JP6411015B2 - Speech synthesizer, speech synthesis method, and program - Google Patents

Description

  The present invention relates to a speech synthesizer, a speech synthesis method, and a program.

  A speech synthesis technology is known that converts Japanese sentences (kanji, kana, mixed sentences, etc.) input by the user into synthetic speech and reads it. In such speech synthesis technology, language processing converts Japanese notation into an intermediate notation in which prosodic symbols related to the expression of synthetic speech are written, and synthetic speech is generated according to the intermediate notation.

  Intermediate notation is automatically generated from Japanese notation by language processing, but often the accuracy is not sufficient. Therefore, when converting from Japanese-language notation to synthetic speech, there is an example in which pronunciation information is interactively created by the interaction between automatic conversion by a speech synthesizer and correction work by a user. There is also an example in which a speech synthesis is performed by inserting a changing part into a plurality of fixed example patterns. There is also known a method of analyzing an input sentence, generating a first intermediate language character string according to a generation rule, and weighting. In this method, the adjustment adjustment of the intermediate language character string is performed to generate the second intermediate language character string and the weighting is performed, and the higher one of the reliability and the weighting is selected among the first and second intermediate language character strings. It aims at outputting a natural voice by outputting it. Furthermore, based on the input synthetic speech signal, and the character arrangement, the number of moras, and the accent type, a pitch pattern of speech closest to the speech to be synthesized is searched from a plurality of naturally uttered speech, and this pitch pattern is There is also known a technique for performing synthesis on the basis. (For example, refer to patent documents 1-4)

Unexamined-Japanese-Patent No. 9-171392 JP-A-9-134190 JP 2000-56786 A JP-A-9-34492

  However, in situations where high quality is required, such as audio for teaching materials using information technology or audio for explaining museums and exhibitions, it is necessary for the user to listen to the synthesized speech many times and correct the intermediate notation, The burden on the user is large. Moreover, in the above-described conventional techniques, the accent may be unnatural, or may not fit the correction pattern too little, or the speech depending on the global meaning depending on plural sentences may not be synthesized. There's a problem.

  In addition, after correcting the intermediate notation, the Japanese notation may be changed due to the correction of the document. In this case, it is simply necessary to automatically generate an intermediate notation anew from the changed Japanese notation by language processing, but the information on the intermediate notation previously corrected by the user disappears, so the same process is repeated again. A correction is required, which again places a burden on the user.

  According to one aspect, the object of the present invention is to reduce the burden of the user's modification to improve the quality of synthesized speech in speech synthesis.

  The speech synthesis apparatus according to one aspect includes an input unit, a language processing unit, a storage unit, a correction unit, and a speech synthesis unit. The input unit receives an input of a language description of a natural language to be subjected to speech synthesis. The language processing unit analyzes the language notation based on dictionary information in which information including at least a reading, part of speech, accent phrase, and prosody regarding the accent phrase corresponding to a morpheme of the natural language is registered. Also, the language processing unit includes: a morpheme notation including a morpheme included in the linguistic notation and a part-of-speech information corresponding to the morpheme; accent phrase information indicating aggregation of accent phrases; and prosody information indicating prosody of the accent phrases Generate an intermediate notation that contains The storage unit stores correction information indicating correction content of prosody information in association with part-of-speech information and accent phrase information. The correction unit relates the prosody information generated by the language processing unit to the prosody information when the correction information is stored in the storage unit in association with the part-of-speech information generated by the language processing unit and the accent phrase information. Correct based on the correction information. The speech synthesis unit synthesizes the speech corresponding to the language notation based on the intermediate notation including the prosody information reflecting the correction by the correction unit.

  Note that even the speech processing method by the speech synthesis device according to the aspect described above and the program for causing a computer to execute the speech processing method exhibit the same effects as the speech synthesis device according to the aspect described above, so Problems are solved.

  According to the voice synthesis apparatus, the voice synthesis method, and the program of the embodiment, it is possible to reduce the burden of the user's correction for improving the quality of the synthesized voice.

It is a block diagram which shows an example of the function of the speech synthesizer by 1st Embodiment. It is a figure which shows notionally correction of the accent intensity by 1st Embodiment. It is a figure showing an example of language processing by a 1st embodiment. It is a figure which shows an example of correction of the accent intensity by 1st Embodiment. It is a figure which shows an example of a change of the Japanese-language description by 1st Embodiment. It is a figure which shows an example of the language processing of the Japanese expression after the change by 1st Embodiment. It is a figure which shows an example of the change morpheme search by 1st embodiment. It is a figure which shows an example of matching with the morpheme notation and the intermediate notation by 1st Embodiment. It is a figure which shows an example of correction determination of the accent intensity | strength by 1st Embodiment. It is a flowchart which shows the process by the speech synthesizer by 1st Embodiment. It is a functional block diagram which shows an example of a structure of the speech synthesizer by 2nd Embodiment. It is a figure which shows an example of the template by 2nd Embodiment. It is a figure which shows notionally the example which refers template DB by 2nd Embodiment. It is a flowchart which shows an example of a process of the speech synthesizer by 2nd Embodiment. It is a functional block diagram which shows an example of a structure of the speech synthesizer by 3rd Embodiment. It is a figure which shows the modification of prosody information. It is a figure which shows the modification of prosody information. It is a figure which shows an example of the hardware constitutions of a standard computer.

First Embodiment
Hereinafter, the speech synthesizer 1 according to the first embodiment will be described with reference to FIGS. 1 to 7. For convenience of explanation, although an explanation will be given focusing on accent intensity as an example of the prosodic symbols, other prosodic symbols can be applied. Accent intensity is one of prosodic symbols that is determined in correspondence with an accent phrase and that controls the strength and weakness of the accent phrase. The other prosodic symbols are symbols indicating, for example, pitch high and low, intonation size, speech speed and speed, volume level and the like. The notation of the prosodic symbols shown below is an example and is not limited thereto.

  In the present embodiment, it is assumed that the accent intensity is corrected in the synthesized speech corresponding to the input Japanese notation (Kanji Kana mixed sentence and the like). In this case, it is automatically determined whether or not the correction of accent intensity is to be performed on the changed Japanese notation after changing a part of the input Japanese notation, as in the correction on the Japanese notation before the change. Do. Note that performing correction of the accent intensity in the same way as the correction in the Japanese notation before the change is expressed as taking over the accent strength or the prosody information to the Japanese notation after the change. Hereinafter, the Japanese notation in which the correction of the accent intensity is performed is referred to as a first Japanese notation, and the Japanese notation in which a part of the Japanese notation is changed in the first Japanese notation is referred to as a second Japanese notation .

  FIG. 1 is a block diagram showing an example of the function of the speech synthesizer 1 according to the first embodiment. As shown in FIG. 1, the speech synthesizer 1 includes an input / output unit 5, a language processing unit 7, a speech synthesis unit 9, an intermediate notation correction unit 11, and a storage unit 21. The intermediate notation correction unit 11 includes a change search unit 13, a morpheme matching unit 15, a correction determination unit 17, and a prosody correction unit 19. Each of the above-described functions of the speech synthesizer 1 may be realized, for example, by a processor reading and executing a program stored in the storage unit 21. In addition, at least a part of the functions may be realized by, for example, a semiconductor integrated circuit. The speech synthesizer 1 may store, for example, in the storage unit 21 a dictionary for performing language processing, and speech to be used for speech synthesis.

