] in the first library L1 is copied to the second library L2.

(2) Vowel + consonant or consonant + vowel: [C, V] ([CV], [VC])
FIG. 3B is an explanatory diagram of the unit data Q of the speech unit P [C, V] in which vowels and consonants are arranged in the second library L2. As illustrated in FIG. 3B, the generation unit 32 replaces the vowels other than the specific vowel among the speech units P [C, V] with the specific vowel [M]. The segment data Q [C, M] stored in the first library L1 is duplicated and stored as the segment data Q of the speech segment [C, V] in the second library L2. For example, for the speech unit P [sa] of the second library L2, the unit data Q [stored in the first library L1 for the speech unit [sM] obtained by replacing the vowel [a] with the specific vowel [M]. sM] is copied to the second library L2.

  As described above, FIGS. 3 (a) and 3 (b) exemplify processing when the segment data Q of the speech segment P including the specific vowel [M] is present in the first library L1. However, the segment data Q of the specific speech segment P including the specific vowel [M] may not exist in the first library L1. The exception processing executed in the above case will be described.

(3) Exception handling 1
As illustrated in FIG. 3 (c), consonants whose acoustic characteristics (for example, phonemes perceived by the listener) approximate to specific vowels [M] (hereinafter referred to as “specific consonants”) [w] other than specific vowels [M] Speech unit P ([wa], [wi], [we], [wo]) and unit data Q ([wa], [wi], [we], [wo])) followed by each vowel Is stored in the first library L1. On the other hand, it is assumed that the speech unit P [wM] and the segment data Q [wM] in which each vowel other than the specific vowel [M] follows the specific consonant [w] is not stored in the first library L1.

  Of the segment data Q [wV] of the speech segment P, the generating unit 32 includes the specific consonant [w] and the vowel [V] ([a] in the example of FIG. 3C) in the first library L1. Then, the second library L2 including the segment data Q [w] indicating the section of the specific consonant [w] for the speech segment P [wV] is generated. Specifically, as illustrated in FIG. 3C, the segment representing the segment of the specific consonant [w] in the segment data Q [wa] of the speech segment P [wa] in the first library L1. Data Q [w] is duplicated as unit data Q of speech unit P [wV] in the second library L2.

  As understood from the above description, in the first embodiment, for the speech unit P [wV] including the specific consonant [w] and the vowel [V], the unit data of the speech unit P in the first library L1. A second library L2 including the segment data Q of the specific consonant [w] section of Q is generated. Therefore, even when the segment data Q of the speech segment P including the specific vowel [M] does not exist in the first library L1, the segment data Q [w] indicating the section of the specific consonant [w] is used, It is possible to generate synthesized speech corresponding to ambiguous pronunciation.

(4) Exception handling 2
As illustrated in FIG. 3D, the first library L1 includes speech units including consonants (hereinafter referred to as “first consonants”) [h] and vowels [a], [e], [o]. For P ([ha], [he], [ho]), segment data Q ([ha], [he], [ho]) is stored, but the first consonant [h] and other vowels For the speech unit P ([hi], [hM]) including [i] and [M], it is assumed that the unit data Q is not stored. On the other hand, the first library L1 contains consonants (hereinafter referred to as “second consonants”) [p \] and vowels [a], [M], [e], [o] that are different from the first consonant [h]. Segment data Q ([p \ -a], [p \ -M], [p \ -e], [p \ -o]) [p \ -M], [p \ -e], [p \ -o]) are stored.

  The generation unit 32 generates the second library L2 including the segment data Q [p \ -M] including the second consonant [p \] and the specific vowel [M] in the first library L1. Specifically, as illustrated in FIG. 3D, the second consonant [p \] and the specific vowel [M] in the segment data Q (for example, [p \ -a]) of the first library L1. Is duplicated as segment data Q of the speech unit P [hV] of the second library L2.

