JP2007212884A - Speech synthesizer, speech synthesizing method, and computer program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a speech synthesizer, a speech synthesizing method, and a computer program that expand a range of a predetermined variable part up to a position where the variable part can be connected to a pitch pattern of a fixed part with consistency based upon how high the accent at the tail of a word of the variable part is and/or how high the accent of the following fixed part is. <P>SOLUTION: Template data and text data of the variable part are acquired and reading, accent, and meter information is extracted from the acquired template data; and the acquired text data is inserted into the variable part and reading, accent information, and meter information on a synthesized speech including the fixed part is generated. When the accent extracted from the template data does not match the generated accent at the head of the initially set fixed part, the variable part is expanded until the extracted accent matches the generated accent to generate the synthesized speech. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a speech synthesizer, a speech synthesis method, and a computer program that can generate synthesized speech with natural accents based on text data. In particular, speech that can generate natural synthesized speech for text data composed of variable parts that replace arbitrary words such as place names and names, and fixed parts that are always the same without depending on the variable parts. The present invention relates to a synthesis device, a speech synthesis method, and a computer program.

  The quality of synthesized speech in a TTS (TextToSpeech) system that reads out arbitrary text data by synthesized speech is improving day by day. Of course, prosody information that shows features such as conversation rhythm, intonation, and accents when the text data that is the target of speech synthesis has been determined in advance, although it has not reached the same quality as a human natural voice. Is extracted in advance from the real voice that is actually generated, and voice synthesis is performed according to the extracted prosodic information, so that it is possible to synthesize a natural voice close to the real voice.

  Such a speech synthesis system is suitable for generating a voice response message automatically output from a machine such as an automatic teller machine of a bank or an automatic voice response system used for primary reception of various call centers. Yes. In other words, the voice response message output from the machine is often a text composed of a variable part for specifying the amount of money and other fixed parts, for example, "Are you sure you want to use XX yen?" . Therefore, a high-quality voice response message can be generated at low cost by applying the TTS system to the variable part and generating synthesized speech using prosodic information extracted from the real voice for the fixed part.

  However, since synthesized speech is generated by different methods for the variable part and the fixed part, it is necessary to adjust the prosody information so that the prosody is smooth and natural at the boundary between the variable part and the fixed part. If not adjusted, there is a problem that the prosody becomes unnatural at the boundary between the variable part and the fixed part.

  In order to solve such a problem, for example, in Patent Document 1, the prosody of the variable part is appropriately adjusted and the prosody of the fixed part is adjusted in advance so as to match the prosody of the variable part. A method is disclosed. Further, Patent Document 2 discloses a method of connecting response speech naturally and smoothly by describing in detail the combination order and combination method (control information) of text data combined with a response template. .

Furthermore, in Patent Document 3, TTS is executed including the character string around the variable part, only the prosody of the variable part is extracted, and only a part of the sentence connected to the fixed part can be changed by rule synthesis. This part discloses a speech synthesis method for synthesizing using synthesis parameters or speech waveform data obtained by analysis synthesis.
Japanese Patent Application Laid-Open No. 11-126087 Japanese Patent No. 3578961 Japanese Patent Laid-Open No. 11-038989

  However, when the above-described method is used, the accent may be unnatural even when the prosody is smoothly connected at the boundary between the variable part and the fixed part. For example, if you have prepared the prosody information of the fixed part using the voice of “(Tokyo) is sunny. (The place name part in parentheses is the variable part)” as a template. ) Is inserted, the accents become unnatural when inserted. FIG. 5 shows the difference in accents when saying "The weather in (Tokyo) is sunny" or "The weather in (Yamagata) is sunny".

  That is, the accent when uttering (Tokyo) ends up with a high end, and the leading portion of the subsequent “no weather is” has a higher pitch pattern (see FIG. 5A). In particular, the accent of “no” is classified as “high”. On the other hand, when “Yamagata” occurs, the accent ends with a lower end, and the head portion of the subsequent “No weather is” also has a lower pitch pattern (see the broken line in FIG. 5B). In this case, the accent of “no” is classified as “low”. In this way, the pitch pattern when uttering “no weather is” that is a fixed part differs depending on the type of accent of the variable part that exists immediately before, for example, the classification of the accent of “no” is “high” It is different from “Low”.

  Therefore, as disclosed in Patent Documents 1 and 3, even when only the pitch pattern of the variable portion is adjusted, the pitch pattern of the fixed portion is different from the first place. It is a synthesized voice that produces a sense of incongruity based on. In particular, there are words whose meaning changes depending on the accent in Japanese, and the meaning changes depending on the accent, such as 飴 (pitch pattern with the first half being low and the second half being high) and rain (pitch pattern with the first half being high and the second half being low). There is a problem that it may be difficult to grasp the meaning of the sentence.

  Further, as disclosed in Patent Document 2, when a pitch pattern of a fixed portion assuming another accent type is deformed, the deformation process becomes complicated, and the deformation method is applied to all template sentences, variable portions, and the like. It takes a lot of time and effort to specify a natural prosody.

  The present invention has been made in view of such circumstances, and the range of the predetermined variable portion is changed to the height of the accent at the end of the word of the variable portion and / or the height of the accent of the fixed portion that follows. Based on this, a speech synthesizer, a speech synthesis method, and a computer program capable of outputting synthesized speech as natural speech by extending to a position where the pitch pattern of the fixed part can be connected with consistency are provided. The purpose is to do.

  In order to achieve the above object, the speech synthesizer according to the first aspect of the present invention is directed to text data composed of a variable part and a fixed part. And a speech synthesizer that generates synthesized speech based on template data in which prosodic information is stored, acquisition means for acquiring the template data and variable portion text data, reading from the acquired template data, accents, and Extraction means for extracting prosodic information, generation means for inserting the acquired text data into the variable part, generating synthesized speech readings, accents, and prosodic information including the fixed part, and fixed part that is initially set The accent extracted from the template data matches the generated accent at the beginning of Determining means for determining whether or not, and variable portion extending means for expanding the variable portion to a position where the extracted accent matches the generated accent when it is determined that they do not match. The extended variable part is a text-to-speech synthesis, and the fixed part excluding the extended variable part is configured to generate a synthesized voice based on the template data.

  The speech synthesizer according to the second aspect of the present invention, for text data composed of a variable part and a fixed part, the variable part is text-to-speech synthesis, the fixed part is read in advance, and accent and prosody information are stored. In the speech synthesizer for generating synthesized speech based on the template data, the acquisition means for acquiring the template data and the text data of the variable part, and reading, accent, and prosody information are extracted from the acquired template data Extraction means, generation means for inserting the acquired text data into the variable part, and generating synthesized speech readings, accents and prosodic information including the fixed part, and the end and fixed part of the variable part that is initially set The accent extracted from the template data matches the generated accent at the beginning of Determining means for determining whether or not, and variable portion extending means for expanding the variable portion to a position where the extracted accent matches the generated accent when it is determined that they do not match. The expanded variable part is a text-to-speech synthesis, and the reduced fixed part is a synthesized voice based on the template data.