  The control unit 3 controls the operation of the speech synthesizer 1. The input / output unit 5 is a function to input / output information to / from the voice synthesizer 1 and includes, for example, an input device such as a touch panel and a keyboard, and an output device such as a display device and a speaker. The language processing unit 7 performs language processing of Japanese notation, for example, based on a dictionary including Japanese part-of-speech information, accent information and the like, and outputs morphological notation, intermediate notation and the like. The morpheme notation is, for example, information including Japanese notation divided for each morpheme and corresponding part of speech information. The intermediate notation is a notation based on, for example, a Japanese notation, accent phrase information indicating a phrase corresponding to one accent phrase in the Japanese notation, and accent strength corresponding to the accent phrase. The speech synthesis unit 9 synthesizes speech on the basis of, for example, intermediate notation and speech recorded in advance, such as for each word, for each phrase, and the like. The storage unit 21 is a device for storing information, and may store a program for controlling the operation of the speech synthesizer 1 or information for performing each process.

  The intermediate notation correction unit 11 determines whether the intermediate notation outputted by the language processing unit 7 needs correction, and automatically corrects the intermediate notation if necessary. Specifically, the change search unit 13 searches the morpheme of the change portion in the Japanese notation from the first Japanese morpheme notation and the second Japanese morpheme notation. The morpheme association unit 15 associates the input first Japanese notation with the corrected intermediate notation performed on the first Japanese notation intermediate notation output from the language processing unit 7.

  The correction determination unit 17 determines whether correction of the accent intensity is necessary in the intermediate notation generated by the language processing unit 7 based on the second Japanese notation. That is, in the first and second Japanese notations, for example, when the part-of-speech information and the accent phrase information correspond to each other in one-to-one correspondence, the correction determination unit 17 determines the accent intensity corresponding to the second Japanese notation. It is determined to correct That is, the correction of the accent intensity performed in the first Japanese notation is inherited to the accent strength of the second Japanese notation after the correction. Thereby, the accent strength of the accent phrase including the morpheme changed in the first Japanese notation in the second Japanese notation is taken as the accent strength after being corrected in the first Japanese notation. In this manner, the correction determination unit 17 determines whether to correct the accent intensity of the Japanese expression of the speech synthesis target to the corrected accent intensity of the Japanese expression that has been subjected to the language processing. The prosody correction unit 19 corrects the intermediate notation based on the determination result of the correction determination unit 17.

  In the following description, each operation in the speech synthesizer 1 includes, for example, a case where an arithmetic processing unit described later executes the program by reading a predetermined program, and for convenience, the above-described functions are described as performing processing. .

  FIG. 2 is a diagram conceptually showing an example of the intermediate notation correction according to the present embodiment by a display example. As shown in the display example 40 a of FIG. 2, in the process (a), the Japanese notation 41 corresponding to the first Japanese notation is input. The Japanese notation 41 shows a Japanese sentence "Lightning alert is ... near Tokyo". As shown in the display example 40b, the Japanese notation 41 is converted into prosody information 43 by the language processing unit 7, for example, in the process (b). Here, the prosody information 43 = "Tokyo iki 'n penni" and "kaminari chui' hoga &".

  As shown in the display example 40 c, in processing (c), the accent intensity in the prosody information 43 is corrected from the strong accent “'” to the weak accent “*” by the user, and is set as the prosody information 45. Here, the prosody information 45 = "Tokyo * N-Penni" and "Kaminari Tuy * Houga &".

  As shown in the display example 40d, in the process (d), a part of the Japanese notation 41 in the first Japanese notation is corrected, and is referred to as the Japanese notation 47. The Japanese language notation 47 is the second Japanese language notation "the dark fog warning is ... around Kyoto". As in the display example 40e, in the process (e), the language processing unit 7 analyzes the Japanese notation 47 and outputs a morphological notation and an intermediate notation. The change search unit 13 compares morpheme notations and searches for morphemes different from the Japanese notation 41 and the Japanese notation 47. The correction determination unit 17 compares part of speech information and accent phrase information corresponding to different retrieved morphemes, and determines whether to correct the accent intensity. As a result of the determination, the accent intensity of the portion of "fog alert" is set to the accent intensity corrected in "lightning alert", and the prosody information 49 is output.

  By the way, when “around Tokyo” is changed to “around Kyoto” as described above, morphological analysis analyzes “Tokyo” (proper noun), “nearby” (common noun) and “ni” (case particle) And, "Kyoto" (proper noun), "periphery" (common noun) and "ni" (case particle). For example, in the example of modifying the prosody based only on the arrangement of parts of speech of morphemes, if there is a precedent of the arrangement of parts of speech that matches “around Tokyo”, the same arrangement of parts of speech matches “around Kyoto”. Will have common accent strength. This method is correct in terms of meaning but not in terms of hearing. Here, when "around Tokyo" and "around Kyoto" are divided into accent phrases, they become "around Tokyo" and "Kyoto" "around", and the number of accent phrases changes from one to two. There is. In this example, the change in the number of accent phrases causes a large auditory change. As described above, since the accent strength is linked to the accent phrase, it is not obvious how to reflect the accent strength of “around Tokyo” in “Kyoto” and “around”. Therefore, as described above, in the present embodiment, one-to-one correspondence of both part-of-speech information and accent phrase information is used as a condition for correction of intermediate notation.

  For example, when changing “To Tokyo” in a sentence to “To Kyoto”, the parts of speech information of “To Tokyo” and “To Kyoto” are (proper nouns) (case particles) and accents, respectively. Since the phrase is also single, the condition of correction is satisfied.

  Hereinafter, an example of the intermediate notation correction will be described with reference to FIGS. 3 to 9. FIG. 3 is a diagram showing that the language processing unit 7 generates the morpheme notation 52 and the intermediate notation 44 based on the Japanese notation 41. The part-of-speech information 51 is information indicating the part-of-speech included in the Japanese notation 41, and is output as a result of morphological analysis of the Japanese notation 41 by the language processing unit 7. The morpheme separation information 42 is information indicating the correspondence between the part of speech information 51 and the Japanese notation 41. The morpheme notation 52 is information in which the part-of-speech information 51 and the morpheme separation information 42 correspond to each other.

  The prosody information 43 is a description relating to the expression of synthetic speech when speech synthesis is performed based on the Japanese expression 41, and accent is output based on the result of language processing of the Japanese expression 41 by the language processing unit 7 It is information including prosody information such as strength. The accent phrase information 54 is information indicating the delimitation of the accent in the Japanese notation 41. The intermediate notation 44 is information in which the prosody information 43 and the accent phrase information 54 correspond to each other.