  As understood from the above description, in the first embodiment, the second consonant [p] that is different from the first consonant [h] for the speech segment P including the first consonant [h] and the vowel [V]. \] And the second library L2 including the segment data Q of the speech segment P including the specific vowel [M] is generated, so the segment data including the first consonant [h] and the specific vowel [M]. Even if Q does not exist in the first library L1, the synthesized speech with ambiguous pronunciation is generated using the segment data Q of the speech segment P including the second consonant [p \] and the specific vowel [M]. Is possible.

  In FIG. 3A to FIG. 3D, the speech segment P including the vowel [V] has been described. For speech units P (for example, [Sil-k], [k-Sil], etc.) that do not contain a vowel [V], such as speech units P in which consonants [C] and silences [sil] are arranged, The segment data Q corresponding to the speech segment P in one library L1 is copied to the second library L2.

  As understood from the above description, the first library L1 and the second library L2 share the same kind of speech element P for which the element data Q is prepared, but are prepared for each speech element P. The voice waveform represented by the piece data Q can be different. The second library L2 generated by the above procedure is used for generating the audio signal V in the same manner as the first library L1. That is, the segment selection unit 34 sequentially selects the segment data Q of the speech segment P corresponding to the phonetic character X1 from the second library L2, and the unit segment data Q selected by the segment selection unit 34 is synthesized. The sound signal V of the sound perceived by the listener as ambiguous is generated by the adjustment and connection by the unit 36. In the first embodiment, clear speech synthesis using the first library L1 and ambiguous speech synthesis using the second library L2 are selected according to an instruction from the user to the input device 16, for example. Is executed automatically.

  FIG. 4 is an explanatory diagram of speech synthesis using the first library L1 and speech synthesis using the second library L2. FIG. 4A illustrates a time series of the speech segment P selected corresponding to the phonetic character X1 “Maya” specified by the synthesis information S, and FIG. 4B illustrates the first library. The time series of the segment data Q read from the first library L1 at the time of speech synthesis using L1 is illustrated, and FIG. 4 (c) is read from the second library L2 at the time of speech synthesis using the second library L2. The time series of the segment data Q is illustrated.

  When speech synthesis (clear speech synthesis) using the first library L1 is instructed, the segment selection unit 34 performs segment data Q ([Sil-m], [ ma], [aa], [aj], [ja], [aa], [a-Sil]) are sequentially selected from the first library L1. The synthesis processing unit 36 uses the segment data Q selected by the segment selection unit 34 to generate synthesized speech in which the phonetic character X1 “maya” is clearly perceived by the listener.

  On the other hand, when speech synthesis using the second library L2 (synthesized ambiguous speech) is instructed, the segment selection unit 34 selects the segment data Q ([Sil-m], [mM] in FIG. ], [MM], [Mj], [jM], [MM], [M-Sil]) are sequentially selected from the second library L2. The synthesis processing unit 36 uses the segment data Q selected by the segment selection unit 34 to make the pronunciation of the phonetic character X1 “maya” ambiguous. muyu) ”is generated.

  FIG. 5 is a flowchart of processing in which the generation unit 32 of the first embodiment generates the second library L2. For example, the process of FIG. 5 is started when the generation of the second library L2 is designated by the user.

  The generation unit 32 selects an arbitrary speech unit (hereinafter referred to as “target unit”) P (SA1), and determines whether the target unit P includes a vowel (SA2). When the target segment P does not include a vowel (SA2: NO), the generation unit 32 uses the segment data Q stored in the first library L1 for the target segment P as an element of the target segment P in the second library L2. Duplicate as single data Q (SA3). On the other hand, when the target segment P includes a vowel (SA2: YES), the generation unit 32 uses the first library to store the segment data Q of the speech segment P obtained by replacing the vowel of the target segment P with the specific vowel [M]. It is determined whether or not it exists in L1 (SA4). When the determination result is affirmative (SA4: YES), the generation unit 32 acquires the segment data Q of the speech segment P obtained by replacing the vowel of the target segment P with the specific vowel [M] from the first library L1. Then, it is stored as segment data Q of the target segment P in the second library L2 (SA5). On the other hand, when the segment data Q of the speech segment P obtained by replacing the vowel of the target segment P with the specific vowel [M] does not exist in the first library L1 (SA4: NO), the generation unit 32 performs the above-described exception processing. By executing 1 or exception processing 2, the segment data Q of the target segment P in the second library L2 is prepared (SA6). The above processing is sequentially repeated for all types of speech units P (SA7: NO), thereby generating the second library L2 including the unit data Q for each speech unit P.

