JP3655808B2 - Speech synthesis apparatus, speech synthesis method, portable terminal device, and program recording medium - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
  The present invention relates to a speech synthesizer and a speech synthesis method for synthesizing speech from character information, a portable terminal device, and a program recording medium.
[0002]
[Prior art]
  Conventionally, there is a regular speech synthesizer as disclosed in Japanese Patent Laid-Open No. 4-160630 as a device for dealing with a Japanese sentence recognition error caused by an analysis error at the time of Japanese analysis. In this regular speech synthesizer, a plurality of Japanese analysis results relating to an input character string are held in a reading sentence buffer together with a fixed order, and the plurality of pieces of Japanese analysis result information are used. According to this rule speech synthesizer, there is a recognition error (reading error in the input character string) in the synthesized speech based on the Japanese analysis result of a certain order of decision. A synthesized sound can be generated based on a Japanese analysis result that is one lower in rank, and a misread portion of the input character string can be synthesized by another reading method. By doing so, when there is an error in reading, the correction work can be performed interactively without requiring complicated labor and work.
[0003]
[Problems to be solved by the invention]
  However, the conventional regular speech synthesizer has the following problems. That is, reading errors are corrected in units of Japanese analysis results held in the reading sentence buffer. Therefore, it is not possible to specify a reading error part of the input sentence (reading sentence) composed of the input character string, read it again from that part, and repeat the voice from the middle of the reading sentence. For this reason, even when a recurrence voice is requested at the last part of a long sentence, the voice is re-uttered from the beginning of the sentence, and there is a problem that it takes time until the desired synthesized voice is obtained.
[0004]
  In addition, although the above-described regular speech synthesizer can cope with an error in reading an input character string (a reading sentence), it cannot cope with a re-listening when the user cannot hear it. In particular, when the user cannot hear only the latter half of the sentence, the user does not want to hear again from the beginning of the sentence, but wants to hear only the latter half of the sentence. In that case, there is a problem that it is impossible to make a voice replay only after a specific position in the read-out sentence.
[0005]
  SUMMARY OF THE INVENTION An object of the present invention is to provide a speech synthesizer capable of performing recurrent voice from a specific position in a speech output sentence.
[0006]
[Means for Solving the Problems]
  In order to achieve the above object, the first invention is a speech synthesizer that analyzes input character string information by an analysis unit, and synthesizes and outputs a speech by a speech synthesis unit based on the analysis result. An instruction input means for inputting an instruction for prompting, and a position specifying means for specifying a start position of a recurrent voice in the character string information based on an analysis result by the analysis means in response to an instruction from the instruction input means And control means for instructing the speech synthesis means to synthesize speech from the specified start position, and the analysis result includes a degree of inter-phrase coupling, and the position specification means is included in the character string information. The phrase that is the starting position is identified based on the number of mora and the degree of inter-phrase coupling from the phrase before the position corresponding to the time point when the instruction is received from the instruction input means. It is characterized in that.
[0007]
  According to the above configuration, when an instruction for prompting the recurrent voice in the synthesized voice is input from the instruction input unit, the position specifying unit specifies the start position of the recurrent voice based on the analysis result by the analysis unit. Therefore, when the user cannot hear the output synthesized speech, the user can hear the recurrent voice from a specific position in the output synthesized speech sentence only by giving an instruction from the instruction input means at that time, and can hear only the latter half of the long sentence. You can listen again in less time even if you did not.
[0008]
  Further, the number of mora is used as an index for specifying the start position of the recurrent voice, and a phrase that is not too close and not too far from the time when the instruction is received is selected as the start position of the recurrent voice. In addition, the inter-phrase coupling degree is used as an index for specifying the starting position of the recurrent voice, and an appropriate phrase that is linguistically weak and easily separated is selected as the starting position of the recurrent voice.
[0009]
  The second invention is an instruction for prompting a recurrent voice in a speech synthesizer that analyzes input character string information by an analysis unit and synthesizes and outputs speech by a speech synthesis unit based on the analysis result. An instruction input means for inputting the position, a position specifying means for specifying a start position of a recurrent voice in the character string information based on an analysis result by the analysis means in response to an instruction from the instruction input means, and the speech synthesis Control means for instructing the means to synthesize speech from the specified starting position, the analysis result includes likelihood, and the position specifying means indicates from the instruction input means in the character string information For the clauses before the position corresponding to the time point received, the total likelihood value is calculated while replacing each analysis result candidate, and the starting position of the recurrent voice is specified based on the obtained total likelihood value. It is characterized that it is way.
[0010]
  According to the above configuration, the start position of the recurrent voice is specified based on the total likelihood value calculated while replacing each analysis result candidate. Therefore, when the synthesized speech “It is also acted by Shijo Kaiho” is uttered based on the input character string information “Bringed by market opening”, the total likelihood value is The second candidate “hence” to be raised is selected and identified as the start position of the recurrent voice. As a result, the synthesized voice of “Thus brought” is replayed.
[0011]
  In the speech synthesizer of the first invention, the analysis result includes the likelihood, and the phrase before the position corresponding to the time point when the instruction is received from the instruction input means in the character string information. It is desirable to provide a result selection means for calculating a total likelihood value while exchanging each analysis result candidate and selecting an analysis result when performing recurrent voice based on the obtained total likelihood value.
[0012]
  According to the above configuration, an analysis result when performing recurrent voice is selected based on a total likelihood value calculated while replacing each analysis result candidate. Therefore, when the synthesized speech “Ichiba” is uttered based on the input character string information “provided by market opening”, the second candidate “Shijo” with a different utterance is selected, and “ The synthesized voice of `` Jokaibo '' is reoccurringThe
[0013]
  MaThe aboveSecondThe speech synthesizer of the invention ofthe aboveLikelihoodIn addition to ,It includes at least one of phonetic intelligibility, word appearance probability, and word importance.LocateMeans,the aboveAnalysis resultofLikelihoodIn addition to the aboveUse at least one of phonetic intelligibility, word appearance probability, and word importanceBased on the starting position candidate for the recurrent voice selected above , Identify the starting position of the recurrent voiceIt is desirable to do so.
[0014]
  According to the above configuration, based on the analysis result likelihood, the position reflecting the intention “reading error” of the recurrent voice instruction by the user is selected as a recurrent voice start position candidate.furtherBased on the phoneme intelligibility or the word appearance probability, a position reflecting the intention “cannot be heard” is selected as a start position candidate of the recurrent voice. Alternatively, a position where the intention “confirmation of important part” is reflected based on the word importance is selected as a start position candidate of the recurrent voice.
[0015]
  The speech synthesizer according to the present invention further includes a speech synthesis information generation unit configured to generate a speech synthesis information sequence based on an analysis result of the analysis unit, and based on the speech synthesis information sequence. In addition to synthesizing speech, a predetermined speech synthesis information sequence after the start position of the recurrent voice in the speech synthesis information sequence generated by the speech synthesis information generation unit according to an instruction from the control unit It is desirable to include speech synthesis information conversion means for converting into a synthesis information string.
[0016]
  According to the above configuration, the predetermined speech synthesis information sequence after the recurrent voice start position is converted into another speech synthesis information sequence by the speech synthesis information conversion means. Therefore, the utterance of the part that could not be heard is changed to “Slightly slowing the utterance speed”, “Slightly increasing the utterance pitch”, “Slightly increasing the utterance power”, or “Longer phoneme length”. It becomes possible to make a recurrent voice.
[0017]
  Further, the speech synthesizer of the present invention is characterized in that the analysis result includes reading sequence information, character sentence storage means for storing characters and corresponding sentences corresponding to the characters, and the analysis means in accordance with instructions from the control means. It is desirable to provide character sentence conversion means for converting a reading sequence of a predetermined analysis result after the start position of the recurrent voice into the analysis result string of the corresponding sentence stored in the character sentence storage means.
[0018]
  According to the above configuration, the text “s”, “z” and “ki” and the corresponding sentences “suzume no su”, “suzume no ni ku” and “stamp” ”Is stored by the text conversion means, the reading sequence“ Suzuki ”of the predetermined sentence“ Suzuki ”after the start position of the recurrent voice is converted into the corresponding sentence string“ Suzuminosu, It is converted to "Dakuten, stamp" and re-voiced.
[0019]
  In the speech synthesizer according to the present invention, it is preferable that the instruction input means is constituted by a speech recognition means for recognizing an instruction inputted by speech.
[0020]
  According to the above configuration, the user's intention to request re-utterance is clearly indicated by the user's voice.
