JP5062178B2 - Audio recording system, audio recording method, and recording processing program - Google Patents

Description

  The present invention relates to an audio recording system and the like, and more particularly to an audio recording system, an audio recording method, and a recording processing program suitable for constructing a natural speech synthesis system adapted to each specialized field.

The speech synthesis technology is roughly divided into two methods: a waveform editing method and a waveform connection method. In either method, the quality of the synthesized speech to be generated is greatly influenced by the quality of the speech data stored in the speech synthesis database.
In particular, in the case of a technology that converts text to speech (TTS: Text-To-Speech), it is a natural performance that any text can be converted into speech. It is requested. In recent years, speech synthesis systems with high naturalness have appeared due to the large scale of databases.

By the way, in order to realize high naturalness, a segment waveform in which a speech signal of the same type as the synthesized speech to be generated and parameters such as a pitch pattern and a time length exist (store) in the speech synthesis database in advance. Must be.
For example, if a speech signal read out from a newspaper article is converted into a speech synthesis database, it can achieve very high naturalness if the text is similar to a newspaper article. Then, since similar data does not exist in the speech synthesis database, naturalness is significantly deteriorated. If the database for speech synthesis is further scaled up to support colloquial style, the naturalness will be improved, but a huge amount of data will be required.

  On the other hand, since computer resources are limited, it is practically impossible to create a database for speech synthesis that can handle any input text data and that all generated synthesized speech is of high quality (eg, non-patented). Reference 1).

  Many database creation methods for efficiently creating a database for speech synthesis have been filed. For example, there is known a method for creating a speech segment database that can generate high-quality synthesized speech for a textual expression that is linguistically and acoustically important (see, for example, Patent Document 1).

According to the technique disclosed in Patent Document 1 described above, prosodic feature quantities such as phoneme series, pitch pattern, tempo, and pause are obtained from each text in text data by morphological analysis processing and prosodic estimation of each text. And the acoustic importance of each text is obtained from the phoneme sequence and the prosodic feature quantity, and a recording list is created from sentences determined to be linguistically and acoustically important. Then, the sentence described in the recording list is recorded as speech, and a speech synthesis database is created by assigning phonological labels.
Sadaaki Furui “Digital Audio Processing”, Tokai University Press, 1985, pages 134-148 JP 2004-138661 A

However, according to the method for creating a speech synthesis database disclosed in Patent Document 1 described above, morphological analysis is performed on text in a dextst database to estimate prosodic features. There is a problem of being dependent.
For example, for a user who uses a speech synthesis system for emergency broadcast applications, a speech synthesis system that can speak clearly at a relatively high speech speed is preferable. However, with the technique disclosed in Patent Document 1, it is not possible to create a speech synthesis database that takes into account the conditions (hereinafter referred to as “tasks”) to be achieved by the speech synthesis system actually used by the user. In addition, there are no assumptions about how to improve the existing speech synthesis system, so it should only be used for the purpose of creating a speech synthesis database from scratch, recording all the sentences listed in the recording list above. There was an inconvenience that it was not possible.

[Object of invention]
The present invention has been made in view of the circumstances described above, and a speech recording system capable of forming a speech synthesis system capable of generating a highly natural synthesized speech corresponding to many requests from a user, and its The object is to provide a method and program.

  In order to achieve the above object, a speech recording system according to the present invention is a speech recording system that records speech signals necessary for creating speech data such as parameters and segment waveforms in a speech synthesis database. A speech synthesizer for generating desired synthesized speech based on speech data preliminarily stored in the speech synthesis database by inputting text data describing the generated content, and based on the feature amount of the synthesized speech And a quality evaluation unit for evaluating the quality of the synthesized speech. Further, the voice recording system has a configuration in which an additional data determination unit that determines voice data to be additionally stored in the voice synthesis database based on the quality evaluation result by the quality evaluation unit is provided.

  Therefore, the quality evaluation unit performs quality evaluation based on the feature amount of the synthesized speech generated by the speech synthesis unit, and the additional data determination unit performs speech data to be additionally stored in the speech synthesis database based on the quality evaluation result. Therefore, the quality of the created synthesized speech can be evaluated even for an existing speech synthesis system, and speech data to be added when updating the existing speech synthesis system based on the evaluation result Can be easily determined.

  In order to achieve the above object, in a speech recording system according to the present invention, a speech synthesis database for storing speech data, processing for synthesizing speech data stored in the speech synthesis database and text data inputted externally Analyzing the speech synthesis system information that defines the operation, and based on this, a quality evaluation unit that determines whether the speech data stored in the speech synthesis database satisfies a desired quality; An additional data determining unit that determines speech data to be additionally stored in the speech synthesis database based on the result is employed.

  For this reason, the quality evaluation unit defines the operation of speech synthesis processing stored in the speech synthesis database, for example, by analyzing and evaluating speech synthesis system information such as morphological analysis models and prosodic information, Easy and high-speed speech data to be stored when updating an existing speech synthesis system from acquired input text data and speech synthesis system information of an existing speech synthesis system Can be determined.

  In order to achieve the above object, a speech recording method according to the present invention is a speech recording method for recording a speech signal in order to create speech data stored in a speech synthesis database, wherein the speech synthesis database is stored in the speech synthesis database. A first step of referring to and generating desired synthesized speech from externally input text data; a second step of evaluating the quality of the synthesized speech based on a feature amount of the synthesized speech; and a result of the quality evaluation And a third step of determining speech data to be additionally stored in the speech synthesis database.

For this reason, the voice recording system executes the second step of performing the quality evaluation based on the feature amount of the synthesized voice generated by executing the first step, and stores the voice synthesis database in the voice synthesis database based on the quality evaluation result. By executing the third step of determining speech data to be additionally stored, the quality of synthesized speech created using an existing speech synthesis system can be evaluated, and an existing speech synthesis system is based on the evaluation result. It is possible to easily determine audio data to be added when updating.

  In order to achieve the above object, a voice recording method according to the present invention is a voice recording method for recording a voice signal in order to create voice data stored in a voice synthesis database, and is input externally. Analyzing and evaluating the first step of acquiring text data, and the input text data thus acquired and the speech synthesis system information that defines the processing operation related to the synthesis of speech data stored in the speech synthesis database. Based on this, a second step of determining whether or not the speech data stored in the speech synthesis database satisfies a desired quality, and additionally storing in the speech synthesis database based on the quality evaluation result And a third step of determining audio data to be performed.

  For this reason, the speech recording system defines the operation of speech synthesis processing stored in the speech synthesis database, for example, a second step of analyzing and evaluating speech synthesis system information such as a morphological analysis model and prosodic information. By executing, the voice data to be additionally stored when the existing voice synthesis system is updated from the acquired input text data and the voice synthesis system information of the existing voice synthesis system without performing the synthesis voice conversion process Can be determined easily and at high speed.

  In order to achieve the above object, the audio recording processing program according to the present invention is a program used in an audio recording system for recording an audio signal to create audio data stored in a database for speech synthesis, A speech synthesis processing function for generating a desired synthesized speech from input text data with reference to a speech synthesis database, a quality evaluation processing function for evaluating the quality of the synthesized speech based on a feature amount of the synthesized speech, The computer is caused to execute an additional data determination processing function for determining speech data to be additionally stored in the speech synthesis database based on the result.

  For this reason, the computer performs quality evaluation based on the feature amount of the synthesized speech, determines speech data to be additionally stored in the speech synthesis database based on the quality evaluation result, and is created using an existing speech synthesis system The quality of synthesized speech can be evaluated, and speech data to be added when updating an existing speech synthesis system can be easily determined based on the evaluation result.

  In order to achieve the above object, the audio recording processing program according to the present invention is a program used for an audio recording system for recording an audio signal in order to create audio data stored in a speech synthesis database. Text data acquisition processing function for acquiring externally input text data, analysis and evaluation of the acquired input text data and speech synthesis system information that defines the speech synthesis processing operation stored in the speech synthesis database And a quality evaluation processing function for determining whether or not the speech data stored in the speech synthesis database satisfies a desired quality based on this, and additionally storing in the speech synthesis database based on the quality evaluation result Causing the computer to execute an additional data determination processing function for determining audio data to be Characterized in that way the.

  For this reason, the computer executes a quality evaluation processing function that analyzes and evaluates speech synthesis system information such as a morphological analysis model and prosodic information, which prescribes the operation of speech synthesis processing stored in the speech synthesis database. This makes it easy to store voice data that should be additionally stored when updating an existing speech synthesis system from the acquired input text data and the speech synthesis system information of the existing speech synthesis system, without going through synthesized speech processing. And can be determined at high speed.

  According to the present invention, the speech synthesis system can be adapted to the task desired by the user and the quality thereof can be improved. Moreover, the present invention can be applied to an existing speech synthesis system and is not required in the speech synthesis database. It is also possible to delete data, thereby providing an unprecedented superior voice recording system, voice recording method, and recording processing program that can prevent the enlargement of the speech synthesis database.

[First Embodiment]
DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, a first embodiment of the invention will be described with reference to the accompanying drawings.
First, as shown in FIG. 1, the audio recording system according to the embodiment is a system that records audio signals necessary for generating audio data such as parameters and segment waveforms in the speech synthesis database 10A. A speech synthesizer 11 for generating desired synthesized speech based on speech data preliminarily stored in the speech synthesis database 10A by inputting text data describing the generated content from the outside; and a feature amount of the synthesized speech And a quality evaluation unit 12 for performing quality evaluation of the synthesized speech based on the above.

  In this voice recording system, an additional data determination unit 13 for determining voice data to be additionally stored in the voice synthesis database 10A based on the result of quality evaluation by the quality evaluation unit 12, and text generation as will be described later. And a recording instruction data generation unit 14 that generates recording instruction data by referring to the generation text data stored in the text generation storage unit 14A. Here, in the present embodiment, the speech synthesis system 10 is configured by the speech synthesis database 10 </ b> A and the speech synthesis unit 11.

  The voice synthesis database 10A stores voice data created based on recorded voice signals. This voice data is composed of data including, for example, various voice parameters and voice segments.

  Here, the above-described speech synthesizer 11 refers to the speech synthesis database 10A, generates desired synthesized speech (intermediate data) from externally input text data (input text data), and outputs it to the quality evaluation unit 12 To do. The quality evaluation unit 12 evaluates the quality of the synthesized speech based on the feature amount of the synthesized speech (intermediate data), and outputs it to the additional data determination unit 13. Further, the additional data determination unit 13 determines voice data to be additionally stored in the speech synthesis database 10A based on the quality evaluation result by the quality evaluation unit 12, and outputs the voice data to the recording instruction data generation unit 14. The recording instruction data generation unit 14 generates recording instruction data including text data describing the utterance content of the audio signal to be recorded corresponding to the audio data determined by the additional data determination unit 13.

Here, the above-described quality evaluation unit 12 analyzes and evaluates intermediate data (synthesized speech) generated by the speech synthesis unit 11 when speech synthesis is performed on the externally input text data, and based on this, the speech is analyzed. A function is provided for determining whether or not the voice data stored in the synthesis database 10A satisfies a desired quality.
For this reason, the quality evaluation unit 12 analyzes and evaluates intermediate data (for example, detailed intermediate data such as phonetic symbols and selected segment waveforms) generated by the speech synthesis unit 11 during speech synthesis. Since it is possible to determine the speech data to be additionally stored, the synthesized speech created using the existing speech synthesis system can be evaluated with high accuracy, and additional storage is performed when the existing speech synthesis system is updated. The voice data to be determined can be easily determined.