  In the speech synthesizer according to a third aspect of the present invention, in the first or second aspect of the invention, the variable portion expanding means is a portion where the accent generated by the generating means matches the accent extracted by the extracting means. Means for extracting an accent phrase having the variable part, and the variable part is extended to the end of the accent phrase closest to the initial variable part among the accent phrases extracted by the means It is characterized by that.

  In the speech synthesis method according to the fourth aspect of the present invention, for text data composed of a variable portion and a fixed portion, the variable portion is text speech synthesis, the fixed portion is read in advance, and accent and prosody information are stored. In the speech synthesis method for generating synthesized speech based on the template data, the template data and the text data of the variable part are acquired, and reading, accent, and prosody information are extracted from the acquired template data, and acquired. Insert text data into the variable part to generate synthesized speech readings, accents, and prosody information including the fixed part, and generate the accents extracted from the template data at the beginning of the fixed part that is initially set It is determined whether or not it matches the accent that has been extracted. The variable part is expanded to the position where the generated accent matches the generated accent, the expanded variable part is text-to-speech synthesis, and the fixed part excluding the expanded variable part generates synthesized speech based on the template data It is characterized by doing.

  In addition, the computer program according to the fifth aspect of the present invention, for text data composed of a variable part and a fixed part, the variable part is text-to-speech synthesis, the fixed part is read in advance, and accent and prosody information are stored. In a computer program that can be executed by a computer that generates synthesized speech based on template data, the computer reads the template data and variable portion text data from the acquired template data. Extraction means for extracting accent and prosody information, generation means for inserting the acquired text data into the variable part and generating synthesized speech reading, accent, and prosody information including the fixed part, are initially set Extracted from the template data at the beginning of the fixed part Judgment means for judging whether or not the accent and the generated accent match, and when the judgment means judges that they do not match, the position is variable up to a position where the extracted accent and the generated accent match. A variable part extending means for extending a part, and the extended variable part is a text-to-speech synthesis, and the fixed part excluding the extended variable part is made to function as a means for generating a synthesized speech based on the template data. To do.

  In the first invention, the fourth invention, and the fifth invention, for text data composed of a variable part and a fixed part, the variable part is text-to-speech synthesis, the fixed part is read in advance, and the accent and prosody information Is generated based on the template data stored. Obtain template data and variable part text data, extract reading, accent, and prosody information from the obtained template data, insert the obtained text data into the variable part, and read the synthesized speech including the fixed part , Accent, and prosody information. It is determined whether or not the accent extracted from the template data matches the generated accent at the beginning of the initial fixed portion. If it is determined that the two do not match, the variable part is expanded to the position where the extracted accent matches the generated accent, and the expanded variable part is text-to-speech synthesis. Except for the fixed part, synthesized speech is generated based on the template data. Thus, if the accent generated at the beginning of the fixed portion and the accent extracted from the template data do not match, for example, the generated accent has a low pitch pattern, and the template data has a high accent pattern. If the variable part is extended to the next part where the accent matches, the continuity of the accent can be ensured from the variable part to the fixed part of the template data. It is possible to generate a synthesized speech having natural accents and prosody similar to the real voice over the entire sentence including.

  In the second invention, for text data composed of a variable portion and a fixed portion, the variable portion is based on text-to-speech synthesis, the fixed portion is read in advance, and based on template data in which accent and prosody information are stored. To generate synthesized speech. Obtain template data and variable part text data, extract reading, accent, and prosody information from the obtained template data, insert the obtained text data into the variable part, and read the synthesized speech including the fixed part , Accent, and prosody information. It is determined whether the accent extracted from the template data matches the generated accent at the initial end of the variable portion and the beginning of the fixed portion. If it is determined that the two do not match, the variable part is expanded to the position where the extracted accent matches the generated accent, the expanded variable part is text-to-speech synthesis, and the reduced fixed part is Generate synthesized speech based on the template data. As a result, if the accents do not match at the end of the variable part and the start of the fixed part, for example, the accent at the end of the variable part has a low pitch pattern and the accent at the start of the fixed part has a high pitch pattern For example, by extending the variable part to the part where the accent matches next, it is possible to ensure a smoother continuity of the accent from the variable part to the fixed part of the template data. It is possible to generate a synthesized speech having natural accents and prosody similar to the real voice over the entire sentence including.

  In the third invention, an accent phrase having a portion in which the generated accent matches the accent extracted from the template is extracted, and the extracted accent phrase is closest to the initially set variable portion. The variable part is extended to the end of the accent phrase. As a result, by extending the variable part to the end of the accent phrase, where there is a high possibility that the accent levels match, it is possible to guarantee a natural change in accent from the variable part to the fixed part of the template data. In addition, since prosodic information included in template data of a fixed part with high quality can be used to the maximum extent, it is possible to generate synthesized speech having natural accents and prosody similar to real voices.

  According to the first invention, the fourth invention, and the fifth invention, when the accent generated at the beginning of the fixed portion does not match the accent extracted from the template data, for example, the generated accent has a low pitch. If the template data has a pitch pattern with a high accent in the template data, etc., the variable part is extended to the part where the accent matches next, so that the accent part continues from the variable part to the fixed part of the template data. Therefore, it is possible to generate synthesized speech having natural accents and prosody similar to real voice over the entire sentence including the variable part and the fixed part.

  According to the second invention, when the accents do not match at the end of the variable portion and the start of the fixed portion, for example, the pitch at the end of the variable portion has a low pitch pattern and the accent at the start of the fixed portion is high. If you have a pattern, etc., you can extend the variable part to the next part where the accent matches, thereby ensuring smoother continuity of the accent from the variable part to the fixed part of the template data. It is possible to generate a synthesized speech having natural accents and prosody similar to the real voice over the entire sentence including the part and the fixed part.

  According to the third aspect of the invention, the variable part is extended to the end part of the accent phrase that is highly likely to match the height of the accent, thereby ensuring a natural change of the accent from the variable part to the fixed part of the template data. In addition, the prosody information included in the fixed portion template data with high quality can be used to the maximum extent, so that it is possible to generate a synthesized speech having a natural accent / prosody close to the real voice.

  Hereinafter, the present invention will be described in detail with reference to the drawings illustrating embodiments thereof.

(Embodiment 1)
FIG. 1 is a block diagram showing a configuration of a computer that embodies the speech synthesis apparatus 1 according to Embodiment 1 of the present invention. A computer related to the speech synthesizer 1 according to the first embodiment of the present invention includes at least an arithmetic processing unit 11 such as a CPU and a DSP, a ROM 12 and a RAM 13, a communication interface unit 14 capable of data communication with an external computer, a fixed type. A sentence is converted into a template, and is read for each template data, and is provided with a storage unit 15 including a template storage unit 151 that stores accents, prosodic information, and the like, and an audio output unit 16 that outputs synthesized speech.