[0021]
  The speech synthesizer according to the present invention further includes a speech recognition dictionary storing a vocabulary that promotes recurrence and an analysis dictionary storing presence / absence of relevance between a word and the vocabulary that prompts recurrence. Recognizes the vocabulary that encourages recurrent voice using the speech recognition dictionary, and the analysis means uses the analysis dictionary to obtain the analysis result including related information indicating the presence or absence of relevance to the vocabulary that promotes recurrent voice Preferably, the position specifying means specifies the start position of the recurrent voice based on the related information from the analyzing means.
[0022]
  According to the above configuration, by registering vocabularies that prompt recurrence such as “who”, “when”, and “where” in the speech recognition dictionary, these vocabularies that ask the user what information the user wants to know are asked. Is recognized, the recurrent voice is made from words representing names and times or places such as “Mr. Suzuki”, “Yesterday”, and “At work” which are highly related to these vocabularies.
[0023]
  Also,ThirdIn the speech synthesis method of analyzing the input character string information and synthesizing and outputting the speech based on the analysis result, the step of inputting an instruction for prompting a recurrent voice, and the input instruction In response, the step of identifying the start position of the recurrent voice in the character string information based on the analysis result of the input character string information, and the step of performing speech synthesis from the identified start position, The analysis result includes inter-phrase connectivity, and in the step of specifying the start position, the number of mora and the above-mentioned from the clause before the position corresponding to the time point when the instruction is received from the instruction input means in the character string information. It is characterized in that the phrase that becomes the start position is specified based on the degree of inter-joint coupling.
[0024]
  According to the above configuration, when an instruction for prompting recurrent voice by synthetic speech is input, the start position of the recurrent voice by synthetic speech is specified based on the analysis result of the input character string information. Therefore, if the user cannot hear the output synthesized speech, the user can hear the recurrent voice from a specific position in the output synthesized speech sentence only by giving an instruction at that time, and even if only the latter half of the long sentence cannot be heard You can listen again in less time.
[0025]
  Further, the number of mora is used as an index for specifying the start position of the recurrent voice, and a phrase that is not too close and not too far from the time when the instruction is received is selected as the start position of the recurrent voice. In addition, the inter-phrase coupling degree is used as an index for specifying the starting position of the recurrent voice, and an appropriate phrase that is linguistically weak and easily separated is selected as the starting position of the recurrent voice.
[0026]
  Also,4thThe portable terminal of the invention is characterized by comprising the speech synthesizer of the invention.
[0027]
  According to the above configuration, when the content of an e-mail with a relatively long text is known by synthetic voice output with a portable terminal with little character information, the recurrent voice location and the next candidate are automatically selected. Accurate recurrence and utterance changes are performed with simple operation.
[0028]
  Also,5thAccording to another aspect of the invention, there is provided a program recording medium comprising: a computer; an analyzing unit that analyzes input character string information; a voice synthesizing unit that synthesizes speech based on an analysis result including a degree of inter-phrase coupling by the analyzing unit; An instruction input means for inputting an instruction for prompting the user, and at the time when the instruction is received from the instruction input means in the character string information based on an analysis result by the analysis means in response to an instruction from the instruction input means. A position specifying means for specifying a phrase that is a starting position of a recurrent voice in the character string information based on the number of mora and the degree of inter-joint connection from the clause before the corresponding position, and for the speech synthesis means A voice synthesis processing program that functions as control means for instructing voice synthesis from the specified start position is recorded.
[0029]
  According to the above configuration, as in the case of the first invention, when the user cannot hear the output synthesized speech, the user simply gives an instruction at that time and makes a recurrent voice from a specific position in the output synthesized speech sentence. You can listen to it, and if you can't hear only the second half of the long sentence, you can listen again in less time.
[0030]
  Further, the number of mora is used as an index for specifying the start position of the recurrent voice, and a phrase that is not too close and not too far from the time when the instruction is received is selected as the start position of the recurrent voice. In addition, the inter-phrase coupling degree is used as an index for specifying the starting position of the recurrent voice, and an appropriate phrase that is linguistically weak and easily separated is selected as the starting position of the recurrent voice.
[0031]
DETAILED DESCRIPTION OF THE INVENTION
  Hereinafter, the present invention will be described in detail with reference to the illustrated embodiments.
<First embodiment>
  FIG. 1 is a block diagram showing the configuration of the speech synthesizer according to the present embodiment. The text analysis unit 1 analyzes the language of the input text, and temporarily stores the obtained text analysis result in the analysis storage unit 3. The analysis dictionary memory 2 stores language data including an analysis dictionary necessary when the text analysis unit 1 performs text analysis. Based on the text analysis result stored in the analysis storage unit 3, the phoneme processing unit 4 serving as the speech synthesis information generating means generates speech synthesis information such as reading sequence information (phoneme symbol string), accent information, and phrase break information. Is generated. Then, the speech synthesis unit 5 synthesizes speech based on the speech synthesis information sequence generated by the phoneme processing unit 4 and outputs it from a speaker (not shown).
[0032]
  The instruction input unit 6 inputs an instruction for the user to urge the user to replay the synthesized voice output from the voice synthesizer 5. The instruction input unit 6 includes buttons or a keyboard. Based on the text analysis result stored in the analysis storage unit 3, the position specifying unit 7 specifies the position where the recurrent voice is started in the text. Upon receiving an instruction from the instruction input unit 6, the control unit 8 causes the position specifying unit 7 to specify the recurrent voice start position, and reads out the subsequent text analysis result from the specified position in the text to the phonological processing unit 4. Instruct and start recurrent voice.
[0033]
  FIG. 2 is a flowchart of the speech synthesis processing operation executed by each unit of the speech synthesizer having the above configuration. In the following, the operation of the speech synthesizer will be described with reference to the flowchart of FIG. 2, taking as an example the case where a kana-kana mixed sentence “Mr. explain.
[0034]
  Here, the text input to the text analysis unit 1 may be input from a sentence input means such as a keyboard, a pen, or a voice recognition device, or may be input via a network such as mail or WWW. Thus, when a text is input to the text analysis unit 1, the speech synthesis processing operation starts.
[0035]
  In step S1, the text analysis unit 1 performs text analysis. That is, the input sentence is divided into morphemes with reference to the analysis dictionary stored in the analysis dictionary memory 2, and information such as readings and parts of speech is given. In the case of the above input sentence, "Suzuki (suzuki: proper noun)", "san (san: suffix)", "kara (kara: particle)", "business talk (shodan: noun)", "no ( ": Particle", "case (ken: noun)", "de (de: particle)", "mail (me: noun)", "ga (ga: particle)", "coming (ki: verb)" , “Te (Te: Verb)”, “I ((I: Verb)”, “Masu (Masu: Auxiliary Verb)” morphemes.
[0036]
  In step S 2, the text analysis result is stored in the analysis storage unit 3 by the text analysis unit 1. FIG. 3 shows a conceptual diagram of the storage state in the analysis storage unit 3. In FIG. 3, numerical information as described later is stored in the columns of analysis result likelihood, accent information, and inter-phrase coupling degree. In the text analysis process in step S1, a plurality of analysis candidates are obtained. Then, when the likelihood is given to each analysis candidate, the analysis candidate having the highest likelihood ranking is stored in the analysis storage unit 3 as a text analysis result. In that case, the likelihood of each analysis candidate is also stored in the analysis storage unit 3.
[0037]
  Further, a clause is generated from the concatenation of the respective analysis results, and the phrase information is stored in the analysis storage unit 3. At the same time, an inter-phrase coupling degree indicating the degree of connectivity between the preceding clause and the subsequent clause is calculated, and accent information for each clause is calculated based on an accent rule or the like. The obtained inter-phrase coupling degree and accent information are also stored together.
[0038]
  The inter-phrase coupling degree is used for controlling the prosody, and the pause length and the phrase command size are determined by the inter-phrase coupling degree. That is, if the degree of coupling between phrases is large, the coupling between both phrases is weak, so a clear pause is put in, and each phrase independently constitutes a phrase. Note that the calculation of the degree of connection between clauses is generally performed using a table (not shown) in which connection values set for each part-of-speech pattern of the preceding clause and the subsequent clause are described.
[0039]
  In step S3, the phonological processing unit 4 performs phonological processing on the word sequence having the highest analysis result likelihood based on the text analysis result stored in the analysis storage unit 3. In step S4, the speech synthesizer 5 synthesizes speech based on the speech synthesis information sequence obtained by the phoneme processing in step S3. Then, the synthesized sound is output from a speaker or the like. In step S5, it is determined whether or not there is an instruction from the instruction input unit 6 that there is a voice that is difficult to hear for the voice output from the speaker. As a result, if there is, the process proceeds to step S6, and if not, the process proceeds to step S8.