  The input text data described above includes format designating data describing a speech format, and the format designating data includes at least one of data indicating pitch pattern and speech rate. . The speech synthesis database 10A described above can store at least one or all of a pitch pattern model, a duration model, and a speech unit waveform as speech data stored therein. Furthermore, the speech data determined by the additional data determination unit 13 described above is targeted for pitch pattern parameters, duration parameters, speech segment waveforms, etc., and may include at least one of them. Good.

FIG. 2 shows a flowchart for explaining the operation of the first embodiment.
The operation of the first embodiment will be described below based on the flowchart shown in FIG.

  First, text data is input from the outside and supplied to the speech synthesizer 11 (step S101). The text data here includes at least text data in which utterance contents are described. The speech synthesizer 11 converts the input text data into a speech waveform (hereinafter referred to as “synthesized speech”) using a known text speech synthesis technique, and outputs the speech waveform to the quality evaluation unit 12 (step S102). Here, the text-to-speech synthesis technology (TTS) is a technology for analyzing input text, estimating prosody and time length, and outputting the synthesized speech as described in Non-Patent Document 1, for example. It is a generic name.

  The quality evaluation unit 12 that has acquired the synthesized speech output by the speech synthesizer 11 as an input evaluates the sound quality of the synthesized sound waveform according to a predetermined procedure disclosed in Example 1 described later, and outputs it to the additional data determination unit 13. (Step S103). The quality evaluation unit 12 compares the evaluation value as the sound quality evaluation result with a preset threshold value (step S104), and ends the process when the evaluation value exceeds the threshold value. On the other hand, when the evaluation value in the quality evaluation unit 12 falls below the threshold value, the additional data determination unit 14 operates, and voice data to be additionally stored in the voice synthesis database 10A is a procedure described later (details in the first embodiment). To be determined) (step S105). Here, “additionally stored speech data” indicates data to be added to the speech synthesis database included in the speech synthesizer 11 in order to convert the input text into synthesized speech.

  Subsequently, the recording instruction data generation unit 14 generates recording instruction data according to a predetermined procedure (see Example 1) (step S106). Here, “recording instruction data” is data for instructing a sentence or a word uttered by a speaker or an utterance style, emotion, speaking speed, intonation, etc. at the time of voice recording. Contains text that indicates

  According to the first embodiment described above, the quality evaluation unit 12 performs quality evaluation based on the feature amount of the synthesized speech generated by the speech synthesis unit 11, and the additional data determination unit 13 performs based on the quality evaluation result. Since the voice data to be additionally stored in the voice synthesis database 10A is determined, the waveform of the synthesized voice created using the existing voice synthesis system 10 (the voice synthesis database 10A and the voice synthesis unit 11) is obtained. In order to evaluate, it becomes possible to easily estimate speech data to be added when updating an existing speech synthesis system. Further, the recording instruction data generation unit 14 generates recording instruction data including text data describing the utterance content of the audio signal to be recorded corresponding to the audio data determined by the additional data determination unit 13. Recording of speech data when updating the existing speech synthesis system 10 is facilitated.

[Second Embodiment]
Next, a second embodiment will be described with reference to FIGS. Here, the same reference numerals are used for the same constituent members as those in the first embodiment described above.
In FIG. 3, the speech recording system according to the second embodiment is substantially the same as in the case of the first embodiment described above. The speech synthesizer 11 and the speech synthesis database 10A constituting the speech synthesis system 10, A quality evaluation unit 22, an additional data determination unit 13, and a recording instruction data generation unit 14 including a text generation storage unit 14A are provided.

In the second embodiment, the quality evaluation unit 22 is configured to receive a predetermined audio signal (audio data) from the outside. The quality evaluating unit 22 compares the synthesized speech (intermediate data) synthesized by the speech synthesizing unit 11 and the externally input speech signal to evaluate the degree of coincidence of the waveform and the feature amount and set in advance. A function of determining whether or not the desired quality is satisfied. In this case, the evaluation of the degree of coincidence of the feature amounts is performed by calculating an evaluation value and comparing based on the evaluation value. As a result, the characteristics of the synthesized speech, which was unclear in the conventional example described above, are clarified.
Other configurations are the same as those in the case of FIG. 1 (first embodiment) described above.

Next, the operation of the second embodiment will be described with reference to FIG.
In the second embodiment, as described above, the quality evaluation unit 22 is greatly different from the first embodiment described above.
That is, step S103 of the quality evaluation in the first embodiment described above becomes a comparative quality evaluation (step S103a) in the second embodiment, and further, in parallel with the text input (step S103) in the second embodiment. The voice input (external input of the voice signal to the quality evaluation unit 22) is added as processing step S107.
That is, in step S107 in FIG. 4, an audio signal is input separately from the input text described above.

  Here, the audio signal has the same utterance content as the content described in the text data input in step S101 in this embodiment. Then, the quality evaluation unit 22 compares the synthesized speech output from the speech synthesis unit 11 and the speech signal input and acquired in step S107, and calculates an evaluation value from the degree of coincidence of the waveform and the feature amount (step S103a). ). In addition, as feature quantities, average pitch frequency, pitch pattern, duration, cepstrum, etc. correspond to this, and as an evaluation method, a plurality of feature quantities are vectorized and a difference between these vectors is evaluated as a scalar. In the embodiment of the present invention, the value is executed.

  Other operation procedures (steps S101, S102, and S104 to S106) are the same as the steps shown in FIG. 2 (first embodiment) described above.

  For this reason, the second embodiment shown in FIGS. 3 to 4 has the same effect as the first embodiment described above, and the quality evaluation unit 22 further synthesizes text data by voice synthesis. Compared with the audio signal and the input audio signal, the feature that cannot be expressed by the existing speech synthesis system is clarified, so that it is possible to create auxiliary recording data adapted to the task.

[Third Embodiment]
Next, a third embodiment will be described with reference to FIGS. Here, the same reference numerals are used for the same components as those of the second embodiment described above.

  The speech recording system in the third embodiment shown in FIG. 5 is substantially the same as in the case of the second embodiment described above, the speech synthesis unit 11 and the speech synthesis database 10A constituting the speech synthesis system 10, A quality evaluation unit 32, an additional data determination unit 13, and a recording instruction data generation unit 14 having a text generation storage unit 14A are provided.

  In the third embodiment, the quality evaluation unit 32 is configured to receive a predetermined audio signal (audio data) from the outside as shown in FIG. Further, the quality evaluation unit 32 extracts segmentation information, which is information necessary to determine the degree of coincidence between the externally input audio signal and the synthesized speech, and sends it to the quality evaluation unit 32 15 is attached. Here, the segmentation information is data including at least duration time information.

The quality evaluation unit 32 then matches the time information using the above-described segmentation information (segmentation data), compares the synthesized speech and the input speech signal, and evaluates the degree of coincidence of the feature amount. It has a function to determine whether or not the quality is satisfied. That is, the quality evaluation unit 32 compares the synthesized speech (intermediate data) synthesized by the speech synthesis unit 11 described above using the extracted segmentation information with the externally input speech signal, and compares the waveform and feature amount. The degree of coincidence is evaluated and whether or not a predetermined desired quality is satisfied is determined. As a result, the characteristics of the synthesized speech, which was unclear in the conventional example described above, are further clarified.
Other configurations are the same as those of the second embodiment shown in FIG.

Next, the operation of the third embodiment will be described with reference to FIG.
In the flowchart of FIG. 6, the segmentation data extraction unit 15 extracts segmentation information from the audio signal input in step S107, and outputs it to the quality evaluation unit 32 (step S108). The quality evaluation unit 32 compares the synthesized speech signals using the extracted segmentation data, and calculates the degree of coincidence between waveforms and feature amounts (step S103b).

  Other operation steps are the same as those in the second embodiment shown in FIG. That is, in FIG. 6, steps (S101, S102, S104 to S107) denoted by the same reference numerals as those shown in FIG. 4 (second embodiment) described above are shown in FIG. 4 (second embodiment). Each step performs the same operation.

  For this reason, according to the third embodiment, in addition to having the same effects as those of the second embodiment described above, the quality evaluation unit 32 further extracts the audio signal extracted by the segmentation data extraction unit 15. Since segmentation data is used for audio signal comparison, it has the advantage that a more precise and detailed comparison is possible.

[Fourth Embodiment]
Next, a fourth embodiment will be described with reference to FIGS. Here, the same reference numerals are used for the same components as those of the second embodiment described above.

  In FIG. 7, the voice recording system according to the fourth embodiment is similar to the second embodiment described above in that the voice synthesis unit 41 and the voice synthesis database 40A constituting the voice synthesis system 40, the quality, An evaluation unit 42, an additional data determination unit 13, a recording instruction data generation unit 14 provided with a text generation storage unit 14A, and a segmentation extraction unit 45 are provided. In the fourth embodiment, the segmentation extraction unit 45 has a function of extracting segmentation information including a phoneme string associated with at least time information of the input speech signal corresponding to the input text data from the speech signal. ing.

  In the fourth embodiment, a predetermined voice signal (voice data) and a synthesized voice that is the output of the voice synthesizer 41 are input to the quality evaluation unit 42 from the outside. It is configured and has the same function as the quality evaluation unit 22 in the second embodiment described above. Further, the above-described speech synthesis unit 41 is provided with the above-described segmentation extraction unit 45 so that the above-described segmentation information (detailed in the fourth embodiment), which is information necessary for speech synthesis, is directly sent.

  Further, the speech synthesizer 41 has a function of generating synthesized speech having time information corresponding to the input speech signal described above by using the segmentation information directly sent from the segmentation extractor 45. . Other configurations are the same as those in FIG. 3 (second embodiment) described above.

Next, the operation of the fourth embodiment will be described with reference to FIG.
In the fourth embodiment, unlike the case of FIG. 3 (second embodiment) described above, since the segmentation data extraction unit 45 is provided, the operation of the speech synthesis unit 41 is the same as the speech synthesis unit 11 described above. The operation is partly different.

That is, in the flowchart shown in FIG. 8, the segmentation data extraction unit 45 extracts the segmentation information of the input speech signal and outputs it to the speech synthesis unit 41 (step S108a). As a result, the speech synthesizer 45 synthesizes the input text according to the duration information indicated by the segmentation data (step S102a).
Each step (S101, S103a, S104 to S107) showing other operations is the same as the operation step shown in FIG. 4 (second embodiment) described above.

  Since the fourth embodiment is configured and functions as described above, according to this, in addition to having the same effects as the second embodiment described above, the speech synthesis unit 41 further includes a segmentation data extraction unit. In order to generate synthesized speech according to the duration length information indicated by the segmentation data of the speech signal extracted by 45, time length matching processing such as DP (Dynamic Programming) matching is not required, and therefore precise and detailed comparison is performed. It is possible to obtain an excellent effect that is possible.

[Fifth Embodiment]
Next, a fifth embodiment will be described with reference to FIGS. Here, the same reference numerals are used for the same components as those of the second embodiment described above.
In FIG. 9, the speech recording system in the fifth embodiment is substantially the same as in the second embodiment described above, and the speech synthesis unit 11 and speech synthesis database 10A constituting the speech synthesis system 10, quality, and the like. An evaluation unit 22, an additional data determination unit 13, and a recording instruction data generation unit 14 including a text generation storage unit 14A are provided.

