JP2004037605A - Method, apparatus, and program for deleting data for speech synthesis - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method, an apparatus, and a program for deleting data for speech synthesis which can make clear a unit speech that a speech database holds and keep the speech database small in capacity to improve a processing speed of speech synthesis. <P>SOLUTION: A speech synthesis data deleting apparatus 1 that deletes data for speech synthesis in the speech database which is used for speech synthesis and in which speaking times of phonemes and words are recorded is equipped with a use frequency recording part 5 for synthesis data in which the use frequency of data for speech synthesis used for speech synthesis is recorded and a deletion part 7 for speech data, which deletes speech synthesis data whose use frequency is less than a previously set use frequency threshold. <P>COPYRIGHT: (C)2004,JPO

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to reduction of data for speech synthesis used for speech synthesis.
[0002]
[Prior art]
2. Description of the Related Art Conventionally, there are the following methods (devices) for speech synthesis, for example.
(1) Speech synthesis method (JP-A-2-47700)
The speech synthesizer disclosed in this publication is provided with a speech database comprising speech data in which the speech times of unit speech (phonemes or words) are recorded, and the speech synthesizer is input to the speech synthesizer. After decomposing the sentence into unit voices, a search of a voice database is performed for each of the decomposed unit voices, and phonetic synthesis and tone correction are performed on the obtained voice synthesis data to perform voice synthesis. .
[0003]
(2) Spontaneous speech waveform signal connection type speech synthesizer (JP-A-10-49193)
The speech synthesizer disclosed in this publication is provided with a speech database in which speech times of phonemes are recorded, and the speech synthesizer decomposes a sentence input into the device into a phoneme sequence, The speech database is searched for each phoneme of the decomposed phoneme sequence to perform speech synthesis.
[0004]
[Problems to be solved by the invention]
However, the conventional speech synthesizer does not specify what unit speech (phoneme or word) the speech database holds. In addition, in any of the speech synthesizers, when the data amount of the speech database is large, the number of speech synthesis data that is candidates for speech synthesis increases, the search time becomes enormous, and the processing speed decreases.
[0005]
Therefore, an object of the present invention is to solve the above-mentioned problems of the conventional technology, to clarify the unit speech held by the speech database, and to reduce the speech database in order to improve the processing speed in speech synthesis. It is an object of the present invention to provide a voice synthesis data reduction method, a voice synthesis data reduction device, and a voice synthesis data reduction program that can be held in a capacity.
[0006]
[Means for Solving the Problems]
The present invention has the following configuration to achieve the above object.
The method for reducing data for speech synthesis according to claim 1, which is used for speech synthesis, is used for speech synthesis of phonemes and words that are used infrequently in a speech database in which speech times of phonemes and words are recorded. A method for reducing data for speech synthesis for reducing data, wherein a use frequency recording step of recording the use frequency of the speech synthesis data used when performing speech synthesis, and a use frequency lower than a preset use frequency threshold And a voice synthesis data reduction step of reducing voice synthesis data.
[0007]
According to this method, in the use frequency recording step, each time the speech synthesis is performed by the speech synthesis device or the like, the usage frequency of the speech synthesis data in the speech database inherent in the device or the like is recorded. Then, in the voice synthesizing data reduction step, voice synthesizing data whose use frequency is lower than a preset use frequency threshold is reduced. The synthesized speech data is composed of phonemes and words, and is the minimum unit for speech synthesis. The preset use frequency threshold is a numerical value that can be set arbitrarily. For example, assuming this use frequency threshold based on the number of times the speech synthesizer has been used, "use count 50 times use frequency threshold 1", that is, , The data for voice synthesis that is never used even if the voice synthesis is executed 50 times can be set as a reduction target. This method of reducing data for speech synthesis is applicable irrespective of the method of speech synthesis in the speech synthesis device as long as at least a speech database exists in the device or the like.
[0008]
According to a second aspect of the present invention, in the method for reducing data for speech synthesis according to the first aspect, the speech database is configured as a set of sentences composed of the data for speech synthesis included in the speech database. In the use frequency recording step, when a sentence in the speech database is subjected to speech synthesis based on speech synthesis data included in all other sentences, use of the used speech synthesis data The frequency is recorded.
[0009]
According to this method, a plurality of sentences composed of speech synthesis data are stored in the speech database, and for each of these sentences, in the use frequency recording step, other sentences stored in the speech database are stored. The usage frequency of the voice synthesis data when voice synthesis is performed using the voice synthesis data included in the voice synthesis data is recorded. In the voice synthesis data reduction step, the voice synthesis use frequency lower than a predetermined usage frequency threshold is used. Data is reduced.