[0040]
  In step S6, the output of the synthesized speech is interrupted by the control unit 8 and the speech synthesis unit 5 at that time. In step S7, the control unit 8 and the position specifying unit 7 refer to the contents of the analysis storage unit 3 to determine the recurrent voice start phrase. Then, the phoneme processing unit 4 is instructed to read the text analysis result after the determined recurrent voice start phrase. After that, the process returns to step S3, and phonological processing and speech synthesis are performed again for the clauses after the recurrent voice start clause. If it is determined in step S5 that there is no instruction from the user, the process proceeds to step S8. The method for determining the recurrent voice start phrase will be described in detail later.
[0041]
  In step S8, it is determined whether or not there is a next sentence input to the text analysis unit 1. As a result, if there is, the process returns to step S1 and proceeds to processing of the next sentence. On the other hand, if not, the speech synthesis processing operation is terminated.
[0042]
  Hereinafter, the processing after step S4 in the above-described speech synthesis processing operation will be described in more detail according to the above-mentioned input sentence “A mail has been received from Mr. Suzuki regarding the negotiation”. Assume that the user has difficulty in hearing the word “negotiation” as shown in FIG. 4 when the speaker outputs “E-mail is coming from Mr. Suzuki regarding the negotiation” in step S4. Then, when the voice is output to the vicinity of the “mail”, when the user gives an instruction of “repeated voice”, the speech synthesis / output is interrupted once, and the process of determining from which phrase the recurrent voice is started Transition.
[0043]
  In this embodiment, as an example of the index for determining the position of a phrase that recurs, the number of mora from the point in time to the head phrase and the degree of inter-phrase coupling are used. Here, the reason why the number of mora is included in the index is to make it easy to select a position that is neither too close nor too far from the indicated time as the position of the recurrent voice start phrase. If a recurrent voice is started from a phrase with a small number of mora counted from the character corresponding to the point in time when the recurrent voice is instructed (that is, an approaching phrase), there is a possibility that it does not return to the point where it originally wants to re-listen. On the other hand, if the number of mora is too large, a recurrent voice is started from a position that goes back considerably, and it takes time to reach a point that is originally desired to be rehearsed, and may be missed again. Therefore, as shown in FIG. 5, the evaluation function f (x) is such that the evaluation function value at the appropriate number of mora becomes the highest value, and the evaluation function value gradually decreases as the number of mora further increases from the number of mora. Define it.
[0044]
  In the above example, if a recurrent voice is instructed at the position of “GA” in the fourth sentence “E-mail”, the number of mora up to the phrase “E-mail” is 4, The number of mora is 7, the number of mora up to the phrase “business negotiations” is 12, and the number of mora up to the phrase “from Mr. Suzuki” is 19. Then, while the difference in the number of mora from the point in time is “4”, it is too small, while the difference in the number of mora is “19”, it is considered that the difference between “matter” and “negotiation” is appropriate. It is.
[0045]
  The reason for using the inter-phrase coupling degree is to start recurrent voice from an appropriate segment break. In the case of the above-mentioned example sentence, the inter-phrase coupling degree B has the following relationship. The inter-phrase coupling degree B indicates that the greater the value is, the weaker the coupling degree is (that is, the more easily divided). In B (a, b), “a” is a phrase number, and “b” is a candidate number.
        B (3,1)> B (1,1)> B (4,1)> B (2,1)> B (5,1)
[0046]
  In other words, the combination of the clauses between the third clause “In case” and the fourth clause “Mail is” is the weakest, and the clause between the fifth clause “Come” and the sixth clause “Is” Strongest coupling. Then, the second phrase “Negotiation” – the third phrase “In case” (B (2,1)) and the first phrase “From Mr. Suzuki” —the second phrase “Negotiation” (B (1,1)) ), The latter is less coupled, so it is better to start recurring voices from the phrase “business negotiations” than to start recurring voices from the phrase “in case”.
[0047]
  As described above, by determining the position of the recurring phrase based on the number of mora from the indicated time to the beginning of the phrase and the degree of inter-phrase coupling, at a position that is neither too close nor too far from the indicated time, Voice can be resumed at a language-appropriate position. Therefore, in the present embodiment, a function g (x, y) is defined that uses as parameters the number of mora “x” from the indicated point to the first phrase and the inter-phrase coupling degree “y” as follows. Thus, by determining the function value for each phrase and comparing it, it is possible to optimally determine the phrase at which the recurrent voice starts.
                g (x, y) = α · f (x) + β · y
[0048]
  As described above, in the present embodiment, the analysis storage unit 3 that stores the text analysis result obtained by the text analysis unit 1 is provided, and the constituent word, analysis result likelihood, Accent information and inter-phrase connectivity are stored. If the synthesized speech output based on the text analysis result is difficult to hear, an instruction to that effect is given from the instruction input unit 6 at that time. Then, the output of the synthesized speech is interrupted under the control of the control unit 8, and the position specifying unit 7 uses the number of mora “x” and the inter-phrase coupling degree “y” from the indicated time point to the first phrase as parameters. Based on the defined function g (x, y), the position of the phrase at which the output of the synthesized speech is restarted is determined. Then, the output of the synthesized speech is resumed from the phrase position.
[0049]
  Therefore, if the user cannot hear the output synthesized speech, the user can give an instruction from the instruction input unit 6 at that time, so that the user is not too close to the instruction time and is not too far from the position. Thus, the output of the synthesized speech can be resumed. That is, when a synthesized speech is output based on the kanji-kana mixed sentence “Mr. Suzuki has received an email about a business talk” input to the text analysis unit 1, the user cannot hear the word “business talk” For this reason, when an instruction of “recurrence voice” is given in the vicinity of “mail”, the position identifying unit 7 identifies the second phrase “negotiation” as the recurrence voice start position. Then, the synthesized voice “Receiving an email on a business talk” reappears.
[0050]
  That is, according to the present embodiment, a recurrent voice can be made from a specific position in the voice output sentence, and when only the latter half of the long sentence cannot be heard, it can be rehearsed in a short time.
[0051]
  <Second Embodiment>
  According to the first embodiment, it is possible to perform recurrence from a specific position in the text, but it is not possible to perform recurrence by changing the utterance from a specific position in the text. The present embodiment addresses such a case.
[0052]
  FIG. 6 is a block diagram showing a configuration of the speech synthesizer according to the present embodiment. The analysis dictionary memory 12, the analysis storage unit 13, the phoneme processing unit 15, the speech synthesis unit 16, and the instruction input unit 17 are the analysis dictionary memory 2, the analysis storage unit 3, the phoneme processing unit 4, the speech synthesis in the first embodiment. It has the same configuration as the unit 5 and the instruction input unit 6 and operates in the same manner.
[0053]
  As described in the first embodiment, the text analysis unit 11 stores a plurality of analysis candidates obtained as a result of the text analysis processing in the analysis storage unit 13 in descending order of likelihood ranking. In addition, the position specifying unit 14 includes a result selection unit 18. This result selection means 18 selects one analysis candidate combination from the plurality of analysis candidates stored in the analysis storage unit 13 in descending order of the total analysis result likelihood and sends it to the phoneme processing unit 15. .
[0054]
  The control unit 19 receives the recurrent voice instruction from the instruction input unit 17 and instructs the position specifying unit 14 to specify the recurrent voice start position. In response to an instruction from the control unit 19, the position specifying unit 14 uses the result selection unit 18 to calculate the total value of the analysis result likelihood from the plurality of analysis candidates stored in the analysis storage unit 13. The next combination of analysis candidates is selected after the combination of analysis candidates, and the start position of the recurrent voice is specified based on the selection result.
[0055]
  FIG. 7 is a flowchart of the speech synthesis processing operation executed by each unit of the speech synthesizer having the above configuration. The operation of the speech synthesizer will be described below with reference to FIG.
[0056]
  In step S11 and step S12, text analysis and storage of the text analysis result in the analysis storage unit 13 are performed in the same manner as in step S1 and step S2 in the first embodiment. However, in this embodiment, as shown in FIG. 8, when there are a plurality of analysis candidates in one word, the first candidate, second candidate,... It stores in the column.
[0057]
  In step S13, the control unit 19 initializes the value of the variable i that counts the rank of the total value of the analysis result likelihoods regarding the currently selected combination of analysis candidates to “1”. If there is no utterance change instruction from the instruction input unit 17 in step S14, the result selection means 18 of the position specifying unit 14 determines that the analysis candidate string having the i-th largest total analysis result likelihood is the analysis storage unit 13 Selected from. Then, the phoneme processing unit 15 performs phoneme processing based on the selected analysis candidate string. Therefore, first, phoneme processing is performed on the word sequence having the highest likelihood of analysis result. In step S15, the speech synthesizer 16 synthesizes speech based on the speech synthesis information sequence obtained by the phoneme processing in step S14, and outputs the synthesized sound from a speaker or the like. In step S16, it is determined whether or not the instruction input unit 17 has an utterance change instruction indicating that there is a synthesis error for the sound output from the speaker. As a result, if there is, the process proceeds to step S17, and if not, the process proceeds to step S22.