Further, the recording instruction data generation unit 14 is provided with a recording audio signal evaluation unit 16. The recorded voice signal evaluation unit 16 compares the input voice signal corresponding to the input text data and the newly recorded voice signal, and the evaluation value obtained as a result of the comparison is less than a predetermined value set in advance. Has a function to output an instruction to re-record to the outside.
Other configurations are the same as those of the second embodiment described above.

Next, the operation of the fifth embodiment will be described with reference to FIG.
In the fifth embodiment, as shown in the flowchart of FIG. 10, a speaker performs voice recording work and collects recorded voice signals using the recording instruction data generated by the recording instruction data generation unit 14. (Step S109). In this case, the speaker is preferably the same speaker as the recorded speech signal used when the speech synthesis database 10A of the speech synthesizer 11 is created.

Subsequently, the recorded audio signal evaluation unit 16 evaluates the recorded audio signal with reference to the audio signal input as the request (step S110). Then, as a result of comparing the evaluation value with the threshold value (step S111), if the evaluation value exceeds the threshold value, the process is terminated. If the evaluation value falls below the threshold value, the recorded audio signal evaluation unit 16 instructs the speaker to perform re-recording, and returns to the process of step S109 to perform re-recording.
Other operations are the same as the operation steps (S101, S103a, S104 to S107) in FIG. 4 (second embodiment) described above.

  Since the fifth embodiment is configured and functions as described above, according to this, in addition to having the same operational effects as the second embodiment described above, the recorded audio signal evaluation unit 16 further evaluates the evaluation value to a constant value. Since the recording can be repeated until a recorded audio signal exceeding 1 is obtained, it is possible to collect the recorded audio signal corresponding to the task with high sound quality.

[Sixth Embodiment]
Next, a sixth embodiment will be described with reference to FIGS. Here, the same reference numerals are used for the same constituent members as those in the first embodiment described above.
The sixth embodiment is for creating voice recording instruction data in the voice recording system.

  11, in the sixth embodiment, a speech synthesis database 60A, a quality evaluation unit 62, and an additional data determination unit 13 that function substantially the same as the components disclosed in the first embodiment (FIG. 1) described above. And a recording instruction data generation unit 14 including a text generation storage unit 14A.

In the voice recording system according to the sixth embodiment, the contents of data stored in the voice synthesis database 60A and the operation of the quality evaluation unit 62 are different from those in the first embodiment described above. It has become. In this case, the speech synthesis database 10A stores speech synthesis system information that defines the operation of speech synthesis processing. Here, the speech synthesis system information is information that defines the operation of speech synthesis processing, and at least one of a morphological analysis model, a language dictionary, an accent dictionary, prosodic information, synthesis rule information, and segment waveform information. It is information including.
Other configurations are the same as those of the first embodiment (FIG. 1) described above.

Next, the operation of the sixth embodiment will be described with reference to FIG.
First, in the flowchart of FIG. 12, the quality evaluation unit 62 analyzes system information related to speech synthesis stored in the speech synthesis database 17 and calculates an evaluation value (step S103c). Subsequently, the operation steps (S104 to S106) after the evaluation value is calculated are the same as those in FIG. 2 (first embodiment) described above. Further, other operation procedures, that is, operation steps (S101, S104 to S106) starting from step S101 in FIG. 12 are the same as the steps shown in FIG. 2 (first embodiment) described above.

  As described above, according to the sixth embodiment, the quality evaluation unit 62 functions to analyze the system information related to speech synthesis stored in the speech synthesis database 61A and calculate the evaluation value. The recording instruction data is created from the existing speech synthesis system information and the text data input by the user without passing through the synthesized speech processing by the voice synthesizing unit 10, and high-speed recording instruction data can be created. An unprecedented superior effect of becoming can be obtained.

[Seventh Embodiment]
Next, a seventh embodiment will be described with reference to FIGS. Here, the same reference numerals are used for the same constituent members as those in the first embodiment described above.

In FIG. 13, in the seventh embodiment, a configuration that functions substantially the same as each configuration disclosed in the first embodiment (FIG. 1), that is, a speech synthesizer 11 and a speech that constitute the speech synthesis system 10. A synthesis database 10A, a quality evaluation unit 72, an additional data determination unit 13, and a recording instruction data generation unit 14 including a text generation storage unit 14A are provided.
In the voice recording system according to the sixth embodiment, the contents of the partial data stored in the voice synthesis database 10A and the operation of the quality evaluation unit 72 are the same as those in the first embodiment described above. The parts are different.

In this case, the quality evaluation unit 72 has a function of taking in and analyzing intermediate data used when the speech synthesis unit 11 converts the text into synthesized speech and calculating an evaluation value. Here, the “intermediate data” is generated information of a synthesized speech signal, and includes information including at least one of a phonetic symbol string, duration information of each segment, pitch pattern information, and a selected segment waveform. It is. As a method of calculating the evaluation value, there is a method of using a score in score calculation for unit selection as an evaluation value.
Other configurations are the same as those of the first embodiment (FIG. 1) described above.

Next, the operation of the seventh embodiment will be described based on the flowchart of FIG.
In the audio recording system according to the seventh embodiment, as shown in FIG. 14, the quality evaluation unit 72 analyzes the intermediate data used when the speech synthesis unit 11 synthesizes text into speech, An evaluation value is calculated (step S103d). The subsequent operations (S104 to S106) after calculating the evaluation value are the same as those in the first embodiment shown in FIG.
Here, in FIG. 14, other operation procedures starting from step S101 are the same as the operation steps (S101, S104 to S106) in the first embodiment (FIG. 2) described above.

  As described above, according to the seventh embodiment, the quality evaluation unit 72 evaluates detailed intermediate data at the time of speech synthesis, and the additional data determination unit 13 determines additional data using this evaluation value. More detailed recording instruction data can be created.

[Eighth Embodiment]
Next, an eighth embodiment will be described with reference to FIGS. Here, the same reference numerals are used for the same components as those of the fifth embodiment described above.
In the eighth embodiment shown in FIG. 15, the speech synthesis unit 11 and the speech synthesis database 10A constituting the speech synthesis system 10 and the quality evaluation unit 82 are substantially the same as in the case of the fifth embodiment described above. , An additional data determination unit 13 and a recording instruction data generation unit 14 including a text generation storage unit 14A.

  Further, the recording instruction data generation unit 14 is provided with a recording audio signal evaluation unit 16. The recorded voice signal evaluation unit 16 compares the input voice signal corresponding to the input text data and the newly recorded voice signal, and the evaluation value obtained as a result of the comparison is less than a predetermined value set in advance. Has a function to output an instruction to re-record to the outside.

Furthermore, the speech synthesis unit 11 described above is provided with a speech synthesis system update unit 88. The voice synthesis system update unit 88 takes in the audio signal recorded using the recording instruction data from the recording instruction data generation unit 14 via the recording audio signal evaluation unit 16 described above, and uses the audio signal to output the audio described above. The voice data for the voice synthesis system stored in the synthesis database 10A is updated.
Other configurations are the same as those of the fifth embodiment (FIG. 9) described above.

Next, the operation of the eighth embodiment will be described based on the flowchart of FIG.
In the voice recording system according to the eighth embodiment, as shown in FIG. 16, the voice synthesis system update unit 88 examines the recorded data evaluation value using the voice signal recorded in step S109, and determines the result. Based on this, the speech synthesis system 10 including the speech synthesis database 10A is updated (step S112).
Here, in FIG. 16, the other operation procedures starting from step S101 are the same as the operation steps (S101, S102, S103a, S104 to S111) in the fifth embodiment (FIG. 10) described above. .

  As described above, according to the eighth embodiment, in addition to the same effects as those of the fifth embodiment described above, the speech synthesis system update unit 88 further performs speech synthesis using the recorded speech signal. Since the speech synthesis system including the database 10A for use is updated and the iterative process is performed using the updated speech synthesis system 10, the quality of the synthesized speech and the fitness of the task can be further effectively increased. It becomes.

[Ninth Embodiment]
Next, a ninth embodiment will be described with reference to FIGS. Here, the same reference numerals are used for the same constituent members as those in the eighth embodiment described above.
The ninth embodiment shown in FIG. 17 is substantially the same as in the case of the eighth embodiment described above, and includes a speech synthesis unit 11 and a speech synthesis database 10A that constitute the speech synthesis system 10, and a quality evaluation unit 82. , An additional data determination unit 13 and a recording instruction data generation unit 14 including a text generation storage unit 14A.

In addition, the recording instruction data generation unit 14 is provided with a recording audio signal evaluation unit 96 that functions substantially the same as the recording audio signal evaluation unit 16 in the above-described eighth embodiment. Further, the above-described quality evaluation unit 82 estimates unnecessary data in the speech synthesis database based on the evaluation result in the quality evaluation unit 82, and unnecessary speech data to be deleted from the speech synthesis database. An unnecessary data determining unit 19 for determining
In this case, as unnecessary data, a pitch pattern, a duration, a segment waveform, and the like are determined.

Furthermore, the speech synthesis unit 11 described above includes a speech synthesis system update unit 98 that updates speech synthesis data stored in the speech synthesis database based on unnecessary data determined by the unnecessary data determination unit 19 described above. It is attached.
Other configurations are the same as those of the above-described eighth embodiment (FIG. 15).

Next, the operation of the ninth embodiment will be described based on the flowchart of FIG.
In the voice recording system according to the ninth embodiment, as shown in FIG. 18, first, the unnecessary data determination unit 19 is unnecessary to be deleted from the voice synthesis database 10 based on the evaluation result in the quality evaluation unit 82. Data is determined and output to the speech synthesis system update unit 98 (step S113). The speech synthesis system update unit 98 adds the speech signal recorded in step S109 and deletes the data indicated by the unnecessary data determined by the unnecessary data determination unit 19 to include the speech synthesis database 10. The system is updated (step S112a).

  Other operation steps (steps S101, S102, S103a, S104 to S107, and S109 to S111) are the same as the operation steps shown in the above-described eighth embodiment (FIG. 16).

  As described above, according to the ninth embodiment, in addition to the same effects as those of the eighth embodiment described above, the speech synthesis system update unit 98 further includes a voice synthesis database of the speech synthesis system 10. Since data is added to 10A and unnecessary data is deleted, there is an advantage that enlargement of the speech synthesis database 10A can be prevented.

Next, examples corresponding to the above-described embodiments will be described.
[Example 1]
Example 1 corresponds to the first embodiment (FIGS. 1 and 2) according to the present invention and is shown in FIG. Here, in FIG. 19, the same components as those in FIG. 1 described above are denoted by the same reference numerals.

  In Example 1 shown in FIG. 19, the components included in the first embodiment shown in FIG. 1, that is, the speech synthesis unit 11 and the speech synthesis database 10A constituting the speech synthesis system 10, A quality evaluation unit 12, an additional data determination unit 13, and a recording instruction data generation unit 14 including a text generation storage unit 14A are provided. Further, the recording instruction data generation unit 14 is provided with a sentence generation text database 17.

  The voice synthesizer 11 obtains text data as a voice recording request and converts it to synthesized voice. In the first embodiment, a waveform connection type speech synthesis method is employed as the speech synthesis method. Here, the waveform connection type speech synthesis means that a large number of segment waveforms and phoneme information called labels corresponding to each segment waveform are stored in advance in the speech synthesis database 10, and the segment waveforms are connected. It is a general term for speech synthesis methods that generate speech by doing so.