  “Template creation” means an operation of generating template data by the following procedure. First, template text in which appropriate words are inserted in the variable part for a fixed sentence composed of a variable part and a fixed part (for example, “The weather of XX is fine.” XX: variable part, other: fixed part) (For example, “The weather in Tokyo is sunny.”) Next, based on the recorded voice, the phoneme length of each phoneme, the pitch data representing the pitch of the voice, and the prosodic information corresponding to the amplitude of each phoneme are extracted. Finally, the syllable accent information (accent level, accent phrase position, etc.) is added to the template text and stored as template data. The template data includes at least a variable portion (eg, “Tokyo”) reading and accent at the time of template creation, a fixed portion reading, accent, phoneme length, and pitch data.

  The arithmetic processing unit 11 is connected to the above-described hardware units of the speech synthesizer 1 via the internal bus 17, and controls the above-described hardware units and also stores a processing program stored in the ROM 12, for example, A program for extracting reading, accent, and prosody information from the template data from the template storage unit 151, and inserting text data into the variable part, and generating synthesized speech readings, accents, and prosody information including the fixed part. Program, program that determines whether the extracted accent matches the generated accent at the end of the default variable part and the start of the fixed part, whether the accents match Execute various software functions according to programs that expand variable parts based on .

  The ROM 12 is composed of a flash memory or the like, and stores a processing program necessary for functioning as the speech synthesizer 1. The RAM 13 is composed of SRAM or the like, and stores temporary data generated when software is executed. The communication interface unit 14 receives information for identifying text data and fixed portion template data to be synthesized from the outside, or downloads the above-described program from an external computer.

  The storage unit 15 is a fixed storage device such as a hard disk, and stores information necessary for generating synthesized speech. For example, a template sentence is converted into a template and read for each template data, and a template storage unit 151 for storing accents, prosodic information, etc., a word dictionary 152 required for reading and generating accents by TTS after the beginning of the variable part, etc. It has. Note that the storage unit 15 is not limited to a fixed storage device, and may be an auxiliary storage device using a portable storage medium such as a CD or a DVD, or may be connected via the communication interface unit 14. It may be a storage device on a possible external computer. The audio output unit 16 is an audio output device such as a speaker, and outputs synthesized audio.

  FIG. 2 is a flowchart showing the procedure of the speech synthesis process of the arithmetic processing unit 11 of the speech synthesizer 1 according to Embodiment 1 of the present invention. The speech synthesizer 1 according to the first embodiment determines whether or not to expand the range set in advance as the variable part, and when it is determined to extend, the accent and mismatch of the template data of the fixed part are determined. It is characterized in that the synthesized part is generated by extending the variable part to a position where it can be connected without becoming.

  The arithmetic processing unit 11 of the speech synthesizer 1 acquires identification information for identifying template data including text data to be read out and information on a fixed part, for example, a template ID (step S201). Here, the text data to be read out may be text with only a variable part, or is composed of text including both parts on condition that the variable part and the fixed part can be identified. May be. Further, instead of the identification information for identifying the template data, a configuration in which reading of the template data, accent and prosody information is directly obtained may be used. The method for acquiring identification information for identifying text data and template data to be read is not particularly limited, and may be input by a user or acquired as data from an application that outputs synthesized speech. It may be a thing.

  The arithmetic processing unit 11 inquires the template storage unit 151 based on the acquired template ID, and extracts reading of the corresponding template data, accent, and prosody information (step S202). However, instead of acquiring the identification information for identifying the template data in step S201, if the reading of the template data, the accent, and the prosody information are acquired directly, the template storage unit 151 is not queried and the step S201 is performed. Use the acquired information as it is. FIG. 3 is a diagram illustrating an example of a data configuration of template data stored in the template storage unit 151 of the speech synthesizer 1 according to Embodiment 1 of the present invention. As shown in FIG. 3A, the template data stores information related to the reading and accent of the variable part and the fixed part in association with the template ID for identifying the template data. For example, “′” indicates a position where the accent changes from “high” to “low”, and “_” indicates the boundary of the accent phrase.

  Here, the accent phrase is the smallest unit indicating the accent of a Japanese common word phrase. For example, as an accent type, the first accent is “low”, the type is “high”, and the accent is “high”. ”To“ low ”only in one part (if the first syllable is“ high ”, the second and subsequent syllables are“ low ”; otherwise, the first syllable is“ low ”, 2 It means the smallest unit of words delimited by one of the types represented by syllable numbers that change from “high” to “low”.

  The prosodic information of the fixed part is stored in alphabetical units in association with the template IDs as shown in FIG. 3 (b), and the time length of each phoneme is in milliseconds and the time series change of the pitch is constant. It is indicated by a numerical string expressed by the frequency for each time interval.

  The data configuration stored in the template storage unit 151 is not limited to the configuration shown in FIG. FIG. 4 is a diagram showing another example of the data configuration stored in the template storage unit 151 of the speech synthesizer 1 according to Embodiment 1 of the present invention. As shown in FIG. 4A, the accent information may have a data configuration in which “high” or “low” information is given to each phoneme. Further, as shown in FIG. 4B, the fixed portion prosody information may store a pitch value at the center of each phoneme as a representative value.

  The arithmetic processing unit 11 reads the text data after the beginning of the variable part, that is, the text data obtained by connecting the input variable part text data and the fixed part of the template data with reference to the word dictionary 152, , And prosodic information (phoneme length and pitch pattern) are generated (step S203). The word dictionary 152 stores readings and accents in units of words, and the processing for generating text data readings, accents, and prosody information after the beginning of the variable portion in units of words is the same as that of normal TTS. For example, morpheme analysis processing and accenting processing are executed for text data in Kana-Kanji notation, and based on this, phoneme length and pitch pattern of each phoneme are generated by prosody generation processing.

  FIG. 5 is a diagram illustrating a state in which the continuity of accents changes due to a difference in text included in the variable part. In FIG. 5, it is assumed that the pitch pattern of the initial fixed portion of the template data is as indicated by the initial fixed portion 52 (solid line). FIG. 5A is a diagram illustrating a change in accent when “Tokyo” is input to the initially set variable portion 51. Since the ending of “Tokyo” has a “high” accent, there is no accent shift at the boundary between the initial variable portion 51 and the initial fixed portion 52. Therefore, it is possible to generate a synthesized voice without a sense of incongruity.

