JPH0679228B2 - Japanese sentence / speech converter - Google Patents

Description

[Detailed Description of the Invention] <Industrial field of application> This invention is a Japanese sentence that can automatically obtain a synthesized speech corresponding to the input Japanese sentence.
The present invention relates to a voice conversion device.

<Prior Art> A conventional Japanese sentence / speech conversion device converts a normal Japanese word into synthetic speech and generates it (for example, “Research Practical Report” vol.32 no.11 (1983)). ). Therefore, when converting an onomatopoeic word into a synthetic voice by this method, an onomatopoeic word that imitates a sound or a voice is stored in a Japanese dictionary in advance, and for example, the input voice signal is referred to the above Japanese dictionary. By dividing it into words and combining the voice data of the synthesis unit corresponding to the reading of the divided words, according to the time series of the voice synthesis parameters obtained, using a normal voice synthesis means for generating an actual voice waveform. To generate and output synthetic speech.

<Problems to be Solved by the Invention> However, the conventional Japanese sentence / speech conversion device described above is
Since the input onomatopoeia is converted into synthetic speech based on the onomatopoeia stored in the Japanese dictionary in advance, only the synthetic speech of the input onomatopoeia is output, and a more realistic expression is possible. I can't.

Therefore, an object of the present invention is to output an onomatopoeia corresponding to the onomatopoeic word or code as synthetic speech (hereinafter referred to as synthetic onomatopoeia) when the input Japanese character string contains a code representing an onomatopoeic word or an onomatopoeia. , Providing a Japanese sentence / speech conversion device that outputs synthetic speech including more realistic onomatopoeia.

<Means for Solving Problems> In order to achieve the above-mentioned object, the Japanese sentence / speech conversion apparatus of the present invention converts an onomatopoeia contained in an input Japanese character string into an onomatopoeia data dictionary code. An onomatopoeia dictionary,
An onomatopoeia data dictionary that stores onomatopoeia data associated with the above onomatopoeia data dictionary code, and an onomatopoeia synthesizer that converts the above onomatopoeia data dictionary code into onomatopoeia data based on the above onomatopoeia data dictionary and synthesizes a synthetic onsound waveform. With
It is characterized by converting Japanese character strings containing onomatopoeia data dictionary codes or onomatopoeic words into synthetic speech containing onomatopoeia.

<Operation> When a Japanese character string containing an onomatopoeia data dictionary code or an onomatopoeia word is input to the word segmentation processing unit, the above onomatopoeia word is converted into an onomatopoeia data dictionary code by the word segmentation processing unit with reference to the onomatopoeia dictionary. Is output. The converted onomatopoeia data dictionary code or the onomatopoeia data dictionary code included in the Japanese character string is
It is converted into onomatopoeia data by referring to the onomatopoeia data dictionary to generate a synthetic onomatopoeia, and combined with the synthetic speech waveform output from the speech synthesizer based on the Japanese character strings other than onomatopoeic words, as a synthetic speech containing onomatopoeia. Is output. Therefore, the onomatopoeic word or the onomatopoeia data dictionary code in the Japanese character string is automatically converted into a synthetic onomatopoeia, and the synthetic voice enables a realistic expression.

<Embodiment> Hereinafter, the Japanese sentence / speech conversion apparatus of the present invention will be described in detail with reference to the illustrated embodiment.

FIG. 1 is a block diagram of an embodiment of the Japanese sentence / speech conversion apparatus of the present invention, and FIG. 2 is a block diagram of FIG. An example of processing in the case of being performed is shown. Fig. 1 and 2
As shown in the figure, from the input unit 101 to the word division processing unit 102,
When a Japanese kana-mixed sentence containing an onomatopoeic word, "Inu wa wa wa wo wa wa." Is input, the word segmentation processing unit 102 refers to the Japanese dictionary 103 and the onomatopoeic word dictionary 104 to execute the longest matching method that has been conventionally performed. The word segmentation process is performed to divide a sentence into words by using the minimum phrase number method that selects words so that the number of phrases in a sentence or sentence is minimized. At this time, if the divided words include an onomatopoeic word, the onomatopoeia data dictionary code “1” corresponding to the onomatopoeia reading “one-one” stored in the above onomatopoeia word dictionary 104 is output to the onomatopoeia synthesis unit 111. Here, a configuration example of the onomatopoeia dictionary 104 is shown below.