[0058]
  In step S17, the control unit 19 and the speech synthesizer 16 perform the time (j<sub>p</sub>: Phrase number at the time of utterance change instruction) (1 ≦ j<sub>p</sub>≦ j<sub>n</sub>: J is the phrase number and n is the number of phrases), and the output of the synthesized speech is interrupted. In step S18, j ≦ j stored in the analysis storage unit 13 by the result selection unit 18<sub>p</sub>The total value ΣK (j, p (j)) for all the clauses of the analysis result likelihood K is increased to “i + 1” th while replacing the result candidates of all the clauses of (j + 1) (j<sub>a</sub>, p (j<sub>a</sub>)) (1≤j<sub>a</sub>≦ j<sub>n</sub>) Is required. However, "j<sub>a</sub>"Is the number of the phrase that was exchanged when the total value ΣK of the (i + 1) th largest analysis result likelihood K was obtained, and" p (j<sub>a</sub>) "Is the candidate number.
[0059]
  In step S19, the result selection means 18 analyzes candidate information (j of the replacement analysis candidate obtained in step S18).<sub>a</sub>, p (j<sub>a</sub>))<sub>a</sub>"P (j<sub>a</sub>) "Th analysis candidate (hereinafter, analysis candidate (j<sub>a</sub>, p (j<sub>a</sub>)))) And "j<sub>a</sub>The candidate analysis sequence “(j<sub>a</sub>+ 1, p (j<sub>a</sub>+1)) to (j<sub>n</sub>, p (j<sub>n</sub>)) ”Is selected and sent to the phoneme processing unit 15. In step S20, the content of the ranking i is incremented by the control unit 19. In step S21, it is determined whether the order i is greater than the total number N of combinations of all analysis candidates. If greater than “N”, the process proceeds to step S22. On the other hand, if “N” or less, the process returns to step S14. Thereafter, “j” used by the phoneme processing unit 15 at the time of the previous speech synthesis.<sub>a</sub>"J" in the analysis candidate string after the "th" clause<sub>a</sub>The analysis candidate of the “th” clause is the analysis candidate (j<sub>a</sub>, p (j<sub>a</sub>The synthesized speech is output based on the analysis candidate sequence replaced with)).
[0060]
  In this way, the processes in steps S17 to S21 and steps S14 to S16 are performed until it is determined in step S16 that there is no utterance change instruction or in step S21 it is determined that the order i is greater than “N”. Repeated.
[0061]
  In step S22, it is determined whether or not there is a next sentence input to the text analysis unit 11. As a result, if there is, the process returns to step S11 to shift to the processing of the next sentence. On the other hand, if not, the speech synthesis processing operation is terminated.
[0062]
  Hereinafter, the processing after step S18 in the above-described speech synthesis processing operation will be described in more detail according to the input sentence “provided by market opening”. FIG. 8 shows the storage state of the analysis storage unit 13 for the text sentence “provided by market opening”. There are two analysis candidates for the word “market”, an analysis candidate whose reading is “Ichiba” and an analysis candidate whose reading is “Shijo”. In addition, there are two analysis candidate strings of “being brought by / being brought about” and “being brought by / being brought about” by the difference in the phrase delimiter positions with respect to the phrase chain “being brought by”.
[0063]
  As a result of the text analysis by the text analysis unit 11, the analysis result likelihood K (1,1) of the first candidate “Ichiba” of the first phrase and the second candidate “Shijo” of the first phrase are analyzed. Between the result likelihood K (1, 2)
                K (1,1)> K (1,2)
This relationship is established. This is an example of misanalysis of text analysis that the character string “market” is more likely to be read as “Ichiba”.
[0064]
  In addition, the analysis result likelihoods K (3,1), K (4,1) of the first candidate in the phrase chain of the third and fourth phrases and the second candidate “according / Between the analysis result likelihood K (3,2) and K (4,2)
        K (3,1) + K (4,1)> K (3,2) + K (4,2)
This relationship is established.
[0065]
  Now, as shown in FIG. 9, when the text “bred by market opening” is synthesized and the first synthesized speech “by the first person” is output, the utterance “That” is performed. (J<sub>p</sub>= 3) When the user inputs an utterance change instruction, the analysis result candidates for the first syllable to the third syllable are sequentially replaced, and the total value ΣK (j, p (j)) of the analysis result likelihood K Is calculated. In this case, the total value of the analysis result likelihoods K of the first to third syllables is “K (1,1) + K (2,1) + K (3,1)” is the largest, and next is “K ( “1,2) + K (2,1) + K (3,1)” is the second largest, so that the candidate analysis information presents the total value ΣK of the “i + 1” th largest, that is, the second largest analysis result likelihood K. (1,2) is obtained. As a result, the result selection means 18 selects the “2” -th analysis candidate “market” in the first phrase, together with the analysis candidate strings after the second phrase used in the immediately preceding speech synthesis. It is sent to the phoneme processing unit 15. In this way, the synthesized speech “provided by Shijokaiho” is output.
[0066]
  Next, at the time when the utterance “it is done” is performed at the time of the second sound output (j<sub>p</sub>= 4), when the user inputs an utterance change instruction, the total value of the analysis result likelihood K of the first to fourth syllables is “K (1,2) + K (2,1) + K (3, 1) + K (4,1) ”is the second largest, and“ K (1,2) + K (2,1) + K (3,2) + K (4,2) ”is the third largest (see“ By selecting “Shijo”, the analysis candidate of the first phrase is fixed to K (1, 2)). Therefore, the analysis candidate information strings (3, 2) and (4, 2) are selected, and the synthesized speech “and thus” is output.
[0067]
  If the time point for inputting the utterance change instruction is delayed, there may be a mistake in the phrase for replacing the analysis result candidates. For example, since the instruction to change the utterance of the first phrase “Ichiba” is given at the time when the utterance “to be done” is performed, the utterance change of the fourth phrase “bred” is performed. This is the case. Even in such a case, since the next candidate is uttered by instructing the utterance change again, the utterance change instruction may be repeated until the desired synthesized speech is output.
[0068]
  As described above, in the present embodiment, the position specifying unit 14 is provided with the result selecting unit 18 that selects the text analysis result stored in the analysis storage unit 13. If the synthesized speech output based on the text analysis result is incorrect, a voice change instruction is given from the instruction input unit 17 at that time. Then, under the control of the control unit 19, the output of the synthesized speech is interrupted, and the result selection means 18 of the position specifying unit 14 replaces the result candidates of all the phrases before the phrase at the position where the utterance change instruction is given. However, the replacement analysis candidate (j<sub>a</sub>, p (j<sub>a</sub>)) Is obtained, and an analysis candidate string (j<sub>a</sub>, p (j<sub>a</sub>)) To (j<sub>n</sub>, p (j<sub>n</sub>)) Is selected. The phoneme processing unit 15 then analyzes the candidate sequence (j<sub>a</sub>, p (j<sub>a</sub>)) To (j<sub>n</sub>, p (j<sub>n</sub>)), The synthesized speech is output again.
[0069]
  Therefore, if the user finds an error in the output synthesized speech, the user can output the next candidate synthesized speech by issuing an utterance change instruction from the instruction input unit 17 at that time. That is, when the synthesized speech is output based on the kanji-kana mixed sentence “provided by market opening” input to the text analysis unit 11, the user wants to change the utterance of “It was done,” When an instruction to change the utterance is given, the result selection means 18 replaces the first candidate “according” and “done” in the third and fourth clauses with the second candidate “according” and “provided”. . Then, the synthesized speech of the phrase string “provided by the phrase” after the phrase in which the replacement is uttered is re-uttered.
[0070]
  That is, according to the present embodiment, it is possible to change the utterance from a specific position in the voice output sentence and perform the re-utterance, and when only the second half of the long sentence is wrong, the correction can be made in a short time.
[0071]
  As will be described later, when the voice recognition unit is used in combination, the instruction input unit 17 utters “No” to the next candidate and utters “next” to repeat the same utterance. It becomes possible. In this way, when the user explicitly gives an instruction, it is possible to prevent a synthesized speech that is different from the user's intention from being output.