  For example, the morpheme division information 42 of the morpheme unit for the Japanese notation 41 = "The lightning alert was issued in the vicinity of Tokyo" is "Tokyo" "Near" "Ni" "Lightning" "Warning" " It is information such as "decree" "sa" "re" "ta". The part-of-speech information 51 is information such as “Tokyo” as a proper noun, “nearby” as a common noun, “ni” as a case particle, and the like. Proper nouns, common nouns, etc. are types of part of speech. For example, in the part of speech information 51, the number of parts of speech is six. The prosody information 43 is information such as "Tokyo iki 'n' penni", "Kaminari chui 'hoga &" and the like. In the prosody information 43, '' indicates a strong accent. Further, in the example of FIG. 3, according to the part-of-speech information 51 and the accent phrase information 54, the number of parts of speech included in one accent phrase is three each.

  FIG. 4 shows an example of accent intensity correction performed on the intermediate notation 44. In the intermediate notation 55, the accent intensity is corrected in the prosody information 43 and is set as the prosody information 45. In the prosody information 45, the weak accent “*” is corrected in place of the strong accent “'” in the “kaminari chui' hoga &” of the prosody information 43. Thus, the prosody information corresponding to the Japanese-language notation 41 is, for example, “Tokyo ki 'n' penni, kaminari chui * hoga hatsure saleta.”

  Note that “'” and “*” are prosodic symbols that designate an accent position and an accent intensity. "'" Means that the accent strength of the accent phrase is strong, "*" means that it is weak. "," "." Designate exhalation paragraph boundaries, and "," "." And "" (full-width space) designate accent phrase boundaries. A string between accent phrase boundaries is called an accent phrase. "&" Is a prosodic symbol indicating nasal noise.

  FIG. 5 shows an example of changing the Japanese notation 41. As shown in FIG. 5, Japanese notation 41 = "To light in the vicinity of Tokyo, lightning warning ..." is changed to Japanese notation 47 = "(around Kyoto) ((dark fog alert) is ..." There is. Parentheses indicate changed parts.

  FIG. 6 shows that the language processing unit 7 generates the morpheme notation 58 and the intermediate notation 61 based on the Japanese notation 47. The part-of-speech information 57 is information indicating the part-of-speech included in the Japanese notation 47, and is output as a result of morphological analysis of the Japanese notation 47 by the language processing unit 7. The morpheme separation information 48 is information indicating the correspondence between the part of speech information 57 and the Japanese notation 47. The morpheme notation 58 is information in which the part-of-speech information 57 and the morpheme separation information 48 correspond to each other.

  The prosody information 59 is a notation related to the expression of synthetic speech based on the Japanese notation 47, and is information including accent intensity that is output by the language processing unit 7. The accent phrase information 60 is information indicating the delimitation of the accent in the Japanese notation 47. The intermediate notation 61 is information in which the prosody information 59 and the accent phrase information 60 correspond to each other.

  FIG. 7 is a diagram showing an example of searching for a modified morpheme. As shown in FIG. 7, the morpheme notation 52 and the morpheme notation 58 are compared, and different morphemes are searched for. The modified morpheme notation 53 is information indicating a morpheme different from the morpheme notation 58 in the morpheme notation 52. The modified morpheme notation 62 is information indicating a morpheme different from the morpheme notation 52 in the morpheme notation 58.

  FIG. 8 is a diagram showing an example in which the modified morpheme notation 53 and the intermediate notation 55 are associated with each other. The correspondence information 56 is information in which a modified morpheme notation 53 indicating a change in the Japanese notation 41 is associated with an intermediate notation 55 after the prosody information is corrected.

  FIG. 9 shows a correction determination example of correcting the Japanese notation 47 after the change according to the prosody information 45 after the correction. As shown in FIG. 9, the change information 63 is information in which the intermediate representation 61 and the modified morphological representation 62 are associated. The correction determination information 64 and the correction reference information 65 are information indicating accent phrases corresponding to each other in the accent phrase information 54, the accent phrase information 60, the part-of-speech information 51, and the part-of-speech information 57. One-to-one correspondence means that the number and type of parts of speech corresponding to one accent phrase are the same. The accent intensity in the correction reference information 65 is applied to the accent phrase corresponding to this one-to-one, the correction is taken over, and the prosody information 66 is generated. The intermediate notation 67 is information in which the prosody information 66 and the accent phrase information 60 correspond to each other.

  In this example, prosody information 66 is output with respect to the Japanese notation 47, and the speech synthesis unit 9 synthesizes speech based on the prosody information 66. In the above example, the accent intensity of “around Tokyo” is not applied to “around Kyoto”, but the accent intensity corrected by “lightning alert” is applied to “dark fog alert” become.

  Hereinafter, the operation of the speech synthesizer 1 according to the present embodiment will be further described with reference to the flowchart. FIG. 10 is a flowchart showing the operation of the speech synthesizer 1. As shown in FIG. 10, the input / output unit 5 receives an input written in Japanese (S71). For example, the input / output unit 5 receives the Japanese notation 41 corresponding to the first Japanese notation. The input / output unit 5 may display the received Japanese notation 41, for example, as in the display example 40a. At this time, it is preferable to be configured to present the user with a graphical user interface (GUI) or the like so that the Japanese notation 41 can be corrected.

  The language processing unit 7 performs, for example, morphological analysis and dependency analysis on the accepted Japanese notation 41, and generates a morphological notation 52 and an intermediate notation 55 (S72). For example, as shown in FIG. 3, the morphological notation 52 includes a part-of-speech information 51 and a morpheme separation information 42. For example, the intermediate notation 44 includes prosody information 43 and accent phrase information 54.

  Here, the speech synthesis unit 9 synthesizes speech based on the generated intermediate notation 55 and outputs the synthesized speech from the input / output unit 5. For example, in the display example 40b of FIG. 2, the reproduction button may be displayed, and when the reproduction button is selected, the sound may be output. The sound may be output by, for example, a speaker. The input / output unit 5 accepts the correction when the user's correction is input in the intermediate notation 44. For example, as shown in the process (c) of FIG. 2, a change in accent intensity is accepted (S73). Further, as shown in FIG. 4, in the intermediate notation 44, the prosody information 43 is corrected to the prosody information 45.

  As shown in the process (d) of FIG. 2, the input / output unit 5 accepts, for example, a correction of a part of the Japanese notation 41. The corrected Japanese notation is Japanese notation 47 (S74). As shown in FIG. 6, the language processing unit 7 language-processes the Japanese notation 47 to generate a morpheme notation 58 and an intermediate notation 61 (S75).

  The change search unit 13 associates the morpheme notation 52 with the morpheme notation 58 (S76). Further, as shown in FIG. 7, the change search unit 13 generates the changed morpheme notation 53 and the changed morpheme notation 62 based on the correspondence between the morpheme notation 52 and the morpheme notation 58. That is, the change search unit 13 compares the morpheme division information 42 with the morpheme division information 48, extracts different morphemes, and generates the modified morpheme notations 53 and 62. The morpheme association unit 15 generates the association information 56 based on the correspondence between the modified morpheme notation 53 and the intermediate notation 55, as shown in FIG. 8 (S77). The correspondence information 56 is stored in the storage unit 21. At this time, the storage unit 21 can be, for example, a semiconductor memory or the like built in the voice synthesizer 1.

