JP3089715B2 - Speech synthesizer - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizing apparatus combining a plurality of different synthesizing means.

[0002]

2. Description of the Related Art FIG. 11 shows an example of a conventional speech synthesizer for providing a voice message by combining rule synthesis and analysis synthesis or recording / reproduction. Reference numeral 1 denotes a sentence generation unit, to which contents to be combined are input. A synthesizing means selecting means 2 selects a synthesizing means according to the contents of the text. Reference numeral 3 denotes a rule synthesizing unit, which is composed of a language processing unit 2a, a prosody control unit 2b, a parameter creation unit 2c, and a synthesis processing unit 2d. Numeral 4 denotes a recording / reproducing means, and a waveform data storage means 4
a and read control means 4b. 5 is D
A / A section creates a synthesized voice from the voice waveform obtained in steps 3 and 4. Reference numeral 6 denotes a synthesized sound output terminal for outputting a voice, and indicates a speaker, headphones, a receiver, and the like.

[0003]

SUMMARY OF THE INVENTION Rule synthesis and analytical synthesis
There is still a considerable difference in sound quality for synthesized sounds of different synthesis means such as recording and playback. Also, when using the voice for recording and reproduction, it is not possible to change all of the pronunciation speed, accent, etc. as they were at the time of recording. For this reason, when synthesizing speech using a plurality of synthesizing means, some processing is required so that the portion connecting the synthesized sounds becomes smoother. Conventional speech synthesizers synthesize and connect voices without performing any processing.Therefore, there is a difference in sound quality at the joints, making it easy to feel uncomfortable. There is a problem that an unnatural impression is given due to different intonation.

[0004] The present invention has been made in view of the points involved, and it is possible to overlap portions connecting synthetic sounds,
It is an object of the present invention to provide a synthesized sound having a smoother and higher quality at a portion where the synthesized sound is connected by adjusting parameters such as a pronunciation speed, a volume, and an intonation.

[0005]

Means for Solving the Problems (1) Sentence generating means for generating a sentence, a plurality of different synthesizing means for synthesizing a speech waveform, and synthesizing means selecting means for selecting the synthesizing means according to the contents of the sentence; A speech synthesizing apparatus comprising an overlap unit for adding and processing waveforms of a plurality of synthesized sounds output from the synthesis unit selected by the synthesis unit selection unit.

[0006]

[0007] (2) If any of the plurality of synthesized speech is a parametric synthesis means, a parameter adjustment means for adjusting together synthesis parameters created by the combining means to the other voice, And overlapping means for adding and processing the waveforms of the plurality of synthesized sounds.

( 3 ) Further, when adding overlapping audio waveforms, an overlap means is provided for overlapping and outputting each amplifier while adjusting each amplifier using the control means.

[0009]

According to the above configuration of the present invention, (1) a sentence is segmented by the synthesizing means selecting means so that the part where the synthesizing method is switched overlaps, and a speech is created by the respectively selected synthesizing means. I do. Thereafter, the overlapped portion of the speech is overlapped to provide a smoother synthesized sound of the connected portion.

(2) The input sentence is cut by the synthesizing means selecting means, and a voice is created by the selected synthesizing means. At this time, if any of the plurality of synthesized sounds is a parametric synthesis means, the parameter adjustment means adjusts the synthesis parameters created by the synthesis means in accordance with other voices.
Provides a smoother synthetic sound at the joint.

( 2 ) Sentences are divided by the synthesizing means selecting means so that the portions where the synthesizing methods are switched overlap each other, and speech is created by the respectively selected synthesizing means. At that time, the parameter adjustment means
If any of the synthesized sounds is a parametric synthesis means, the synthesis parameters created by the synthesis means are adjusted according to the other voices, and then the overlapping parts of the voices are overlapped. Provides a smoother synthetic sound at the junction. ( 3 ) In the overlap processing in the first invention,
When adding the overlapping voice waveforms, the respective amplifiers are superimposed and output while being adjusted by using the control means, thereby providing a smoother synthesized sound at the connecting portion.