  FIG. 5B is a diagram illustrating a change in accent when “mountain shape” is input to the initially set variable portion 51. Since the ending of “Yamagata” is a “low” accent, in the original natural utterance, the initial accent of the fixed portion must be “low” as shown by the broken line. However, since the accent of the first portion of the fixed portion that is initially set on the template data is “high”, there is an accent shift at the boundary between the variable portion 51 that is initially set and the fixed portion 52 that is initially set. Has occurred. That is, where the accent should be changed as in the broken line portion 53, the solid line pitch pattern is connected at the boundary between the initially set variable portion 51 and the initially set fixed portion 52. There is a frequency divergence due to the difference, and if it is left as it is, a sense of incongruity occurs in the synthesized speech.

  Therefore, the arithmetic processing unit 11 sets the initially set variable part as the variable part (step S204), and the accent acquired from the template storage unit 151 in step S202 at the boundary between the variable part and the remaining fixed part. Then, it is determined whether or not the accent levels newly generated in step S203 match (step S205). When the arithmetic processing unit 11 determines that the two accents do not match (step S205: NO), the arithmetic processing unit 11 determines whether or not the variable part can be further expanded (step S206). When it is determined that the unit 11 can be expanded (step S206: YES), the arithmetic processing unit 11 expands the variable part (step S207) and returns the process to step S205.

  For example, the variable portion may be expanded as follows. The difference between the pitch newly generated in step S203 and the pitch of the template data extracted in step S202 is calculated while shifting to the right by one syllable from the boundary portion between the current variable portion and the fixed portion, and the calculated pitch difference What is necessary is just to make the position which became below the threshold value into the boundary of the newly expanded variable part and fixed part.

  If the arithmetic processing unit 11 determines that the two accents match (step S205: YES), the arithmetic processing unit 11 determines that the synthesized speech can be generated without any sense of incongruity and has expanded. The prosodic information newly generated in step S203 and the prosodic information of the template data extracted in step S202 are connected to each of the variable part and other remaining fixed parts (step S208). The arithmetic processing unit 11 generates audio data based on the connected prosodic information (step S209), and outputs the audio from the audio output unit 16 (step S210). For the generation of audio data, a waveform editing method, a waveform connection method, or the like conventionally used in the TTS system may be used.

  On the other hand, if the arithmetic processing unit 11 determines that the variable unit cannot be expanded any more (step S206: NO), the arithmetic processing unit 11 determines that the accent cannot be connected to the fixed portion of the template data without a sense of incongruity. Then, in order to generate all text after the start of the initially set variable part with TTS, all text after the start of the initially set variable part is set as a variable part (step S211), and the process is stepped. The process proceeds to S208.

  Here, in order to make the prosody more smooth in step 208, the prosody information of the variable part and the fixed part may be modified. In this case, the arithmetic processing unit 11 specifies the magnification for changing the pitch of the fixed portion so that the frequency at the end of the variable portion matches the frequency at the start of the fixed portion, and the time until the next pause break is determined. Multiply the fixed part pitch by the proportional factor along with the pitch change factor. As a result, synthesized speech can be generated in a state where continuity is ensured, and speech output without a sense of incongruity can be performed.

  As described above, according to the first embodiment, at the boundary between the end of the initially set variable portion and the end of the initially set fixed portion, the accent of the template data and the start end intimidation of the initially set variable portion. If the newly generated accent does not match the text data of, the newly created accent after the beginning of the variable part and the fixed part of the template Accents can be connected continuously, and a synthesized speech having natural accents and prosody close to the real voice can be generated.

(Embodiment 2)
Hereinafter, the speech synthesis apparatus 1 according to Embodiment 2 of the present invention will be described with reference to the drawings. Since the configuration of the computer that embodies the speech synthesizer 1 according to the second embodiment is the same as that of the first embodiment, the detailed description is omitted by attaching the same reference numerals.

  The speech synthesizer 1 according to the second embodiment determines whether or not to extend the range set in advance as the variable part, and if it is determined to extend, the extension specified in the template data in advance It is characterized in that the initially set variable part is extended to a variable part extension candidate position that can be a boundary between a later variable part and a fixed part that is reduced accordingly.

  FIG. 6 is a diagram illustrating an example of a data configuration stored in the template storage unit 151 of the speech synthesizer 1 according to Embodiment 2 of the present invention. In the second embodiment, as shown in FIG. 6, the variable part extension candidate position that can be the end of the variable part after extension is provided as a syllable number string assigned in ascending order from the position near the end of the initially set variable part. Has been granted. Here, the variable partial extension candidate position is represented by the right boundary of the syllable indicated by the syllable number, and the right boundary of “no” of the first syllable and the right boundary of “wa” of the fifth syllable are variable. Partial extension candidate positions (“′” and “_” are symbols indicating the place where the accent changes from high to low and the symbol indicating the accent phrase boundary, and are not included in the number of syllables).

  FIG. 7 is a flowchart showing a procedure of speech synthesis processing of the arithmetic processing unit 11 of the speech synthesizer 1 according to Embodiment 2 of the present invention. The speech synthesizer 1 according to the second embodiment determines whether or not to expand the range that is initially set as the variable part. If it is determined that the range is to be expanded, the variable part expansion candidate position set in the template data This is characterized in that the variable part is sequentially extended to the extension candidate position to the extension candidate position where the continuity of the accent can be ensured.

  The arithmetic processing unit 11 of the speech synthesizer 1 acquires identification information for identifying text data and template data to be read out, for example, a template ID (step S701). Here, the text data to be read out may be text of only the variable part, and is composed of text of both parts on condition that the variable part and the fixed part can be identified. Also good. Further, instead of the identification information for identifying the template data, a configuration in which the reading of the template data, the accent, and the prosody information are directly acquired may be used. The method of acquiring identification information for identifying text data and template data to be read out is not particularly limited, and may be input by a user or acquired as data from an application that outputs synthesized speech. There may be.

  The arithmetic processing unit 11 inquires of the template storage unit 151 based on the acquired template ID, and extracts the syllable number string describing the reading of the corresponding template data, the accent, the prosodic information, and the stored extended candidate position. (Step S702). The arithmetic processing unit 11 reads the text data after the beginning of the variable part with reference to the word dictionary 152, and generates accent and prosody information (step S703). However, instead of acquiring the identification information for identifying the template data in step S701, if the template data reading, accent, and prosody information are to be acquired directly, the template storage unit 151 is not queried, and step S701 is performed. Use the acquired information as it is.

  The arithmetic processing unit 11 sets the initially set variable part as an extended variable part candidate (step S704), and the newly generated accent and template in the vicinity of the boundary between the extended variable part candidate and other remaining fixed parts. It is determined whether or not the accent extracted from the data matches (step S705). When the arithmetic processing unit 11 determines that the two accents do not match (step S705: NO), the arithmetic processing unit 11 determines whether or not the next extension candidate position exists (step S706).

  If the arithmetic processing unit 11 determines that the next extension candidate position exists (step S706: YES), the arithmetic processing unit 11 sets the next extension candidate position as an extension variable part candidate (step S707), and performs the process step. Returning to S705, the above-described processing is repeated based on the expanded variable part candidate. When the arithmetic processing unit 11 determines that there is no next extension candidate position (step S706: NO), the arithmetic processing unit 11 determines that the accent could not be smoothed and was initially set. All the text after the beginning of the variable part is set as the variable part (step S709).