  The correction determination unit 17 determines whether there is an unprocessed accent phrase including the changed morpheme extracted by the modified morpheme notation 62 (S78). If there is no unprocessed accent phrase (S78: NO), the prosody modification unit 19 outputs the intermediate notation at that time (S81). Further, the speech synthesis unit 9 synthesizes speech based on the outputted intermediate notation (S 82).

  If there is an unprocessed accent phrase (S78: YES), the correction determination unit 17 determines whether each accent phrase satisfies the prosody information correction condition, including the changed morpheme extracted by the modified morpheme notation 62 (S78: YES) It is determined whether or not (S79). In the prosody information correction condition, here, for example, both the part-of-speech information 57 and the accent phrase information 60 corresponding to the accent phrase including the morpheme are in one-to-one correspondence with the part-of-speech information 51 and the accent phrase information 54. is there.

  If the correction determination unit 17 determines that the prosody information correction condition is satisfied (S79: YES), the prosody correction unit 19 determines the accent intensity of the accent phrase in the intermediate notation 61 as the corresponding accent in the intermediate notation 55. Correct to the accent strength of the phrase (S80). Furthermore, the prosody modification unit 19 returns the process to S78. If the correction determination unit 17 determines that the prosody information correction condition is not satisfied (S79: NO), the process returns to S78.

  As described above, according to the speech synthesizer 1 according to the first embodiment, the language processing unit 7 performs language processing such as Japanese notation 41 input through the input / output unit 5. By this language processing, for example, a part-of-speech information 51, a morpheme notation 52 including morpheme division information 42, a prosody information 43, and an intermediate notation 44 including accent phrase information 54 are generated.

  The speech synthesis unit 9 synthesizes and outputs speech based on the generated intermediate notation 44. When the correction of the prosody information 43 is input based on the output voice, the input / output unit 5 accepts the correction.

  When the input / output unit 5 receives, for example, the Japanese notation 47 obtained by correcting a part of the Japanese notation 41, the language processing unit 7 performs language processing of the Japanese notation 47. By this language processing, for example, the intermediate notation 61 including the part-of-speech information 57, the morpheme notation 58 including the morpheme division information 48, the prosody information 59, and the accent phrase information 60 is generated.

  The change search unit 13 acquires the changed morpheme notation 53 and the changed morpheme notation 62 based on the correspondence between the morpheme notation 52 and the morpheme notation 58. That is, the change search unit 13 compares the morpheme division information 42 with the morpheme division information 48 to extract different morphemes. The morpheme relating unit 15 corrects the intermediate notation output from the language processing unit 7 after the linguistic processing of the first Japanese notation and the first Japanese notation input 52 being the first Japanese notation. Correspondence with the reflected intermediate representation 55 is generated, and correspondence information 56 is generated. The correspondence information 56 is stored, for example, in the storage unit 21.

  Based on the extracted modified morpheme notation 62, the correction determination unit 17 shares part-of-speech information and accent phrase information with the correspondence information 56 for each accent phrase that includes the changed morphemes (one-to-one correspondence) It is determined whether or not The change search unit 13 performs determination in units of morphemes on or around all morphemes that are changed by the morpheme division information 42 and the morpheme division information 48. Whether or not one-to-one correspondence can be determined can be realized by using, for example, Dynamic Programming (DP) matching, which is widely performed in matching processing. In the common case, the accent intensity of the corresponding accent phrase of the prosody information 59 is corrected to the accent intensity of the prosody information 45.

  As described above in detail, according to the speech synthesizer 1, when correction of prosody information is performed on synthesized speech generated based on the result of language processing in the first Japanese notation, information on the correction is stored. Store in section 21. The information on correction is information including correction content on prosody information and at least part-of-speech information and accent phrase information corresponding to the corrected portion. It is determined whether or not there is an accent phrase corresponding to the corrected portion of the prosody information of the first Japanese notation in the language processing result of the second Japanese notation different from the first Japanese notation. If there is a corresponding accent phrase, the correction of the prosody information is automatically reflected by correcting the prosody information of the corresponding accent phrase of the second Japanese notation in the same manner as the correction for the first Japanese notation. Do.

  The corresponding accent phrase is an accent phrase in which the part-of-speech information and the accent phrase information correspond to each other in the first and second Japanese notations. This automatically corrects the prosody information of the corrected accent phrase and the corresponding accent phrase to the prosody information after correction, and takes over the correction. For example, in the first Japanese notation, when a part is changed to become a second Japanese notation, this correction is performed on an accent phrase including all the change morphemes involved in the change. Furthermore, when the user repeatedly corrects or changes Japanese notation or intermediate notation, the above processing is repeated.

  As described above, according to the speech synthesizer 1 according to the first embodiment, when correction of prosody information is performed, the speech synthesis device 1 has part-of-speech information and accent phrase information corresponding to the corrected prosody information. The correction can be handed over to the Japanese notation. That is, when the number and type of parts of speech of the accented phrase are the same, the speech synthesizer 1 can determine to correct the prosody information as in the correction information stored in the storage unit. When the user's correction or change is made in Japanese, prosody already corrected or changed by the user can be automatically reflected in the portion where the correction or change is made. As a result, in order to generate natural synthetic speech, the time and effort required for the user to repeatedly modify the prosody repeatedly is greatly reduced, and high-quality synthetic speech can be easily generated.

  According to the present embodiment, it is determined whether or not the part-of-speech information and the accent phrase information correspond to each other in order to reflect the correction of the prosody information that has already been made on the changed part of the Japanese notation. And can take over the correction if it corresponds. Specifically, the morpheme association unit 15 associates the part-of-speech information on the changed Japanese notation with the intermediate notation. Thereby, the correction determination unit 17 determines whether the parts-of-speech of all morphemes of the accent phrase including the corrected morpheme searched by the change search unit 13 have part-of-speech information corresponding to the accent phrase before correction one-on-one Whether or not can be determined.

  By using part-of-speech information in this manner, it is possible to determine whether the meanings of morphemes are similar before and after the change. In addition, by using the accent phrase, it is possible to determine whether or not the speech is similar when synthesized speech before and after the change. In this way, the speech synthesizer 1 can perform the same prosody correction as the accent phrase that has already been corrected, only when the meanings are similar and aurally similar. Therefore, it is possible to generate an intermediate notation taking into consideration context and prosody, and to prevent the unintended accent intensity from being forcibly taken over by the accent intensity etc., so that the synthetic speech is appropriate. Correction is possible. In addition, speech can be made more natural as compared with an example in which the accent is similarly corrected when only the part of speech or the accent phrase is matched.

  Even when high sound quality is required, such as teaching material voice and explanatory voice, synthetic speech can be generated with sufficient accuracy. In addition, even if the Japanese notation is changed due to the correction of an original after correction of the intermediate notation, or even if the correction is repeated, it is possible to reuse the information on the intermediate notation that the user has previously corrected. Therefore, it is possible to prevent the user from repeatedly putting a burden of correction and reduce costs.

  It is also avoided that the constraints are too severe and in many cases inability to take over the accent strength, as is the case for taking over on the basis of mora numbers and accent types and as well.

  According to the above configuration, the user's corrected prosodic symbols are properly taken over, which contributes to the reduction of the correction cost burdening the user.