[0012]

【Example】

(Embodiment 1) FIG. 1 is a block diagram of a first embodiment of the present invention. The following describes an example of a bank transfer confirmation message by a combination of the two combining methods. The transfer confirmation message of the bank by the combination of the two combining methods is used in the same manner in all the following embodiments. In the figure, reference numeral 1 denotes a text generation unit, such as a keyboard. Numeral 2 is a synthesizing means selecting section for selecting a synthesizing method according to the text content, so that the proper noun and the following two or three characters are used for rule synthesis, and the part other than the proper noun is used for recording / reproducing. Separate sentences. 3 divides the input character string into words and phrases, and
A language processing unit 3a for determining the part of speech and the like, and a prosody control unit 3 for determining the position and length of the pose and the intonation
b, a parameter creation unit 3c for creating a synthesis parameter according to the results of the language processing unit and the prosody control unit, and a synthesis processing unit 3d for creating a synthesized sound from the parameters created by the parameter creation unit. Department. Reference numeral 4 denotes a recording / playback unit including a waveform data storage unit 4a and a reading control unit 4b. Reference numeral 5 denotes an overlap section for adding overlapping portions of the rule-synthesizing waveform data and recording / reproducing waveform data, and 6 denotes a D / A section for producing a synthesized sound from the completed waveform data. Reference numeral 7 denotes a synthesized voice output terminal that outputs voice, and indicates a speaker, headphones, a receiver, and the like.

Next, the details of each processing will be described using actual examples. "There was a transfer from Mr. Ito of Matsushita Electric."
Consider the case where the sentence is input. When a sentence to be synthesized by the sentence input unit 1 enters, the input sentence by the synthesizing means selecting unit 2 is a proper noun "Matsushita Electric's Ito" followed by a rule synthesis, and a part other than a proper noun " There was a transfer from Mr .. "is separated from recording and playback. The character string sent to the rule synthesizing unit 3 is
a, the prosody control unit 3b, and the parameter creation unit 3c divide the data as follows, and obtain information on synthesis parameters such as accent type, part of speech, and reading, and then generate a synthesized sound by the synthesis processing unit 3d. (Input sentence) "From Matsushita Electric's Ito-sama" (word division) Matsushita Electric / no / Ito / sama / kara (reading) Matsushita Tenki No Ito Sama Kara (accent type) 5B02A (part of speech) Proper noun case Auxiliary proper nouns Nouns Auxiliary Here, the accent type B given to "no" and the accent type A given to "kara" are N
HK Accent Dictionary and Commentary Appendix (Japan Broadcasting Corporation, 1985)
This shows a syllable position with a connecting accent nucleus when combining with an independent word to form a phrase.

The character string sent to the recording / reproducing unit 4 is extracted from the waveform data storage unit 4a, and the read sound is reproduced by the reading control unit 4b. After the overlapping portion 5 adds the overlapping portions of the regular synthesis waveform data and the recording / reproduction waveform data, the D / A portion 6 creates a synthesized sound from the waveform and provides a sound from the synthesized sound output terminal 7. Is done. As described above, by performing the overlap processing on the overlapping portion of the two waveform data,
It is possible to provide a sound with a more natural connection and less discomfort. In the present embodiment, as an example, “proper noun” is set as rule composition, but this does not restrict the present invention at all.

(Embodiment 2) FIG. 2 is a block diagram of a second embodiment of the present invention. In this embodiment, the recording / reproducing unit 4 having the configuration of the first embodiment is replaced with an analysis / synthesis unit. In the figure, 1 sentence generation section, 2 synthesis means selection section, 3 rule synthesis section,
The 5 overlap section, 6D / A section, and 7 synthesized voice output terminal perform the same processing as in the first embodiment. Reference numeral 4 denotes an analysis / synthesis unit including a parameter storage unit 4a and a parameter control unit 4b for generating parameters from input sentences, and a synthesis processing unit 4c for generating a synthesized sound from the parameters generated by the parameter storage unit and the parameter control unit. .