[0072]
  <Third Embodiment>
  According to the first embodiment, the recurrent voice can be performed from a specific position in the voice output sentence, but if the same synthesized sound as the first is output at the time of the recurrent voice, there is a possibility that it cannot be heard again. . The present embodiment deals with such a case, and changes the speech synthesis information sequence such as the utterance speed of the portion that was difficult to hear at the time of recurrence.
[0073]
  FIG. 10 is a block diagram showing the configuration of the speech synthesizer according to the present embodiment. The text analysis unit 21, the analysis dictionary memory 22, the analysis storage unit 23, the phoneme processing unit 24, the speech synthesis unit 25, the instruction input unit 26, and the position specifying unit 27 are the text analysis unit 1, analysis dictionary in the first embodiment. The memory 2, the analysis storage unit 3, the phoneme processing unit 4, the speech synthesis unit 5, the instruction input unit 6, and the position specifying unit 7 have the same configuration and operate in the same manner.
[0074]
  The speech synthesis information conversion unit 29 converts the speech synthesis information sequence generated by the phoneme processing unit 24 into another speech synthesis information sequence in accordance with an instruction from the control unit 28. In this way, by changing the speech synthesis information sequence at a location that is difficult for the user to hear, a synthesized speech with a different utterance property can be generated, making it easier to hear the location that could not be heard before.
[0075]
  Here, as an example of conversion into an audible utterance, there are the following conversion methods.
(1) Slightly slow down the utterance speed.
(2) Slightly increase the utterance pitch.
(3) Increase the utterance power a little.
(4) Make the phoneme length a little longer.
[0076]
  For example, when changing the power, the following is performed. That is, factors affecting the vowel power include the type of the vowel, the type of adjacent phoneme, the position, the pitch, and the like, and the power value can be estimated from these values. Therefore, by setting a numerical value slightly higher than the power value estimated from the factor values, synthesized speech different from the original can be generated. In addition, when changing the utterance speed, an appropriate speed value is determined in advance, and by setting it to be slower than the appropriate value by a predetermined value at the time of recurrence, synthesized speech different from the initial one can be obtained. It generates.
[0077]
  As described above, according to the present embodiment, a synthesized speech having a speech synthesis information sequence different from the previous synthesized speech can be obtained by slightly changing an estimated value or a preset value in the speech synthesis information. Can be generated and output. Therefore, it is possible to improve the easiness of hearing at the time of recurrent voice.
[0078]
  <Fourth embodiment>
  As in the case of the third embodiment, this embodiment outputs a synthesized sound that is different from the first when a recurrent voice is produced. To communicate.
[0079]
  FIG. 11 is a block diagram showing the configuration of the speech synthesizer according to the present embodiment. The text analysis unit 31, the analysis dictionary memory 32, the analysis storage unit 33, the phonological processing unit 34, the speech synthesis unit 35, the instruction input unit 36, and the position specifying unit 37 are the text analysis unit 1, analysis dictionary in the first embodiment. The memory 2, the analysis storage unit 3, the phoneme processing unit 4, the speech synthesis unit 5, the instruction input unit 6, and the position specifying unit 7 have the same configuration and operate in the same manner.
[0080]
  The text sentence storage unit 39 is a memory that stores one character and a sentence corresponding to the letter, and stores, for example, a sentence used for a Japanese call. In addition, the above-mentioned “Japanese-language call” corresponds to a sentence such as “Asahi no A” for the letter “A”, and “Iroha no I” for the letter “I”, and accurately transmits the voice. It is used for the purpose.
[0081]
  Analysis result of the part (determined by the control unit 38 based on the recurrent voice start position specified by the position specifying unit 37) where the text conversion unit 40 determines that the user has difficulty in hearing according to the instruction from the control unit 38 Is converted into a corresponding sentence stored in the character sentence storage unit 39. For example, when it is determined that the user has difficulty in hearing the synthesized speech “Suzuki”, it is converted into a sentence used in the above Japanese call “Sparrow's, Sparrow's cloudy point, stamp”. Then, a text analysis result of this converted sentence is generated and sent to the phonological processing unit 34.
[0082]
  Thereafter, the phoneme processing unit 34 generates a speech synthesis information sequence based on the text analysis result generated by the character sentence conversion unit 40, and the speech synthesis unit 35 generates the speech synthesis information sequence based on the generated speech synthesis information sequence. Synthetic speech is output.
[0083]
  As described above, according to the present embodiment, in accordance with the instruction from the instruction input unit 36, for example, when it is determined that the user has difficulty in hearing the synthesized speech “Suzuki”, the control unit 38 and the character sentence conversion unit 40 The reading sequence information “Suzuki” is converted into a sentence used in the above Japanese call, and a text analysis result of the converted sentence is generated. Therefore, when reading the reading sequence information “Suzuki” again, it is possible to say “Suzume no Suzume, Suzume no Nisaku, Postage stamp”, and further clarify the parts that are difficult for users to hear. Can do it.
[0084]
  <Fifth embodiment>
  In the first embodiment, the recurrent voice start position is specified based on the inter-phrase coupling degree and the number of mora. In the second embodiment, the recurrent voice start position is specified based on the analysis result likelihood. In the present embodiment, a position reflecting the user's intention to request re-utterance is set as a re-utterance start position.
[0085]
  FIG. 12 is a block diagram showing the configuration of the speech synthesizer according to the present embodiment. The text analysis unit 41, the analysis dictionary memory 42, the analysis storage unit 43, the phoneme processing unit 45, the speech synthesis unit 46, the instruction input unit 47, and the speech synthesis information conversion unit 51 are the text analysis unit 21 in the third embodiment, The analysis dictionary memory 22, the analysis storage unit 23, the phoneme processing unit 24, the speech synthesis unit 25, the instruction input unit 26, and the speech synthesis information conversion unit 29 have the same configuration and operate in the same manner. The result selection unit 44 has the same configuration as the result selection unit 18 in the second embodiment and operates in the same manner.
[0086]
  The position specifying unit 48 includes start position candidate selection means 49, and specifies the start position of the recurrent voice reflecting the intention of the user requesting the recurrent voice. Hereinafter, the function of the position specifying unit 48 will be described.
[0087]
  In order to decide from which part of the input text to the text analysis unit 41 the recurrence is to be made, (1) the factor that you want to hear again, (2) the item that you want to hear again, (3) It is necessary to consider sex. Furthermore, factors that are desired to be re-listed include (a) misreading, (b) unclear phonology, and (c) presence of words with low appearance frequency. As an item to be re-listened, there is a word that determines the meaning of a sentence such as 5W1H. In addition, as a measure of relevance as a part where the voice recurs, (A) the number of mora between the phrase that is output at the time when the instruction is given and each phrase, and (B) the strength of the coupling between the phrases Is raised.
[0088]
  The position specifying unit 48 comprehensively determines these, and orders the recurring phrases. In this case, the ordering is performed by obtaining the following function values for each phrase.
        f (p1, p2, p3, p4, p5, p6)
      Where p1: analysis result parameter
              p2: Phoneme clarity parameter
              p3: Word appearance probability parameter
              p4: Word importance parameter
              p5: Inter-phrase connectivity parameter
              p6: Number of mora parameters
[0089]
  When there are a plurality of analysis results, it can be said that the smaller the likelihood of these analysis results, the higher the possibility that the analysis is erroneous. Therefore, as an example of the analysis result parameter p1, the reciprocal of the difference in the analysis result likelihood between the candidate having the first analysis result likelihood and the second candidate is considered.
[0090]
  In addition, it is highly possible that an unclear phoneme cannot be heard. Therefore, the phoneme clarity parameter p2 is used as the parameter of the function f. That is, the degree of clarity is digitized (clarity) in advance for each phoneme. In that case, the intelligibility value of a phoneme having low intelligibility is increased. Therefore, high intelligibility is given to unvoiced frictional sound and the like.
[0091]
  In addition, words that are not often used are more difficult to hear. Therefore, the word appearance probability parameter p3 is used as the parameter of the function f. In addition, if the word cannot be heard, the more important words that the meaning of the sentence itself is not understood at all are more likely to be reheard. Therefore, the word importance parameter p4 is used as the parameter of the function f. It should be noted that words related to 5W1H such as “who”, “where”, and “when”, which will be described later, are set to have a high importance.
[0092]
  The inter-phrase connectivity parameter p5 is used to start a recurrent voice from a linguistically appropriate clause break, as already described in the first embodiment. The greater the inter-phrase coupling degree, the weaker the inter-phrase coupling degree, and the more appropriate the recurring voice is. The mora number parameter p6 is also as described in the first embodiment.
[0093]
  Furthermore, the character position output when the recurrent voice instruction is input from the instruction input unit 47 includes the difference between the utterance start time and the instruction utterance time, and the difference between the instruction utterance time and the utterance end scheduled time. From the ratio, it can be estimated roughly.