The onomatopoeia reading in the above onomatopoeia dictionary 104 is performed by the word division processing unit 1
It is used when detecting onomatopoeic words in 02.

The word string “/” divided by the word division processing unit 102
The dog / ga / one-one / crowd / "is the word reading processing unit 105.
An isomorphic selection process is performed to distinguish words that are the same but have different readings and accents. At that time, the marker 1 is inserted in the part of the onomatopoeic word "one-one" as a mark that the onomatopoeia is to be inserted later. The word string "/ output from the word reading processing unit 105
Based on the reading of dog / moth / (marker 1) / nak / "
The voice parameter generation unit 107 refers to the voice data dictionary 110 in which the voice data of the synthesis unit is stored to search the voice data of the synthesis unit corresponding to the reading of the word,
The voice data of the synthesis unit are interpolated and combined,
Finally, a voice synthesis parameter time series corresponding to "/ dog / moth / (marker 1) / naku /" is obtained. At that time, the marker 2 for instructing the onomatopoeia synthesis unit 111 to operate is inserted at a position corresponding to the portion of the marker 1.

On the other hand, in the prosody processing unit 106, accents and poses of bunsetsus when the words are chained are set based on the output from the word division processing unit 102. At this time, the marker 1 indicating the onomatopoeia part is inserted like the word reading processing unit 105, and short pauses are inserted before and after the marker 1. The prosody parameter generation unit 108 generates a duration, a pitch pattern, and a power pattern for a synthesis unit for speech synthesis that constitutes the speech synthesis parameter time series output from the speech parameter generation unit 107.

In the voice synthesis unit 109, based on the voice synthesis parameter time series output from the voice parameter generation unit 107 and the duration, pitch pattern and power pattern output from the prosody parameter generation unit 108, the Synthetic speech waveforms of "," Naku "are sequentially generated, and output section 11
Output to 3. On the other hand, in the onomatopoeia synthesis unit 111, the onomatopoeia data dictionary code “1” output from the word division processing unit 102 is converted into onomatopoeia data by referring to the above onomatopoeia data dictionary 112, and corresponds to the onomatopoeia “one-one”. An actual synthetic pseudo-sound waveform is generated and output to the output unit 113.
At that time, the output timings of the voice synthesis unit 109 and the onomatopoeia synthesis unit 111 are set as follows. That is, the voice synthesis unit 109 outputs the synthesized voice waveform of “Inuga” according to the voice synthesis parameter time series output from the voice parameter generation unit 107, and in the next marker 2 (§ in FIG. 2) part. The operation of the voice synthesis unit 109 is stopped, and a signal instructing the operation is output to the onomatopoeia synthesis unit 111. The onomatopoeia synthesizing unit 111 starts its operation in response to this instruction, and generates and outputs a synthetic pseudo-sound waveform corresponding to “one-one”. When the onomatopoeia synthesizing unit 111 finishes the output, it outputs a signal indicating that the output is finished to the voice synthesizing unit 109, and the voice synthesizing unit 109 continues to generate the synthesized speech waveform of "naku" again. In this way, the part of the onomatopoeia in the input Japanese sentence can be replaced with the onomatopoeia, and the synthesized speech can be output. Therefore, the onomatopoeia included in the input Japanese sentence is output as a synthetic onomatopoeia, and more realistic voice expression can be performed.

The second embodiment not only replaces the onomatopoeia with onomatopoeia but also automatically generates onomatopoeia that matches the situation of the sentence according to the content of the sentence, and its block diagram is shown in FIG. In FIG. 3, the difference from FIG. 1 is that a situation analysis unit 314 is added. In addition to the configuration of the first embodiment, the Japanese dictionary 303 includes, in addition to the configuration of the first embodiment, a noun as a subject, a verb as a predicate, an adjective and an adjective as a word related to sound or voice (for example, a noun is dog, car, For verbs, sing, run, make noise, if adjectives are noisy, and if adjective verbs are lively, add a code indicating the type of onomatopoeia to each word.