Next, the details of each processing will be described using actual examples. "There was a transfer from Mr. Ito of Matsushita Electric."
Consider the case where the sentence is input. When a sentence to be synthesized by the sentence input unit 1 enters, the input sentence by the synthesizing means selecting unit 2 is a proper noun "Matsushita Electric's Ito" followed by a rule synthesis, and a part other than a proper noun " There was a transfer from us. "Is separated from analytical synthesis. The character string sent to the rule synthesizing unit 3 is
a, the prosody control unit 3b, and the parameter creation unit 3c divide the data as follows, and obtain information on synthesis parameters such as accent type, part of speech, and reading, and then generate a synthesized sound by the synthesis processing unit 3d. (Input sentence) "From Matsushita Electric's Ito-sama" (word division) Matsushita Electric / no / Ito / sama / kara (reading) Matsushita Tenki No Ito Sama Kara (accent type) 5B02A (part of speech) Proper noun case Auxiliary proper nouns Nouns Auxiliary Here, the accent type B given to "no" and the accent type A given to "kara" are N
HK Accent Dictionary and Commentary Appendix (Japan Broadcasting Corporation, 1985)
This shows a syllable position with a connecting accent nucleus when combining with an independent word to form a phrase.

The character string sent to the analysis / synthesis unit 4 is stored in a parameter storage unit 4a, a parameter control unit 4b, and a synthesis processing unit 4
The synthesized sound is reproduced by c. In the overlap part 5,
After the overlapping portion of the rule-synthesizing waveform data and the analysis-synthesis waveform data are added together, a synthetic sound is formed by the D / A section 6 from the waveform, and a sound is provided from the synthetic sound output terminal 7. As described above, by performing the overlapping process on the overlapping portion of the two waveform data, it is possible to provide a sound with a more natural connection portion and less discomfort. In the present embodiment, as an example, “proper noun” is set as rule composition, but this does not restrict the present invention at all.

(Embodiment 3) FIG. 3 is a block diagram showing an embodiment 3 of the present invention. The rule synthesizing unit 3 of the second embodiment is replaced with a recording / reproducing unit. In the figure, 1 sentence generation section, 2 synthesis means selection section, 4 analysis synthesis section,
The 5 overlap section, 6D / A section, and 7 synthesized voice output terminal perform the same processing as in the second embodiment. Reference numeral 3 denotes a recording / reproducing unit including a waveform data storage unit 3a and a reading control unit 3b.

Next, the details of each process will be described using actual examples. "There was a transfer from Mr. Ito of Matsushita Electric."
Consider the case where the sentence is input. When a sentence to be synthesized by the sentence input unit 1 comes in, the input sentence is input by the synthesizing means selection unit 2 into the proper noun “Ito Matsushita Electric” and the subsequent “samakara” are analyzed and synthesized. There was a transfer from Mr .. "is separated from recording and playback. The character string sent to the recording / reproducing unit 3 is extracted from the waveform data storage unit 3a, and the read control unit 3b reproduces the synthesized sound. The character string sent to the analysis / synthesis unit 4 is reproduced as a synthesized sound by the parameter storage unit 4a, the parameter control unit 4b, and the synthesis processing unit 4c. After the overlapping portion 5 adds together the overlapping portions of the recording / playback waveform data and the analysis / synthesis waveform data,
A synthetic sound is formed from the waveform by the D / A unit 6, and a sound is provided from a synthetic sound output terminal 7. As described above, by performing the overlap processing on the overlapping portion of the recording / reproducing waveform data and the analysis / synthesis waveform data, it is possible to provide a sound with a more natural connection portion and less discomfort. In this embodiment, as an example, “proper noun” is set to analysis synthesis, but this does not restrict the present invention at all.