  As described above, in the first embodiment, vowels other than the specific vowel [M] ([a], [i], [e], [o]) among the plurality of speech units P in the first library L1. ), The segment data Q of the speech unit P in which the vowel is replaced with the specific vowel [M] is copied to the second library L2. That is, using the segment data Q of the speech segment P of the existing first library L1, the second library L2 used for the synthesized speech with ambiguous pronunciation is generated. Therefore, compared with the case where the first library L1 corresponding to clear speech and the second library L2 corresponding to ambiguous speech are individually generated from actual recorded sounds, the burden of creating a speech synthesis library is reduced. While mitigating, it is possible to generate synthesized speech that is perceived as ambiguous by the listener.

Second Embodiment
In the first embodiment, the second library L2 is generated in advance from the existing first library L1 and used for speech synthesis. In the second embodiment, in the scene in which the audio signal V corresponding to the synthesis information S is generated, the segment is used under the same rules as in the case where the second library L2 is generated from the first library L1 in the first embodiment. The selection unit 34 dynamically selects the segment data Q from the existing library L, thereby generating an ambiguous audio signal V without generating the second library L2. In addition, about the element in which an effect | action and a function are the same as that of 1st Embodiment in each aspect illustrated below, the detailed description of each is abbreviate | omitted suitably using the code | symbol referred by description of 1st Embodiment.

  FIG. 6 is a block diagram of the speech synthesizer 100 in the second embodiment. In the second embodiment, the generation unit 32 and the second library L2 of the first embodiment are omitted. The library L in FIG. 6 corresponds to the first library L1 of the first embodiment, and includes clear speech segment data Q for each speech segment P. When the user gives an instruction to generate clear pronunciation synthesized speech, the segment selection unit 34 of the second embodiment uses the pronunciation character X1 as in the first embodiment (illustrated in FIG. 4B). The segment data Q of each corresponding speech segment P is sequentially selected from the library L. On the other hand, when the user gives an instruction to generate an ambiguous pronunciation synthesized speech, the segment selection unit 34 is substantially the same as the replacement process described above with reference to FIGS. 3 (a) to 3 (d). 4 is sequentially selected from the library L in accordance with the above rule. The segment selection by the segment selection unit 34 will be specifically described.

(1) Vowel + Vowel: [VV]
When the speech segment P corresponding to the phonetic character X1 includes a vowel other than the specific vowel [M] ([a], [i], [e], [o]), the segment selection unit 34 selects the vowel. The segment data Q of the speech segment P replaced with the specific vowel [M] is selected from the library L.

(2) Vowel + consonant or consonant + vowel: [C, V] ([VC], [CV])
When the speech segment P corresponding to the phonetic character X1 includes a vowel ([a], [i], [e], [o]) other than the specific vowel [M] and a consonant, The segment data Q of the speech segment P in which the vowels other than the specific vowel [M] are replaced with the specific vowel [M] is selected from the library L.

(3) Exception handling 1
When the speech unit P corresponding to the phonetic character X1 is a speech unit P [wV] including the specific consonant [w] and the vowel [V] other than the specific vowel [M], the unit selection unit 34 A segment data Q [wV] of a speech segment P [wV] including a specific consonant [w] and a vowel [V] other than the specific vowel [M] is selected from the library L, and the segment data Q [wV] The segment data Q [w] indicating the interval of the specific consonant [w] is generated.

(4) Exception handling 2
When the speech segment P corresponding to the phonetic character X1 is a speech segment P [hV] including the first consonant [h] and the vowel [V] other than the specific vowel [M], the segment selection unit 34 The segment data Q of the speech segment P including the second consonant [p \] different from the first consonant [h] and the specific vowel [M] is selected from the library L.