[0010]
A speech synthesis data reduction device according to claim 3, which is used for speech synthesis, is used for speech synthesis of phonemes and words that are used infrequently in a speech database in which speech times of phonemes and words are recorded. A data reduction device for voice synthesis for reducing data, wherein a usage frequency recording means for recording the frequency of use of voice synthesis data used when performing voice synthesis, and a usage frequency lower than a preset usage frequency threshold And voice synthesis data reduction means for reducing voice synthesis data.
[0011]
According to this configuration, the use frequency recording unit records the use frequency of the data for voice synthesis in the voice database included in the voice synthesizer or the like each time the voice synthesis is performed by the voice synthesizer or the like. Then, the voice synthesizing data reducing unit reduces voice synthesizing data whose use frequency is lower than the preset use frequency threshold.
[0012]
According to a fourth aspect of the present invention, in the data reduction apparatus for speech synthesis according to the third aspect, the speech database is configured as a set of sentences composed of speech synthesis data included in the speech database. When the use frequency recording unit executes speech synthesis based on the speech synthesis data included in all the other sentences in the sentences in the speech database, the use of the speech synthesis data used is The frequency is recorded.
[0013]
According to such a configuration, a plurality of sentences composed of speech synthesis data are stored in the speech database, and each of these sentences is stored in the other sentences stored in the speech database by the use frequency recording unit. The frequency of use of the voice synthesis data when voice synthesis is performed using the included voice synthesis data is recorded, and the voice synthesis data reduction unit detects voice synthesis data having a frequency of use lower than a preset use frequency threshold. Be reduced.
[0014]
A speech synthesis data reduction program according to claim 5, which is used for performing speech synthesis, and is used for a speech synthesis consisting of phonemes and words which are used infrequently in a speech database in which speech times of phonemes and words are recorded. It is characterized in that an apparatus for reducing data functions as the following means. Means for causing the apparatus to function include a use frequency recording means for recording the use frequency of the speech synthesis data used when performing speech synthesis, and reducing the speech synthesis data having a use frequency lower than a preset use frequency threshold. Means for reducing data for speech synthesis.
[0015]
According to this configuration, the use frequency recording unit records the use frequency of the data for voice synthesis in the voice database included in the voice synthesizer or the like each time the voice synthesis is performed by the voice synthesizer or the like. Then, the voice synthesizing data reducing unit reduces voice synthesizing data whose use frequency is lower than the preset use frequency threshold.
[0016]
The data reduction program for speech synthesis according to claim 6 is the data reduction program for speech synthesis according to claim 5, wherein the speech database is configured as a set of sentences composed of speech synthesis data included in the speech database. When the use frequency recording unit executes speech synthesis based on the speech synthesis data included in all the other sentences in the sentences in the speech database, the use of the speech synthesis data used is The frequency is recorded.
[0017]
According to such a configuration, a plurality of sentences composed of speech synthesis data are stored in the speech database, and each of these sentences is stored in the other sentences stored in the speech database by the use frequency recording unit. The frequency of use of the voice synthesis data when voice synthesis is performed using the included voice synthesis data is recorded, and the voice synthesis data reduction unit detects voice synthesis data having a frequency of use lower than a preset use frequency threshold. Be reduced.
[0018]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings.
(Configuration of data reduction device for speech synthesis)
FIG. 1 is a block diagram of a speech synthesis data reduction system. As shown in FIG. 1, the voice synthesis data reduction system includes a voice synthesis data reduction device 1 and a voice synthesis device 2. The voice synthesis data reduction device 1 includes an input / output unit 3, a synthesis data The apparatus includes a use frequency recording unit 5, a synthesis data reduction unit 7, and a recording unit 9.
[0019]
The speech synthesizer 2 includes a speech database 4, and outputs a speech data string (speech synthesis result) from input text data. The unit speech (speech synthesis data) held in the speech database 4 of the speech synthesizer 2 is based on "words". In this embodiment, a "sentence" including a plurality of words is configured in the database. It is a unit. Each sentence is assigned a “sentence number”, and the speech time of each word is stored. It should be noted that other configurations of the speech synthesizer 2, for example, the manner of speech synthesis are not directly related to the present invention, and therefore, illustration and description are omitted.