Second Embodiment
The speech synthesizer 100 according to the second embodiment will be described below with reference to FIGS. 11 to 14. In the speech synthesizer 100 according to the second embodiment, the same components and operations as those of the speech synthesizer 1 according to the first embodiment are denoted by the same reference numerals and the description thereof will be omitted. In this embodiment, when designating an intermediate notation to be used regardless of the intermediate notation generated by the language processing unit 7, the intermediate notation to be used and the corresponding morpheme notation are stored as a template. Then, the speech synthesizer 100 refers to the template when generating the intermediate notation of the Japanese notation of the speech synthesis target.

  FIG. 11 is a functional block diagram showing an example of the configuration of the speech synthesizer 100 according to the second embodiment. As shown in FIG. 11, the speech synthesizer 100 includes an input / output unit 5, a language processing unit 7, a speech synthesis unit 9, an intermediate notation correction unit 103, and a storage unit 21. The storage unit 21 may store a template Data Base (DB) 117. The intermediate notation correction unit 103 includes a DB registration unit 105, a template search unit 113, a template correspondence unit 115, and a prosody correction unit 19. Each of the above-described functions of the speech synthesizer 100 may be realized, for example, by a processor reading and executing a program stored in the storage unit 21. Further, part or all of the functions may be realized by, for example, a semiconductor integrated circuit or the like.

  The intermediate notation correction unit 103 refers to the template DB 117 to determine whether the intermediate notation output by the language processing unit 7 needs correction, and automatically corrects the intermediate notation if necessary. The DB registration unit 105 registers, in the template DB 117, information on an intermediate notation different from the intermediate notation generated by the language processing unit 7. The intermediate notation to be registered is input, for example, through the input / output unit 5. The DB registration unit 105 registers the part of speech information, the accent phrase information, and the prosody information in association with each other. The template DB 117 is, for example, a database in which part-of-speech information, accent phrase information and prosody information are stored in association with each other.

  The template search unit 113 searches whether or not the corrected Japanese expression or the newly registered intermediate expression of the Japanese expression matches the information registered in the template DB 117. When there is a matching template, the template association unit 115 associates the Japanese notation of the target with the template. When there is a matching template, the prosody modification unit 19 modifies the corresponding accent intensity associated by the template association unit 115 to the accent intensity stored in the template DB 117.

  In the following description, each operation in the speech synthesizer 1 includes, for example, a case where an arithmetic processing unit to be described later executes the operation by reading a predetermined program, and each function described above performs processing for convenience. explain.

  FIG. 12 shows an example of the template 125. As shown in FIG. As shown in FIG. 12, the template 125 includes accent phrase information 121, part-of-speech information 122, Japanese spelling examples 123, and accent strength information 124. The Japanese notation example 123 is information indicating an example of Japanese notation determined to match the template 125. In Japanese typographical example 123, "#" indicates that any morpheme in which part of speech information matches may be used. In the Japanese typographical example 123, in this example, the case particle in the first half of Japanese transcription and the second half of the Japanese typography are designated, but it is not limited thereto. Accent strength information 124 is an example of prosody information. Although the information "not specified" is registered as the accent intensity information 124, this indicates that no correction is necessary.

  FIG. 13 conceptually shows an example of referring to the template DB 117 when Japanese language notation 47 is input. As shown in FIG. 13, in the template DB 117, for example, at least one template such as the templates 135, 145 and 155 is registered. The template 135 includes accent phrase information 131, part-of-speech information 132, Japanese typographical examples 133, and accent strength information 134. The other templates 145 and 155 are the same. Accent intensity information 134 is an example of prosody information.

  In FIG. 13, accent phrase information 60, part-of-speech information 57, and morpheme separation information 48 generated from the Japanese notation 47 are shown. In this example, the Japanese phrase 47 matches the template 135 because the accent phrase information 60 corresponds to the accent phrase information 131 one to one, and the part of speech information 57 corresponds to the part of speech information 132 one to one. Therefore, the accent intensity information 134 is applied to the intermediate notation.

  Specifically, accent phrase information 131 = “(accent phrase) (accent phrase) (accent phrase)”, part of speech information 132 = “(proper noun) (common noun) (case particle) (common noun) (case particle) On the other hand, accent intensity information 134 = "(strong accent (')) (weak accent (*)) (weak accent (*))" is registered. As shown in 6, the prosody information 59 is generated by the language processing, and the prosody information 59 is corrected based on the accent strength information 134 as "Shuhenni" as "Shuhenni *", and "Nomoke '-Hooga &" Will be corrected as "Nomke *-Hooga &".

  FIG. 14 is a flowchart showing processing of the speech synthesizer 100 according to the second embodiment. Before the process shown in FIG. 14, for example, the DB registration unit 105 registers in the template DB 117 based on the morpheme analysis result and the intermediate notation of Japanese notation in which the prosody information has been corrected. For example, the template is registered based on the morpheme notation 52 shown in FIG. 3 and the intermediate notation 55 after the correction shown in FIG. Alternatively, the user may manually input the registration content via the input / output unit 5.

  As shown in FIG. 14, the input / output unit 5 receives, for example, an input of Japanese notation 47 (S161). The language processing unit 7 language processes the Japanese notation 47 and outputs the morphological notation 58 and the intermediate notation 61 (S 162). The template search unit 113 searches whether the part of speech information 57 in the output morpheme notation 58 and the accent phrase information 60 in the intermediate notation 61 match the template in the template DB 117 (S163). At this time, the template search unit 113 searches whether there is a template in which the part-of-speech information 57 and the accent phrase information 60 correspond to each other. In addition, when a part of the Japanese text is designated as a template, it is searched whether or not the Japanese notation also matches.

  If there is no matching template (S163: NO), the prosody modification unit 19 outputs the intermediate notation at that time as it is (S166). If there is a matching template (S163: YES), the template associating unit 115 associates the template in the template DB 117 with the prosody information 59, for example (S164). The prosody modification unit 19 modifies the prosody information 59 based on, for example, the matched template 135 (S165), and outputs an intermediate notation based on the template (S166). The speech synthesis unit 9 synthesizes speech based on the intermediate notation (S167).

  Here, the operation of the template search unit 113 will be further described with reference to the example of the template DB 117 of FIG. Here, in addition to the part-of-speech information and the accent intensity, a part of the Japanese notation is registered together, but this may not be necessary.

  For example, Japanese words 47 “Dog fog alert around Kyoto” matches template 135 but not template 145. This is because “around Kyoto” is the two accent phrases “Kyoto” and “around” and does not match the accent phrase information 141 of the template 145. Such a search can be realized, for example, by using DP matching on a morpheme basis.