The situation analysis unit 314 refers to information such as word reading and part-of-speech output from the word division processing unit 302, and a code indicating the type of the onomatopoeia, from a combination of words satisfying the relationship between the subject and the predicate, The onomatopoeia dictionary 304 is searched for a corresponding onomatopoeia data dictionary code, and the obtained onomatopoeia data dictionary code is output to the onomatopoeia synthesis unit 311. On the other hand, in the word reading processing unit 305 and the prosody processing unit 306, the word division processing unit 30
Data in which the marker 1 is inserted at the end of the reading of the word output from 2 is output, and thereafter, the same processing as in the first embodiment is performed. When outputting synthetic speech, the speech synthesis unit 30
When the marker 1 is read after the voice synthesis output process of the sentence is finished in 9, the voice synthesis unit 309 outputs a signal indicating that the voice synthesis output process is finished to the onomatopoeia synthesis unit 311.
Upon receiving this signal, the onomatopoeia synthesizer 311 synthesizes and outputs onomatopoeia.

For example, when the noun "dog" and the verb "bark" are entered, the code indicating the type of the onomatopoeia added to "dog" and "bark" as described above becomes "one-one-one" of the onomatopoeia data dictionary code "1". Is specified. Then, after synthesizing the Japanese sentence using the input "dog" and "bark" into the synthesized voice, put a short pause and synthesize and output the onomatopoeia corresponding to the onomatopoeic word "wan-wan", and then repeat the short pause. Then, the speech synthesis of the next input Japanese sentence is started. The output example is shown below.

Input sentence "I heard a dog howling" Output "Inunotooboe gaikoeta (pose) (dog howling onomatopoeia) (pose) (next sentence)" The third embodiment is input in Fig. 3 Japan Insert a symbol indicating the type of onomatopoeia such as the above onomatopoeia data dictionary code directly in the word sentence, specify the onomatopoeia in the input Japanese sentence in advance, and detect the symbol in the word division processing unit 302,
The situation analysis unit 314 measures the output timing of the onomatopoeia and activates the onomatopoeia synthesis unit 311 to output the onomatopoeia. At that time, the following two types are prepared as the output timing of the synthetic onomatopoeia. That is, one is a method of outputting the synthetic onomatopoeia at the same time as the synthetic speech, and a method of outputting the synthetic onomatopoeia as the background of the outputted synthetic voice. The other is a method of interrupting a designated portion in the synthetic voice and outputting a synthetic onomatopoeia. In the former method of outputting a synthetic onomatopoeia as a background of output synthetic speech, it is carried out by inserting a marker 3 in a portion indicating the onomatopoeia. Marker 3 in the time series of speech synthesis parameters output from the speech parameter generation unit 307
In this part, the voice synthesizer 309 outputs a control signal for instructing the onomatopoeia synthesizer 311 to start up, and the voice synthesizer 30
9 continues to operate in parallel with the onomatopoeia synthesis unit 311 without waiting for the onomatopoeia synthesis unit 311 to complete onomatopoeia synthesis, and the synthesized voice is synthesized and output to the output unit 313. Therefore, the output unit 313 can mix the outputs of the voice synthesis unit 309 and the onomatopoeia synthesis unit 311 and output the mixed voice waveform. On the other hand, in the latter method of outputting a synthetic onomatopoeia by interrupting the output synthetic voice, the marker 3 is similarly inserted in a portion where an onomatopoeia is desired to be inserted. As in the first embodiment, the voice synthesis unit 309 outputs a signal instructing the onomatopoeia synthesis unit 311 to activate at the marker 3 and at the same time stops the operation, and when the output of the onomatopoeia synthesis unit 311 ends, the voice synthesis is performed again. Start output operation.