  The synthesis processing unit 36 generates synthesized speech using the unit data Q of the speech unit P selected by the unit selection unit 34. In the second embodiment, the segment data Q is selected under the same rules as in the first embodiment. Therefore, as in the first embodiment, the synthesized speech with clear pronunciation and the synthesized speech with unclear pronunciation are It is possible to generate both.

  FIG. 7 is a flowchart of processing in which the segment selection unit 34 of the second embodiment selects the segment data Q. For example, when the start of speech synthesis is instructed by the user, the processing of FIG. 7 is sequentially executed for each speech segment (target segment) P corresponding to each phonetic character X1 specified by the synthesis information S. .

  The segment selection unit 34 determines whether or not the target segment P includes a vowel other than the specific vowel [M] (SB1). When the target segment P does not include a vowel other than the specific vowel [M] (SB1: NO), the segment selection unit 34 selects the segment data Q of the target segment P from the library L (SB2). On the other hand, when the target segment P includes a vowel other than the specific vowel [M] (SB1: YES), the segment selection unit 34 replaces the segment data of the speech segment P by replacing the vowel with the specific vowel [M]. It is determined whether or not Q exists in the library L (SB3). When the determination result is affirmative (SB3: YES), the segment selection unit 34 selects the segment data Q of the speech segment P in which the vowel of the target segment P is replaced with the specific vowel [M] from the library L. (SB4). On the other hand, when the corresponding segment data Q does not exist in the library L (SB3: No), the segment selector 34 prepares the segment data Q by the exception processing 1 or the exception processing 2 described above (SB5). Since the subsequent steps are the same as those in the first embodiment, detailed description thereof is omitted.

  As understood from the above description, in the second embodiment, the segment selection unit 34 does not include vowels ([a], [i], [e], [o]) other than the specific vowel [M]. For the speech unit P, the unit data Q of the speech unit P is selected, and the speech unit P including vowels ([a], [i], [e], [o]) other than the specific vowel [M]. , The segment data Q of the speech segment P in which the vowel is replaced with the specific vowel [M] is selected. That is, it is not necessary to prepare separate libraries corresponding to clear and ambiguous sounds. Therefore, in the second embodiment, similarly to the first embodiment, it is possible to generate a synthesized speech with ambiguous pronunciation while reducing the burden of creating a speech synthesis library. Further, in the second embodiment, generation and storage of the second library L2 are unnecessary, so that both the first library L1 and the second library L2 need to be stored in the storage device 12 as compared with the first embodiment. Thus, there is an advantage that the capacity required for the storage device 12 is reduced.

<Third Embodiment>
In the second embodiment, the speech element V corresponding to the phonetic character X1 is connected to each other and the speech data V corresponding to the phonetic character X1 is generated. An operation of generating an ambiguous speech signal V by selectively connecting each segment data Q of speech units obtained by replacing a segment of vowels with a specific vowel [M] was performed. In the third embodiment, for each speech segment P corresponding to the phonetic character X1, the former segment data Q (hereinafter referred to as "first segment data Q1") and the latter segment data Q (hereinafter referred to as "second segment"). Is referred to as “single data Q2”).

  FIG. 8 is a block diagram of the speech synthesizer 100 of the third embodiment. In the third embodiment, a variable setting unit 37 and an element mixing unit 38 are added to the configuration of the second embodiment. The segment selection unit 34 of the third embodiment selects, from the library L, first segment data Q1 and second segment data Q2 for each speech segment P corresponding to the phonetic character X1 specified by the synthesis information S. To do. As described above, the first unit data Q1 is the unit data Q stored in the library L for each speech unit P corresponding to the phonetic character X1, and the second unit data Q2 is the same as in FIG. For example, the unit data Q of the speech unit P is obtained by replacing the vowel of the speech unit P corresponding to the phonetic character X1 with the specific vowel [M]. In other words, the first unit data Q1 represents the speech waveform of the speech unit P that is clearly pronounced, and the second unit data Q2 represents the speech waveform when the speech unit P is pronounced vaguely. .

