JP2000250572A - Device and method for preparing voice database, device and method for preparing singing voice database - Google Patents

Abstract

PROBLEM TO BE SOLVED: To correctly and efficiently record a voice of required text, accent intonation, tempo and voice quality or a singing voice of required text, melody and voice quality. SOLUTION: The text, accent intonation, temp and voice quality corresponding to the required voice are inputted from a text input editor 1, a rhythm design editor 3 and a voice quality design editor 6, and a required synthesized voice is made by a reading giving part 2, a rhythm parameter generation part 4, a voice synthetic part 5 and a voice quality conversion part 7 to be outputted from a speaker 8, and that is let an utterer hear, and is let the same person imitate, and is let the same utter to be recorded from a micro-phone 9. Further, that is listened on the spot, and when that is the intended voice, by operating a switch 10 to be registered in a database 12.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech sample used for learning of a speech recognition device, a speech unit used for a waveform connection type speech synthesis device, a response speech used for a voice response device, animation / movie / multimedia contents. Like a voice or a singing voice for creation, the creator images and desires in a database operation of a voice or a singing voice in which utterance content, accent intonation, tempo, voice quality, etc. or lyrics, melody, voice quality, etc. are predetermined. The present invention relates to a voice database creation device and method for constructing a voice or singing voice database by accurately and efficiently recording voice or singing voice, and a singing voice database creation device and method.

[0002]

2. Description of the Related Art A conventional speech database creating apparatus is composed of a microphone for recording a voice uttered by a speaker and a recording medium for recording the voice, and the uttered content is notified to the speaker by a memo. , Prosody, kanji reading, voice quality, etc. were communicated by the database creator explaining to the speaker.

Further, the conventional singing voice database creating apparatus is composed of a microphone for recording the singing voice and a recording medium for recording the singing voice. , Voice quality, etc. were communicated by the database creator explaining to the singer.

[0004]

However, in such a conventional voice database creating apparatus, a memo describing the content of the voice to be registered is handed over to the speaker, and a database is prepared for reading, accent / intonation, tempo, voice quality, and the like. After the creator spends a lot of time explaining, the speaker speaks according to it, so much labor, manpower and time are required, and the intonation, tempo, voice quality etc. are explained to the speaker. And it is very difficult for the database creator to collect the desired voice.

In the conventional singing voice database creating apparatus, a memo describing the lyrics and melody of the singing voice to be registered is given to the singer, and the creator of the database spends a great deal of time explaining the voice quality and the like, and then the singer Singing in accordance with it requires a lot of trouble, manpower and time, and it is difficult to explain the voice quality etc. to the speaker, and it is very difficult for the database creator to collect the desired singing voice. there were.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a speech database creation apparatus and method capable of accurately and efficiently recording text, accent intonation, tempo, and voice quality desired by a database creator and registering the speech in a speech database. It is another object of the present invention to provide a singing voice database creating apparatus and method capable of accurately and efficiently recording singing voices having desired lyrics, melody, and voice quality and registering the singing voices in a singing voice database.

[0007]

In order to achieve the above object, a speech database creating apparatus according to the present invention comprises: a text input unit for inputting a text corresponding to a speech desired by a speech database creator; It has a prosody input unit for inputting, a voice quality input unit for inputting information indicating desired voice quality, and a voice synthesis unit for synthesizing voice exactly according to parameters input from those input units.

In an actual recording site, immediately after the speaker hears the synthesized voice, the speaker imitates the synthesized voice,
Text, accent,
A sound sample set of intonation, tempo, and voice quality can be accurately and efficiently recorded and registered in a sound database.

Further, the singing voice database creation device of the present invention indicates a lyric input unit for inputting lyrics corresponding to a singing voice desired by the singing voice database creator, a melody input unit for inputting a desired melody, and a desired voice quality. A voice quality input unit for inputting information; and a singing voice synthesizing unit for synthesizing a singing voice exactly according to the parameters input from the input units.

In an actual recording site, immediately after the singer hears the synthesized singing voice, the singer imitates the singing voice,
A singing voice sample set of the lyrics, melody, and voice quality desired by the database creator can be accurately and efficiently recorded and registered in the singing voice database.

[0011]

Next, the present invention will be described with reference to the drawings.