In DP matching, first, the template search unit 113 substitutes, for each morpheme, for example, the morpheme notation 58 generated by language processing of the Japanese notation input by the user and the morpheme notation of the template stored in the template database. Define the cost as follows: Then, the template search unit 113 performs DP matching in units of morphemes, where the insertion cost and the deletion cost are 1, regarding different portions between the morpheme notation 58 generated by the language processing unit 7 and the morpheme notation of the template. The insertion cost can be set to “1”, for example, when there is a morpheme in the morpheme notation 58, not in the template of the template DB 117. The deletion cost can be set to “1”, for example, when there is a morpheme that exists in the template of the template DB 117 and is not in the morpheme notation 58.
1) If the part of speech of the two morphemes and the string itself match, the replacement cost = 0
2) If the part of speech of two morphemes match and the morpheme of the template is "#", the replacement cost = 0
3) If the above two conditions are not met, replacement cost = 1

  Next, it is checked whether there is a part where the sum of replacement, insertion and deletion costs is 0 in the morphological notation 58, and further, the inclusion relation of the accent phrase and the morpheme corresponding to the part is the accent phrase of the template. Make sure that it matches the inclusion relation of and morphemes. If they match, it is determined that the template matches. By repeating the above, it is possible to search for a matching template from the template database.

  As described above, according to the speech synthesizer 100 according to the present embodiment, for example, a template including an intermediate notation different from the intermediate notation generated by the language processing unit 7 is created. When the Japanese notation of the speech synthesis target is input, the template DB 117 searches the morpheme notation and the intermediate notation of the result of the language processing, and determines whether or not they match.

  Whether or not the template is matched is determined by at least part of speech information and accent phrase information. You may specify and search specifically what morpheme it is. The template DB 117 stores one or more templates.

  For example, a part of Japanese notation can be registered in the information registered in the template DB 117 as in Japanese notation examples 133, 143, 153, and the like. As a result, the accent intensity can be appropriately set depending not only on the local meaning such as the arrangement of words and morphemes but also on the global meaning depending on a plurality of sentences. By making template DB 117 freely rewritable, it is possible to generate synthesized speech that matches the user's request.

  With the above configuration, if a template is registered in advance in the template DB 117, when the user inputs Japanese notation and automatically generates an intermediate notation by language processing, prosody information is registered based on the registered template. It can be corrected. In addition, by using part-of-speech information and accent phrase information as a template, prosody information can be appropriately corrected in consideration of context and prosody, which contributes to reduction of the time for the user to correct the intermediate notation again.

Third Embodiment
The speech synthesizer 200 according to the third embodiment will be described below with reference to FIG. In the speech synthesizer 200 according to the third embodiment, the same components and operations as those of the speech synthesizers 1 and 100 according to the first or second embodiment are designated by the same reference numerals and the description thereof will be omitted. In the present embodiment, in the speech synthesis apparatus 200, when the intermediate notation is corrected, the prosody information after correction, the corresponding part-of-speech information, and the accent phrase information are automatically registered in the template DB 117, and the template Update DB117.

  FIG. 15 is a functional block diagram showing an example of the configuration of the speech synthesizer 200 according to the third embodiment. As shown in FIG. 15, the speech synthesizer 200 includes an input / output unit 5, a language processing unit 7, a speech synthesis unit 9, an intermediate notation correction unit 103, and a storage unit 21. Preferably, the template Data Base (DB) 117 is stored in the storage unit 21. The intermediate notation correction unit 203 includes a DB registration unit 105, a template search unit 113, a template correspondence unit 115, a prosody correction unit 19, and a template update unit 205. Each of the above-described functions of the speech synthesizer 200 may be realized, for example, by a processor reading and executing a program stored in the storage unit 21. Further, part or all of the functions may be realized by, for example, a semiconductor integrated circuit or the like.

  The template updating unit 205 automatically associates the modified intermediate notation prosody information and at least the corresponding part-of-speech information and the accent phrase information with each other when the intermediate expression is corrected in the speech synthesizer 200. Register or update in DB117. When speech synthesis is performed, the intermediate transcription correction unit 203 refers to the latest template DB 117 to determine matching with a template. The other configuration and operation are similar to those of the speech synthesizer 100 according to the second embodiment.

  As described above, according to the speech synthesis apparatus 200 according to the third embodiment, in addition to the effects of the speech synthesis apparatus 100, the template DB 117 can be automatically registered and updated. Therefore, it is possible to more finely reflect the prosody information correction, and it is possible to synthesize high-quality speech while reducing the time and effort of the user's correction.

(Modification)
Hereinafter, with reference to FIGS. 16 and 17, modifications applicable to the first to third embodiments will be described. The speech synthesis apparatus to which this modification is applied may be any of the speech synthesis apparatuses 1, 100, and 200. The description of the configurations and operations of the speech synthesizers 1, 100 and 200 described in the first to third embodiments will be omitted.

  FIG. 16 is an example of using speech speed information as prosody information. As shown in FIG. 16, as the prosody information 211, it is assumed that "Tokyo’ ki ペ ン ペ ン ペ ン ペ ン ペ ン ペ ン ペ ン カ ミ ナ ガ ガ ガ ツ ツ ツ ツ ツ ツ ツ. Here, “▽” is a prosodic symbol indicating that the speech speed is to be slowed down. At this time, the part-of-speech information corresponding to "Kaminaritui 'Hoga" underlined in FIG. 16 = "(common noun) (common noun) (case particle)", accent phrase information = "(accent phrase)" The prosodic symbol "▽" will be applied to the intermediate notation for the accent phrase corresponding to. For example, it is possible to use a prosodic symbol “Δ” for increasing the speech speed.

  FIG. 17 is an example using volume information as prosody information. As shown in FIG. 17, as the prosody information 215, it is assumed that "Tokyo’ 'ペ ン ペ ン ペ ン ペ ン チ ュ ー ペ ン カ ペ ン ホ ー ホ ー ハ ガ ツ ツ ツ ツ レ ー レ ー レ ー. Here, “↑” is a prosodic symbol indicating that the volume is increased. At this time, the part-of-speech information corresponding to "Kaminaritui 'Hoga" underlined in FIG. 17 = "(common noun) (common noun) (case particle)", accent phrase information = "(accent phrase)" The prosodic character "" will be applied to the intermediate notation with respect to the accent phrase corresponding to. For example, it is possible to use a prosodic symbol "↓" which reduces the volume.

  As described above, according to this modification, it is possible to automatically correct the prosodic symbols for various prosodic symbols other than the accent intensity. This further reduces the effort for generating higher quality synthetic speech.

  In the above first to third embodiments and modifications, the intermediate notation correction units 11, 103 and 203 are an example of correction units, and the input / output unit 5 is an input unit, a prosody correction reception unit, and a language change reception unit. It is an example of a part. The input / output unit 5 and the DB registration unit 105 are an example of a registration reception unit. The morpheme association unit 15 and the template update unit 205 are examples of the association unit.

  The present invention is not limited to the embodiments described above, and various configurations or embodiments can be adopted without departing from the scope of the present invention. For example, although the Japanese notation has been described in the first to third embodiments and the modifications described above, the present invention is also applicable to other natural languages capable of language processing as described above.

  The morpheme association unit 15 associates the modified morpheme notation 53 with the intermediate notation 55, as shown in FIG. However, since the intermediate notation 55 is an intermediate notation corrected by the user, in some cases, the reading etc. may be changed and the correspondence may not be simply taken. In such a case, for example, only the surroundings of the changed morpheme may be taken. As a method of determining the range to take correspondence, the morpheme of the same part-of-speech as the morpheme that is continuous before and after the morpheme may be used. Alternatively, all morphemes can be matched by performing DP matching on all morphemes of corresponding Japanese notation.