  When the arithmetic processing unit 11 determines that the newly generated accent matches the accent extracted from the template data in the vicinity of the boundary between the extended variable part candidate and the other remaining fixed part (step S705: YES), the arithmetic processing unit 11 determines that it is possible to generate a synthesized speech without a sense of incongruity, extends the variable part to the current extension variable part candidate (step S708), and expands the variable part and other For the remaining fixed parts, the prosodic information newly generated in step S703 and the prosodic information of the template data extracted in step S702 are connected (step S710). The arithmetic processing unit 11 generates audio data based on the connected prosodic information (step S711), and outputs the audio from the audio output unit 16 (step S712).

  Here, the determination of the accent match in step S705 may be as follows, for example. In the syllables on both sides of the boundary between the extension variable part candidate and the remaining remaining fixed part, if the accent newly generated in step S703 and the accent height of the template data extracted in step S702 match, “match”. If any one of them is different, it is determined that they do not match. Alternatively, in the syllables on both sides of the boundary, the difference between the prosodic information (pitch data value) generated in step S703 and the prosodic information (pitch data value) of the template data extracted in S702 is less than or equal to the threshold value. In this case, it is determined that they are “matched”. When the threshold value is exceeded, it is determined that “does not match”. Here, the calculation method of the difference in pitch value may use the difference in pitch at typical time, or the average value of the difference in pitch values described at regular intervals in the syllable. May be.

  As described above, according to the second embodiment, the variable part extension candidate position in which the heights of the accents are surely matched from the variable part to the fixed part of the template data, and the variable part in which the extension range is minimized The variable part can be expanded to the expansion candidate position. The expanded variable part is text-to-speech synthesis, and the remaining fixed part is generated with synthesized speech based on prosodic information prepared in advance as template data, so the continuity of accents between the variable part and the fixed part Can be generated, and it is possible to generate a synthesized speech having natural accents and prosody close to the real voice over the whole sentence.

(Embodiment 3)
Hereinafter, the speech synthesizer 1 according to Embodiment 3 of the present invention will be described with reference to the drawings. Since the configuration of the computer that embodies the speech synthesizer 1 according to the third embodiment is the same as that of the first embodiment, the detailed description is omitted by attaching the same reference numerals.

  The speech synthesizer 1 according to the third embodiment determines whether or not to expand the range of the initially set variable part, and when it determines to extend, newly generates text data after the start of the variable part. For each accent phrase, an accent phrase in which the accent of the template data prepared in advance and the newly generated accent match in the vicinity of the end of the accent phrase. Extend the variable part set in advance to the end of the adjacent accent phrase, the expanded variable part is text-to-speech synthesis, and the remaining fixed part is synthesized based on the prosodic information of the template data prepared in advance It is characterized in that voice is generated by voice.

  FIG. 8 is a diagram illustrating an example of accent heights and accent phrases. Fig. 8 (a) shows the morpheme for all text data (notifying the weather of Yamagata) after the beginning of the variable part when "Yamagata (Yamagata)" is inserted as the initial variable part. The result of the analysis and the accent assignment is shown by the height of the accent and the accent phrase boundary. FIG. 8B shows the accent height and accent phrase boundary of the fixed portion “Notenki Hahaledes (the weather is sunny)” prepared as template data (in this case variable) Template data with "Tokyo" inserted in the part). In FIGS. 8A and 8B, the accent phrase boundaries are indicated by alternate long and short dash lines A and B, and the boundary between the initially set variable portion and the initially set fixed portion is indicated by the solid line Y.

  The arithmetic processing unit 11 of the speech synthesizer 1 newly performs morphological analysis and accent assignment in a state including the initially set variable portion “Yamagata”, and sets the initially set variable portion and the initially set fixed portion. At the boundary Y with the part, the heights of the accents on both sides of the boundary Y match between the newly generated accent (FIG. 8A) and the accent of the template data (FIG. 8B). Determine whether or not. In this case, according to the newly generated accent shown in FIG. 8A, the accent of the syllable adjacent to the boundary Y between the initially set variable portion and the initially set fixed portion has a low “ta”. On the other hand, the accent of the syllable adjacent to the boundary Y between the initial variable part and the initial fixed part of the template data shown in FIG. 8B is “U”. “High” and “No” are high, and the accents on both sides of the boundary Y are different between the generated accent and the accent of the template data.

  Therefore, the arithmetic processing unit 11 continues the determination at the boundary A that is the next accent phrase boundary. First, the arithmetic processing unit 11 determines whether the boundary A is the position of the accent phrase boundary even in the template data. Since the boundary A is not an accent phrase boundary in the template data of FIG. 8B, the arithmetic processing unit 11 continues the determination at the boundary B that is the next accent phrase boundary.

  The arithmetic processing unit 11 determines whether the accent phrase boundary B is the position of the accent phrase boundary even in the template data, as in the previous time. Since the boundary B in FIG. 8 (b) coincides with the position of the accent phrase boundary of the template data, next, on both sides of the accent phrase boundary B, FIG. 8 (a) and FIG. 8 (b) It is determined whether or not the accent levels of the two match. On both sides of the accent phrase boundary B, the accent of “wa” and “ha” is “low” in FIG. 8A, and the accent of “wa” and “ha” in FIG. 8B. Is “low”, the accent heights on both sides of the accent phrase boundary B are the same in FIGS. 8A and 8B. Therefore, the arithmetic processing unit 11 extends the variable part to the accent phrase boundary B.

  9 and 10 are flowcharts showing the procedure of the speech synthesis process of the arithmetic processing unit 11 of the speech synthesis apparatus 1 according to Embodiment 3 of the present invention. The arithmetic processing unit 11 of the speech synthesizer 1 acquires identification information for identifying text data and template data to be read out, for example, a template ID (step S901). Here, the text data to be read out may be text of only the variable part, and is composed of text of both parts on condition that the variable part and the fixed part can be identified. May be. Further, instead of the identification information for identifying the template data, a configuration in which the reading of the template data, the accent, and the prosody information are directly acquired may be used. The method of acquiring identification information for identifying text data and template data to be read out is not particularly limited, and may be input by a user or acquired as data from an application that outputs synthesized speech. There may be.

  The arithmetic processing unit 11 queries the template storage unit 151 based on the acquired template ID, and extracts the reading of the corresponding template data, the accent, and the position of the stored accent phrase boundary (step S902). However, instead of acquiring the identification information for identifying the template data in step S901, if the reading of the template data, the accent, and the prosody information are acquired directly, the template storage unit 151 is not queried and the step S901 is performed. Use the acquired information as it is. The arithmetic processing unit 11 refers to the word dictionary 152 for the text data after the beginning of the variable part that has been initially set, newly reads, generates accent and prosodic information (step S903), and sets an accent phrase boundary. (Step S904).