The unit mixing unit 38 mixes the first unit data Q1 and the second unit data Q2 selected by the unit selection unit 34 for one speech unit P in accordance with the mixing ratio K. Data Q is generated. Specifically, the unit mixing unit 38, as expressed by the following mathematical formula (1), applies the first unit data Q1 and the second unit data Q2 to which a weight value corresponding to the mixing ratio K is applied. A weighted sum is generated as segment data Q.

  The symbol α in the formula (1) is a predetermined constant. As understood from the equation (1), the larger the mixing ratio K is, the more the ratio of the second segment data Q2 in the segment data Q increases, resulting in the audible ambiguity of the speech of the speech signal V. Become. The synthesis processing unit 36 generates the audio signal V by adjusting and connecting the unit data Q after mixing by the unit mixing unit 38 in the same manner as in the first embodiment.

  The variable setting unit 37 in FIG. 8 sets the mixing ratio K between the first segment data Q1 and the second segment data Q2 in accordance with the feature amount of the speech to be synthesized. Here, a tendency is assumed that the higher the sound, the vaguely the phoneme perceived by the listener. In consideration of the above tendency, the variable setting unit 37 of the third embodiment sets the mixing ratio K corresponding to the pitch X, using the pitch X3 specified for the note to be synthesized as a feature amount. Specifically, the mixing ratio K is controlled such that the mixing ratio K increases as the pitch X3 of the note increases (that is, the voice of the voice signal V becomes ambiguous).

  FIG. 9 is an explanatory diagram for setting the mixture ratio K by the variable setting unit 37. In the note sequence image 42 in the editing screen 40 illustrated in FIG. 9, the case where the pronunciation character X1 “asayakeno” designated by the user is assigned to five notes is illustrated. . In the third embodiment, the mixing ratio K between the first segment data Q1 and the second segment data Q2 is set for each note in a specific section t on the time axis. The section t is a section for which the mixing ratio K is set, and is an arbitrary section such as a section selected by the user by an operation on the input device 16 or all sections of the synthesized music.

数式(2)の記号βは所定の定数（典型的には正数）である。数式(2)から理解される通り、音高Ｘ3が区間ｔ内の最小値ＮLである音符（Ｘ3＝ＮL）の混合比Ｋは最小値０となり、音高Ｘ3が区間ｔ内の最大値ＮHである音符（Ｘ3＝ＮH）の混合比Ｋは最大値βとなる。すなわち、混合比Ｋは、音高Ｘ3が高いほど増加するように最小値０と最大値βとの間の範囲内で音高Ｘ3に応じて変化する。図９に例示される通り、編集画面４０の変数領域４４には、音符毎の混合比Ｋが表示（グラフ表示）される。 The symbol R in FIG. 9 means the difference between the maximum value NH and the minimum value NL of the pitch X3 in the section t (that is, the distribution width of the pitch X3 in the section t) (R = NH-NL). The variable setting unit 37 calculates the mixing ratio K corresponding to the pitch X3 of the note for each note in the section t by the calculation of the following formula (2).

The symbol β in the equation (2) is a predetermined constant (typically a positive number). As understood from the equation (2), the mixing ratio K of the notes (X3 = NL) whose pitch X3 is the minimum value NL in the interval t is the minimum value 0, and the pitch X3 is the maximum value NH in the interval t. The mixing ratio K of the notes (X3 = NH) is the maximum value β. That is, the mixing ratio K changes in accordance with the pitch X3 within the range between the minimum value 0 and the maximum value β so that it increases as the pitch X3 increases. As illustrated in FIG. 9, the mixing ratio K for each note is displayed (graph display) in the variable area 44 of the editing screen 40.

  As understood from the above description, in the third embodiment, for the speech unit P corresponding to the phonetic character X1, the first unit data Q1 of the speech unit in the library L and the vowel in the library L are used. The second unit data Q2 of the speech unit P replaced with the specific vowel [M] is mixed. Therefore, similarly to the first embodiment and the second embodiment, it is possible to generate a synthesized speech with an ambiguous pronunciation while reducing the burden of creating a speech synthesis library.