(Embodiment 1 ) FIG. 4 is a block diagram of Embodiment 1 of the present invention. Instead of the overlap processing of the first embodiment, the parameter is adjusted by a parametric synthesis means. Consider the case where the sentence "There was a transfer from Mr. Ito of Matsushita Electric" was input. Here, it is assumed that the only parameter having a difference between the two sounds is the pitch. When a sentence to be synthesized by the sentence input unit 1 enters, the input sentence by the synthesizing means selection unit 2 is a proper noun, "Matsushita Electric's Ito" is rule-synthesized, and "other than a proper noun, there was a transfer from the state." Is divided into recording and playback. The sentence sent to the rule synthesis unit is the language processing unit 3
a, the prosody control unit 3b, and the parameter creation unit 3c divide the data as follows, and obtain information on the synthesis parameters such as accent type, part of speech, and reading. (Input sentence) "Ito of Matsushita Electric" (word division) Matsushita Electric / no / Ito (reading) Matsushita Nokito Ito (accent type) 5B0 (part of speech) proper noun Gasuke proper noun Here for "no" Accent type B given
Is an NHK accent dictionary and commentary appendix (Japan Broadcasting Corporation)
1985), and shows the syllable position with the connecting accent nucleus when combining with independent words to form a phrase.

The character string sent to the recording / reproducing unit 4 is extracted from the waveform data storage unit 4a, and the read sound is reproduced by the reading control unit 4b. The parameter adjuster 5 refers to the recording / reproducing waveform data and slightly lowers and adjusts the parameter of the pitch different from the rule synthesizing parameter, and creates a synthesized sound from the adjusted parameter by the synthesizing processor 6. . A synthetic sound is created by the D / A unit 7 from the regular synthesizing waveform data and the recording / reproducing waveform data, and a sound is provided from a synthetic sound output terminal 8. In this way, by adjusting the rule-synthesizing acoustic parameters in accordance with the recording / reproducing waveform data, it is possible to provide a sound with a more natural connection portion and less discomfort. In the present embodiment, as an example, “proper noun” is set as rule composition, but this does not restrict the present invention at all. Also,
Although the parameter adjustment unit sets the parameter to adjust only the “pitch”, this is not intended to restrict the present invention in any way, and all parameters used for synthesis can be adjusted.

(Embodiment 2 ) FIG. 5 is a block diagram of Embodiment 2 of the present invention. Reference example
In this example , analysis and synthesis are used in place of recording / reproducing of No. 1 .
In the figure, 1 sentence generating section, 2 synthesizing means selecting section, and 3 rule synthesizing section perform the same processing as in the first embodiment . Reference numeral 4 denotes an analysis / synthesis unit including a parameter storage unit 4a and a parameter control unit 4b. Reference numeral 5 denotes a parameter adjustment unit that compares parameters created by the rule synthesis unit 3 and the analysis synthesis unit 4 and adjusts parameters having differences. Reference numeral 6 denotes a synthesis processing unit for generating a synthesized sound from the adjusted parameters, and reference numeral 7 denotes a D / A unit. Reference numeral 8 denotes a synthesized voice output terminal that outputs voice, and indicates a speaker, headphones, a receiver, and the like.

Next, the details of each process will be described using actual examples. "There was a transfer from Mr. Ito of Matsushita Electric."
Consider the case where the sentence is input. Here, it is assumed that the only parameter having a difference between the two sounds is the pitch. When a sentence to be synthesized by the sentence input unit 1 comes in, the input sentence by the synthesizing means selection unit 2 is a proper noun "Matsushita Electric's Ito" is rule-synthesized, and the part other than the proper noun "There was a transfer from Sama." Separated from analytical synthesis. The sentence sent to the rule synthesizing unit is divided as follows by the language processing unit 3a, the prosody control unit 3b, and the parameter creation unit 3c, and information on synthesis parameters such as accent type, part of speech, and reading is obtained. (Input sentence) "Ito of Matsushita Electric" (word division) Matsushita Electric / no / Ito (reading) Matsushita Nokito Ito (accent type) 5B0 (part of speech) proper noun Gasuke proper noun Here for "no" Accent type B given
Is an NHK accent dictionary and commentary appendix (Japan Broadcasting Corporation)
1985), and shows the syllable position with the connecting accent nucleus when combining with independent words to form a phrase.

