KR20050041749A - Voice synthesis apparatus depending on domain and speaker by using broadcasting voice data, method for forming voice synthesis database and voice synthesis service system - Google Patents

Abstract

The present invention provides a region and speaker-dependent speech synthesizer using broadcast speech data that can replace voice recording, which is relatively expensive, when a large-scale corpus-based speech synthesizer is implemented, and a database for speech synthesis used in such a speech synthesizer. The present invention relates to a method and a voice synthesis service system. In the present invention, instead of separate text design for speech synthesis database, speaker selection, and voice recording, a method for automating each database for speech synthesis using voice data recorded by a specific area and speaker is recorded. After that, we implement a speech synthesis device and speech synthesis service system using it. According to the present invention, it is easy to construct and expand a database for speech synthesis depending on the service area, and to improve the naturalness and affinity of the synthesized speech.

Description

VOICE SYNTHESIS APPARATUS DEPENDING ON DOMAIN AND SPEAKER BY USING BROADCASTING VOICE DATA, METHOD FOR FORMING VOICE SYNTHESIS DATABASE AND VOICE SYNTHESIS SERVICE SYSTEM }

The present invention relates to a region and speaker dependent speech synthesis apparatus using broadcast speech data, a method of constructing a database for speech synthesis used in such speech synthesis apparatus, and a speech synthesis service system. More specifically, the present invention relates to a corpus-based speech synthesis apparatus constructed by constructing a database for speech synthesis depending on a service area and a speaker according to a specific service field. In other words, a technology for selecting and recording a specific broadcast voice that is friendly to the general public according to a specific service field, and then using it to construct a database for speech synthesis by region and speaker, and to implement a speech synthesis apparatus using the database thus constructed. It is about.

Conventional corpus-based speech synthesizers are designed for speech synthesis for unlimited texts or rely on specialized service areas such as news, weather forecasts, traffic information, and stock information in each field to improve synthesized speech quality for specific service areas. The aim was to synthesize speech.

The speech synthesis method for unlimited text has the advantage of generating synthesized sounds for all the texts, but due to the physical limitations of the database size for speech synthesis, it is difficult to include all possible speeches. Therefore, there is a disadvantage in that the quality of the synthesized sound is insufficient for commercialization. Along with these problems, as the demand for the use of a speech synthesizer increases in a specific service area, a region-dependent speech synthesis method using a database for region-dependent speech synthesis optimized for the corresponding area has been proposed to improve the quality of the synthesized voice in a specific service area. As a related patent, Korean Patent Application No. 1998-51342 (filed date: November 27, 1998) has been filed a "speech synthesis method using a database for region dependent speech synthesis".

Such a conventional corpus-based speech synthesis method generally includes constructing a speech synthesis database and generating a synthesized sound by optimally combining necessary synthesis units using the constructed database. Among them, a process of building an existing database for speech synthesis is shown in FIG. 1.

As shown in FIG. 1, the process of constructing an existing database for speech synthesis includes selecting a speaker such as selecting a voice actor (S10), designing a text for synthesis (S11), and Performing utterance and recording with the synthesized text (S12), verifying whether there is a mismatch between the recorded content and the text content, correcting the text or performing re-recording (S13), and collecting through the steps (S10 to S12) And performing a phoneme and pitch labeling operation by synthesizing the synthesized voice data and the text (S14), and constructing a voice synthesis database (S15). As a prior patent related to the step of designing the text for synthesis in the process of building such a conventional speech synthesis database (S11), the patent application 2002-1353 (application date: January 10, 2002) "based on the corpus Method for selecting a recorded sentence for speech synthesis has been filed.

However, in the existing speech synthesis database construction method configured as described above, the speaker selection step (S10), the design step of the synthesis text (S11), the step of performing speech and recording with the designed synthesis text (S12), the recording content When the text content and the discrepancy are inconsistent, the step of correcting or re-recording the text (S13) is a process requiring a high cost and a long time. Therefore, the construction and expansion of the database is not easy, and due to the directed recording process in the step of recording the text for synthesis (S12), the naturalness tends to be reduced or processed compared to the voice of the actual situation. In addition, there is a difficulty in that the same voice actors should be recruited for additional recording and voices collected in a similar state as before.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-described technical problem, and is an area and speaker-dependent speech synthesis apparatus capable of facilitating the construction and expansion of a speech synthesis database and improving the affinity of synthesized speech, which is used in such speech synthesis apparatus. Its purpose is to provide a method for constructing a database for speech synthesis and a speech synthesis service system.