  The quality evaluation unit 12 evaluates the synthesized speech output from the speech synthesis unit 11 and calculates the numerical value of the evaluation data. As evaluation criteria, smoothness at the connection point of the segment waveform, naturalness of the prosody (whether there is a steep change in the prosody, etc.), spectral divergence from the standard phoneme, and the like are taken up.

  Based on the evaluation data output from the quality evaluation unit 12, the additional data determination unit 13 estimates speech data that is insufficient to synthesize externally input text data and adds it to the speech synthesis database 10A. The audio data to be stored is determined. Specific contents of the audio data to be additionally stored include a pitch pattern, a duration length, and a segment waveform.

Next, a specific method for determining additional data will be described with reference to FIG.
As shown in FIG. 20 (A), for example, it is assumed that a text having the content “I live in Yunnan Province, China” has been synthesized into speech and a synthesized speech is output.

  In this case, the synthesized sound is evaluated by the quality evaluation unit 12, and a duration length evaluation value, a pitch pattern evaluation value, and a waveform connection evaluation value are calculated for each accent phrase. For the first accent phrase “Chinese” and the third accent phrase “I live”, the above three evaluation values are all “10”, but the second accent phrase “To Yunnan” continues The time length evaluation value “80”, the pitch pattern evaluation value “65”, and the waveform connection evaluation value are “20”.

For the accent phrase “in Yunnan”, the duration length and the pitch pattern can be expressed by what already exists in the speech synthesis database 10A. This indicates that it does not exist in the speech synthesis database 10A.
Here, if the threshold value for the evaluation value is set to “50”, the additional data determination unit 13 determines that the audio data to be additionally stored is a segment waveform, and the additional stored audio data is Then, the recording instruction data generation unit 14 is activated to record the waveform data uttered "Unnanshoni".

  The recording instruction data generation unit 14 indexes the sentence generation text database 17 from the additional data determined by the additional data determination unit 13, and generates recording instruction data from the output sentence generation text. The recording instruction data includes at least an utterance list describing the utterance contents to be recorded, and may further include a text indicating the utterance style and a chart indicating the correct prosody. An example of the format of the utterance list is shown in FIG.

Here, the externally input text data may include auxiliary text data for instructing speech speed, emotion, inflection and the like in addition to the utterance content text data in which the utterance content is described. In this case, the auxiliary text data may be used when the speech synthesis unit 11 performs speech synthesis of the utterance content text data, or may be recording instruction data.
In the above-described example, the recording instruction data generation unit 14 generates a meaningful sentence using the sentence generation text database 17, but generates a meaningless sentence or a meaningless word as a recorded sentence list. Also good. Further, the text that becomes the voice synthesis request may be used as it is as the recorded sentence list, or the text of the voice synthesis request and the text of the sentence generation text database 17 may be mixed.

Furthermore, the recording instruction data generation unit 14 can estimate the total time of the audio signal recorded from the utterance list included in the recording instruction data, and change the satisfaction rate of the additional stored audio data accordingly. For example, when the filling rate of the additional data is 10%, the total time of the voice signal becomes very long, and the voice data greatly increases when added to the voice synthesis database 10A in the voice synthesis system. It is assumed that
In such a case, for example, by allowing the sufficiency rate to be 60%, it is possible to prevent a significant increase in audio data. At this time, 40 [%] of the additional data must be deleted. As a method for determining the data to be deleted, a method of selecting as much additional data as possible in the same sentence from the sentence generation text database 17. And a method for estimating the importance of data from the frequency of requests.

[Example 2]
Example 2 corresponds to the second embodiment (FIGS. 3 to 4) according to the present invention and is shown in FIG. Here, in FIG. 21, the same components as those in FIG. 3 described above are denoted by the same reference numerals.

  In Example 2 shown in FIG. 21, the components included in the second embodiment shown in FIG. 3 described above, that is, the speech synthesis unit 11 and the speech synthesis database 10A constituting the speech synthesis system 10, A quality evaluation unit 22, an additional data determination unit 13, and a recording instruction data generation unit 14 including a text generation storage unit 14A are provided. Further, the recording instruction data generation unit 14 is provided with a sentence generation text database 17.

  Further, it is assumed that text data and a voice signal are input as a voice synthesis request. Here, it is desirable that the text data and the audio signal have a correspondence relationship, and it is more preferable that they have the same content. For this reason, in the second embodiment, description will be made assuming that the text data and the audio signal have the same content.

  The quality evaluation value 12a output by the quality evaluation unit 22 compares the synthesized speech and the corresponding speech signal, and calculates an evaluation value based on the comparison result of the feature amount. Here, as a feature quantity to be compared, a fundamental frequency (F0) pattern, a spectrum, a waveform, a duration length, and the like can be considered, and as a numerical value to be calculated, a similarity such as a square distance can be considered. Here, it is assumed that the similarity of F0, spectrum, and duration is calculated.

The additional data determination unit 13 determines additional data by comparing the numerical value of the comparison result data output from the quality evaluation unit 12 with a preset threshold value. Further, the recording instruction data generation unit 14 generates recording instruction data based on the additional stored audio data determined by the additional data determination unit 13.
As in the first embodiment, the recording instruction data may create an utterance list with text describing the utterance content, or may be an example of a speaker generating an input voice signal. Of course, both text and audio signals may be used, and a chart indicating the prosody may be added.

[Example 3]
Example 3 corresponds to the fourth embodiment (FIGS. 5 to 6) and the fifth embodiment (FIGS. 7 to 8) according to the present invention. FIGS. 22 (A) and 22 (B) show this example. Indicates. Here, in FIG. 22A, the same components as those in FIG. 5 or FIG.

  In Example 3 shown in FIGS. 22 (A) and 22 (B), each component included in each of the third to fourth embodiments described above, that is, the speech synthesis unit 11 constituting the speech synthesis system 10 and A speech synthesis database 10A, a quality evaluation unit 32 (42), an additional data determination unit 13, a recording instruction data generation unit 14 having a text generation storage unit 14A, and a segmentation extraction unit 15 are provided. Further, the recording instruction data generation unit 14 is provided with a sentence generation text database 17. Also in the third embodiment, it is assumed that text data and a voice signal are input as in the second embodiment.

Here, the segmentation data extraction unit 15 (45) extracts the duration time information of each phoneme of the speech signal as the segmentation data. As an extraction method, there is segmentation by HMM (Hidden Markov Model) in DP matching. The description format of segment tension data includes, for example, a method of describing the duration of each phoneme (ch, u1,...) In “msec units” as shown in FIG.
As described in the third embodiment of the present invention (see FIG. 5), when segmentation data is output to the quality evaluation unit 32, segmentation data may be used for phoneme matching in comparison. is there.

In addition, as described in the fourth embodiment of the present invention (see FIG. 7) described above, when segmentation data is output to the speech synthesizer 41, each phoneme included in the segmentation data is generated during synthesis speech generation. Create synthesized speech that faithfully reproduces the duration time of. As a result, a synthesized speech in which the duration of each phoneme is the same as the speech signal is generated, and it is not necessary for the quality evaluation unit 32 to perform phoneme matching.
The segmentation data extracted by the segmentation data extraction unit 15 (45) may be output to both the speech synthesis unit 11 and the quality evaluation unit 32.

[Example 4]
Example 4 corresponds to the above-described sixth embodiment shown in FIG. 11, and is shown in FIG. Here, in FIG. 23, the same components as those in FIG. 11 described above are denoted by the same reference numerals.

  In Example 4 shown in FIG. 23, each component included in the above-described sixth embodiment, that is, a speech synthesis database 60A, a quality evaluation unit 62, an additional data determination unit 13, and a text generation storage unit. And a recording instruction data generation unit 14 having 14A. Further, the recording instruction data generation unit 14 is provided with a sentence generation text database 17. Here, the speech synthesis database 60A stores speech synthesis system information such as synthesis rules, pitch pattern models, duration models, and segment waveforms.

  The quality evaluation unit 62 analyzes the speech synthesis system information stored in the speech synthesis database 60A based on the externally input text data, and calculates an evaluation value. Here, as speech synthesis system information, syllable strings accompanied by quality degradation are extracted from the segment waveforms.

  For example, it is assumed that a word such as “Irkutsk” is included in text data that is a speech synthesis request. In the speech synthesis database 60A, each syllable constituting "Irukutsu" exists, but no syllable string "Rukutsu" exists. When expressing a syllable string such as “Irukutsu” in an existing speech synthesis system including such a speech synthesis database 60A, a combination of “ru”, “ku”, and “tsu” with different environments before and after the syllable is generated. Must. For this reason, it becomes clear that this data is a syllable string accompanied by quality deterioration.

  Therefore, in the additional data determination unit 13, a segment waveform including the syllable string “Rukutsu” is listed as the target of the additional stored voice data. Recording instruction data for recording a syllable signal including the is generated.

[Example 5]
Next, Example 5 will be described with reference to FIG. Example 5 corresponds to the seventh embodiment shown in FIG. 13 described above. Here, in FIG. 24, the same components as those in FIG. 13 described above are denoted by the same reference numerals.

In Example 5 shown in FIG. 24, each component included in the seventh embodiment, that is, the speech synthesis unit 11 and the speech synthesis database 10A constituting the speech synthesis system 10, and the quality evaluation unit 72 are provided. And an additional data determination unit 13 and a recording instruction data generation unit 14 including a text generation storage unit 14A. The recording instruction data generation unit 14 is provided with a sentence generation text database 17.
The quality evaluation unit 72 extracts intermediate data used when the input text data is converted into synthesized speech by the speech synthesis unit 11. Here, it is assumed that the change amount of F0, the pitch pattern, and the number of segment waveform candidates are extracted as intermediate data.

Here, the intermediate data will be described in detail using the change amount of F0 as an example.
Generally, it is known that the sound quality of synthesized speech deteriorates when the change amount of F0 is large. For example, assuming that the original average F0 of the speech unit U1 is 150 [Hz], the speech unit U1 is changed to 300 [Hz]. Further, it is assumed that the additional data determination unit 13 sets a rule that “a speech unit whose F0 change amount is 1.5 times or more is set as additional stored speech data”.

  In this case, since the rate of change of the speech unit U1 is twice, the speech unit U1 becomes a target of additional stored speech data, and the recording instruction data generation unit 14 uses the same utterance content as the speech unit U1 and averages it. Recording instruction data is generated so as to record an audio signal having F0 of 300 [Hz]. As an instruction method, when uttering with a single syllable, a method of instructing a tone sound of 300 [Hz] as a model voice signal is conceivable.

[Example 6]
Next, Example 6 will be described. Example 6 corresponds to the fifth embodiment (FIG. 9) and the eighth embodiment (FIG. 15) described above, and FIG. 25 shows the details thereof.
Here, in FIG. 25, the same reference numerals are used for the same constituent members as those in FIG. 9 or FIG.

  The sixth embodiment shown in FIG. 25 includes a speech synthesis unit 11 and a speech synthesis database 10A that constitute the speech synthesis system 10, a quality evaluation unit 22 (82), an additional data determination unit 13, and a text generation storage unit. And a recording instruction data generation unit 14 having 14A. Further, the recording instruction data generation unit 14 is provided with a recording audio signal evaluation unit 16. Furthermore, the speech synthesis unit 11 described above is provided with a speech synthesis system update unit 88.