The character string sent to the analysis / synthesis unit 4 is used to create parameters by a parameter storage unit 4a and a parameter control unit 4b. The parameter adjusting unit 5 compares the parameters for rule synthesis and the parameters for analysis and synthesis, adjusts both parameters to adjust the pitch, and then synthesizes the synthesized sound from the synthesized parameters by the synthesis processing unit 6 / D / A unit 7. A sound is provided from the structure synthesis sound output terminal 8. As described above, by adjusting the parameters when using the two synthesis means of the rule synthesis and the analysis synthesis, it is possible to provide a voice with less discomfort. In the present embodiment, as an example, “proper noun” is set as rule composition, but this does not restrict the present invention at all. In the parameter adjustment section, the parameter was set to adjust only "pitch".
This does not restrict the present invention in any way, and all parameters used for the synthesis can be adjusted.

(Embodiment 3 ) FIG. 6 is a block diagram of Embodiment 3 of the present invention. Reference example
The recording / playback is used instead of the rule synthesis of 2 .
In the figure, a one-sentence generation unit, a two-synthesis-means selection unit, and a four-analysis-synthesis unit perform the same processing as in the second embodiment . Reference numeral 3 denotes a recording / reproducing unit including a waveform data storage unit 3a and a reading control unit 3b. Reference numeral 5 denotes a parameter adjustment unit that adjusts a parameter having a difference from the parameter created by the analysis / synthesis unit 4 in the parameters for analysis / synthesis with reference to the waveform data obtained by the recording / reproduction unit 3. Reference numeral 6 denotes a synthesis processing unit that generates a synthesized sound from the adjusted parameters, and reference numeral 7 denotes a D / A unit that generates a synthesized sound from recording / reproducing waveform data and analysis / synthesis waveform data. Reference numeral 8 denotes a synthesized voice output terminal that outputs voice, and indicates a speaker, headphones, a receiver, and the like.

Next, the details of each process will be described using actual examples. "There was a transfer from Mr. Ito of Matsushita Electric."
Consider the case where the sentence is input. Here, it is assumed that the only parameter having a difference between the two sounds is the pitch. When the sentence to be synthesized by the sentence input unit 1 comes in, the input sentence is analyzed and synthesized by the synthesizing means selection unit 2 for the proper noun "Matsushita Denki Ito", and the part other than the proper noun "There was a transfer from Sama." Separated from recording and playback. The sentence sent to the recording / playback unit is the waveform data storage unit 3a.
Reproducing the synthesized sound by the read control unit 3b. The character string sent to the analysis / synthesis unit 4 is stored in the parameter storage unit 4a.
The parameter is created by the parameter control unit 4b.
The parameter adjustment unit 5 refers to the recording / playback waveform data and adjusts the analysis / synthesis parameters so as to match the pitch. Then, the synthesis processing unit 6 generates the analysis / synthesis waveform data from the synthesis parameters. The analysis / synthesis waveform data and the recording / playback waveform data pass through the D / A unit 7, and a sound is provided from a synthesized sound output terminal 8. As described above, by adjusting the parameters of the analysis and synthesis of the parametric synthesis means so as to match the recording / reproducing waveform data, it is possible to provide a sound having a more natural connection portion and less discomfort. In this embodiment, as an example, “proper noun” is set to analysis synthesis, but this does not restrict the present invention at all. Further, although the parameter adjustment section sets only the parameter for adjusting the “pitch”, the parameter is not limited to the present invention at all, and all parameters used for the synthesis can be adjusted.