  In the first to third embodiments and the modification, the morpheme notation includes part-of-speech information, but the part-of-speech information is not limited to the above example. For example, rough classification such as nouns and verbs, or detailed classification such as proper nouns and independent verbs, or more proper and relevant nouns such as place names and product names may be used. . Further, a combination may be defined in which the types are determined to be the same. Since the correction condition of the prosody information becomes stricter as it becomes more detailed, it becomes possible to select whether or not to make correction depending on the degree to which the prosody information is to be taken over, and the accuracy of the correction of the prosody symbol is improved.

  When the change search unit 13 searches for the changed morpheme, the changed morpheme may be searched for part or all of the morphemes before and after the user changed the Japanese notation. This makes it possible to efficiently search for modified morphemes and contribute to the improvement of the calculation speed.

  The designation of the accent intensity may be "not designated", which uses the result of language processing as it is, in addition to the strength and weakness. The template registered in the template DB 117 may be a template such as an accent paragraph unit or an exhalation paragraph unit separated by punctuation marks, etc., in addition to the sentence unit template. This is expected to further improve the quality of synthetic speech.

  The template updating unit 205 may update the template, for example, each time there is a correction of the intermediate notation. Also, the DB registration unit 105 and the template update unit 205 may receive, for example, manual registration input by the user using a GUI or the like.

  Here, an example of a computer commonly applied to cause a computer to perform the operation of the speech synthesis method according to the first to third embodiments and the modification will be described. FIG. 18 is a block diagram showing an example of a standard computer hardware configuration. As shown in FIG. 18, in the computer 300, a central processing unit (CPU) 302, a memory 304, an input device 306, an output device 308, an external storage device 312, a medium drive device 314, a network connection device, etc. via a bus 310. It is connected.

  The CPU 302 is an arithmetic processing unit that controls the overall operation of the computer 300. The memory 304 is a storage unit for storing in advance a program for controlling the operation of the computer 300 or for using it as a work area as needed when executing the program. The memory 304 is, for example, a random access memory (RAM), a read only memory (ROM), or the like. The input device 306 is a device that, when operated by the user of the computer, acquires inputs of various information from the user associated with the operation content, and sends the acquired input information to the CPU 302, for example. It is a keyboard device, a mouse device, etc. The output device 308 is a device that outputs the processing result of the computer 300, and includes a display device and the like. For example, the display device displays a text or an image according to the display data sent by the CPU 302.

  The external storage device 312 is, for example, a storage device such as a hard disk, and is a device for storing various control programs executed by the CPU 302, acquired data, and the like. The medium drive device 314 is a device for writing to and reading from a portable storage medium 316. The CPU 302 can also perform various control processes by reading out and executing a predetermined control program recorded in the portable recording medium 316 via the recording medium drive device 314. The portable recording medium 316 is, for example, a Compact Disc (CD) -ROM, a Digital Versatile Disc (DVD), a Universal Serial Bus (USB) memory, or the like. The network connection device 318 is an interface device that manages exchange of various data performed with the outside by wire or wirelessly. The bus 310 is a communication path for mutually connecting the above-described devices and the like and exchanging data.

  A program that causes a computer to execute the speech synthesis method according to the first to third embodiments and the modification is stored, for example, in the external storage device 312. The CPU 302 reads a program from the external storage device 312 and causes the computer 300 to perform a voice synthesis operation. At this time, first, a control program for causing the CPU 302 to perform speech synthesis processing is created and stored in the external storage device 312. Then, a predetermined instruction is given from the input device 306 to the CPU 302, and the control program is read from the external storage device 312 and executed. Also, this program may be stored in the portable storage medium 316. Also, for example, the template DB 117 may be provided in a storage device such as another computer connected via the network connection device 318. In addition, at least a part of the speech synthesis process can be performed by another computer connected via the network connection device 318.

Further, the following appendices will be disclosed regarding the above embodiment.
(Supplementary Note 1)
An input unit that receives an input of a language expression of a natural language to be subjected to speech synthesis;
The linguistic notation is analyzed based on dictionary information in which information including at least a reading, part of speech, accent phrase and prosody concerning the accent phrase corresponding to a morpheme of the natural language is analyzed, and the morpheme and the morpheme included in the linguistic notation A morpheme notation including part-of-speech information corresponding to a word processing section for generating an intermediate notation including accent phrase information indicating grouping of accent phrases and prosody information indicating prosody of the accent phrases;
A storage unit that stores correction information indicating correction content of prosody information in association with part-of-speech information and accent phrase information;
When the correction information is stored in the storage unit in association with the part-of-speech information generated by the language processing unit and the accent phrase information, the prosody information generated by the language processing unit is based on the correction information A correction unit to correct,
A speech synthesis unit that synthesizes speech corresponding to the language notation based on an intermediate notation including prosody information reflecting the correction by the correction unit;
A speech synthesizer characterized by having.
(Supplementary Note 2)
The correction unit is configured to match the number and type of parts of speech included in the accent phrases indicated by the accent phrase information generated by the language processing unit and the accent phrase information stored in the storage unit. The speech synthesizer according to claim 1, wherein the correction is performed.
(Supplementary Note 3)
A prosody modification receiving unit for accepting modification of prosody information generated by the language processing unit;
An associating unit that associates the prosody information after the correction with the part-of-speech information generated by the language processing unit and the accent phrase information;
And have
The storage unit stores correction information indicating the content of correction received by the prosody correction receiving unit in association with the part-of-speech information and the accent phrase information associated with each other. The speech synthesizer as described in.
(Supplementary Note 4)
A language change receiving unit that receives a change in language notation to be subjected to the speech synthesis;
A change search unit for searching morphemes related to the change;
And have
The language processing unit analyzes the changed language notation,
The correction unit performs correction when the correction information is stored in association with part-of-speech information corresponding to an accent phrase including a morpheme related to the searched change. The voice synthesizer according to any of the above.
(Supplementary Note 5)
It further has a registration acceptance unit that accepts registration of the correction information,
The voice synthesis apparatus according to any one of appendices 1 to 4, wherein the storage unit registers the correction information accepted by the registration accepting unit.
(Supplementary Note 6)
The speech synthesizer
Accepts input of natural language linguistic expressions targeted for speech synthesis,
The linguistic notation is analyzed based on dictionary information in which information including at least a reading, part of speech, accent phrase and prosody concerning the accent phrase corresponding to a morpheme of the natural language is analyzed, and the morpheme and the morpheme included in the linguistic notation Generating an intermediate notation including morpheme notation including part-of-speech information corresponding to, and accent phrase information indicating aggregation of accent phrases and prosody information indicating prosody of the accent phrases;
When correction information indicating correction content of the prosody information is stored in the storage unit in association with the generated part of speech information and the accent phrase information, the generated prosody information is corrected based on the correction information And
The speech corresponding to the language notation is synthesized based on an intermediate notation including prosody information reflecting the correction.
A speech synthesis method characterized in that.
(Appendix 7)
In the correction process, if the generated accent phrase information and the accent phrase information stored in the storage unit respectively indicate the number of types and the number of parts of speech included in the accent phrase, the correction is performed. The speech synthesis method as set forth in claim 7, characterized in that:
(Supplementary Note 8)
further,
Accept modification of the generated prosody information,
The corrected prosody information is associated with the generated part-of-speech information and accent phrase information, and the correction information indicating the content of the received correction is associated with the associated part-of-speech information and the accent phrase information The voice synthesis method according to Supplementary Note 6 or 7, characterized in that
(Appendix 9)
further,
Accept changes in the language representation targeted for the speech synthesis,
Analyzing the changed language notation to search for morphemes associated with the change;
8. A correction is performed when the correction information is stored in association with a part-of-speech information corresponding to an accent phrase including a morpheme related to the retrieved change. Speech synthesis method.
(Supplementary Note 10)
The voice synthesis method according to any one of appendices 6 to 9, characterized in that registration of the correction information is accepted and the accepted correction information is stored in the storage unit.
(Supplementary Note 11)
Accepts input of natural language linguistic expressions targeted for speech synthesis,
The linguistic notation is analyzed based on dictionary information in which information including at least a reading, part of speech, accent phrase and prosody concerning the accent phrase corresponding to a morpheme of the natural language is analyzed, and the morpheme and the morpheme included in the linguistic notation Generating an intermediate notation including morpheme notation including part-of-speech information corresponding to, and accent phrase information indicating aggregation of accent phrases and prosody information indicating prosody of the accent phrases;
When correction information indicating correction content of the prosody information is stored in the storage unit in association with the generated part of speech information and the accent phrase information, the generated prosody information is corrected based on the correction information And
The speech corresponding to the language notation is synthesized based on an intermediate notation including prosody information reflecting the correction.
A program that causes a computer to execute a process.
(Supplementary Note 12)
In the correction process, if the generated accent phrase information and the accent phrase information stored in the storage unit respectively indicate the number of types and the number of parts of speech included in the accent phrase, the correction is performed. The program according to appendix 11, wherein the program is performed.
(Supplementary Note 13)
further,
Accept modification of the generated prosody information,
The corrected prosody information is associated with the generated part-of-speech information and accent phrase information, and the correction information indicating the content of the received correction is associated with the associated part-of-speech information and the accent phrase information APPENDIX 11 or the program according to Appendix 12, which is characterized by storing.
(Supplementary Note 14)
further,
Accept changes in the language representation targeted for the speech synthesis,
Analyzing the changed language notation to search for morphemes associated with the change;
15. A correction is performed when the correction information is stored in association with a part-of-speech information corresponding to an accent phrase including a morpheme related to the retrieved change, and the correction is performed as described in any of supplementary notes 11 to 13. program.
(Supplementary Note 15)
15. The program according to any one of appendices 11 to 14, wherein the registration of the correction information is accepted, and the accepted correction information is stored in the storage unit.