  The arithmetic processing unit 11 sets the initially set variable part as an extended variable part candidate (step S905), and the template extracted in step S902 in the syllables on both sides of the boundary between the extended variable part candidate and the remaining fixed part. It is determined whether the data accent matches the accent newly generated in step S903 (step S906). When the arithmetic processing unit 11 determines that the accent of the template data does not match the height of the newly created accent in the vicinity of the boundary (step S906: NO), the arithmetic processing unit 11 performs a new extension variable. It is determined whether or not there is a next accent phrase boundary as a partial candidate (step S907).

  When the arithmetic processing unit 11 determines that there is a next accent phrase boundary that can newly expand the variable part (step S907: YES), the arithmetic processing unit 11 sets the next accent phrase as a new expansion variable part candidate. It is reset (step S908), and it is determined whether or not the right boundary of the accent phrase matches the position of the accent phrase boundary stored in the template data (step S909). When the arithmetic processing unit 11 determines that the position of the right boundary of the extension variable part candidate newly set in step S908 does not match the position of the accent phrase boundary stored in the template data (step S909: NO) ) The arithmetic processing unit 11 returns the process to step S907 and repeats the above-described process.

  When the arithmetic processing unit 11 determines that the position of the right boundary of the newly set extension variable part candidate matches the position of the accent phrase boundary stored in the template (step S909: YES), the arithmetic processing unit 11 returns the process to step S906.

  The arithmetic processing unit 11 sets the accent of the template data extracted in step S902 and the height of the accent newly created in step S903 on both sides of the boundary between the currently set extension variable portion candidate and the other remaining fixed portions. Are matched (step S906: YES), the variable part is expanded to the currently set extension variable part candidate (step S910), and the extended variable part and other remaining fixed parts are respectively The prosodic information newly generated in step S203 is connected to the prosodic information of the template data extracted in step S202 (step S911). The arithmetic processing unit 11 generates voice data based on the connected prosodic information (step S912), and outputs the voice from the voice output unit 16 (step S913). In step S907, when the arithmetic processing unit 11 determines that there is no next accent phrase boundary that can newly expand the variable part (step S907: NO), all the elements after the start of the initial variable part are set. Is set as a variable part (step S914), and the processing after step S911 described above is performed.

  As described above, according to the third embodiment, template data from the variable portion can be set without previously setting variable portion extension candidate positions that cannot be set unless the person has expertise in speech. It is possible to extend the variable part to the end of the accent phrase boundary where there is a high possibility that the accent level matches the fixed part of the, and to make the best use of the prosodic information contained in the template data of the fixed part with good quality Therefore, even a person who does not have speech expertise can generate synthesized speech having natural accents and prosody similar to real voices.

(Embodiment 4)
Hereinafter, the speech synthesizer 1 according to Embodiment 4 of the present invention will be described with reference to the drawings. Since the configuration of the computer that embodies the speech synthesizer 1 according to the fourth embodiment is the same as that of the first embodiment, detailed description thereof is omitted by attaching the same reference numerals. The speech synthesizer 1 according to the fourth embodiment includes the number of accent phrases of the input variable part that is the target of TTS and the template data stored in the template storage unit 151 for the variable part that is initially set. Is characterized in that it is determined whether or not to expand the initially set variable portion in accordance with the difference between the number of accent phrases in the initially set variable portion assumed in FIG.

  FIG. 11 is a diagram showing an example of a data configuration stored in the template storage unit 151 of the speech synthesizer 1 according to Embodiment 4 of the present invention. As shown in FIG. 11, not only the reading and accent (including the position of the accent phrase) of the initial fixed part “Noten Kaharedes” are stored, but also the variable part “Tokyo” used for templating. And the accent (including the position and number of accent phrases) are also stored.

  For example, when “Osaka” is input to the default variable part, the text data “Osaka weather is sunny” after the start of the variable part is subjected to morphological analysis and accenting processing, and the initial value is set. "Osaka" is obtained as an accent of the variable part. Here, when the numbers of accent phrases of “Osaka” and “Tokyo” are compared, the number of accent phrases is both “1”, so the initially set variable part is not expanded.

  Next, for example, when “Osaka City in Osaka Prefecture” is input to the initially set variable portion, morphological analysis and accenting processing are performed, and “Osaka 'Funo Osaka'shi'” is set as the default variable portion accent. Get. The number of accent phrases of “Osaka 'Funo Osaka' is” is “2”, and the number of accent phrases in the initial variable portion set in the template data is different from “1”. Therefore, for example, when the criterion value for the number of accent phrases is set to ‘1’, the variable portion is expanded because the difference in the number of accent phrases is equal to or greater than the criterion value ‘1’. Note that the reference value for determining whether or not to expand the variable portion may be set in advance or may be set in the template data. As the expansion method, the method described in Embodiments 1 to 3 may be used.

  Furthermore, even when the number of accent phrases is the same, the generated accent and the accent of the template data do not match near the boundary between the initially set variable part and the fixed part. The variable part may be expanded by the method described in 1 to 3.

  According to the fourth embodiment, even if the initially set number of words of the variable part is significantly different from the number of words of the variable part assumed in the template data, the template t-data is fixed from the variable part. It is possible to generate a synthesized speech having a natural accent that does not cause a sense of incongruity due to the difference in the height of the accent to the part.

  Note that the number of accent phrases is not limited to the number of accent phrases. For example, if the number of mora is different and the number of mora is different even if the number of accent phrases is the same, The variable part may be expanded.

  Regarding the above first to fourth embodiments, the following additional notes are disclosed.

(Appendix 1)
For text data composed of a variable part and a fixed part, the variable part is text-to-speech synthesis, the fixed part is read in advance, and synthesized speech is generated based on template data that stores accent and prosodic information In the speech synthesizer
Obtaining means for obtaining the template data and variable portion text data;
Extraction means for extracting reading, accent, and prosody information from the acquired template data;
Generating means for inserting the acquired text data into the variable part and generating the synthesized speech reading including the fixed part, accent, and prosody information;
Determination means for determining whether or not the accent extracted from the template data matches the generated accent at the beginning of the fixed portion that is initially set;
When the determination means determines that they do not match, the variable means includes variable portion expansion means for expanding the variable portion to a position where the extracted accent matches the generated accent, and the expanded variable portion is a text voice A speech synthesizer characterized in that a synthesized speech is generated on the basis of the template data for fixed portions other than the expanded variable portion in synthesis.