FIG. 1 shows an example of an embodiment of a speech database creating apparatus according to the present invention. In the figure, 1 is a text input editor, 2 is a reading addition section, 3 is a prosody design editor, and 4 is a prosody parameter generation. Section, 5 is a speech synthesis section, 6
Is a voice quality design editor, 7 is a voice quality conversion unit, 8 is a speaker, 9 is a microphone, 10 is a switch, 11 is an instruction display monitor, and 12 is a storage device.

The text input editor 1 is for inputting text corresponding to a voice desired by a voice database creator. The reading giving unit 2 analyzes the pronunciation (reading kana) of the input text, associates the analyzed pronunciation with the text (gives the reading kana), and outputs the text.

The prosody design editor 3 designs (edits) and inputs a desired prosody pattern such as desired accent intonation and tempo. The prosody parameter generation unit 4 generates a prosody parameter of the voice from the input prosody pattern. The speech synthesizer 5 synthesizes speech based on the text to which the reading kana is added and the prosodic parameters.

The voice quality design editor 6 designs (edits) information indicating a desired voice quality and inputs the information. The voice quality conversion unit 7 converts the synthesized voice into a desired voice quality according to the information indicating the voice quality.

The speaker 8 outputs a synthesized voice obtained by converting the voice quality. The microphone 9 records the voice uttered by the speaker while imitating the synthesized voice output from the speaker 8.

The switch 10 inputs the result of determining whether the voice database creator has listened to the voice recorded from the microphone 9 and uttered as intended by the creator sufficiently to register the voice in the database. It is for. The instruction display monitor 11 notifies the output of the switch 10 to the speaker.

The storage device 12 holds a voice database, and simultaneously registers the voice recorded from the microphone 9 and the corresponding text (input from the text input editor 1) in this database.

Next, how to create a voice database using the above-described apparatus will be described together with its operation.

Before the recording operation, the database creator
A text corresponding to a desired voice is input using a text input editor 1 incorporated in the present voice database creating apparatus, and a prosody pattern such as accent / intonation and tempo of the text is desired using a prosody design editor 3. Design and input.

As a prosody design editor, for example,
Mizuno, Abe, Nakashima, "Speech Production Tool with Improved Speech Expression and Production Efficiency -Signing98-" Proceedings of the Fall Meeting of the Acoustical Society of Japan 1998, 2-P-1
2, pp. 309-310, 1998.10 and the like.

Further, the database creator uses the voice quality design editor 6 to design and input information indicating a desired voice quality.

The input text is provided with a reading by a reading providing unit 2, and the input prosody pattern is converted into a prosody parameter of a voice by a prosody parameter generation unit 4. 5 is converted into a synthesized speech, and further converted by a voice quality conversion unit 7 into a voice quality according to the information indicating the input voice quality.

As the voice conversion unit (device), for example, Abe, "Voice conversion device (VarioVoi) by waveform processing
ce) "The Acoustical Society of Japan 1997 Spring Research Conference Proceedings, 3-7-6, pp. 269-270, 1997.
3 and the like.

Through the above processing, the creator of the speech database can obtain a synthesized speech having the same acoustic properties as the image and the desired speech. The voice database creator creates such a synthesized voice in advance.

At the recording site, the synthesized voice is output from the speaker 8 to be heard by the speaker in order, and the speaker is made to imitate the synthesized voice and uttered.

A recording operator (which may or may not be the same as the database creator) listens to the recorded voice, and is acoustically close to the immediately preceding synthesized voice and sufficiently matches the intention of the database creator. If it is determined that the switch is operating, the switch 10 is operated. Thus, the voice recorded from the microphone 9 is stored in the storage device 12 together with the input text.
(Voice database).

The operation output of the switch 10 is displayed on the instruction display monitor 11 so that the speaker can immediately know whether or not the utterance is sufficient.

FIG. 2 shows an example of an embodiment of a singing voice database creation apparatus according to the present invention. In the figure, reference numeral 21 denotes a lyric input editor, 22 denotes a reading addition section, 23 denotes a melody design editor, and 24 denotes a melody parameter generation. Section, 25 is a singing voice synthesis section, 26 is a voice quality design editor, 27 is a voice quality conversion section,
28 is a speaker, 29 is a microphone, 30 is a switch, 31
Is an instruction display monitor, and 32 is a storage device.

The lyrics input editor 21 is for inputting the lyrics corresponding to the desired singing voice by the creator of the singing voice database. The reading giving unit 22 analyzes the pronunciation (reading kana) of the input lyrics, associates the analyzed pronunciation with the lyrics (adds reading kana), and outputs the lyrics.