  Among these, the recorded audio signal evaluation unit 16 compares the recorded audio signal with the input audio signal, and evaluates how much the recorded audio signal is adapted to the voice synthesis request by the input audio signal. As an evaluation method, as in the case of the second embodiment in FIG. 21 described above, an evaluation based on a distance of feature amounts such as an F0 pattern, a spectrum, a waveform, and a duration length can be considered.

  Further, it is assumed that a threshold value is set in advance for the evaluation value in the recorded audio signal evaluation unit 16. If the evaluation value is lower than the threshold value, an instruction for re-recording is output to the speaker. Specifically, a method of displaying an utterance list to be rerecorded via a monitor or the like can be considered. The re-recorded audio signal is evaluated again by the recorded audio signal evaluation unit 16, and if the evaluation value is lower than the threshold value, it is recorded again. When the evaluation value is higher than the threshold value, the recorded voice signal is output to the voice synthesis system update unit 88.

  The voice synthesis system update unit 88 updates the voice data in the voice synthesis database 10A using the input recorded voice signal. It is assumed that a synthesis rule, a pitch pattern model, a duration model, a segment waveform, and the like are stored in the speech synthesis database 10A. Further, it is assumed that the additional waveform data and the pitch pattern are set as additional stored audio data. In this case, a pitch pattern and a segment waveform are created from the input recorded audio signal.

  The creation method may be manual or automatic. However, as a method for automatically creating a pitch pattern, a method such as F0 automatic extraction by an autocorrelation method or the like, and as a method for automatically creating a fragment waveform, segmentation by HMM or A method by cutting out can be considered. The speech synthesis system update unit 88 can update the above-described speech synthesis database 10A using the segment waveform and the pitch pattern created here.

Here, FIG. 28 shows a specific system configuration example of the speech synthesis system.
As shown in FIG. 28, users # 1 to # 3 store a speech synthesis request based on text and speech signals in the speech recording system 20 using a Web (World Wide Web) site or the like. The voice recording system 20 creates recording instruction data by the method described above, and outputs a recording instruction to a voice actor, an announcer, or the like through a monitor 31 or the like. The speaker records voice using a voice recording device 30 including a microphone and a personal computer, and stores the voice in the voice recording system 20.

  Then, the audio recording system 20 evaluates the recorded audio signal, and if the evaluation value is equal to or less than a certain value, issues a re-recording instruction through the monitor 31 or the like. If the evaluation value is equal to or greater than a certain value, the speech synthesis database 100 is updated and a new updated speech synthesis system 21 is created.

[Example 7]
Next, Example 7 will be described.
Example 7 corresponds to the ninth embodiment described above, and will be described in detail with reference to FIG. 17 shown in the ninth embodiment.
In FIG. 17, the unnecessary data determination unit 19 determines unnecessary data to be deleted from the speech synthesis database 10 </ b> A of the speech synthesis unit 11 from the evaluation result in the quality evaluation unit 82. As unnecessary data, a pitch pattern, a duration, a segment waveform, and the like are treated as target data.

  Here, the pitch pattern and the segment waveform will be described as an example. It is assumed that there are multiple text data and speech signals as speech synthesis requests, and there are many sentences with the sentence endings such as “How are you?” As shown in FIG. Furthermore, the pitch pattern at the end of "How is it?" Is not finished at all with the pitch rising like the general question sentence (Fig. 27 (A)), and all ends with the pitch at the end of the ending. Suppose that it is what is (FIG. 27 (B)).

In this case, in the speech synthesis engine (speech synthesizer 11) adapted to the speech synthesis request described above, “How is it?” Should be expressed by a pitch pattern in which the pitch at the end of the sentence is lowered.
Therefore, the pitch pattern and the segment waveform of “How is it?” That the pitch at the end of the sentence increases as shown in FIG. 27A in the speech synthesis database 10A are found to be data to be deleted.

  Here, the speech synthesis system update unit 98 adds data such as pitch patterns and segment waveforms extracted from the recorded speech signal input from the recorded speech signal evaluation unit 96 to the speech synthesis database 10A and is unnecessary. The unnecessary data determined by the data determination unit 19 is deleted from the speech synthesis database 10A.

  Here, unnecessary data is determined from the evaluation result in the quality evaluation unit 82, but in other embodiments, based on the evaluation result in the quality evaluation unit 12, 22, 32, 42, 62, 72, 82, respectively. In the embodiment, unnecessary data may be determined. In the seventh embodiment, the speech synthesis system update unit 88 adds the recorded speech signal and parameters created from the speech signal, and deletes speech data in the speech synthesis database 10A based on unnecessary data. However, it is also possible to employ a configuration in which the speech synthesis system update unit prepared separately performs the addition process and the deletion process. In this way, it is possible to speed up the audio data update process.

  As described above, each of the embodiments and examples described above, according to the present invention, it is possible to adapt an existing speech synthesis system to a task desired by a user, improve the quality, and to improve the quality of the speech synthesis database 10. It can prevent enlargement.

  2, 4, 6, 8, 10, 12, 14, 16, and 18 are the operations of the audio recording system according to the first to ninth embodiments of the present invention. In addition, each step of the audio recording method according to the present embodiment is also shown.

  That is, when the voice recording system according to the present embodiment is operated, a first synthesized voice is generated from input text data by referring to at least the voice synthesis database 10 in sequence control for voice recording. Steps (for example, S101 and S102 in FIG. 1), a second step (S103) for evaluating the quality of the synthesized speech based on the feature amount of the synthesized speech, and the speech synthesis database 10 based on the quality evaluation result. A third step (S104, S105) for determining audio data to be additionally stored is sequentially executed.

  In the second step (S103), the synthesized speech and the input speech signal corresponding to the input text data are compared to evaluate the similarity of the feature amount, and determine whether or not the desired quality is satisfied. A sub-step (for example, S103a in FIG. 4) may be included.

  Further, in the operation of the audio recording system according to the present embodiment, at the time of operation, at least a first step of acquiring input text data (for example, S101 in FIG. 12) and an acquired input in sequence control for audio recording. Analyzes and evaluates text data and speech synthesis system information that defines speech synthesis processing operations stored in the speech synthesis database, and whether the speech data stored in the speech synthesis database satisfies the desired quality The second step (S103c, S104) for determining the voice data and the third step (S105) for determining voice data to be additionally stored in the voice synthesis database 10 based on the result of the quality evaluation are sequentially executed. It has become.

  Further, a fourth step (for example, FIG. 2) for generating recording instruction data including text data describing the utterance content of the audio signal to be recorded corresponding to the audio data determined in the third step (S105). S106).

  The second step (S103) described above is a sub-step of acquiring an input speech signal corresponding to the input text data and extracting segmentation data including at least a phoneme string associated with time information from the input speech signal ( For example, S108) in FIG. 6 is matched with time information using segmentation data, the synthesized speech and the input speech signal are compared, the similarity of the feature quantity is evaluated, and whether or not the desired quality is satisfied. It is good also as a structure including the substep (S102, S104) to determine.

  The second step (S103) described above is a sub-step of acquiring an input speech signal corresponding to the input text data and extracting segmentation data including at least a phoneme string associated with the time information from the input speech signal ( For example, the first step (S102) described above including S108a in FIG. 8 includes a sub-step (S102a) for generating synthesized speech having time information corresponding to the input speech signal using the extracted segmentation data. It is good also as a structure including.

  Further, the fourth step (S106) described above is a sub-step for evaluating the recorded audio signal by comparing the input audio signal corresponding to the input text data and the recorded audio signal (for example, S109 and S110 in FIG. 10). ) And sub-steps (S111, S109) for outputting recording instruction data for performing re-recording when the evaluation value obtained as a result of the evaluation is less than the threshold value.

Further, a fifth step (for example, FIG. 16) of updating the voice data stored in the voice synthesis database 10 using the voice signal recorded by using the recording instruction data output in the fourth step (S106). S112a) is executed.
Further, the fifth step (S112a) described above is a sub-step (for example, FIG. 18) for determining unnecessary speech data to be deleted from the speech synthesis database 10 based on the evaluation result in the fourth step (S106). S113) may be included.

  Further, the fifth step (S112a) described above is a sub-step (for example, FIG. 18) for updating the speech data stored in the speech synthesis database 10 based on the unnecessary data determined in the fourth step (S113). S112b) may be included.

  According to the audio recording method according to the above-described embodiment, the audio recording system executes the second step of performing quality evaluation based on the feature amount of the synthesized speech generated by executing the first step, It is possible to evaluate the quality of synthesized speech created using an existing speech synthesis system by executing the third step of determining speech data to be additionally stored in the speech synthesis database 10 based on the quality evaluation result. It is possible to easily determine speech data to be added when updating an existing speech synthesis system based on the evaluation result.

  In addition, when the voice recording system executes the second step, the existing step is performed by executing a sub-step of comparing the synthesized voice obtained by voice synthesis of the input text data and the voice signal corresponding to the input text data. This makes it possible to create voice data that is more adapted to the task.

Further, according to the speech recording method according to the present embodiment, the speech recording system defines the operation of speech synthesis processing stored in the speech synthesis database 10A, for example, a speech synthesis system such as a morphological analysis model and prosodic information. By executing the second step of analyzing and evaluating the information, the existing speech synthesis is performed from the acquired input text data and the speech synthesis system information of the existing speech synthesis system without performing the synthesized speech processing. The voice data to be additionally stored when the system is updated can be determined easily and at high speed.
In addition, the voice recording system generates recording instruction data including text data describing the utterance content of the voice signal to be recorded corresponding to the voice data determined by executing the third step. By performing step 4, it becomes easy to record voice data when updating an existing voice synthesis system.

  In addition, when the audio recording system executes the second step described above, it extracts segmentation data such as duration information and uses it for comparison between the synthesized audio and the input audio signal. A more precise and detailed quality assessment of synthesized speech is possible. In addition, since the audio recording system generates synthesized speech according to the duration length information indicated by the segmentation data of the audio signal, it is possible to eliminate the need for comparison operation between the synthesized speech and the input audio signal at the time of quality evaluation.

  Further, when the voice recording system executes the fourth step, the input voice signal corresponding to the input text data is compared with the recorded voice signal, and the evaluation value obtained as a result of the comparison satisfies a predetermined value. By executing the sub-step that outputs an instruction to perform re-recording outside when there is not, the recording operation can be repeated until the evaluation value of the recorded voice by the speaker exceeds a predetermined value, so that the sound quality is high, and Recorded audio signal adapted to the task can be acquired.

  In addition, the voice recording system performs the fifth step of updating the voice synthesis system using the recorded voice signal, thereby performing an iterative process using the updated voice synthesis system, and the quality of the synthesized voice and the task It becomes possible to increase the fitness of.

  Further, the voice recording system executes a sub-step in the fifth step of determining unnecessary voice data to be deleted from the voice synthesis database based on the evaluation result output by executing the fourth step. As a result, the speech data is additionally stored in the speech synthesis database 10 and unnecessary data is deleted, so that the enlargement of the speech synthesis database 10 can be avoided. Further, the voice recording system executes a sub-step in the fifth step of updating the voice data stored in the voice synthesis database 10 based on the unnecessary data determined by executing the fourth step. Since unnecessary data is deleted, it is possible to avoid the enlargement of the speech synthesis database.