  By the way, as a configuration for mixing the first segment data Q1 and the second segment data Q2, for example, a configuration for executing mixing at a mixing ratio K in accordance with an instruction from the user (hereinafter referred to as “comparative”). Can also be employed. However, in contrast, there may be a problem that the burden on the user who indicates the mixture ratio K is large. In the third embodiment, since the mixing ratio K is variably set according to the feature quantity to be synthesized (specifically, the pitch X3 of each note), the burden on the user is reduced compared to the proportionality. It is possible. In particular, in the first embodiment, the mixing ratio K is set so that the ratio of the second segment data Q2 increases as the pitch X3 of each note is higher. There is an advantage that it is possible to generate natural synthesized speech that reproduces.

<Fourth embodiment>
In the fourth embodiment, on the premise of the configuration of the first embodiment in which the second library L2 is generated in advance from the first library L1, the first segment data Q1 selected from the first library L1 and the second library L2 The second unit data Q2 of the speech unit P selected from the above is mixed as in the third embodiment to generate a synthesized speech.

  FIG. 10 is a block diagram of the speech synthesizer 100 of the fourth embodiment. The generating unit 32 in FIG. 10 generates the second library L2 from the existing first library L1 by the same method as in the first embodiment. Therefore, according to the fourth embodiment, the same effect as the first embodiment is realized. On the other hand, the segment selection unit 34 obtains the first segment data Q1 of the first library L1 and the second segment data Q2 of the second library L2 for each speech segment P corresponding to the synthesized character X1. Select sequentially.

  As illustrated in FIG. 10, in the fourth embodiment, a variable setting unit 37 and a segment mixing unit 38 similar to those in the third embodiment are added to the speech synthesizer 100 of the first embodiment. As in the third embodiment, the variable setting unit 37 sets the mixing ratio K for each note in accordance with the pitch X3 that is the feature value for each note. Specifically, the mixing ratio K is calculated by, for example, the calculation of Equation (2) so that the mixing ratio K increases as the pitch X3 increases. Similar to the third embodiment, the unit mixing unit 38 sets the first unit data Q1 and the second unit data Q2 selected by the unit selection unit 34 to the mixing ratio K set by the variable setting unit 37. The segment data Q is generated by mixing them accordingly. Therefore, according to the fourth embodiment, similarly to the third embodiment, it is possible to reduce the burden on the user by comparing the mixing ratio K with the proportionality instructed by the user.

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

(1) In the configuration in which the mixing ratio K between the first segment data Q1 and the second segment data Q2 is set for each note as in the third embodiment, the mixing ratio K is stepwise (discontinuous) for each note. Can vary. Therefore, a configuration in which the mixing ratio K is continuously changed in time by interpolating the mixing ratio K of each note is also suitable. For example, as illustrated in FIG. 11, the mixing ratio K of each note is set so that a numerical value corresponding to the pitch X3 of the note (calculated value of Equation (2)) is obtained at an intermediate point of the note generation period X2. The mixing ratio K may be continuously changed over a plurality of notes by linear interpolation.

  For example, as illustrated in FIG. 12, the mixing ratio K corresponding to the pitch X3 of the note is maintained in a part of the note sound generation period X2 (hereinafter referred to as “steady period”) σ. It is also possible to interpolate the mixing ratio K between the end point of the stationary period σ in the period X2 and the starting point of the stationary period σ in the immediately following sounding period X2 (linear interpolation in the example of FIG. 12). The steady period σ is a period in which, for example, a point when a predetermined length (for example, ¼ of the continuation length) has elapsed from the start point of the sound generation period X2 is a start point, and a time point a predetermined length before the end point of the sound generation period X2 . In addition, although linear interpolation was illustrated in the illustration of FIG.11 and FIG.12, the specific method of interpolation is arbitrary, For example, you may utilize well-known curve interpolation.

(2) In the third embodiment, the maximum value β of the mixing ratio K is set as a predetermined constant. However, the maximum value β can be variably set according to an instruction from the user to the input device 16. In the third embodiment, the configuration in which the mixing ratio K increases as the pitch X3 is higher is exemplified. However, the relationship between the pitch X3 and the mixing ratio K is not limited to the above example. For example, it is possible to calculate the mixing ratio K so that the mixing ratio K decreases (the synthesized speech becomes clear) as the pitch X3 is higher.