(Embodiment 4 ) FIG. 7 is a block diagram of a fourth embodiment of the present invention. The overlap processing is performed in place of the D / A unit having the configuration of the third embodiment. In the figure, one sentence generation unit, 4 recording and reproduction unit, 5 parameter adjustment unit, and 6 synthesis processing unit perform the same processing as in the third embodiment. Consider the case where the sentence "There was a transfer from Mr. Ito of Matsushita Electric" was input. Here, it is assumed that the only parameter having a difference between the two sounds is the pitch. In the figure,
According to the combining means selecting unit 2, the input sentence is divided into the proper noun "Matsushita Denki Ito" and the subsequent part of "Samakara" is rule-based synthesis, and the part other than the proper noun "Some transfer was made" is divided into recording and playback. . The character string sent to the rule synthesizing unit 3 is
Language processing unit 3a, prosody control unit 3b, parameter creation unit 3
After being divided by c as described below and obtaining information on synthesis parameters such as accent type, part of speech, and reading, a synthesis sound is created by the synthesis processing unit 3d. (Input sentence) "From Matsushita Electric's Ito-sama" (word division) Matsushita Electric / no / Ito / sama / kara (reading) Matsushita Tenki No Ito Sama Kara (accent type) 5B02A (part of speech) Proper noun case Auxiliary proper nouns Nouns Auxiliary Here, the accent type B given to "no" and the accent type A given to "kara" are N
HK Accent Dictionary and Commentary Appendix (Japan Broadcasting Corporation, 1985)
This shows a syllable position with a connecting accent nucleus when combining with an independent word to form a phrase.

The character string sent to the recording / reproducing unit 4 is extracted from the waveform data storage unit 4a, and the read-out control unit 4b reproduces the synthesized sound. The parameter adjuster 5 refers to the recording / reproducing waveform data, adjusts the parameter of the pitch different from the rule synthesis parameter, and creates a synthesized sound from the adjusted parameter by the synthesis processing unit 6. Further, in the overlap section 7, the overlapped portion of the regular synthesis waveform data and the recording / playback waveform data is added to produce a synthetic sound, and the sound is provided from the synthetic sound output terminal 8. In this way, the acoustic parameters for rule synthesis are adjusted in accordance with the waveform data for recording / reproduction, and the overlapping portions of the waveform data for rule synthesis and the recording / reproduction waveform data are output in an overlapping manner, so that the connection portion is obtained. Can provide a more natural sound with less discomfort. In the present embodiment, as an example, “proper noun” is set as rule composition, but this does not restrict the present invention at all. Further, although the parameter adjustment section sets only the parameter for adjusting the “pitch”, the parameter is not limited to the present invention at all, and all parameters used for the synthesis can be adjusted.

(Embodiment 5 ) FIG. 8 is a block diagram of a fifth embodiment of the present invention. The overlap processing is performed in place of the D / A unit having the configuration of the third embodiment. In the figure, one sentence generation part, four analysis synthesis parts, five parameter adjustment parts, and six synthesis processing parts perform the same processing as in the fourth embodiment. Consider the case where the sentence "There was a transfer from Mr. Ito of Matsushita Electric" was input. Here, it is assumed that the only parameter having a difference between the two sounds is the pitch. In the figure,
According to the combining means selecting unit 2, the input sentence is divided into the proper noun "Matsushita Electric's Ito" and the subsequent part of "from the form" is rule composition, and the part other than the proper noun "there was a transfer from the form" is divided into analysis and composition. . The character string sent to the rule synthesizing unit 3 is
Language processing unit 3a, prosody control unit 3b, parameter creation unit 3
After being divided by c as described below and obtaining information on synthesis parameters such as accent type, part of speech, and reading, a synthesis sound is created by the synthesis processing unit 3d. (Input sentence) "From Matsushita Electric's Ito-sama" (word division) Matsushita Electric / no / Ito / sama / kara (reading) Matsushita Tenki No Ito Sama Kara (accent type) 5B02A (part of speech) Proper noun case Auxiliary proper nouns Nouns Auxiliary Here, the accent type B given to "no" and the accent type A given to "kara" are N
HK Accent Dictionary and Commentary Appendix (Japan Broadcasting Corporation, 1985)
This shows a syllable position with a connecting accent nucleus when combining with an independent word to form a phrase.