The melody design editor 23 designs (edits) and inputs a desired melody pattern. The melody parameter generation unit 24 generates a melody parameter of the singing voice from the input melody pattern. The singing voice synthesizing unit 25 synthesizes a singing voice based on the lyrics to which the reading kana is added and the melody parameter.

The voice quality design editor 26 designs (edits) and inputs information indicating a desired voice quality. The voice quality conversion unit 27 converts the synthesized singing voice into a desired voice quality according to the information indicating the voice quality.

The speaker 28 outputs a synthesized singing voice whose voice quality has been converted. The microphone 29 records the singing voice sung by the singer by imitating the synthesized singing voice output from the speaker 28.

The switch 30 inputs the result of the singing voice database creator listening to the singing voice recorded from the microphone 29 and determining whether or not the singing voice is sung as intended by the creator sufficiently to register it in the database. It is for. The instruction display monitor 31 notifies the singer of the output of the switch 30.

The storage device 32 holds a singing voice database, and simultaneously registers the singing voice recorded from the microphone 29 and the corresponding lyrics (input from the lyrics input editor 21) in this database.

Next, the operation of creating a singing voice database using the above-described apparatus will be described along with its operation.

Before the recording operation, the database creator
The lyrics corresponding to the desired singing voice are input using the lyrics input editor 21 incorporated in the present singing voice database creation device, and the melody pattern of the lyrics is designed and input as desired using the melody design editor 23. .

As a melody design editor, for example,
Mizuno, Abe, Nakashima, "Speech Production Tool with Improved Speech Expression and Production Efficiency -Signing98-" Proceedings of the Fall Meeting of the Acoustical Society of Japan 1998, 2-P-1
2, pp. 309-310, 1998.10 and the like.

Further, the database creator uses the voice quality design editor 26 to design and input information indicating the desired voice quality.

The input lyric is given a reading by a reading giving unit 22, and the inputted melody pattern is converted into a melody parameter of a singing voice by a melody parameter generating unit 24. 25, the singing voice is converted into a synthesized singing voice, and further converted into a voice quality according to the input information indicating the voice quality by a voice quality converting unit 27.

As the voice conversion unit (device), for example, Abe, “Voice conversion device (VarioVoi) by waveform processing
ce) "The Acoustical Society of Japan 1997 Spring Research Conference Proceedings, 3-7-6, pp. 269-270, 1997.
3 and the like.

By the above processing, the singing voice database creator can obtain a synthesized singing voice having the same acoustic properties as the image and the desired singing voice. The singing voice database creator creates such a synthetic singing voice in advance.

At the recording site, the synthesized singing voice is output from the speaker 28 and is sequentially heard by the singer, and the singing person is caused to imitate the synthesized singing voice and sing the singing voice.

A recording operator (which may or may not be the same as the database creator) listens to the recorded singing voice, and is acoustically close to the immediately preceding synthesized singing voice and sufficiently matches the intention of the database creator. If the switch 30 is determined to be operating, the switch 30 is operated. Thereby, the singing voice recorded from the microphone 29 is stored in the storage device 32 together with the inputted lyrics.
(Singing voice database).

The operation output of the switch 30 is displayed on the instruction display monitor 31, so that the singer can immediately know whether or not the singing is sufficient.

[0046]

As described above, according to the present invention,
Speech samples uttered with accent intonation, tempo, voice quality that the database creator imagines,
Alternatively, a singing voice sample sung with desired lyrics, melody, and voice quality can be recorded accurately and in a short time as desired by a database creator to construct a voice database or a singing voice database. Furthermore, as long as the database creator only creates a synthetic voice or a synthetic singing voice, the agent can also perform the recording operation without the creator himself present at the recording site.

The speaker or singer can immediately know whether his or her utterance or singing has sufficiently met the intention of the database creator, and therefore the accuracy of the utterance or singing can be improved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an example of an embodiment of a voice database creation device of the present invention.

FIG. 2 is a block diagram showing an example of an embodiment of a singing voice database creation device of the present invention.

[Explanation of symbols]

1: text input editor, 2, 22 :: reading addition unit,
3: prosody design editor, 4: prosody parameter generation unit, 5: speech synthesis unit, 6, 26: voice design editor, 7, 27: voice conversion unit, 8, 28: speaker, 9,
29: microphone, 10, 30: switch, 11, 31: instruction display monitor, 12, 32: storage device, 21: lyrics input editor, 23: melody design editor, 24: melody parameter generator, 25: singing voice synthesizer.