(Appendix 2)
For text data composed of a variable part and a fixed part, the variable part is text-to-speech synthesis, the fixed part is read in advance, and synthesized speech is generated based on template data that stores accent and prosodic information In the speech synthesizer
Obtaining means for obtaining the template data and variable portion text data;
Extraction means for extracting reading, accent, and prosody information from the acquired template data;
Generating means for inserting the acquired text data into the variable part and generating the synthesized speech reading including the fixed part, accent, and prosody information;
Determining means for determining whether the accent extracted from the template data matches the generated accent at the initial end of the variable portion and the initial end of the fixed portion;
When the determination means determines that they do not match, the variable means includes variable portion expansion means for expanding the variable portion to a position where the extracted accent matches the generated accent, and the expanded variable portion is a text voice A speech synthesizer characterized in that a synthesized speech is generated based on the template data for the fixed portion reduced by synthesis.

(Appendix 3)
The template data includes information on variable part extension candidate positions that can be a boundary between a variable part and a fixed part that change due to the extension of the variable part,
The variable part expanding means includes
Means for extracting the variable partial extension candidate position from the template data;
Means for determining whether the accent generated by the generating means matches the accent extracted by the extracting means at the extracted variable partial extension candidate position;
Of the variable partial extension candidate positions determined to match by the means, the variable part is extended to the variable partial extension candidate position closest to the initially set variable part. The speech synthesizer according to supplementary note 1 or 2.

(Appendix 4)
The variable part expanding means includes
Means for extracting an accent phrase having a portion in which the accent generated by the generating means matches the accent extracted by the extracting means;
The variable part is extended to the end of the accent phrase closest to the initially set variable part among the accent phrases extracted by the means. Speech synthesizer.

(Appendix 5)
The variable part expanding means includes
Means for extracting an accent phrase having a portion in which the accent generated by the generating means matches the accent extracted by the extracting means;
Means for determining whether the accent generated by the generating means matches the accent extracted by the extracting means at both the end of the extracted accent phrase and the beginning of the next accent phrase; ,
Supplementary note 1 or 2, wherein the variable part is extended to the end of the accent phrase closest to the initially set variable part among the accent phrases determined to match by the means. The speech synthesizer according to 2.

(Appendix 6)
The variable part expanding means includes
Means for determining whether the number of extracted accent phrases is greater than a predetermined value;
The speech synthesizer according to any one of appendices 1 to 5, wherein the variable portion that is initially set is expanded only when it is determined that the means is large.

(Appendix 7)
For text data composed of a variable part and a fixed part, the variable part is text-to-speech synthesis, the fixed part is read in advance, and synthesized speech is generated based on template data that stores accent and prosodic information In the speech synthesis method to
Obtaining the template data and variable part text data;
Extract reading, accent, and prosody information from the acquired template data,
Insert the acquired text data into the variable part, generate the synthesized speech reading, accent, and prosody information including the fixed part,
It is determined whether or not the accent extracted from the template data matches the generated accent at the beginning of the fixed portion that is initially set,
If it is determined that they do not match, the variable part is expanded to the position where the extracted accent matches the generated accent,
An extended variable part is a text-to-speech synthesizer, and a fixed part excluding the extended variable part generates a synthesized speech based on the template data.

(Appendix 8)
For text data composed of a variable part and a fixed part, the variable part is text-to-speech synthesis, the fixed part is read in advance, and synthesized speech is generated based on template data that stores accent and prosodic information In the speech synthesis method to
Obtaining the template data and variable part text data;
Extract reading, accent, and prosody information from the acquired template data,
Insert the acquired text data into the variable part, generate the synthesized speech reading, accent, and prosody information including the fixed part,
It is determined whether the accent extracted from the template data matches the generated accent at the initial end of the variable part and the start of the fixed part,
If it is determined that they do not match, the variable part is expanded to the position where the extracted accent matches the generated accent,
A speech synthesis method characterized in that the expanded variable part is a text-to-speech synthesis, and the reduced fixed part is a synthesized speech based on the template data.

(Appendix 9)
The template data includes information on variable part extension candidate positions that can be a boundary between a variable part and a fixed part that change due to the extension of the variable part,
Extracting the variable partial extension candidate position from the template data;
It is determined whether or not the generated accent matches the accent extracted from the template data at the extracted variable partial extension candidate position,
The variable part is extended to the variable part extension candidate position that is closest to the initially set variable part among the variable part extension candidate positions determined to be coincident with each other, Speech synthesis method.

(Appendix 10)
An accent phrase having a portion in which the generated accent matches the accent extracted from the template data;
9. The speech synthesis method according to appendix 7 or 8, wherein, among the extracted accent phrases, the variable part is extended to the end of the accent phrase closest to the initially set variable part.

(Appendix 11)
An accent phrase having a portion in which the generated accent matches the accent extracted from the template data;
Determining whether the generated accent and the accent extracted from the template data match at both the end of the extracted accent phrase and the beginning of the next accent phrase;
9. The speech synthesizing method according to appendix 7 or 8, wherein the variable part is expanded to the end of the accent phrase closest to the initially set variable part among the accent phrases determined to match.

(Appendix 12)
Determine whether the number of extracted accent phrases is greater than a predetermined value,
The speech synthesis method according to any one of appendices 7 to 11, wherein the variable portion that is initially set is expanded only when it is determined that the value is large.

(Appendix 13)
For text data composed of a variable part and a fixed part, the variable part is text-to-speech synthesis, the fixed part is read in advance, and synthesized speech is generated based on template data that stores accent and prosodic information In a computer program that can be executed on a computer,
The computer,
Obtaining means for obtaining the template data and text data of the variable part;
Extraction means for extracting reading, accent, and prosody information from the acquired template data,
Generating means for inserting the acquired text data into the variable part and generating the reading of the synthesized speech including the fixed part, the accent, and the prosody information;
Determination means for determining whether or not the accent extracted from the template data and the generated accent match at the beginning of the fixed portion that is initially set;
If it is determined by the determination means that the extracted accent and the generated accent match, the variable portion extending means that extends the variable portion to a position where the extracted accent matches, and the expanded variable portion is a text-to-speech synthesizer. The computer program causing the fixed part excluding the extended variable part to function as means for generating synthesized speech based on the template data.

(Appendix 14)
For text data composed of a variable part and a fixed part, the variable part is text-to-speech synthesis, the fixed part is read in advance, and synthesized speech is generated based on template data that stores accent and prosodic information In a computer program that can be executed on a computer,
The computer,
Obtaining means for obtaining the template data and text data of the variable part;
Extraction means for extracting reading, accent, and prosody information from the acquired template data,
Generating means for inserting the acquired text data into the variable part and generating the synthesized speech reading, accent, and prosody information including the fixed part;
Judgment means for judging whether or not the accent extracted from the template data matches the generated accent at the initial end of the variable part and the start of the fixed part;
If it is determined by the determination means that the extracted accent and the generated accent match, the variable portion extending means that extends the variable portion to a position where the extracted accent matches, and the expanded variable portion is a text-to-speech synthesizer. The computer program characterized in that the reduced fixed portion functions as means for generating synthesized speech based on the template data.