The character string sent to the recording / reproducing section 4 is extracted from the waveform data storage section 4a, and the synthesized voice is reproduced by the reading control section 4b. The parameter adjuster 5 refers to the recording / reproducing waveform data, adjusts the parameter of the pitch different from the rule synthesis parameter, and creates a synthesized sound from the adjusted parameter by the synthesis processing unit 6. Further, the overlapping section 7 adds the overlapped portions of the rule-synthesizing waveform data and the analysis-synthesizing waveform data to form a synthesized sound, and a sound is provided from the synthesized sound output terminal 8. In this way, by comparing and adjusting the rule synthesizing parameter and the analysis synthesizing parameter, and overlapping and outputting the overlapping portion of the rule synthesizing waveform data and the analysis synthesizing waveform data, the connection portion can be further improved. It is possible to provide natural and less uncomfortable voice. In the present embodiment, as an example, “proper noun” is set as rule composition, but this does not restrict the present invention at all. Further, although the parameter adjustment section sets only the parameter for adjusting the “pitch”, the parameter is not limited to the present invention at all, and all parameters used for the synthesis can be adjusted.

[0031] (Embodiment 6) FIG. 9 is a block diagram of an embodiment of the present invention of claim 2 wherein. The overlap processing is performed in place of the D / A unit having the configuration of the third embodiment. In the figure, one sentence generation unit, 4 analysis and synthesis unit, 5 parameter adjustment unit, and 6 synthesis processing unit perform the same processing as in the fifth embodiment. Reference numeral 3 denotes a recording / reproducing unit including a waveform data storage unit 3a and a reading control unit 3b. Reference numeral 7 denotes an overlap unit for adding overlapping portions of the recording / reproducing waveform data and the analysis / synthesis waveform data, and reference numeral 8 denotes a synthesized voice output terminal for outputting a voice, such as a speaker, a headphone, and a receiver.

Next, the details of each processing will be described using actual examples. "There was a transfer from Mr. Ito of Matsushita Electric."
Consider the case where the sentence is input. Here, it is assumed that the only parameter having a difference between the two sounds is the pitch. According to the synthesizing means selection unit 2, the input sentence is divided into the proper noun "Matsushita Electric's Ito" and the part of "samakara" following it is recorded and played back, and the part other than the proper noun "there was a transfer from sama" is divided into analysis and synthesis. . The sentence sent to the recording / reproducing unit 3 is reproduced by the waveform data storage unit 3a and the reading control unit 3b. For the character string sent to the analysis / synthesis unit 4, parameters are created by the parameter storage unit 4a and the parameter control unit 4b. The parameter adjustment unit 5 refers to the recording / playback waveform data and adjusts the analysis / synthesis parameters so as to match the pitch. Then, the synthesis processing unit 6 generates the analysis / synthesis waveform data from the synthesis parameters. The analysis and synthesis waveform data and the recording / reproduction waveform data are added together by an overlap unit 7 to perform overlapping processing, and a sound is provided from a synthesized sound output terminal 8. In this way, the parameters of the analysis and synthesis of the parametric synthesis means are adjusted to match the recording and playback waveform data, and the recording and playback waveform data and the analysis and synthesis waveform data are overlapped and output, so that the connection It is possible to provide a sound with a more natural part and a less uncomfortable feeling. In this embodiment, as an example, “proper noun” is set to analysis synthesis, but this does not restrict the present invention at all. Further, although the parameter adjustment section sets only the parameter for adjusting the “pitch”, the parameter is not limited to the present invention at all, and all parameters used for the synthesis can be adjusted.

[0033] (Embodiment 7) FIG. 10 is a block diagram of an embodiment of the present invention described in claim 3. 3 shows details of an overlap portion according to claim 1. In the figure, reference numerals 1 and 2 denote synthesized sound input terminals A and
Reference numerals B and 3 denote overlapping sections each including a control section 3a, an amplifier A3b, and an amplifier B3c. Reference numeral 4 denotes a synthesized voice output terminal for outputting voice, which indicates a speaker, headphones, a receiver, and the like.