According to an aspect of the present invention, there is provided a region and speaker dependent speech synthesis apparatus, including: a broadcast speech recording unit configured to record broadcast speech data of a fixed speaker for each service region and to provide broadcast speech data for each region and a speaker; A plurality of database for speech synthesis in which the broadcasting voice data for each area and the speaker provided by the broadcasting voice recording unit are constructed as a database for each area and the speaker; A synthesizer configured to perform speech synthesis according to a predetermined speech synthesis algorithm using broadcast speech data stored in the respective speech synthesis databases; And a synthesized sound reproducing unit for reproducing the synthesized sound from the synthesizer.

According to an aspect of the present invention, there is provided a method for constructing a database for a region and speaker-dependent speech synthesis, the method comprising: collecting broadcast voice data for each service region and a speaker through a process of designating and recording a specific broadcast; Performing a text automatic transcription and verification process using a continuous speech recognizer to obtain text corresponding to the broadcast speech data collected in the first step; A third step of receiving phonetic voice data and the transcribed text collected in the step and performing phoneme labeling in a triphone unit using a speech recognizer and pitch labeling using a pitch extraction tool; And a fourth step of synthesizing the phoneme and pitch labeling results of the third step and the broadcast voice data of the first step and constructing a voice synthesis database depending on the area and the speaker.

According to an aspect of the present invention, there is provided a region and speaker-dependent speech synthesis service system comprising: a plurality of web contents each including news information, weather forecast information, and traffic information as a service region; A content update processor for updating information of each service area in real time; A plurality of database for speech synthesis storing service area and speaker-specific broadcast voice data; It is connected to the plurality of web contents and the content update processing unit through an internet network, and is connected to the plurality of voice synthesis databases, and uses the plurality of web contents provided by the content update processing unit to synthesize voices for the area and the speaker. And a voice synthesis server configured to perform a voice synthesis function using broadcast voice data stored in each database when a service request is made through a wired or wireless information terminal, and provide the result to the information terminal. It features.

Hereinafter, with reference to the accompanying drawings will be described in detail a preferred embodiment of the present invention.

First, a region / speaker dependent speech synthesis apparatus using broadcast speech data according to an embodiment of the present invention will be described with reference to FIG. 2. 2 is a block diagram of a region / speaker dependent speech synthesis apparatus using broadcast speech data according to an embodiment of the present invention.

As shown in FIG. 2, the speech synthesis apparatus according to the embodiment of the present invention includes a broadcast speech recording unit 20, a plurality of speech synthesis databases 21 to 24, a synthesizer 25, and a synthesized sound reproduction unit 26. ). The speech synthesis apparatus according to the embodiment of the present invention uses a fixed speaker's broadcast voice as synthesis speech data in place of a process of recording voice texts and pre-designed synthesis texts in the prior art. For example, a broadcast voice of a fixed speaker for each area, such as "Um Ki-young" as a speaker in a news area, "Ioung-seon" as a speaker in a weather forecast area, and "Hwang Jung-min" as a speaker in a traffic information area, is the broadcast voice recording unit 20. Recorded by). As described above, a plurality of zones / speaker-specific speech synthesis databases are recorded for the broadcast voices recorded by the broadcast voice recorder 20 using characteristics of a fixed speaker and a broadcast voice that can be classified into a specific service area. (21-24) are established. Next, speech synthesis is performed by the synthesizer 25 according to a predetermined speech synthesis algorithm by using the speech data stored in each speech synthesis database 21 to 24, and the final synthesized sound is synthesized by the synthesized sound reproducing unit 26. Is played.

This method is easier to construct and expand a database for speech synthesis than the conventional method, and at the same time, the broadcast voice represented by the general public and used for the synthesizer 25 is used in the synthesizer 25 so that the listener of the final synthesized speech The affinity of can be greatly improved. In addition, it is effective to improve the naturalness of the voice data and the synthesized sound by using the actual broadcast voice of the service area to be utilized, rather than using artificially designed text produced voice as in the conventional method.

Next, a method of constructing a speech synthesis database used in the speech synthesis apparatus will be described with reference to FIGS. 3 and 4. 3 is a flowchart illustrating a method for constructing a database for speech synthesis according to an embodiment of the present invention.

As shown in FIG. 3, when the operation of the voice synthesis database construction method starts, broadcast voice data for each region / speaker is collected through a process of designating and recording a specific broadcast (S30). Then, in order to obtain a text corresponding to the collected broadcast voice data, a text automatic transcription and verification step using a continuous speech recognizer is performed (S31). A more detailed process for the automatic text transfer and verification step S31 is shown in FIG. 4, which will be described later.