(Appendix 15)
The template data includes information on variable part extension candidate positions that can be a boundary between a variable part and a fixed part that change due to the extension of the variable part,
The computer,
Means for extracting the variable partial extension candidate position from the template data;
Means for determining whether or not the accent generated by the generating means and the accent extracted by the extracting means match at the extracted variable partial extension candidate position; 15. The computer according to appendix 13 or 14, wherein the computer is caused to function as means for extending the variable part to a variable part extension candidate position closest to the initially set variable part among the variable part extension candidate positions. program.

(Appendix 16)
The computer,
Means for extracting an accent phrase having a portion in which the accent generated by the generating means and the accent extracted by the extracting means match, and an initial variable set among the accent phrases extracted by the means 15. The computer program according to appendix 13 or 14, wherein the computer program functions as means for extending the variable part to the end of the accent phrase closest to the part.

(Appendix 17)
The computer,
Means for extracting an accent phrase having a portion in which the accent generated by the generating means matches the accent extracted by the extracting means;
Means for determining whether the accent generated by the generating means matches the accent extracted by the extracting means at both the end of the extracted accent phrase and the beginning of the next accent phrase; and Appendices 13 or 14 which function as means for extending the variable part to the end of the accent phrase closest to the initial variable part among the accent phrases determined to match by the means The computer program described.

(Appendix 18)
The computer,
A means for determining whether or not the number of extracted accent phrases is greater than a predetermined value, and a function for expanding the variable part that is initially set only when it is determined that the means is large. The computer program according to any one of appendices 13 to 17.

It is a block diagram which shows the structure of the computer which embodies the speech synthesizer concerning Embodiment 1 of this invention. It is a flowchart which shows the procedure of the speech synthesis process of the arithmetic processing part of the speech synthesizer concerning Embodiment 1 of this invention. It is a figure which shows an example of the data structure memorize | stored in the template memory | storage part of the speech synthesizer which concerns on Embodiment 1 of this invention. It is a figure which shows the other example of the data structure memorize | stored in the template memory | storage part of the speech synthesizer which concerns on Embodiment 1 of this invention. It is a figure which shows the state from which the continuity of an accent changes by the difference in the text contained in a variable part. It is a figure which shows an example of the data structure memorize | stored in the template memory | storage part of the speech synthesizer which concerns on Embodiment 2 of this invention. It is a flowchart which shows the procedure of the speech synthesis process of the arithmetic processing part of the speech synthesizer concerning Embodiment 2 of this invention. It is a figure which shows an example of the height of an accent, and an accent phrase. It is a flowchart which shows the procedure of the speech synthesis process of the arithmetic processing part of the speech synthesizer concerning Embodiment 3 of this invention. It is a flowchart which shows the procedure of the speech synthesis process of the arithmetic processing part of the speech synthesizer concerning Embodiment 3 of this invention. It is a figure which shows an example of the data structure memorize | stored in the template memory | storage part of the speech synthesizer which concerns on Embodiment 4 of this invention.

Explanation of symbols

1 Speech Synthesizer 11 Arithmetic Processing Unit 12 ROM
13 RAM
14 Communication Interface 15 Storage Device 16 Audio Output 17 Internal Bus

Claims (5)

For text data composed of a variable part and a fixed part, the variable part is text-to-speech synthesis, the fixed part is read in advance, and synthesized speech is generated based on template data that stores accent and prosodic information In the speech synthesizer
Obtaining means for obtaining the template data and variable portion text data;
Extraction means for extracting reading, accent, and prosody information from the acquired template data;
Generating means for inserting the acquired text data into the variable part and generating the synthesized speech reading including the fixed part, accent, and prosody information;
Determination means for determining whether or not the accent extracted from the template data matches the generated accent at the beginning of the fixed portion that is initially set;
When the determination means determines that they do not match, the variable means includes variable portion expansion means for expanding the variable portion to a position where the extracted accent matches the generated accent, and the expanded variable portion is a text voice A speech synthesizer characterized in that a synthesized speech is generated on the basis of the template data for fixed portions other than the expanded variable portion in synthesis. For text data composed of a variable part and a fixed part, the variable part is text-to-speech synthesis, the fixed part is read in advance, and synthesized speech is generated based on template data that stores accent and prosodic information In the speech synthesizer
Obtaining means for obtaining the template data and variable portion text data;
Extraction means for extracting reading, accent, and prosody information from the acquired template data;
Generating means for inserting the acquired text data into the variable part and generating the synthesized speech reading including the fixed part, accent, and prosody information;
Determining means for determining whether the accent extracted from the template data matches the generated accent at the initial end of the variable portion and the initial end of the fixed portion;
When the determination means determines that they do not match, the variable means includes variable portion expansion means for expanding the variable portion to a position where the extracted accent matches the generated accent, and the expanded variable portion is a text voice A speech synthesizer characterized in that a synthesized speech is generated based on the template data for the fixed portion reduced by synthesis. The variable part expanding means includes
Means for extracting an accent phrase having a portion in which the accent generated by the generating means matches the accent extracted by the extracting means;
3. The accent part extracted by the means is extended to the end of the accent phrase closest to the initially set variable part, wherein the variable part is expanded. Voice synthesizer. For text data composed of a variable part and a fixed part, the variable part is text-to-speech synthesis, the fixed part is read in advance, and synthesized speech is generated based on template data that stores accent and prosodic information In the speech synthesis method to
Obtaining the template data and variable part text data;
Extract reading, accent, and prosody information from the acquired template data,
Insert the acquired text data into the variable part, generate the synthesized speech reading, accent, and prosody information including the fixed part,
It is determined whether or not the accent extracted from the template data matches the generated accent at the beginning of the fixed portion that is initially set,
If it is determined that they do not match, the variable part is expanded to the position where the extracted accent matches the generated accent,
An extended variable part is a text-to-speech synthesizer, and a fixed part excluding the extended variable part generates a synthesized speech based on the template data. For text data composed of a variable part and a fixed part, the variable part is text-to-speech synthesis, the fixed part is read in advance, and synthesized speech is generated based on template data that stores accent and prosodic information In a computer program that can be executed on a computer,
The computer,
Obtaining means for obtaining the template data and text data of the variable part;
Extraction means for extracting reading, accent, and prosody information from the acquired template data,
Generating means for inserting the acquired text data into the variable part and generating the synthesized speech reading, accent, and prosody information including the fixed part;
Determination means for determining whether or not the accent extracted from the template data and the generated accent match at the beginning of the fixed portion that is initially set;
If it is determined by the determination means that the extracted accent and the generated accent match, the variable portion extending means that extends the variable portion to a position where the extracted accent matches, and the expanded variable portion is a text-to-speech synthesizer. The computer program causing the fixed part excluding the extended variable part to function as means for generating synthesized speech based on the template data.

Images