For example, as the marker 3, a symbol ▽ that mixes and outputs a synthetic onomatopoeia and a synthetic voice at the same time, and a symbol ▼ that interrupts and outputs the synthetic onomatopoeia to the synthetic voice are set. The data dictionary code is shown below. Then, in the input example sentence "▽ 20 sea noises. ▽", the starting point of the onomatopoeic is designated by the preceding symbol ▽, and the subsequent symbol ▽.
Is used to specify the end point of the onomatopoeia, and the onomatopoeia data dictionary code “20”
Specifies an onomatopoeia. Therefore, in the output synthetic voice in this case, while the "Uminarigasuru" and the synthetic voice are output, the synthetic quasi-sound of the sea noise is simultaneously output as the background. In addition, in the input example sentence “Umimeri ▼ 20,1.5”, the symbol ▼ indicates the insertion position of the onomatopoeia, the onomatopoeia data dictionary code “20” specifies the onomatopoeia, and the next number after the comma continues the onomatopoeia. Time is indicated. Therefore,
In this case, the output synthetic speech is "Uminaliga" and "Suru".
Synthetic sound of sea roar is interrupted for 1.5 seconds between synthetic voices of "Uminaliga (synthetic onomatopoeia of sea sound for 1.5 seconds) through"
Is output.

<Effects of the Invention> As is clear from the above, the Japanese sentence / speech conversion device of the present invention includes the onomatopoeia word dictionary, the onomatopoeia data dictionary, and the onomatopoeia synthesis unit, and includes the onomatopoeia words included in the input Japanese character string. Is converted into an onomatopoeia data dictionary code using the above onomatopoeia dictionary, and the converted onomatopoeia data dictionary code or the directly input onomatopoeia data dictionary code is converted into an onomatopoeia data dictionary by the above onomatopoeia synthesis section using the above onomatopoeia data dictionary. Since it is converted to a code to generate a synthetic pseudophone, the onomatopoeic word in the Japanese character string or the onomatopoeia data dictionary code is automatically converted to a synthetic onomatopoeia, which enables a realistic expression by the synthetic voice. .

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention, and FIG.
FIG. 3 is a diagram showing the flow of processing in the above embodiment, and FIG. 3 is a block diagram in an embodiment different from the above embodiment. 101,301 ... Input section, 102,302 ... Word division processing section, 103,303 ... Japanese dictionary, 104,304 ... Onomatopoeia dictionary, 105,305 ... Word reading processing section, 106,306 ... Prosodic processing section, 107,307 ... Voice parameter generating section, 108,308 ... Prosodic parameter generating section , 109, 309 ... Voice synthesis unit, 110, 310 ... Voice data dictionary, 111, 311 ... Onomatopoeia synthesis unit, 112, 312 ... Onomatopoeia data dictionary, 113, 313 ... Output unit, 314 ... Situation analysis unit.

Claims (1)

[Claims] 1. A language processing unit having a word division processing unit, a word reading processing unit, a prosody processing unit and a Japanese dictionary, and a speech synthesis having a prosody parameter generation unit, a speech parameter generation unit, a speech synthesis unit and a speech data dictionary. A Japanese sentence-to-speech conversion device for converting an input Japanese character string into a synthetic voice with a section, and obtains an onomatopoeia data dictionary code, which is a code indicating the type of onomatopoeia data, from the input onomatopoeia Therefore, in order to obtain the onomatopoeia data from the onomatopoeia word dictionary in which the onomatopoeia reading and the above onomatopoeia data dictionary code are stored in association with each other, and the above onomatopoeia data dictionary code,
The onomatopoeia data dictionary in which the onomatopoeia data dictionary code and the onomatopoeia data are stored in association with each other, the above onomatopoeia data dictionary code converted by using the above onomatopoeia word dictionary, or the above-mentioned onomatopoeia data dictionary code which is directly input is the above onomatopoeia data. An onomatopoeia synthesizing unit for converting into onomatopoeia data using a dictionary to generate a synthetic waveform of the onomatopoeia,
A Japanese sentence / speech conversion device characterized by converting an input Japanese character string containing an onomatopoeia data dictionary code or an onomatopoeic word into a synthetic speech containing an onomatopoeia.