[0020]
The speech synthesis data reduction device 1 is used to reduce the data amount of the speech database 4 used each time speech synthesis is performed in the speech synthesis device 2, that is, speech synthesis data (words). By reducing the data amount of the speech database 4 by the data reduction device 1, the search time for speech synthesis candidates (word candidates) at the time of speech synthesis can be significantly reduced.
[0021]
The input / output unit 3 is for transmitting and receiving information (synthesis data use frequency information and synthesis data reduction information described later) to and from the speech synthesizer 2. The input / output unit 3 is configured to transmit and receive information via a communication network such as the Internet (not shown).
[0022]
The synthesis data use frequency recording unit 5 records synthesis data use frequency information, which is information on the use frequency of unit speech (speech synthesis data) used each time speech synthesis is performed in the speech database 4 of the speech synthesizer 2. This is recorded in the unit 9. In other words, it can be said that the synthesis data use frequency recording unit 5 acquires the synthesis data use frequency information from the speech synthesis device 2 when the speech is synthesized by the speech synthesis device 2.
[0023]
In the synthesis data use frequency recording unit 5, for example, a test set for reducing the voice database (text data that is frequently synthesized) is prepared in the voice synthesis device 2, and the text (text data) of this test set is prepared. The result of speech synthesis (synthesis data use frequency information) when is input is recorded. In this embodiment, since the constituent unit of the speech database 4 of the speech synthesizer 2 is a sentence, the synthesis data use frequency information records which word of which sentence was used.
[0024]
When the speech synthesis data in the speech database 4 of the speech synthesizer 2 is recorded in units of sentences, the synthesis data use frequency recording unit 5 converts one sentence into another sentence other than the sentence. The synthesis unit usage frequency information, which is information on the usage frequency of the voice synthesis data when the voice synthesis is performed, is recorded in the recording unit 9. The synthesizing data use frequency recording unit 5 corresponds to a voice synthesizing data use frequency recording unit described in claims.
[0025]
The combining data reducing unit 7 compares the combining data use frequency information recorded in the recording unit 9 by the combining data use frequency recording unit 5 with a use frequency threshold (total use frequency threshold), and If it is smaller than the total usage frequency threshold), the synthesis data reduction information, which is information for reducing the voice synthesis data included in the synthesis data usage frequency information, is transmitted via the input / output unit 3 to the voice. This is output to the synthesizing device 2. In the voice synthesizing device 2 which has received the synthesis data reduction information, the corresponding voice synthesis data in the voice database 4 is reduced. The synthesizing data reducing section 7 corresponds to the voice synthesizing data reducing means described in the claims.
[0026]
The use frequency threshold is a numerical value that can be arbitrarily set in advance. For example, assuming that the use frequency threshold is based on the number of uses of the speech synthesizer, “use number 50 use frequency threshold 1”, that is, 50 times Even if voice synthesis is executed, data for voice synthesis that is never used can be set as a reduction target.
[0027]
In addition, the total use frequency threshold is set such that, when the synthesis data in the speech database 4 of the speech synthesis device 2 is recorded in units of sentences, one sentence is included in the speech synthesis data included in other sentences other than the sentence. This is for comparison with the synthesis data use frequency information, which is information on the use frequency of the speech synthesis data when the speech synthesis is performed.
[0028]
The recording unit 9 is composed of a general hard disk or the like, and records synthesis data use frequency information and a use frequency threshold (total use frequency threshold). Note that the synthesis data use frequency information is updated in the recording unit 9 in a format in which a numerical value is added to the data use frequency set for each voice synthesis data (+1 when used once). .
[0029]
According to the speech synthesis data reduction device 1, the speech data in the speech database 4 existing in the speech synthesis device 2 is used every time speech synthesis is performed by the speech synthesis device 2 in the synthesis data use frequency recording unit 5. The usage frequency of the data is recorded. Then, the synthesis data reducing unit 7 reduces the voice synthesis data whose use frequency is lower than the preset use frequency threshold. For this reason, when the unit speech (speech synthesis data) held in the speech database 4 of the speech synthesis device 2 is a word, the speech database 4 used for speech synthesis is used in the speech database 4 which is used less frequently. By appropriately deleting the synthesizing data (words), the voice database 4 can be kept compact (small volume), and the processing speed when performing voice synthesis by the voice synthesizing device 2 can be improved.