In the audio recording system described above, each of the first to fifth steps executed by each component and each sub-step in the first to fifth steps is executed in the respective steps. The contents are programmed as a speech synthesis processing function, a quality evaluation processing function, an additional data determination processing function, a recording instruction data generation function, a speech synthesis system update processing function, and respective sub-processing functions, and are executed by a computer. May be.
Even in this way, it is possible to perform processing equivalent to the processing contents in the first to fifth steps and the sub-steps in each step by the computer, thereby having the same purpose as the above-described audio recording method. Can be achieved efficiently.

Since the present invention is configured and functions as disclosed in the above-described embodiments and examples, according to this, the speech synthesis system can be adapted to the task desired by the user and the quality thereof can be improved. In addition to being applicable to the speech synthesis system, it is possible to delete unnecessary data in the speech synthesis database, thereby preventing the speech synthesis database from becoming bloated. Therefore, it is possible to construct a speech synthesis system adapted to the most demanded task. More specifically, the content of the speech synthesis system can be updated while satisfying the requirements of a large number of users at a constant level, and it is possible to construct a recording system for speech that is optimal for speech synthesis applications. It is possible to provide an audio recording system, an audio recording method, and a recording processing program that have excellent effects.

Here, the above-described quality evaluation unit compares the synthesized speech with the input speech signal corresponding to the input text data, evaluates the degree of coincidence of the feature amounts, and satisfies whether a predetermined desired quality is satisfied. It is good also as a structure provided with the function to determine.
Therefore, according to this, the quality evaluation unit compares the synthesized speech obtained by synthesizing the input text data with the speech signal corresponding to the input text data, so that the feature cannot be expressed by the existing speech synthesis system. Since it becomes clear, it becomes possible to create voice data more adapted to the task.

  The quality evaluation unit analyzes and evaluates intermediate data generated by the speech synthesizer during speech synthesis for the input text data, and based on this, the speech stored in the speech synthesis database It is good also as a structure provided with the function to determine whether data satisfy | fill the desired quality.

  Therefore, according to this, the quality evaluation unit analyzes and evaluates detailed intermediate data generated by the speech synthesis unit at the time of speech synthesis, such as phonetic symbols and selected segment waveforms. Since speech data to be additionally stored can be determined, synthesized speech created using an existing speech synthesis system can be evaluated with high accuracy and should be additionally stored when updating an existing speech synthesis system Audio data can be easily determined.

  Here, the input text data described above may further include format designation data describing a voice format, and the format designation data may include at least one of data indicating a pitch pattern and a speech rate.

The above-described speech synthesis database may be configured to store at least one of a pitch pattern model, a duration model, and a speech segment waveform as the stored speech data.
Furthermore, the audio data determined by the additional data determination unit described above may include at least one of a pitch pattern parameter, a duration parameter, and a speech segment waveform.

In addition, the additional data determining unit described above generates recording instruction data including text data describing the utterance content of the audio signal to be recorded, corresponding to the audio data determined by the additional data determining unit. An instruction data generation unit may be provided.
In this way, the recording instruction data generation unit generates the recording instruction data including the text data describing the utterance content of the audio signal to be recorded corresponding to the audio data determined by the additional data determination unit. This makes it easier to record speech data when updating an existing speech synthesis system.

Further, the recording instruction data generating unit described above calculates the total time of the audio signal recorded from the text data describing the utterance content of the audio signal to be recorded, included in the recording instruction data, and the calculation result It is good also as a structure provided with the function which suppresses the quantity of the audio | voice signal which should be recorded according to.
Further, the recording instruction data generation unit described above has a text generation storage unit and a function of generating the recording instruction data with reference to the generation text data stored in the text generation storage unit It is good.

In addition, the recording instruction data generation unit described above has a function of calculating the amount of audio signals to be recorded from the recording instruction data and changing the satisfaction rate of the additional data according to the calculated amount of audio signals. It is good also as a structure provided.
Further, the recording instruction data generation unit described above may have a function of selecting the additional data and creating recording instruction data.
The recording instruction data generation unit described above includes at least one of the input text data input to the speech synthesis unit, the generation text data stored in the text generation storage unit, and the additional data. Thus, a configuration may be provided that has a function of generating the recording instruction data that satisfies all or part of the additional data.

Further, the recording instruction data generated by the recording instruction data generating unit includes utterance content instruction data for instructing utterance content, and the utterance content instruction data is text data describing the utterance content, or generation content It is good also as a structure containing at least one of the synthetic | combination speech |
The recording instruction data generated by the recording instruction data generating unit further includes utterance style instruction data for instructing a voice form, and the utterance style instruction data indicates a pitch pattern, a duration length, an inflection, and an emotion. It may be configured to include at least one of the data. Further, the recording instruction data generated by the above-described recording instruction data generation unit may be included in an input voice signal corresponding to the input text data.

  Further, the quality evaluation unit includes a segmentation data extraction unit that extracts segmentation data including at least a phoneme string associated with time information of the input speech signal corresponding to the input text data. The unit matches the time information using the segmentation data, compares the synthesized speech and the input speech signal, evaluates the similarity of the feature amount, and determines whether or not the desired quality is satisfied It is good also as a structure provided with the function to perform.

  Therefore, according to this, the segmentation extraction unit extracts the segmentation data of the input speech signal corresponding to the input text data, the quality evaluation unit uses the extracted segmentation data to match the time information, and synthesize The quality evaluation unit inputs segmentation data such as duration information and the synthesized speech by comparing the speech with the input speech signal and evaluating the similarity of the feature quantity to determine whether the desired quality is satisfied. It can be used for comparison with a speech signal, and as a result, a more precise and detailed quality assessment of synthesized speech becomes possible.

Furthermore, a segmentation data extraction unit for extracting segmentation data including at least a phoneme string associated with time information of the input speech signal corresponding to the input text data is provided in the speech synthesis unit, and the speech synthesis The unit may have a function of generating synthesized speech having time information corresponding to the input speech signal using the segmentation data.
Therefore, according to this, since the synthesized speech according to the duration length information indicated by the segmentation data of the speech signal is generated in the speech synthesizer, the comparison operation between the synthesized speech and the input speech signal in the quality evaluation unit is unnecessary. can do.

Further, when the input voice signal corresponding to the input text data is compared with the recorded voice signal in the voice recording system described above, and the evaluation value obtained as a result of the comparison is less than a predetermined value set in advance, On the other hand, a configuration may be adopted in which a recording audio signal evaluation unit that outputs an instruction to perform re-recording is provided in the recording instruction data generation unit.
Therefore, according to this, the recorded audio signal comparison unit compares the input audio signal corresponding to the input text data with the recorded audio signal, and processing is performed when the evaluation value obtained as a result of the comparison exceeds a predetermined value. , And if it does not reach the specified value, the recording operation can be repeated until the evaluation value of the recorded voice by the speaker exceeds the specified value. It is possible to acquire recorded audio signals that are sound quality and adapted to the task.

In addition, a speech synthesis system update unit that updates speech data for a speech synthesis system stored in the speech synthesis database by a speech signal recorded using the instruction data described above may be provided alongside the speech synthesis unit. Good.
Therefore, according to this, since the speech synthesis system update unit updates the speech synthesis system using the recorded speech signal, and further performs an iterative process using the updated speech synthesis system, the quality of synthesized speech and It becomes possible to increase the fitness of the task.

In addition, unnecessary data determination for estimating unnecessary data in the speech synthesis database and determining unnecessary speech data to be deleted from the speech synthesis database based on the evaluation result in the quality evaluation unit is performed in the quality evaluation unit described above. It is good also as a structure which added the part.
Therefore, according to this, the unnecessary data determining unit additionally stores the speech data in the speech synthesis database in order to determine unnecessary speech data to be deleted from the speech synthesis database based on the evaluation result in the quality evaluation unit. In addition, since unnecessary data is deleted, it is possible to avoid the enlargement of the speech synthesis database.

  Here, the unnecessary data determined by the above-described unnecessary data determination unit may include at least one of a pitch pattern parameter, a duration parameter, and a speech segment waveform.

In addition, a speech synthesis system updating unit that updates speech synthesis speech data stored in the speech synthesis database based on the above-described unnecessary data may be provided in the speech synthesis unit.
Therefore, according to this, the speech synthesis system update unit updates the speech data stored in the speech synthesis database based on the unnecessary data determined by the unnecessary data determination unit, thereby avoiding the enlargement of the speech synthesis database. can do.

In the second step described above, the synthesized speech and the input speech signal corresponding to the input text data are compared to evaluate the degree of coincidence of the feature quantities, and based on this, the synthesized speech is obtained with a desired quality. It is good also as a structure including the substep which determines whether it satisfy | fills.
Therefore, according to this, there is a feature that the voice recording system cannot express in the existing voice synthesis system by comparing the synthesized voice obtained by voice synthesis of the input text data and the voice signal corresponding to the input text data. Since it becomes clear, it becomes possible to create voice data more adapted to the task.

Further, in the voice recording method described above, a fourth step of generating recording instruction data including text data describing the utterance content of the voice signal to be recorded corresponding to the voice data determined in the third step. , May be added.
For this reason, the audio recording system generates recording instruction data including text data describing the utterance content of the audio signal to be recorded corresponding to the audio data determined by executing the third step. By executing the fourth step, it becomes easy to record voice data when updating an existing voice synthesis system.

  Here, in the second step described above, a sub-step of obtaining an input speech signal corresponding to the input text data and extracting segmentation data including at least a phoneme string associated with time information from the input speech signal; , Matching with time information using the segmentation data, comparing the synthesized speech and the input speech signal to evaluate the degree of coincidence of the feature quantity, and based on this, whether or not the desired quality is satisfied And a sub-step for determination.

  For this reason, according to this, when the audio recording system executes the second step, the sub-step of extracting the segmentation data of the input audio signal corresponding to the input text data, and the segmentation data extracted here, The duration length information is obtained by matching the information, comparing the synthesized speech and the input speech signal, evaluating the similarity of the feature quantity, and executing a sub-step for determining whether or not the desired quality is satisfied. Or the like can be used for comparison between the synthesized speech and the input speech signal. As a result, a more precise and detailed quality assessment of the synthesized speech is possible.

  The second step includes a sub-step of acquiring an input speech signal corresponding to the input text data and extracting segmentation data including at least a phoneme string associated with time information from the input speech signal. Further, the first step may include a sub-step of generating synthesized speech having time information corresponding to the input speech signal using the extracted segmentation data.

In the first step described above, a sub-step of acquiring an input speech signal corresponding to the input text data and extracting segmentation data including at least a phoneme string associated with time information from the input speech signal; A sub-step of generating synthesized speech having time information corresponding to the input speech signal using the extracted segmentation data.
Therefore, according to this, since the audio recording system generates synthesized speech according to the duration length information indicated by the segmentation data of the audio signal, it is unnecessary to perform a comparison operation between the synthesized speech and the input audio signal at the time of quality evaluation. can do.