(3) The feature amount applied to the calculation of the mixing ratio K is not limited to the pitch X3 exemplified in the third embodiment. For example, a configuration in which the mixing ratio K is set according to the volume of each note may be employed. For example, assuming that the sound volume perceived by the listener becomes ambiguous as the volume decreases, a configuration in which the variable setting unit 37 sets the mixing ratio K so that the mixing ratio K increases as the volume decreases is preferable. It is.

(4) For the configurations of the third embodiment and the fourth embodiment in which the mixing ratio K is set according to the feature quantity to be synthesized, the first embodiment in which the generation unit 32 generates the second library L2 from the first library L1. The configuration of the second embodiment in which the segment selection unit 34 selects the segment data L2 of the speech segment in which the vowel is replaced with the specific vowel [M] is not essential. For example, in a configuration in which the first library L1 and the second library L2 are prepared in advance from recorded sounds of different voice qualities, the first segment data Q1 selected from the first library L1 and the second library L2 are selected. It is also possible to mix the second segment data Q2 with a mixing ratio K corresponding to the pitch X3, for example.

(5) In the first embodiment, the speech synthesizer 100 that generates the speech signal V using the second library L2 generated by the generation unit 32 is illustrated, but the existing first library L1 is changed to the first embodiment. The present invention can also be realized as an apparatus (speech synthesis library generation apparatus) that generates the second library L2 by the exemplified method. In the speech synthesis library generation device, the presence or absence of functions for speech synthesis (segment selection unit 34, synthesis processing unit 36) is not questioned.

DESCRIPTION OF SYMBOLS 100 ... Voice synthesizer, 10 ... Arithmetic processing unit, 12 ... Memory | storage device, 14 ... Display apparatus, 16 ... Input device, 18 ... Sound emission device, 32 ... Generation | occurrence | production part, 33 ... Display control , 34... Segment selection unit, 35... Instruction receiving unit, 36... Composition processing unit, 37... Variable setting unit, 38. Image, 44... Variable area, 54.

Claims (5)

An element for generating a second library from a first library for speech synthesis that includes unit data for each speech unit, the speech unit including a vowel other than a specific vowel among a plurality of speech units A speech synthesis library generating apparatus comprising: a generating unit that generates a second library including speech segment data obtained by replacing the vowel with the specific vowel in one library. An element for generating a second library from a first library for speech synthesis that includes unit data for each speech unit, the speech unit including a vowel other than a specific vowel among a plurality of speech units A generating unit that generates a second library including unit data of a speech unit obtained by replacing the vowel with the specific vowel in one library;
A unit selection unit for selecting first unit data of a speech unit corresponding to a phonetic character to be synthesized in the first library and second unit data of the speech unit in the second library; ,
A unit mixing unit for mixing the first unit data and the second unit data;
A speech synthesizer comprising: a synthesis processing unit that generates synthesized speech using the segment data after mixing by the segment mixing unit. An element for selecting speech segment data corresponding to a phonetic character to be synthesized from a speech synthesis library including speech segment data for each speech segment, and for speech segments containing vowels other than specific vowels Is a segment selection unit that selects segment data of a speech segment in which the vowel is replaced with the specific vowel;
A speech synthesis apparatus comprising: a synthesis processing unit that generates synthesized speech using the speech unit selected by the unit selection unit. The unit selection unit includes a first unit data of the speech unit and a second unit data of the speech unit obtained by replacing the vowel with the specific vowel for a speech unit including a vowel other than the specific vowel. And select
A unit mixing unit for mixing the first unit data and the second unit data;
The speech synthesis apparatus according to claim 3, wherein the synthesis processing unit generates synthesized speech using the segment data after mixing by the segment mixing unit. It has a variable setting unit that sets the mixing ratio according to the feature quantity to be combined,
The speech synthesis apparatus according to claim 2, wherein the unit mixing unit mixes the first unit data and the second unit data at a mixing ratio set by the variable setting unit.

Images