DESCRIPTION OF SYMBOLS 1 speech synthesis apparatus 3 control unit 5 input / output unit 7 language processing unit 9 speech synthesis unit 11 intermediate notation correction unit 13 change search unit 15 morpheme association unit 17 correction determination unit 19 prosody correction unit 21 storage unit 41, 47 Japanese notation 42, 48 morpheme division information 43, 45, 49, 59, 66 prosody information 51, 57 part-of-speech information 52, 58 morpheme notation 53, 62 modified morpheme notation 54, 60 accent phrase information 55, 61, 67 intermediate notation 56 correspondence information 63 Change information 64 Correction determination information 65 Correction reference information

Claims (6)

An input unit that receives an input of a language expression of a natural language to be subjected to speech synthesis;
The linguistic notation is analyzed based on dictionary information in which information including at least a reading, part of speech, accent phrase and prosody concerning the accent phrase corresponding to a morpheme of the natural language is analyzed, and the morpheme and the morpheme included in the linguistic notation A morpheme notation including part-of-speech information corresponding to a word processing section for generating an intermediate notation including accent phrase information indicating grouping of accent phrases and prosody information indicating prosody of the accent phrases;
A storage unit that stores correction information indicating correction content of prosody information in association with part-of-speech information and accent phrase information;
When the number and type of part of speech included in each of the accent phrases indicated by the accent phrase information generated by the language processing unit and the accent phrase information stored in the storage unit match, the language processing unit A correction unit that corrects prosody information generated by the correction based on the correction information;
A speech synthesis unit that synthesizes speech corresponding to the language notation based on an intermediate notation including prosody information reflecting the correction by the correction unit;
A speech synthesizer characterized by having. A prosody modification receiving unit for accepting modification of prosody information generated by the language processing unit;
An associating unit that associates the prosody information after the correction with the part-of-speech information generated by the language processing unit and the accent phrase information;
And have
Wherein the storage unit, 請 Motomeko 1 modification information indicating the content of the correction accepted by the prosody modification accepting unit, and to store in association with the part of speech information associated with said accent phrase information The speech synthesizer as described in. A language change receiving unit that receives a change in language notation to be subjected to the speech synthesis;
A change search unit for searching morphemes related to the change;
And have
The language processing unit analyzes the changed language notation,
The adjustment unit, according to claim 1 or claim, characterized in that to correct if the correction information in association with part of speech information corresponding to the accent phrase including morpheme associated with retrieved the changes are stored The speech synthesizer according to 2 . It further has a registration acceptance unit that accepts registration of the correction information,
The speech synthesis apparatus according to any one of claims 1 to 3 , wherein the storage unit registers the correction information received by the registration reception unit. The speech synthesizer
Accepts input of natural language linguistic expressions targeted for speech synthesis,
The linguistic notation is analyzed based on dictionary information in which information including at least a reading, part of speech, accent phrase and prosody concerning the accent phrase corresponding to a morpheme of the natural language is analyzed, and the morpheme and the morpheme included in the linguistic notation Generating an intermediate notation including morpheme notation including part-of-speech information corresponding to, and accent phrase information indicating aggregation of accent phrases and prosody information indicating prosody of the accent phrases;
Accent phrase information stored in the storage unit when the number and type of parts of speech included in each of the accent phrases indicated by the generated accent phrase information and the accent phrase information stored in the storage unit match based on the correction information associated with the correctness of the generated said prosodic information Osamu,
The speech corresponding to the language notation is synthesized based on an intermediate notation including prosody information reflecting the correction.
A speech synthesis method characterized in that. Accepts input of natural language linguistic expressions targeted for speech synthesis,
The linguistic notation is analyzed based on dictionary information in which information including at least a reading, part of speech, accent phrase and prosody concerning the accent phrase corresponding to a morpheme of the natural language is analyzed, and the morpheme and the morpheme included in the linguistic notation Generating an intermediate notation including morpheme notation including part-of-speech information corresponding to, and accent phrase information indicating aggregation of accent phrases and prosody information indicating prosody of the accent phrases;
The accent phrase stored in the storage unit when the number and type of part of speech included in each of the accent phrases indicated by the generated accent phrase information and the accent phrase information stored in the storage unit match. based on the correction information associated with the information, correct the generated said prosodic information Osamu,
The speech corresponding to the language notation is synthesized based on an intermediate notation including prosody information reflecting the correction.
A program that causes a computer to execute a process.