[0030]
The speech database 4 of the speech synthesizer 2 stores a plurality of sentences composed of speech synthesis data (words). Each of these sentences is stored in the speech database 4 by the synthesis data use frequency recording unit 5. The usage frequency of the voice synthesis data when voice synthesis is performed using the other sentence is recorded, and the voice data synthesis unit 7 uses the voice synthesis data having a usage frequency lower than a preset total usage frequency threshold. Data is reduced. For this reason, the speech synthesis data (word) in one sentence of the speech database 4 is synthesized by speech synthesis using another sentence, and the use frequency is recorded. If the use frequency is low, it is deleted. It is possible to automatically (self-learning) maintain the data amount of the voice database 4 at a small capacity, and improve the processing speed when the voice synthesizer 2 synthesizes voice.
[0031]
(Operation of the data reduction device for speech synthesis)
Next, the operation of the data reduction device 1 for speech synthesis will be described with reference to the flowchart shown in FIG. 2 (see FIG. 1 as appropriate).
First, data use frequency information for synthesis is input to the input / output unit 3 (S1). In the description of the operation of the speech synthesis data reducing device 1, the speech synthesis is performed in the speech synthesis device 2 for a certain period, and the synthesis data use frequency information held in the speech synthesis device 2 is periodically updated. It is to be input to the reduction device 1. When the synthesis data use frequency information is output from the input / output unit 3 to the synthesis data use frequency recording unit 5, the synthesis data use frequency recording unit 5 records the synthesis data use frequency information in the recording unit 9. Is performed (S2).
[0032]
Then, the data-for-synthesis reduction unit 7 compares the frequency-of-use data for use with the frequency of use with the frequency-of-use threshold, and determines whether or not the frequency-of-use data for use for synthesis (use frequency in FIG. 2) is less than the frequency-of-use threshold. Is performed (S3). If it is determined that the synthesis data use frequency information is less than the use frequency threshold (S3, Yes), the synthesis data reduction information is output to the input / output unit 3, and the input / output unit 3 sends the information to the speech synthesizer 2. It is output (S4). After that, the operation of the data reduction device 1 for speech synthesis ends. If it is not determined that the synthesis data use frequency information is less than the use frequency threshold (S3, No), the operation of the speech synthesis data reduction device 1 is terminated.
[0033]
(Specific example of speech synthesis data reduction by speech synthesis data reduction device)
Next, a case in which the data for speech synthesis is reduced (operated) by the data reduction system for speech synthesis will be specifically described.
In an actual operation example, a “speech database reduction test set” is input to the speech synthesizer 2 as a sentence of text data, and speech synthesis is performed. This "test set for reducing voice database" is dozens of types of sentences (text data) randomly extracted from text data that is frequently subjected to voice synthesis in news programs and the like. That is, the speech synthesis data (words) in the speech database 4 used when the “test set for speech database reduction” is input is frequently used even when speech is synthesized in ordinary Japanese. Is likely to be
[0034]
First, when the “speech database reduction test set” is input by the voice synthesis device 2, voice synthesis is performed, and the synthesis data use frequency recording unit 5 of the voice synthesis data reduction device 1 uses the voice synthesis device 2. Which sentence of which sentence in the voice database 4 is used is obtained from the synthesis data use frequency information, and is recorded in the recording unit 9.
[0035]
For example, if "<Sentence> Next news <End of sentence>" is input as "Test set for voice database reduction",
"<First sentence>next" ... 0ms to 1000ms of sentence number 50
"News": Sentence number 8, 2125ms-2840ms
"Is <End of sentence>" ... Sentence number 32: 1500ms to 2000ms
Is used (assuming that the data is included in the synthesis data use frequency information).
[0036]
At this time, the data-for-synthesis use frequency recording unit 5 reads "" at the beginning of the sentence number, "Next", "" News "at the sentence number 8, and""" at the sentence number 32. "1 is added to the data use frequency (recording unit 9) of the". ". Each time the next test set (text) is input, the data use frequency recording unit 5 When all test sets are completed, the frequency of data use is low, that is, the frequency less than the frequency of use threshold is reduced by the data-for-synthesis reduction unit 7.
[0037]
According to this specific example, the speech database 4 whose speech synthesis data has been reduced by the speech synthesis data reduction device 1 based on the synthesis data use frequency information generated by the “speech database reduction test set”. Although the range of the variation of the speech database, that is, the variation of the Japanese language to be synthesized is narrowed, the data amount can be suppressed to be small (small volume) and the processing speed at the time of speech synthesis can be improved.