[0094]
  Various methods can be considered for determining the function f. Hereinafter, a determination method based on observation data will be described. Note that the calculation method of the function f is a method of multiplying each parameter by a weighting coefficient and adding the values in order to simplify the description. The observation data includes the information stored in the analysis storage unit 43 as shown in FIG. 13, the character position where the synthesized speech was output when the recurrent voice instruction was input to the instruction input unit 47, and It is composed of the reading start position expected by the user and the reading. A large amount of these observation data is prepared, and the weighting factor is estimated so that the probability of the solution expected by the user is maximized. The weighting factor is generally estimated using a multivariate analysis method such as multiple regression analysis.
[0095]
  If there are many instructions in the observation data that there are multiple analysis results and each reading is different, if the value of the analysis result parameter p1 is large, the weight is learned so that the possibility of reading errors is high. Will be. Therefore, when there is a next candidate whose analysis result likelihood is close, it is determined that there is a reading error, and the probability of recurrence of the next candidate for reading increases.
[0096]
  The starting position candidate selecting means 49 of the position specifying unit 48 has a function f (p1, p2, p3, p4, p5, p6) for ordering recurrent voice phrases having the weighting coefficient obtained as described above. ) Is installed. When a recurrent voice instruction is input to the instruction input unit 47, the analysis result likelihood K, phonological clarity C, and word appearance of each phrase stored in the analysis storage unit 43 according to the instruction from the control unit 50 Substituting the values of the frequency F, the word importance S, the inter-phrase connectivity B, and the number of mora M, the value of the function f of each clause is calculated. In this way, a candidate position for the start of recurrent voice is selected. Then, the position specifying unit 48 specifies the phrase having the maximum value of the function f as the recurrence voice starting position, and returns the specific position to the control unit 50.
[0097]
  Then, the control unit 50 stores the predetermined number of phrases from the recurrent voice start position in the analysis storage unit 43 in the same manner as in the second embodiment with respect to the result selection unit 44. It is instructed to select, from among a plurality of analysis candidates, a combination of analysis candidates having the next largest analysis result likelihood after the combination of analysis candidates currently being output in synthesized speech. Further, the voice synthesis information conversion unit 51 is instructed to convert the voice synthesis information string in the same manner as in the third embodiment. As a result, regarding the predetermined phrase after the recurrent vocalization start phrase with the maximum value of the function f, when the parameter that is the largest factor is the analysis result parameter p1, the result selection unit 44 corrects the reading error. Will be. Furthermore, when the phoneme clarity parameter p2 or the word appearance probability parameter p3 is used, the speech synthesis information conversion unit 51 re-utters the speech by increasing the utterance power a little or decreasing the utterance speed a little. become.
[0098]
  Thus, in the present embodiment, the analysis storage unit 43 stores the analysis result likelihood K, phonological clarity C, word appearance frequency F, word importance S, inter-phrase coupling degree B, and mora for each phrase. The number M is stored. Further, the position specifying unit 48 is provided with start position candidate selecting means 49 equipped with a function f (p1, p2, p3, p4, p5, p6) for ordering recurrent voice phrases, and when there is a recurrent voice instruction. Each phrase is analyzed using the analysis result likelihood K, phonological clarity C, word appearance frequency F, word importance S, inter-phrase coupling degree B, and mora number M stored in the analysis storage unit 43. The value of the function f is calculated, and the recurrent voice start position is specified based on the calculation result. Then, the control unit 50 instructs the selection of the next analysis candidate combination for the result selection unit 44 and the conversion of the speech synthesis information sequence to the phonological conversion unit 51 with respect to the identified recurrent voice start position. .
[0099]
  Therefore, according to the present embodiment, the recurrent voice position is automatically identified by reflecting the user's recurrent voice request intention such as “reading is different”, “separation is different”, “difficult to hear”, etc. Can do it.
[0100]
  <Sixth embodiment>
  According to the fifth embodiment, even if the user does not explicitly indicate “difference in reading”, “difference in separation”, “difficult to hear”, etc., the speech synthesizer Can automatically determine the recurrent voice start position. However, by explicitly instructing such information, it is possible to more reliably determine the recurrent voice position and the recurrent voice method. The present embodiment relates to explicitly instructing the user's intention to request reoccurrence.
[0101]
  FIG. 14 is a block diagram showing a schematic configuration in the speech synthesizer of the present embodiment. The text analysis unit 61, analysis storage unit 63, phoneme processing unit 64, and speech synthesis unit 65 have the same configuration as the text analysis unit 1, analysis storage unit 3, phoneme processing unit 4, and speech synthesis unit 5 in the first embodiment. And operate similarly. The position specifying unit 68 has the same configuration as the position specifying unit 48 in the fifth embodiment and operates in the same manner.
[0102]
  The voice recognition unit 66 includes a microphone and recognizes an input voice using a voice recognition dictionary stored in the voice recognition dictionary memory 67. The speech recognition dictionary memory 67 stores a speech recognition vocabulary as a speech recognition dictionary. Here, in the speech recognition dictionary memory 67, words such as “E”, “what”, “again”, etc. that prompt a recurrence voice can be registered. It is also possible to register a vocabulary that asks what information the user wants to know, such as “who”, “when”, and “where”.
[0103]
  In addition, the analysis dictionary memory 62 is provided with information on whether each word is related to the registered vocabulary in the speech recognition dictionary memory 67 for each word. For example, the word “Suzuki” is a personal name and is associated with the registered vocabulary “who”. Alternatively, the word “6 o'clock” is time, associated with the registered vocabulary “when”, and so forth. Therefore, the text analysis unit 61 can obtain related word information as described above as an analysis result, and is stored in the analysis storage unit 63.
[0104]
  The position specifying unit 68 has start position candidate selecting means 69 equipped with a function f (p1, p2, p3, p4, p5, p6) for ordering recurrent vocal phrases. When the control unit 70 instructs to specify the recurrent voice position, the analysis result likelihood K, phonological clarity C, word appearance frequency F, and word importance S of each phrase stored in the analysis storage unit 63 , The value of the function f is obtained using the values of the inter-phrase coupling degree B and the mora number M. Then, the phrase exhibiting the largest function f value is identified as the recurrent voice position.
[0105]
  In this case, when the vocabulary registered in the speech recognition dictionary memory 67 is recognized by the speech recognition unit 66, the related word information which is one of the analysis results stored in the analysis storage unit 63 is referred to. Search for related words related to the recognition vocabulary. Then, when calculating the phrase function f (p1, p2, p3, p4, p5, p6) relating to the related word, the value of the word importance parameter p4 is made larger than usual (for example, the word importance parameter p4 And the value of the word importance S of the related word are multiplied by an integer). By doing this, if there is a word related to the recognized vocabulary in the input text sentence, the value of the function f of the phrase including the related word is increased, and recurrent voice is started from the phrase.
[0106]
  In that case, the analysis storage unit 63 keeps the analysis result of the entire text of the input text sentence, so that it is not limited to the text sentence that is currently being synthesized speech output. It becomes possible to recite back to the output text. Further, the end position is specified by using the values of the inter-phrase coupling degree B and the number of mora M in the same manner as in the case where the start position of the recurrent voice is specified by the position specifying unit 7 in the first embodiment, and the start is started. By using it together with specifying the position, it becomes feasible to re-speak only the phrase related to the person name, such as “From Mr. Suzuki” in response to the question “who” from the microphone.
[0107]
  Similarly, when the speech recognition unit 66 recognizes a vocabulary that prompts re-reading such as “different”, an analysis is performed when calculating the function f (p1, p2, p3, p4, p5, p6). The value of the result parameter p1 is made larger than usual. In this way, a result selection unit similar to the result selection unit 44 in the fifth embodiment can be used in combination to select a candidate having a different reading in a phrase having a plurality of analysis results and make a reoccurring voice. .
[0108]
  As described above, in the present embodiment, the position specifying unit 68 has the same function as the position specifying unit 48 in the fifth embodiment. Then, using the analysis result likelihood K, phonological clarity C, word appearance frequency F, word importance S, inter-phrase connectivity B, and mora number M stored in the analysis storage unit 63, The value of f (p1, p2, p3, p4, p5, p6) is obtained, and the phrase exhibiting the largest function f value is identified as the recurrent voice position.
[0109]
  At this time, when the registered vocabulary in the speech recognition dictionary memory 67 is recognized by the speech recognition unit 66, the function f of the phrase including the related word “6 o'clock” related to the recognition vocabulary “when” is calculated. At this time, the value of the word importance parameter p4 is set larger than usual. Therefore, the user can automatically start re-utterance from the desired phrase “6 o'clock ...” simply by giving a voice instruction “when”.