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Masanobu Abe 3-19-2 Nishi-Shinjuku, Shinjuku-ku, Tokyo Japan Telegraph and Telephone Corporation (72) Inventor Shinya Nakashima 3-192-2, Nishi-Shinjuku, Shinjuku-ku, Tokyo No. Nippon Telegraph and Telephone Corporation F-term (reference) 5D045 AA07 AA08 AA09

Claims (6)

[Claims] 1. A speech database creation device for creating a speech database consisting of text and corresponding speech data by recording speech uttered by a human being, wherein: a text input unit for inputting a text corresponding to the speech; A reading provision unit that analyzes the reading kana of the text, and adds a reading kana to the text, a prosody input unit that inputs a desired prosody pattern, and a prosody parameter generation unit that generates a prosodic parameter of speech from the prosody pattern. A voice synthesis unit that synthesizes a voice based on the text to which the reading kana is given and the prosodic parameter; a voice quality input unit that inputs information indicating a desired voice quality; and the synthesized voice indicates the voice quality. A voice conversion unit that converts the voice into a desired voice according to the information; a voice reproduction unit that reproduces the synthesized voice after the voice conversion; A voice recording unit that records a voice of a speaker who imitates the recorded and reproduced synthesized voice, a determination input unit that inputs a determination result as to whether to register the recorded voice in the voice database, and the determination result. And an instruction display unit that displays the recorded audio and the text corresponding thereto are registered in the audio database in association with each other. 2. The apparatus according to claim 1, wherein said prosody input section and said voice quality input section each have means for editing information indicating a prosody pattern and voice quality. 3. A speech database creation method for creating a speech database comprising text and speech data corresponding to the text by recording a speech uttered by a human, comprising: inputting a text corresponding to the speech; Analyzing the reading kana, assigning the reading kana to the text, inputting a desired prosody pattern, generating a prosodic parameter of speech from the prosody pattern, based on the text to which the reading kana is added and the prosody parameter Synthesizing a voice, inputting information indicating a desired voice quality, converting the synthesized voice to a desired voice quality according to the information indicating the voice quality, reproducing the synthesized voice after the voice quality conversion, the reproducing synthesized Record the voice of the speaker who imitated the voice, and determine whether to register the recorded voice in the voice database. Speech database creation method and registers the voice database in association with the text and the corresponding sound is the recording. 4. A singing voice database creating apparatus for creating a singing voice database composed of lyrics and corresponding singing voice data by recording a singing voice uttered by a human, a lyrics input section for inputting lyrics corresponding to the singing voice, A reading assignment unit that analyzes the reading kana of the given lyrics and adds a reading kana to the lyrics, a melody input unit that inputs a desired melody, a melody parameter generation unit that generates a melody parameter of a singing voice from the melody, A singing voice synthesizing unit that synthesizes a singing voice based on the lyrics to which the reading kana is assigned and the melody parameter; a voice quality inputting unit that inputs information indicating a desired voice quality; and the synthesized singing voice according to the information indicating the voice quality. A voice quality conversion unit for converting to a desired voice quality, a singing voice reproducing unit for reproducing the synthesized singing voice after the voice quality conversion, and a singer imitating the reproduced synthesized singing voice A singing voice recording unit that records a singing voice, a determination input unit that inputs a determination result as to whether or not to register the recorded singing voice in the singing voice database, and an instruction display unit that displays the determination result, A singing voice database creating apparatus, wherein a recorded singing voice is associated with lyrics corresponding thereto and registered in the singing voice database. 5. The singing voice database creating apparatus according to claim 4, wherein said melody input section and said voice quality input section each have means for editing information indicating the melody and voice quality. 6. A singing voice database creating method for creating a singing voice database comprising lyrics and singing voice data corresponding to the singing voice uttered by a human being, wherein the singing voice corresponding to the singing voice is input, and Analyzing the reading kana, giving the reading kana to the lyrics, inputting a desired melody, generating a melody parameter of the singing voice from the melody, and singing the singing voice based on the lyric to which the reading kana is given and the melody parameter. Synthesizing, inputting information indicating a desired voice quality, converting the synthesized singing voice into a desired voice quality according to the information indicating the voice quality, reproducing the synthesized singing voice after the voice quality conversion; The singing voice of the imitated singer is recorded, and it is determined whether or not the recorded singing voice is registered in the singing voice database, and the recorded singing voice and the corresponding singing voice are recorded. Vocal database creation method characterized by the lyrics in association registered in the voice database.