[0038]
As described above, the present invention has been described based on one embodiment, but the present invention is not limited to this.
For example, the processing of each component of the voice synthesis data reduction device 1 is regarded as a voice synthesis data reduction method that is regarded as one process, and the processing of each component is described in a general-purpose computer language. It can be considered a data reduction program. In these cases, the same effect as that of the data reduction device 1 for speech synthesis can be obtained.
[0039]
【The invention's effect】
According to the first, third, and fifth aspects of the present invention, each time speech synthesis is performed by a voice synthesis device or the like, the frequency of use of voice synthesis data in a voice database that is inherent in the voice synthesis device or the like is recorded. Speech synthesis data that is used less frequently than the preset use frequency threshold is reduced. For this reason, when the unit speech held in the speech database of the speech synthesizer or the like is a word, in the speech database used for speech synthesis, it is necessary to appropriately delete the infrequently used speech synthesis data. Thus, the voice database can be maintained in a small capacity, and the processing speed when performing voice synthesis by the voice synthesizer can be improved.
[0040]
According to the second, fourth and sixth aspects of the present invention, when a plurality of sentences composed of speech synthesis data (words) are stored in the speech database, each of these sentences is stored in the speech database. The usage frequency of the voice synthesis data when voice synthesis is performed using other sentences is recorded, and voice synthesis data having a usage frequency lower than a preset total usage frequency threshold is reduced. For this reason, the speech synthesis data in one sentence of the speech database is speech-synthesized using another sentence, and the use frequency is recorded. If the use frequency is low, the data is deleted. The data amount of the speech database can be kept small (self-learning), and the processing speed when speech is synthesized by the speech synthesizer can be improved.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a data reduction system for speech synthesis according to an embodiment of the present invention.
FIG. 2 is a flowchart for explaining the operation of the speech synthesis data reducing device shown in FIG. 1;
[Explanation of symbols]
REFERENCE SIGNS LIST 1 voice synthesis data reduction device 2 voice synthesis device 3 input / output unit 4 voice database 5 synthesis data usage frequency recording unit 7 synthesis data reduction unit 9 recording unit

Claims (6)

A speech synthesis data reduction method for reducing speech synthesis data composed of phonemes and words, which is used less frequently in a speech database in which speech times of phonemes and words used for speech synthesis are recorded. ,
A use frequency recording step of recording the use frequency of the speech synthesis data used when performing speech synthesis;
A voice synthesis data reduction step of reducing voice synthesis data whose use frequency is lower than a use frequency threshold set in advance;
A method for reducing data for speech synthesis, comprising: The speech database is configured as a set of sentences composed of speech synthesis data included in the speech database,
In the use frequency recording step, when a sentence in the speech database is subjected to speech synthesis based on speech synthesis data included in all other sentences, a use frequency of the used speech synthesis data is recorded. 2. The method for reducing data for speech synthesis according to claim 1, wherein: A speech synthesis data reduction device for reducing speech synthesis data consisting of phonemes and words, which is used in a speech database in which the speech times of phonemes and words used for speech synthesis are recorded. ,
Usage frequency recording means for recording the frequency of use of voice synthesis data used when performing voice synthesis,
Voice synthesis data reducing means for reducing voice synthesis data having a frequency of use lower than a preset use frequency threshold,
An apparatus for reducing data for speech synthesis, comprising: The speech database is configured as a set of sentences composed of speech synthesis data included in the speech database,
The use frequency recording unit records the use frequency of the used voice synthesis data when the text in the voice database is subjected to voice synthesis based on the voice synthesis data included in all other texts. 4. The apparatus for reducing data for speech synthesis according to claim 3, wherein: A device for reducing speech synthesis data consisting of phonemes and words, which is used less frequently in a speech database in which speech times of phonemes and words are used when performing speech synthesis,
Usage frequency recording means for recording the frequency of use of voice synthesis data used when performing voice synthesis,
Voice synthesis data reduction means for reducing voice synthesis data having a frequency of use lower than a preset use frequency threshold,
A data reduction program for speech synthesis characterized by functioning as a computer. The speech database is configured as a set of sentences composed of speech synthesis data included in the speech database,
The use frequency recording means, when one sentence in the speech database is subjected to speech synthesis based on the speech synthesis data included in all other sentences, the use frequency of the used speech synthesis data The data reduction program for speech synthesis according to claim 5, wherein the program is recorded.

Images