[0110]
  In addition, when the speech recognition unit 66 recognizes the vocabulary “different” that promotes re-reading, the value of the analysis result parameter p1 is made larger than usual when the function f is calculated. Therefore, the user can automatically select another reading candidate and re-speak by simply giving a voice instruction “different”.
[0111]
  It is effective to install the speech synthesizer in each of the above embodiments in a portable terminal device with relatively little text information such as a cellular phone or an electronic notebook. That is, when the content of an e-mail with a relatively long sentence is known by such a portable terminal with little character information, it is heard by synthetic voice. However, it is difficult to complete the accuracy rate and clarity of text-to-speech synthesis, and it is necessary to have a means of recovery in case of misanalysis or unclearness.
[0112]
  According to the speech synthesizer in each of the above embodiments, since the recurrent voice start position and the selection of the next candidate can be automatically performed, the recurrent voice and utterance change can be performed with a very simple operation. It is very effective as a speech synthesizer.
[0113]
  By the way, the functions as the analysis means, the speech synthesis means, the instruction input means, the position specification means and the control means by the text analysis section, the speech synthesis section, the instruction input section, the position specification section and the control section in the above embodiments are as follows: This is realized by a speech synthesis processing program recorded on a program recording medium. The program recording medium in each of the above embodiments is a program medium composed of a ROM (read only memory). Alternatively, it may be a program medium that is loaded into an external auxiliary storage device and read out. In any case, the program reading means for reading the voice synthesis processing program from the program medium may have a configuration in which the program medium is directly accessed and read, or the random access memory (RAM). You may have the structure which downloads to the provided program storage area (not shown), accesses the said program storage area, and reads. It is assumed that a download program for downloading from the program medium to the program storage area of the RAM is stored in the main unit in advance.
[0114]
  Here, the program medium is configured to be separable from the main body side, and is a tape system such as a magnetic tape or a cassette tape, a magnetic disk such as a floppy disk or a hard disk, or a CD (compact disk) -ROM, MO (magneto-optical). Optical discs such as discs, MDs (mini discs), DVDs (digital video discs), card systems such as IC (integrated circuit) cards and optical cards, mask ROMs, EPROMs (ultraviolet erasable ROMs), EEPROMs (electrical This is a medium that carries a fixed program including a semiconductor memory system such as an erasable ROM) and a flash ROM.
[0115]
  In addition, the speech synthesizer in each of the above embodiments has a configuration that can be connected to a communication network including the Internet, and the program medium is a medium that fluidly carries the program by downloading from the communication network or the like. There is no problem. In this case, it is assumed that a download program for downloading from the communication network is stored in the main device in advance. Or it shall be installed from another recording medium.
[0116]
  It should be noted that what is recorded on the recording medium is not limited to a program, and data can also be recorded.
[0117]
【The invention's effect】
  As is clear from the above, the speech synthesizer of the first invention receives the instruction for prompting the recurrent voice input from the instruction input unit, and the character string is determined by the position specifying unit based on the analysis result by the analyzing unit. Since the start position of the recurrent voice in the information is specified, and the control means instructs the voice synthesis means to synthesize the voice from the specified start position, when the user cannot hear the output synthesized voice, Just by giving an instruction from the instruction input means at the time, a recurrent voice can be heard from a specific position in the output synthesized speech sentence, and even if only the latter half of the long sentence cannot be heard, it can be heard again in a short time.
[0118]
  Further, the analysis result is configured to include inter-phrase connectivity, and the position specifying unit is changed from a phrase before the position corresponding to the time point when the instruction is received from the instruction input unit in the character string information. Since the phrase that is the starting position is specified based on the number of mora and the degree of coupling between the phrases, the phrase is not too close and not too far from the point in time when the instruction is received. An appropriate phrase that is weak and easily separated can be selected as the start position.
[0119]
  Further, the speech synthesizer of the second invention is configured such that the analysis result includes likelihood, and the position specifying means corresponds to a time point when an instruction is received from the instruction input means in the character string information. For the clause before the position to be calculated, the total likelihood value is calculated while replacing each analysis result candidate, and the start position of the recurrent voice is specified based on the obtained total likelihood value. When the synthesized speech “I am also driven by the way” is uttered based on the text string information “provided by market opening”, the next highest total value of the above-mentioned likelihood is raised. Two candidates “according” can be selected to identify the start position of the recurrent voice. Therefore, in that case, the synthesized speech “to be brought” can be replayed from the position of the phrase “to”.
[0120]
  The speech synthesizer according to the first aspect of the present invention is configured so that the analysis result includes a likelihood and corresponds to a time point when the result selection unit receives an instruction from the instruction input unit in the character string information. For the clause before the position to be input, calculate the total likelihood value while replacing the candidates for each analysis result, and select the analysis result when performing recurrence based on the obtained total likelihood value. When the synthesized speech “Ichiba” is uttered based on the character string information “provided by market opening”, the second candidate “Shijo” with different utterances is selected and “Shijokai Can be re-speech "synthesized voice brought about byThe
[0121]
  MaThe aboveSecondThe speech synthesizer of the invention ofthe aboveLikelihoodIn addition toPhonological clarity,At least one of word appearance probability and word importance is included, andLocateMeansthe aboveAnalysis resultofLikelihoodIn addition to the abovePhonological clarity,Using at least one of word appearance probability and word importanceThe start position of the recurrent voice is identified based on the selected start position of the recurrent voiceIf so, based on the analysis result likelihood, the position where the intention “reading error” of the recurrent voice instruction by the user is reflected,Can be selected as a starting position candidate for recurrent voice. furtherBased on phonological intelligibility or word appearance probability, the position reflecting the above intention “cannot be heard” or the position reflecting the above intention “confirm important part” based on the word importance Can be selected as a starting position candidate.
[0122]
  In the speech synthesizer of the present invention, the speech synthesis information generation unit of the speech synthesis unit generates a speech synthesis information sequence based on the analysis result of the analysis unit, and the speech synthesis information conversion unit converts the speech synthesis information. If a predetermined speech synthesis information sequence after the start position of the recurrent voice in the speech synthesis information sequence generated by the generation means is converted into another speech synthesis information sequence, the utterance of the part that could not be heard is expressed as `` speech rate. The re-utterance can be performed by changing to “slightly late”, “slightly higher pitch”, “slightly higher utterance power”, or “longer phoneme length”.
[0123]
  The speech synthesizer according to the present invention is configured so that the analysis result includes the reading sequence information, and the character sentence conversion means stores the reading sequence of the predetermined analysis result after the start position of the recurrent voice as a character sentence storage. If the corresponding sentence analysis result sequence stored in the means is converted into the character sentence storage means, for example, the characters “su”, “zu” and “ki” and the corresponding sentences “suzume no su”, “suzume no ni” If you memorize the “spot” and “stamp”, the reading sequence “Suzuki” of the given sentence “Suzuki” after the start position of the above recurrent voice will be used as the corresponding sentence column It can be re-voiced after converting to "Dakuten, stamp".
[0124]
  In the speech synthesizer according to the present invention, if the instruction input means is composed of a voice recognition means for recognizing an instruction inputted by voice, the user's voice request intention can be clearly indicated by the user's voice.
[0125]
  In the speech synthesizer of the present invention, the speech recognition means recognizes a vocabulary that promotes recurrence using a speech recognition dictionary, and the analysis means uses the analysis dictionary to determine the relevance of the vocabulary that prompts recurrence. If the analysis result including the related information indicating the presence / absence is generated and the position specifying unit specifies the start position of the recurrent voice based on the related information from the analyzing unit, the voice recognition dictionary stores “who” ”,“ When ”,“ where ”, etc., by registering vocabulary that prompts recurrence, if these vocabularies that ask the user what information they want to know are recognized by voice, It is possible to make a recurrence voice from words representing time and place, such as “Mr. Suzuki”, “Yesterday”, “In the company”, etc.
[0126]
  Also,ThirdThe speech synthesis method of the invention includes a step of giving an instruction for prompting a recurrent voice, a step of specifying a start position of a recurrent voice in the character string information based on an analysis result of the input character string information, A step of performing speech synthesis from the specified start position, wherein the analysis result includes a degree of inter-phrase coupling, and in the step of specifying the start position, an instruction is received from the instruction input means in the character string information Since the phrase that becomes the start position is specified based on the number of mora and the degree of inter-phrase coupling from the phrase before the position corresponding to the time point, when the user cannot hear the output synthesized speech, Just by giving an instruction at that time, a recurrent voice can be heard from a specific position in the output synthesized speech sentence. Therefore, even if only the second half of the long sentence cannot be heard, it can be re-listened in a short time. Furthermore, an appropriate phrase that is not too close and not too far from the point in time when the instruction is received and that is linguistically weak and easily separated can be selected as the start position.