In the fourth step described above, a sub-step of evaluating the recorded voice signal by comparing the input voice signal corresponding to the input text data and the recorded voice signal, and an evaluation value obtained as a result of the evaluation And a sub-step for outputting recording instruction data for performing re-recording when the threshold value is less than the threshold value.
Therefore, according to this, when the voice recording system executes the fourth step, the substep for comparing the input voice signal corresponding to the input text data and the recorded voice signal, and the evaluation value obtained as a result of the comparison The process ends when the value exceeds a predetermined value, and when the value does not reach the predetermined value, the sub-step of outputting an instruction to perform re-recording to the outside is executed. Since the recording operation can be repeated until the value exceeds a predetermined value, it is possible to acquire a recorded sound signal with high sound quality and adapted to the task.

Further, it may be configured to include a fifth step of updating the voice data stored in the voice synthesis database by the voice signal recorded using the recording instruction data output in the fourth step described above. .
For this reason, according to this, the voice recording system performs the iterative process using the updated voice synthesis system by executing the fifth step of updating the voice synthesis system using the recorded voice signal, and performs the synthesis. It is possible to increase the quality of voice and the fitness of tasks.

Further, the fifth step described above may include a sub-step of determining unnecessary speech data to be deleted from the speech synthesis database based on the evaluation result in the fourth step.
For this reason, according to this, the sub step in the fifth step in which the voice recording system determines unnecessary voice data to be deleted from the voice synthesis database based on the evaluation result output by executing the fourth step. By executing the above, since the speech data is additionally stored in the speech synthesis database and unnecessary data is deleted, the enlargement of the speech synthesis database can be avoided.

Further, the fifth step described above may include a sub-step of updating the voice data stored in the voice synthesis database based on the unnecessary data determined in the fourth step.
For this reason, according to this, the speech synthesis system updates the speech data stored in the speech synthesis database based on the unnecessary data determined by executing the fourth step. By executing, unnecessary data is deleted, so that it is possible to avoid the enlargement of the speech synthesis database.

Further, in the quality evaluation processing function described above, the synthesized speech and the input speech signal corresponding to the input text data are compared to evaluate the degree of coincidence of the feature amount, and based on this, the desired quality is obtained. A sub-processing function for determining whether or not the condition is satisfied may be included, and this may be executed by a computer.
Therefore, according to this, by comparing the synthesized speech obtained by synthesizing the input text data with the speech signal corresponding to the input text data, the characteristics that cannot be expressed by the existing speech synthesis system are clarified. Therefore, it becomes possible to create voice data more adapted to the task.

In addition, in the additional storage data determination function described above, the recording instruction data including text data describing the utterance content of the audio signal to be recorded corresponding to the audio data determined by the additional storage data determination function is generated. An instruction data generation function is provided, and the computer executes the function.
Therefore, according to this, the computer generates the recording instruction data including the text data describing the utterance content of the audio signal to be recorded corresponding to the audio data determined to be additionally stored. Recording of audio data when updating the synthesis system becomes easy.

  In the quality evaluation processing function described above, the input speech signal corresponding to the input text data is acquired, and the segmentation data including at least the phoneme string associated with the time information is extracted from the input speech signal. Matching with time information using the processing function and the segmentation data, comparing the synthesized speech and the input speech signal, evaluating the similarity of the feature amount, and determining whether or not the desired quality is satisfied. Sub processing functions may be included and executed by the computer.

  Therefore, according to this, the computer extracts segmentation data including a phoneme string associated with time information from the input speech signal corresponding to the input text data, and uses the segmentation data to match the time information. By comparing the synthesized speech and the input speech signal to evaluate the similarity of the feature amount, segmentation data such as duration information can be used for comparison between the synthesized speech and the input speech signal, and as a result This makes it possible to evaluate the quality of synthesized speech more precisely and in detail.

Further, in the quality evaluation processing function described above, a sub-voice that acquires an input speech signal corresponding to the input text data and extracts segmentation data including at least a phoneme string associated with time information from the input speech signal. And the above-described speech synthesis processing function includes a sub-step of generating synthesized speech having time information corresponding to the input speech signal using the extracted segmentation data. You may make it perform.
Therefore, according to this, since the computer generates the synthesized speech according to the duration length information indicated by the segmentation data of the audio signal, the comparison operation between the synthesized speech and the input audio signal at the time of quality evaluation is not required. Can do.

  Further, in the recording instruction data generation function described above, a sub-processing function for evaluating the recorded voice signal by comparing the input voice signal corresponding to the input text data with the recorded voice signal, and the evaluation A sub-processing function that outputs recording instruction data for re-recording when the evaluation value obtained as a result is less than the threshold may be included, and this may be executed by a computer.

  Therefore, according to this, the computer compares the input voice signal corresponding to the input text data with the recorded voice signal, and ends the process when the evaluation value obtained as a result of the comparison exceeds a predetermined value, Since the recording operation can be repeated until the evaluation value of the recorded voice by the speaker exceeds the predetermined value by outputting an instruction to perform re-recording outside when the predetermined value is not reached, the sound quality is high, and Recorded audio signal adapted to the task can be acquired.

A speech synthesis system update processing function for updating speech data stored in the speech synthesis database using speech signals recorded according to the recording instruction data generated by the recording instruction data generation function; You may make it make a computer perform these along with the generation function.
Therefore, according to this, when the computer updates the speech synthesis system using the recorded speech signal, it performs an iterative process using the updated speech synthesis system, thereby improving the quality of the synthesized speech and the fitness of the task. It becomes possible to increase.

  Further, the above-described speech synthesis system update processing function includes a sub-processing function for determining unnecessary speech data to be deleted from the speech synthesis database based on the evaluation result in the recording instruction data generation function, May be executed by a computer.

  Therefore, according to this, the computer determines unnecessary speech data to be deleted from the speech synthesis database based on the evaluation result, and additionally stores the speech data in the speech synthesis database and deletes the unnecessary data. Therefore, the enlargement of the speech synthesis database can be avoided.

The speech synthesis system update processing function includes a sub-processing function for updating speech data stored in the speech synthesis database based on unnecessary data determined by the recording instruction data generation function. May be executed by a computer.
Therefore, according to this, unnecessary data is deleted by updating the speech data stored in the speech synthesis database by the computer based on the determined unnecessary data, thus avoiding the enlargement of the speech synthesis database. can do.

While the present invention has been described with reference to the embodiments (and examples), the present invention is not limited to the above embodiments (and examples). Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

  This application claims the priority on the basis of Japanese application Japanese Patent Application No. 2006-300824 for which it applied on November 6, 2006, and takes in those the indications of all here.

  The present invention is effectively applied to, for example, an application for constructing a speech synthesis system with an emphasis on a specific task such as emergency broadcasting or reading a story. In addition, since it is possible to synthesize voice sounds in accordance with the place of use, it is possible to make phrases adapted to each of tourist guidance, emergency reports from government offices, event guidance, museum guidance, etc. Further, since it is possible to periodically upgrade the speech synthesis system in response to a user request, there is a high versatility.

It is a block diagram which shows the structure of the audio | voice recording system concerning 1st Embodiment of this invention. It is a flowchart which shows operation | movement of 1st Embodiment disclosed in FIG. It is a block diagram which shows the structure of the audio | voice recording system concerning 2nd Embodiment of this invention. It is a flowchart which shows operation | movement of 2nd Embodiment disclosed in FIG. It is a block diagram which shows the structure of the audio | voice recording system concerning the 3rd Embodiment of this invention. It is a flowchart which shows operation | movement of 3rd Embodiment disclosed in FIG. It is a block diagram which shows the structure of the audio | voice recording system concerning 4th Embodiment of this invention. It is a flowchart which shows operation | movement of 4th Embodiment disclosed in FIG. It is a block diagram which shows the structure of the audio | voice recording system concerning 5th Embodiment of this invention. It is a flowchart which shows operation | movement of 5th Embodiment disclosed in FIG. It is a block diagram which shows the structure of the audio | voice recording system concerning 6th Embodiment of this invention. It is a flowchart which shows operation | movement of 6th Embodiment disclosed in FIG. It is a block diagram which shows the structure of the audio | voice recording system concerning 7th Embodiment of this invention. It is a flowchart which shows operation | movement of 7th Embodiment disclosed in FIG. It is a block diagram which shows the structure of the audio | voice recording system concerning 8th Embodiment of this invention. It is a flowchart which shows operation | movement of 8th Embodiment disclosed in FIG. It is a block diagram which shows the structure of the audio | voice recording system concerning 9th Embodiment of this invention. It is a flowchart which shows operation | movement of 9th Embodiment disclosed in FIG. It is a block diagram which shows the specific structure of Example 1 in this invention. FIG. 20A is an explanatory diagram illustrating an example of evaluation values, and FIG. 20B is an explanatory diagram illustrating an example of text assembling. It is a block diagram which shows the specific structure of Example 2 in this invention. FIGS. 22A and 22B are diagrams showing a third embodiment of the present invention, FIG. 22A is a block diagram showing a specific configuration thereof, and FIG. 22B is an explanation showing an example of a description method of duration of each phoneme in a segmentation description method. FIG. It is a block diagram which shows the specific structure of Example 4 in this invention. It is a block diagram which shows the specific structure of Example 5 in this invention. It is a block diagram which shows the specific structure of Example 6 in this invention. It is explanatory drawing which shows the further specific example of Example 6 in this invention. It is explanatory drawing quoted in order to demonstrate operation | movement of Example 7 corresponding to 9th Embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10,40 Speech synthesis system 10A, 40A, 60A Speech synthesis database 11 Speech synthesis unit 12, 22, 32, 42, 62, 72, 82 Quality evaluation unit 13 Additional data determination unit 14 Recording instruction data generation unit 14A For text generation Storage unit 15, 45 Segmentation data extraction unit 16 Recorded speech signal evaluation unit 17 Text database for sentence generation 18, 88, 98 Speech synthesis system update unit 19 Unnecessary data determination unit 20 Speech recording system 21 Updated speech synthesis system 30 Speech recording machine

Claims (43)