Next, the phoneme labeling of the triphone unit using a speech recognizer and the pitch labeling using a pitch extraction tool are performed using the inputted broadcast voice data and the transferred text as inputs (S32). In the final step, the phoneme and pitch labeling results and the broadcast voice data are synthesized to construct a database for region / speaker dependent speech synthesis (S33). Through the process of constructing the voice synthesis database, it is effective to simplify and automate the process of selecting and recording existing speakers, designing text for synthesis, and correcting errors, thereby facilitating the construction and expansion of the database for speech synthesis. .

FIG. 4 is a flow chart showing in more detail the text automatic transcription and verification step shown in FIG.

As shown in FIG. 4, when the automatic text transcription and verification step is started, the broadcasting voice data for each region / speaker collected in the upper step is input (S40), and the continuous voice recognition step is performed (S41). The recognizer calculates a textual recognition result and a recognition score value for each recognition unit such as a morpheme or a word. Next, it is determined whether the recognition score value is greater than or equal to the threshold value by comparing the calculated recognition score value with a threshold value (S42). When the value of the recognition score is greater than or equal to the threshold in the step S42, it is determined that the correct recognition is performed and the recognized text is used as the text corresponding to the broadcast voice (S43). If the recognition score is less than the threshold in step S42, the text is stored in the verification target list (S47), and the verification and error correction process for the text is performed (S48). When the error correction process is completed, this text is applied as the text corresponding to the broadcast voice. The threshold value used in step S42 may be arbitrarily determined by the user. The verification target list in step S47 is automatically generated by comparing the recognition score and the threshold. In the step S48, when the two contents are different by comparing the actual broadcasting voice data with the recognition result text with respect to the verification result list, a method of correcting the text is used.

Meanwhile, in order to improve the recognition rate of the continuous speech recognizer used in the present invention, a language model updating method and a speaker adaptation method are used. That is, in step S44, the text of each service area is automatically collected from the already recognized recognition result and the web. In the next step S45, a language model is extracted from the collected text and provided to the speech recognizer. . Therefore, in step S41, the language model may be automatically updated in the speech recognizer. In addition, in the speaker adaptation step S46, the broadcast voice data input in the step S40 and the transcribed text of the step S43 are input to generate an acoustic model adapted to the speaker, and then apply the same to the recognizer. . The above two methods are applied repeatedly each time the text auto-transcription and verification step of FIG. 2 is performed. Through this, speech recognition optimized for the corresponding region and the speaker may be performed.

5 shows a configuration of a speech synthesis service system using a speech synthesis database according to an embodiment of the present invention.

The database for speech synthesis according to the embodiment of the present invention described above is advantageous in the improved nature and affinity of the synthesized sound with ease of database construction, and can be used in the speech synthesis service system as shown in FIG. have.

The voice synthesis service system shown in FIG. 5 includes a plurality of web contents 52 such as news information, weather forecast information, traffic information, and the like, a content update processing unit 53, and a plurality of area / speaker dependent speech synthesis databases ( 50) and an area / speaker dependency that is connected to the plurality of web contents 52 and the content update processing unit 53 through the Internet network and is connected to the plurality of speech synthesis databases 50 to perform a speech synthesis function. The speech synthesis server 51 is configured. In addition, the area / speaker dependent speech synthesis server 51 may be connected to an information terminal 54 such as a personal computer through an internet network or to a mobile information terminal 55 such as a PDA or a wireless telephone through a wireless network. .

The speech synthesis server 51 constructs a speech synthesis database 50 for each region / speaker using a plurality of web contents provided by a content update processing unit through an internet network, and an information terminal through an internet network or a wireless network. When the voice synthesis sound service request is made by (54, 55), the synthesis sound is generated from the plurality of region / speaker-specific speech synthesis databases 50 and provided to the corresponding information terminal. As described above, in the present invention, since the area / speaker dependent speech synthesis database 50 is constructed from the service field, it is possible to provide a user with a high-quality synthesized sound with high affinity. In particular, in the case of a mobile information terminal 55 such as a PDA and a wireless telephone, voice service may be more actively utilized due to the characteristics of the device. In this case, it is possible to improve the quality of the voice service and the user's satisfaction through familiar voices generated by dividing by service field, and may be applied as a service to compensate for the weakness of broadcasting which has a limitation in real time information transmission.

When using the apparatus and method proposed in the present invention, it is possible to construct a region-dependent synthesizer capable of generating high-quality synthesized sound for a specific area, and at the same time, general public by specific service fields such as news, weather forecast, traffic information, and stock information. The affinity of the synthesized sound can be greatly improved by using the speaker's voice which is widely known. In addition, it is possible to implement a speech synthesis apparatus with improved naturalness by using the voice of the actual service situation in the corresponding service area. In addition, when building a database for speech synthesis, the process of selecting a separate speaker, designing the text for synthesis, uttering and recording the designed synthesis text, and correcting and re-recording the text in the case of inconsistency between the recording content and the text content are simplified or It is replaced by an automated method, and it is very easy to build and expand a large-scale voice database for a single speaker in a specific area due to the nature of the broadcast, such as a regular broadcasting time and a fixed speaker in a specific area. Based on the above characteristics of the present invention, it is possible to more easily implement a region / speaker dependent speech synthesis apparatus in which the naturalness and affinity of the synthesized speech is increased as compared with the conventional speech synthesis method.