[0127]
  Also,4thSince the portable terminal device of the invention is equipped with the speech synthesizer of the present invention, when the content of an e-mail with a relatively long sentence is known by speech synthesis output by a portable terminal device with little character information, The next candidate can be automatically selected. Therefore, accurate recurrence voices and utterance changes can be performed with a very simple operation.
[0128]
  Also,5thSince the program recording medium of the invention records the speech synthesis processing program that causes the computer to function as the analysis means, speech synthesis means, instruction input means, position specifying means, and control means in the first invention, As in the case of the first invention, if the output synthesized speech cannot be heard, the user can hear the recurrent voice from a specific position in the output synthesized speech sentence only by giving an instruction at that time. Therefore, even if only the second half of the long sentence cannot be heard, it can be re-listened in a short time. Furthermore, an appropriate phrase that is not too close and not too far from the point in time when the instruction is received and that is linguistically weak and easily separated can be selected as the start position.
[Brief description of the drawings]
FIG. 1 is a block diagram of a speech synthesizer according to the present invention.
FIG. 2 is a flowchart of a speech synthesis processing operation executed by the speech synthesizer shown in FIG.
3 is a conceptual diagram showing a storage state of an analysis storage unit in FIG. 1. FIG.
FIG. 4 is a diagram showing an example of a synthesized voice output when an instruction of “recurrent voice” is given.
FIG. 5 is a diagram illustrating a relationship between the number of mora and an evaluation function.
FIG. 6 is a block diagram of a speech synthesizer different from FIG.
7 is a flowchart of a speech synthesis processing operation executed by the speech synthesizer shown in FIG.
8 is a conceptual diagram showing a storage state of an analysis storage unit in FIG.
FIG. 9 is a diagram showing an example of synthesized speech output when an instruction “change utterance” is given.
10 is a block diagram of a speech synthesizer different from those in FIGS. 1 and 6. FIG.
11 is a block diagram of a speech synthesizer different from those shown in FIGS. 1, 6 and 10. FIG.
12 is a block diagram of a speech synthesizer different from those shown in FIGS. 1, 6, 10 and 11. FIG.
13 is a conceptual diagram showing a storage state of an analysis storage unit in FIG.
14 is a block diagram of a speech synthesizer different from those shown in FIGS. 1, 6, and 10 to 12. FIG.
[Explanation of symbols]
  1,11,21,31,41,61 ... text analysis part,
  2, 12, 22, 32, 42, 62 ... analysis dictionary memory,
  3, 13, 23, 33, 43, 63 ... analysis storage unit,
  4, 15, 24, 34, 45, 64 ... phonological processing unit,
  5, 16, 25, 35, 46, 65 ... speech synthesis unit,
  6, 17, 26, 36, 47 ... instruction input unit,
  7, 14, 27, 37, 48, 68 ... position specifying part,
  8, 19, 28, 38, 50, 70 ... control unit,
18 ... result selection means,
29, 51 ... speech synthesis information conversion unit,
39 ... a text storage unit,
40 ... character sentence conversion part,
44 ... result selection part,
49, 69 ... start position candidate selection means,
66 ... voice recognition unit,
67: Voice recognition dictionary memory.

Claims (11)

In the speech synthesizer that analyzes the input character string information by the analysis means, and synthesizes and outputs the speech by the speech synthesis means based on the analysis result,
An instruction input means for inputting an instruction for prompting a recurrence voice;
In response to an instruction from the instruction input means, based on an analysis result by the analysis means, a position specifying means for specifying a start position of a recurrent voice in the character string information;
Control means for instructing the voice synthesis means to synthesize voice from the specified start position,
The above analysis results include inter-phrase connectivity,
The position specifying means is a phrase that becomes the start position based on the number of mora and the degree of inter-phrase coupling from a phrase before the position corresponding to the time point when the instruction is received from the instruction input means in the character string information. A speech synthesizer characterized in that the voice synthesizer is specified. In the speech synthesizer that analyzes the input character string information by the analysis means, and synthesizes and outputs the speech by the speech synthesis means based on the analysis result,
An instruction input means for inputting an instruction for prompting a recurrence voice;
In response to an instruction from the instruction input means, based on an analysis result by the analysis means, a position specifying means for specifying a start position of a recurrent voice in the character string information;
Control means for instructing the voice synthesis means to synthesize voice from the specified start position,
The analysis result includes likelihood,
The position specifying means calculates a total likelihood value while replacing candidates of each analysis result with respect to a clause before the position corresponding to the time point when the instruction is received from the instruction input means in the character string information. A speech synthesizer characterized in that the start position of the recurrent voice is specified based on the total likelihood value. The speech synthesis apparatus according to claim 1,
The analysis result includes likelihood,
For the clause before the position corresponding to the time point when the instruction is received from the instruction input means in the character string information, the total likelihood value is calculated while replacing each analysis result candidate, and the obtained total likelihood value A speech synthesizer comprising: a result selection means for selecting an analysis result when performing recurrent voice based on the above. The speech synthesis apparatus according to claim 2 ,
The analysis result includes at least one of phonological clarity , word appearance probability, and word importance in addition to the likelihood ,
It said position specifying means, the phoneme clarity in addition to the likelihood of the analysis results, on the basis of the start position candidates of selection was the re-utterance using at least one word occurrence probability and word significance, the recurrence voice A speech synthesizer characterized in that the start position of the voice is specified. In the speech synthesizer according to any one of claims 1 to 4,
The speech synthesis means has speech synthesis information generation means for generating a speech synthesis information sequence based on the analysis result by the analysis means, and synthesizes speech based on the speech synthesis information sequence,
Speech synthesis for converting a predetermined speech synthesis information sequence after the start position of the recurrent voice into another speech synthesis information sequence in the speech synthesis information sequence generated by the speech synthesis information generation unit in accordance with an instruction from the control unit A speech synthesizer characterized by comprising information conversion means . The speech synthesizer according to any one of claims 1 to 4 ,
The analysis result includes reading sequence information,
A text storage means for storing a character and a corresponding sentence corresponding to the character;
In accordance with an instruction from the control means, a reading sequence of a predetermined analysis result after the start position of the recurrent voice in the analysis result by the analysis means is converted into an analysis result string of a corresponding sentence stored in the character sentence storage means. A speech synthesizer comprising a character sentence conversion means. The speech synthesizer according to any one of claims 1 to 6 ,
The voice synthesizing apparatus according to claim 1, wherein the instruction input means includes voice recognition means for recognizing an instruction inputted by voice. The speech synthesizer according to claim 7 .
A speech recognition dictionary that stores vocabulary that encourages recurrence,
An analysis dictionary that memorizes the relevance between words and the vocabulary that promotes recurrent voice
With
The voice recognition means recognizes a vocabulary that encourages recurrent voice using the voice recognition dictionary,
The analysis means uses the analysis dictionary to generate the analysis result including related information indicating the presence or absence of relevance with the vocabulary that prompts the recurrence voice,
The position specifying means specifies the start position of the recurrent voice based on the related information from the analyzing means.
Speech synthesis apparatus characterized in that it is so. In the speech synthesis method for analyzing the input character string information and synthesizing and outputting the speech based on the analysis result ,
Inputting instructions for prompting recurrence ;
In response to the input instruction, based on the analysis result of the input character string information, identifying the start position of the recurrent voice in the character string information;
Performing speech synthesis from the identified start position
With
The above analysis results include inter-phrase connectivity,
In the step of specifying the start position, the start position is determined based on the number of mora and the degree of inter-phrase coupling from the clause preceding the position corresponding to the time when the instruction is received from the instruction input means in the character string information. identify and become clause
Speech synthesis wherein a call. A portable terminal device comprising the speech synthesizer according to any one of claims 1 to 8 . Computer
An analysis means for analyzing the input character string information;
Speech synthesis means for synthesizing speech based on the analysis result including the inter-phrase coupling degree by the analysis means;
An instruction input means for inputting an instruction for prompting a recurrence voice;
In response to an instruction from the instruction input means, based on an analysis result by the analysis means, the number of mora from the clause before the position corresponding to the time point when the instruction input means is received in the character string information, and A position specifying means for specifying a phrase that is a starting position of a recurrent voice in the character string information based on the inter-phrase coupling degree;
Control means for instructing the voice synthesis means to synthesize voice from the specified start position
A computer-readable program recording medium on which is recorded a voice synthesis processing program that functions as a computer program .

Images