A voice recording system for recording voice signals necessary to create voice data such as parameters and segment waveforms in a voice synthesis database,
A speech synthesizer that inputs text data describing the generated content and generates a desired synthesized speech based on speech data stored in advance in the speech synthesis database; and the synthesized speech based on the feature amount of the synthesized speech And a quality evaluation unit that performs quality evaluation of
An audio recording system, comprising: an additional data determining unit that determines audio data to be additionally stored in the database for speech synthesis based on a result of the quality evaluation. In the audio recording system according to claim 1,
The quality evaluation unit compares the synthesized speech with an input speech signal corresponding to the input text data, evaluates the degree of coincidence of feature amounts, and determines whether or not a predetermined desired quality is satisfied. An audio recording system characterized by having In the audio recording system according to claim 1,
The quality evaluation unit analyzes and evaluates intermediate data generated by the speech synthesis unit when speech synthesis is performed on the input text data, and based on this, the speech data stored in the speech synthesis database is desired. An audio recording system comprising a function for judging whether or not quality is satisfied. A speech synthesis database for storing speech data;
Analyzing the input text data and the speech synthesis system information defining the processing operation related to the synthesis of speech data stored in the speech synthesis database, and based on this, the speech data stored in the speech synthesis database is A quality evaluation unit that determines whether or not the desired quality is satisfied;
An additional data determination unit that determines voice data to be additionally stored in the speech synthesis database based on the result of the quality evaluation;
An audio recording system characterized by comprising In the audio recording system according to any one of claims 1 to 4,
The input text data includes format specification data describing a speech format, and the format specification data includes at least one of data indicating a pitch pattern and an utterance speed. . In the audio recording system according to any one of claims 1 to 5,
The speech recording system, wherein the speech synthesis database stores at least one of a pitch pattern model, a duration model, and a speech segment waveform as the stored speech data. In the audio recording system according to any one of claims 1 to 6,
The voice recording system, wherein the voice data determined by the additional data determination unit includes at least one of a pitch pattern parameter, a duration parameter, and a voice segment waveform. In the audio recording system according to any one of claims 1 to 7,
Corresponding to the audio data determined by the additional data determining unit, a recording instruction data generating unit for generating recording instruction data including text data in which the utterance content of the audio signal to be recorded is described, the additional data determining unit An audio recording system characterized by being attached to the In the audio recording system according to claim 8,
The recording instruction data generation unit has a function of calculating the total time of the audio signal recorded from the text data in which the utterance content of the audio signal to be recorded is included, included in the recording instruction data, and the calculation result The sound recording system is characterized by having a function of suppressing the amount of sound signals to be recorded according to the conditions. In the audio recording system according to claim 8 or 9,
The recording instruction data generation unit includes a text generation storage unit, and has a function of generating the recording instruction data with reference to the generation text data stored in the text generation storage unit Sound recording system featuring In the audio recording system according to any one of claims 8 to 10,
The recording instruction data generation unit is configured to calculate the amount of audio signals to be recorded from the recording instruction data, and to have a function of changing a satisfaction rate of the additional data according to the calculated amount of audio signals; An audio recording system characterized by The sound recording system according to any one of claims 8 to 11,
The audio recording system, wherein the recording instruction data generation unit has a function of selecting the additional data and creating recording instruction data. In the audio recording system according to any one of claims 10 to 12,
The recording instruction data generation unit includes at least one of the input text data input to the speech synthesizer, the generation text data stored in the text generation storage unit, and the additional data. An audio recording system having a function of generating the recording instruction data satisfying all or part of the additional data. In the audio recording system according to any one of claims 8 to 13,
The recording instruction data generated by the recording instruction data generating unit includes utterance content instruction data for instructing utterance content,
The speech recording system is characterized in that the speech content instruction data includes at least one of text data describing the speech content or synthesized speech having the same generated content. The audio recording system according to any one of claims 8 to 14,
The recording instruction data generated by the recording instruction data generating unit includes utterance style instruction data for instructing a voice style,
The voice recording system is characterized in that the utterance style instruction data includes at least one of data indicating pitch pattern, duration, intonation, and emotion. The sound recording system according to any one of claims 8 to 15,
An audio recording system, wherein the recording instruction data generated by the recording instruction data generating unit includes an input audio signal corresponding to the input text data. In the sound recording system according to any one of claims 2 and 5 to 16,
A segmentation data extraction unit for extracting segmentation data including at least a phoneme string associated with time information of an input speech signal corresponding to the input text data is provided in the quality evaluation unit,
The quality evaluation unit uses the segmentation data to match time information, compares the synthesized speech with the input speech signal, evaluates the degree of coincidence of feature amounts, and determines whether or not a desired quality is satisfied. Audio recording system characterized by having a function to perform. In the sound recording system according to any one of claims 2 and 5 to 16,
A segmentation data extraction unit for extracting segmentation data including at least a phoneme string associated with time information of an input speech signal corresponding to the input text data is provided in the speech synthesis unit,
The voice recording system, wherein the voice synthesizer has a function of generating synthesized voice having time information corresponding to the input voice signal using the segmentation data. The sound recording system according to any one of claims 8 to 18,
An instruction to compare the input voice signal corresponding to the input text data with the newly recorded voice signal, and to perform external recording when the evaluation value obtained as a result of the comparison is less than a predetermined value set in advance The audio recording system is characterized in that a recording audio signal evaluation unit for outputting is attached to the recording instruction data generation unit. The sound recording system according to any one of claims 8 to 19,
A speech synthesis system update unit for updating speech data for a speech synthesis system stored in the speech synthesis database by a speech signal recorded using the recording instruction data is provided in the speech synthesis unit. Audio recording system. The audio recording system according to any one of claims 1 to 19,
An unnecessary data determination unit that estimates unnecessary data in the speech synthesis database and determines unnecessary speech data to be deleted from the speech synthesis database based on the evaluation result in the quality evaluation unit. An audio recording system characterized by the addition of The audio recording system according to claim 21, wherein
The audio recording system, wherein the unnecessary data determined by the unnecessary data determination unit includes at least one of a pitch pattern parameter, a duration parameter, and a speech segment waveform. In the audio recording system according to claim 21 or 22,
A speech recording system, wherein a speech synthesis system updating unit that updates speech synthesis speech data stored in the speech synthesis database based on the unnecessary data is provided in the speech synthesis unit. A voice recording method for recording a voice signal necessary for creating voice data stored in a voice synthesis database,
A first step of referring to the speech synthesis database and generating a desired synthesized speech from externally input text data;
A second step of evaluating the quality of the synthesized speech based on the generated feature amount of the synthesized speech;
A third step of determining speech data to be additionally stored in the speech synthesis database based on the quality evaluation result;
An audio recording method characterized by comprising: The audio recording method according to claim 24,
In the second step, the synthesized speech and the input speech signal corresponding to the input text data are compared to evaluate the degree of feature match, and based on this, whether the synthesized speech satisfies a desired quality. An audio recording method comprising a sub-step for determining whether or not. A voice recording method for recording a voice signal to create voice data stored in a voice synthesis database,
A first step of acquiring externally input text data;
Analyzing and evaluating the acquired input text data and speech synthesis system information that defines processing operations related to synthesis of speech data stored in the speech synthesis database, and based on this, the speech synthesis database A second step of determining whether the stored audio data satisfies a desired quality;
A third step of determining speech data to be additionally stored in the speech synthesis database based on the quality evaluation result;
An audio recording method characterized by comprising: The sound recording method according to any one of claims 24 to 26, wherein:
A voice comprising the fourth step of generating recording instruction data including text data describing the utterance content of the voice signal to be recorded corresponding to the voice data determined in the third step Recording method. The audio recording method according to any one of claims 25 to 27, wherein:
The second step includes
A sub-step of acquiring an input speech signal corresponding to the input text data and extracting segmentation data including at least a phoneme string associated with time information from the input speech signal;
Sub-step for matching the time information using the segmentation data, comparing the synthesized speech and the input speech signal, evaluating the degree of coincidence of the feature values, and determining whether or not the desired quality is satisfied. Including,
An audio recording method characterized by this. The audio recording method according to claim 25 or 27, wherein:
The first step is a sub-step of acquiring an input speech signal corresponding to the input text data and extracting segmentation data including at least a phoneme string associated with time information from the input speech signal. Sub-step of generating synthesized speech having time information corresponding to the input speech signal using the segmentation data;
An audio recording method characterized by including 30. The audio recording method according to any one of claims 27 to 29, wherein:
The fourth step includes a sub-step of evaluating the recorded voice signal by comparing the input voice signal corresponding to the input text data and the recorded voice signal, and an evaluation value obtained as a result of the evaluation is obtained in advance. A sub-step for outputting recording instruction data for re-recording when the set threshold value is not met,
An audio recording method characterized by including: The audio recording method according to any one of claims 27 to 30, wherein
Voice recording characterized by comprising a fifth step of updating the voice data stored in the voice synthesis database with the voice signal recorded using the recording instruction data output in the fourth step. Method. The audio recording method according to claim 31, wherein
5. The voice recording method according to claim 5, wherein the fifth step includes a sub-step of determining unnecessary voice data to be deleted from the voice synthesis database based on the evaluation result in the fourth step. The audio recording method according to claim 32, wherein
The voice recording method according to claim 5, wherein the fifth step includes a sub-step of updating voice data stored in the voice synthesis database based on the unnecessary data determined in the fourth step. A voice recording processing program used in a voice recording system for recording voice signals for creating voice data stored in a voice synthesis database,
A speech synthesis processing function for referring to the speech synthesis database and generating desired synthesized speech from externally input text data;
A quality evaluation processing function for evaluating the quality of the synthesized speech based on the feature amount of the generated synthesized speech;
An additional data determination processing function for determining speech data to be additionally stored in the speech synthesis database based on the result of the quality evaluation;
An audio recording processing program characterized in that a computer is executed. In the sound recording processing program according to claim 34,
The quality evaluation processing function compares the synthesized speech and an input speech signal corresponding to the input text data to evaluate the degree of coincidence of feature amounts, and determines whether or not a desired quality is satisfied based on the comparison. An audio recording processing program characterized by including a sub-processing function. A voice recording processing program for recording a voice signal to create voice data stored in a voice synthesis database,
Text data acquisition processing function to acquire externally input text data,
The acquired input text data and the speech synthesis system information that defines the speech synthesis processing operation stored in the speech synthesis database are analyzed and evaluated, and stored in the speech synthesis database based on the analysis. A quality evaluation processing function for determining whether the audio data satisfies a desired quality;
An additional data determination processing function for determining speech data to be additionally stored in the speech synthesis database based on the quality evaluation result;
An audio recording processing program characterized in that a computer is executed. The audio recording processing program according to any one of claims 34 to 36,
Recording instruction data generation for generating recording instruction data including text data describing the utterance content of the audio signal to be recorded corresponding to the audio data determined by the additional storage data determination function in the additional storage data determination function Along with the function,
An audio recording processing program characterized by causing a computer to execute these. In the sound recording processing program according to any one of claims 35 to 37,
The quality evaluation processing function
A sub-processing function for obtaining an input speech signal corresponding to the input text data, extracting segmentation data including at least a phoneme string associated with time information from the input speech signal, and time information using the segmentation data A sub-processing function that compares the synthesized speech and the input speech signal, evaluates the similarity of the feature amount, and determines whether or not a desired quality is satisfied,
An audio recording processing program characterized by causing a computer to execute these. In the sound recording processing program according to claim 35 or 37,
The quality evaluation processing function includes a sub-step of acquiring an input speech signal corresponding to the input text data and extracting segmentation data including at least a phoneme string associated with time information from the input speech signal;
The speech synthesis processing function includes a sub-step of generating synthesized speech having time information corresponding to the input speech signal using the extracted segmentation data,
An audio recording processing program characterized by causing a computer to execute these. 40. The sound recording processing program according to any one of claims 37 to 39,
The recording instruction data generation function includes a sub-processing function that evaluates the recorded voice signal by comparing an input voice signal corresponding to the input text data and the recorded voice signal, and an evaluation value obtained as a result of the evaluation. Including a sub-processing function that outputs recording instruction data for re-recording when the threshold value is not met,
An audio recording processing program characterized by causing a computer to execute these. In the sound recording processing program according to any one of claims 37 to 40,
A speech synthesis system update processing function for updating speech data stored in the speech synthesis database using speech signals recorded according to the recording instruction data generated by the recording instruction data generation function, and the recording instruction data generation function Attached to the
An audio recording processing program characterized by causing a computer to execute these. In the sound recording processing program according to claim 41,
The speech synthesis system update processing function includes a sub-processing function for determining unnecessary speech data to be deleted from the speech synthesis database based on an evaluation result in the recording instruction data generation function,
An audio recording processing program characterized by causing a computer to execute these. In the sound recording processing program according to claim 42,
The speech synthesis system update processing function includes a sub-processing function for updating speech data stored in the speech synthesis database based on unnecessary data determined in the recording instruction data generation function,
An audio recording processing program characterized by causing a computer to execute these.

Images