Next, the details of each process will be described using actual examples. It is assumed that the synthesized sound input terminal A and the synthesized sound input terminal B have overlapping portions. A case will be described in which sounds output by two different synthesizing means are output in an overlapping manner. When there is an input from the synthetic sound input terminal A and the synthetic sound input terminal B, the signals enter the amplifier A and the amplifier B, respectively. At this time, 3a
The control unit controls not to increase the volume of the overlapped output portion, fades out the audio output first, and fades in the audio output later, and gradually switches between the two audios. The two amplifiers are controlled so that the sound is transmitted, and the sound is provided from the synthetic sound output terminal 4. As described above, when two sounds in the overlapping portion are output in the overlap portion, the sound is gradually switched by controlling the respective amplifiers, thereby providing a sound with a more natural connection portion and less discomfort. be able to.

[0035]

As described above, according to the present invention, when speech is provided by combining a plurality of different synthesizing methods, optional functions such as overlap processing and parameter adjustment are provided at the joint of the speech synthesizing methods. With this arrangement, a more natural speech can be synthesized.

[Brief description of the drawings]

FIG. 1 is a block diagram of a speech synthesizer according to a first embodiment of the present invention.

FIG. 2 is a block diagram of a speech synthesizer according to the first embodiment of the present invention.

FIG. 3 is a block diagram of a speech synthesizer according to the first embodiment of the present invention.

FIG. 4 is an explanatory diagram of a reference example to help understanding of the present invention .

FIG. 5 is an explanatory diagram of a reference example to help understanding of the present invention .

FIG. 6 is an explanatory view of a reference example to help understanding of the present invention .

FIG. 7 is a block diagram of a speech synthesizer according to a third embodiment of the present invention.

FIG. 8 is a block diagram of a speech synthesizer according to a third embodiment of the present invention.

FIG. 9 is a block diagram of a speech synthesizer according to a third embodiment of the present invention.

FIG. 10 is a block diagram of a speech synthesizer according to a fourth embodiment of the present invention.

FIG. 11 is a block diagram of a conventional speech synthesizer.

[Explanation of symbols]

 Reference Signs List 1 sentence generation unit 2 synthesis means selection unit 3 rule synthesis unit 3a language processing unit 3b prosody control unit 3c parameter creation unit 3d synthesis processing unit 4 recording / playback unit 4a waveform data storage unit 4b read control unit 5 overlap processing unit 6 D / Part A 7 Synthetic sound output terminal

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-3-7999 (JP, A) JP-A-62-215299 (JP, A) JP-A-4-263299 (JP, A) JP-A-59-1984 42598 (JP, A) JP-A-4-367000 (JP, A) JP-A-60-63597 (JP, A) JP-A-4-19799 (JP, A) JP-A-1-191900 (JP, A) JP 3-73000 (JP, B2) JP 3-15759 (JP, B2) (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 11/00-13/08 G10L 19/00 -21/06 JICST file (JOIS)

Claims (3)

(57) [Claims] 1. A text generating means for generating a text, a plurality of different synthesizing means for synthesizing a speech waveform, a synthesizing means selecting means for selecting the synthesizing means according to the contents of the text, and a selecting means by the synthesizing means selecting means. And an overlap means for adding waveforms of a plurality of synthesized sounds output from the synthesized means. 2. A sentence generating means for generating a sentence, a plurality of different synthesizing means for synthesizing a speech waveform, a synthesizing means selecting means for selecting the synthesizing means according to the contents of the sentence, and one of the plurality of synthesized sounds. Is a parametric synthesizing means, a parameter adjusting means for adjusting the synthesizing parameters created by the synthesizing means in accordance with another voice, and an overlapping means for adding and processing the waveforms of the plurality of synthesized sounds. A speech synthesizer comprising: 3. A sentence generating means for generating a sentence, a plurality of different synthesizing means for synthesizing a speech waveform, a synthesizing means selecting means for selecting the synthesizing means according to the contents of the sentence, and an overlapping speech waveform. A voice synthesizing device comprising: overlap means for overlapping and outputting the respective amplifiers while adjusting the volume using control means when matching.