What has been described above is only one embodiment for implementing a region / speaker dependent speech synthesis apparatus using broadcast speech data, a method for constructing a speech synthesis database, and a speech synthesis service system according to the present invention. The spirit of the present invention is not limited to the examples, and any person having ordinary knowledge in the field to which the present invention pertains can make various changes without departing from the gist of the present invention as claimed in the following claims. I would say this is crazy.

1 is a flowchart illustrating a process of constructing an existing database for speech synthesis.

2 is a diagram illustrating a configuration of a region / speaker dependent speech synthesis apparatus using broadcast speech data according to an embodiment of the present invention.

3 is a flowchart illustrating a method for constructing a database for speech synthesis according to an embodiment of the present invention.

FIG. 4 is a flow chart illustrating in more detail the text auto-transcription and verification steps shown in FIG.

5 is a diagram illustrating a configuration of a speech synthesis service system using a speech synthesis database according to an embodiment of the present invention.

<Description of the symbols for the main parts of the drawings>

20: broadcast voice recording unit 21 ~ 24: database for synthesis by area / speaker

25: synthesizer 26: synthesized sound playback unit

Claims (8)

A broadcast voice recorder configured to record broadcast voice data of a fixed speaker for each service area and provide broadcast voice data for each speaker; A plurality of database for speech synthesis in which the broadcasting voice data for each area and the speaker provided by the broadcasting voice recording unit are constructed as a database for each area and the speaker; A synthesizer configured to perform speech synthesis according to a predetermined speech synthesis algorithm using broadcast speech data stored in the respective speech synthesis databases; And And a synthesized sound reproducing unit configured to reproduce the synthesized sound from the synthesizer. Region and speaker dependent speech synthesis device. The method of claim 1, The speech synthesis database includes a news area, a weather forecast area, and a traffic information area, and broadcast voice data of a representative speaker in each service area is used. Region and speaker dependent speech synthesis device. A first step of collecting broadcast voice data for each service area and a speaker through a process of designating and recording a specific broadcast; Performing a text automatic transcription and verification process using a continuous speech recognizer to obtain text corresponding to the broadcast speech data collected in the first step; A third step of receiving phonetic voice data and the transcribed text collected in the step and performing phoneme labeling in a triphone unit using a speech recognizer and pitch labeling using a pitch extraction tool; And And a fourth step of synthesizing the phoneme and pitch labeling results of the third step and the broadcast voice data of the first step to form a database for speech synthesis that is dependent on area and speaker. How to build a database for domain and speaker dependent speech synthesis. The method of claim 3, The third step is Receiving the broadcast voice data of each region and the speaker and performing a continuous speech recognition process, and calculating a recognition result and a recognition score in a text form for each recognition unit such as morpheme and word; And If the calculated recognition score is equal to or greater than a predetermined threshold, it is determined that correct recognition has been performed, and the recognized text is applied as text corresponding to the broadcast voice. Storing the verification target list and performing a verification and error correction process on the text; How to build a database for domain and speaker dependent speech synthesis. The method of claim 4, wherein The verification target list is automatically generated by comparing the recognition score with a threshold. How to build a database for domain and speaker dependent speech synthesis. The method of claim 3, Automatically collecting the recognition result and the text of each service area from the web, and extracting the language model from the collected text and applying to the continuous speech recognizer of the second step How to build a database for domain and speaker dependent speech synthesis. The method of claim 3, And receiving the input broadcast voice data and the transcribed text and generating an acoustic model adapted to the speaker, and applying the same to the continuous speech recognizer of the second step. How to build a database for domain and speaker dependent speech synthesis. A plurality of web contents each including news information, weather forecast information, and traffic information as a service area; A content update processor for updating information of each service area in real time; A plurality of database for speech synthesis storing service area and speaker-specific broadcast voice data; It is connected to the plurality of web contents and the content update processing unit through an internet network, and is connected to the plurality of voice synthesis databases, and uses the plurality of web contents provided by the content update processing unit to synthesize voices for the area and the speaker. And a voice synthesis server configured to perform a voice synthesis function using broadcast voice data stored in each database when a service request is made through a wired or wireless information terminal, and provide the result to the information terminal. Zone- and Speaker-dependent Speech Synthesis Service System.