JPH0679230B2 - Japanese sentence / speech converter - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION <Industrial field of application> The present invention relates to a Japanese sentence / speech conversion device capable of automatically obtaining a synthetic speech corresponding to the above Japanese sentence from an input Japanese sentence.

<Prior Art> Conventionally, there are the following Japanese sentence / speech converters (Information Processing Society of Japan, vol.27 No.11 (1986) "Language Processing Method for Japanese Sentence Speech Output"). This Japanese sentence / speech converter assigns phonemes and accents to the words of the input Japanese sentence by referring to the Japanese dictionary, but for unknown words not registered in the above Japanese dictionary, the Kanji dictionary. The unknown word is processed by uniformly adding the reading and the accent in accordance with a simple rule for referring to and adding the reading and obtaining the accent from the reading.

<Problems to be Solved by the Invention> However, in the above-mentioned conventional Japanese sentence / speech conversion device, unknown word processing uniformly adds readings and accents to unknown words based on a simple rule. , When a sentence with various pronunciations and accents that are characteristic of Japanese sentences is input and unknown words are processed, the probability of obtaining the correct reading and accent is matched with the Japanese dictionary. There is a problem that it is lower than the probability that the correct reading and accent of the word is obtained. Therefore, the object of the present invention is to store the readings and accents of unknown words generated based on rules, and to output and check the contents as appropriate, so that they are inaccurate due to special words such as company names. Another object of the present invention is to provide a Japanese sentence-to-speech conversion device that facilitates correction of an incorrect reading or accent for an unknown word that is difficult to be correctly read into a correct reading or accent.

<Means for Solving Problems> In order to achieve the above object, the Japanese sentence / speech conversion apparatus of the present invention uses a kanji dictionary to convert unknown kanji of unknown words to kanji A processing unit, an unknown word accent processing unit that sets an accent from the reading of an unknown word, an unknown word storage unit that stores the obtained reading and accent of the unknown word, and contents stored in the unknown word storage unit. It is characterized by having an unknown word output section for outputting.

<Operation> When an unknown word that is not registered in the Japanese dictionary is input from the language processing unit to the unknown word reading processing unit, the Kanji unknown word of the unknown words is processed by the unknown word reading processing unit.
While referring to the Kanji dictionary, it is converted into Yomi and output,
The unknown words in hiragana and katakana are output as they are by reading them. The output unknown word reading is accented by the unknown word accent processing unit based on a rule. The accented unknown word is stored in the unknown word storage unit and is output by the unknown word output unit in response to the operator's request. Therefore, it is possible to confirm the reading of the unknown word and the correctness of the accent given based on the rule, and it becomes easy to correct the reading or the accent.

<Example> Hereinafter, the present invention will be described in detail with reference to illustrated examples.

FIG. 1 is a block diagram of a Japanese sentence / speech conversion apparatus according to an embodiment of the present invention.

When a Japanese kana mixed sentence is input from the input unit 101 to the word division processing unit 102, the Japanese dictionary 103 and the user dictionary 104
The word segmentation that divides a sentence into accented words using the longest matching method and the phrase minimum method that selects words so that the number of clauses in the sentence is minimized. The processing is performed by the word division processing unit 102, and the divided words are output to the word reading accent processing unit 106. The words divided by the word division processing unit 102 are
Isomorphic word selection that distinguishes between words that have the same character but are pronounced differently and accented by the word reading accent processing unit 106 if they are divided by matching the stored contents of the Japanese dictionary 103 and the user dictionary 104. Processing is performed, and the processed word is output to the prosody processing unit 112. The user dictionary editing unit 105 is to add, convert, or delete a new reading and accent of a word in the user dictionary 104.

On the other hand, unknown words that do not match the Japanese dictionary 103 are processed as follows. An unknown word in the kanji among unknown words is once converted into a reading by the unknown word reading processing unit 107 with reference to the kanji dictionary 108 storing the reading for each kanji character sentence,
The unknown words in hiragana and katakana are output to the unknown word accent processing unit 109 as they are as they are read. In the unknown word accent processing unit 109, a process of generating an accent from reading is performed using a rule described in detail later,
The accented unknown word is the prosody processing unit 112.
Is output to the unknown word storage unit 110 as well as
The reading of the unknown word and the accent are stored in association with each other.
The unknown word output unit 111 displays the reading and accent of the unknown word stored in the unknown word storage unit 110 on the display or synthesizes the synthesized voice and outputs it. Here, when the processing result of the unknown word output by the unknown word output unit 111 is not correct and the word is important, the user dictionary editing unit 105 newly adds it to the user dictionary 104 as described above. To be done. Therefore, the unknown word newly added to the user dictionary 104 is given correct reading and accent from the next time.

From the accent of each word output from the word reading accent processing unit 106 or the unknown word accent processing unit 109, the prosody processing unit 112 sets accents and poses of phrases when the words are chained. Further, the prosody parameter generation unit 113 generates parameters for the duration, pitch pattern, and power pattern for the unit for synthesis. Furthermore, in the voice parameter generation unit 114,
The synthesis unit corresponding to the reading of each input word is searched by referring to the speech data dictionary 115 of the synthesis unit, the synthesis units are interpolated, and the prosody parameter generation unit 113 is also provided.
Duration of the unit input from, pitch pattern,
Combining according to the power pattern, the final speech synthesizer parameter time series is obtained. The voice synthesizing unit 116 generates an actual voice waveform based on the voice synthesizing parameter input from the voice parameter generating unit 114, and the output unit
Output to 117.

The unknown word processing will be specifically described when the sentence "I went from Tokyo to Ikebukuro by subway and got on the JNR from Ikebukuro to Nippori" was entered as a Japanese character string. Here, in the above sentence, there are no place names “Tokyo”, “Ikebukuro”, and “Nippori” in the Japanese dictionary 103, and the rule of unknown word accent generation used by the unknown word accent processing unit 109 is “ The accent type of an unknown word is assumed to be type 0 ". The accent type is classified according to the position of a syllable that is highly sung among the syllables forming a word. For example, “n type” is a high syllable from the second syllable to the nth syllable. All syllables from the n + 1th onwards indicate that they are chanted low (Shinmei Kengoku Kanji Dictionary 3rd Edition Appendix Accent List).

Then, the unknown words “Tokyo”, “Ikebukuro”, and “Nippori” are subjected to unknown word processing by the unknown word reading processing unit 107 and the unknown word accent processing unit 109 as described above, and are stored in the unknown word storage unit 110. Stores the following contents.

1 Tokyo Tokyo Type 0 2 Ikebukuro Ikefukuro Type 0 3 Nippori Nippori Type 0 The unknown word output unit 111 displays the contents of the unknown word storage unit 110 above when the Japanese sentence / speech converter is stopped, such as at the end of work. Display it on the display or output it with synthetic voice to check. As a result, when it is confirmed that the reading and accent of the unknown word are incorrect, the following measures are taken.

(A) When there is a user dictionary For each of the contents of the unknown word storage unit 110 output by the unknown word output unit 111, if the reading or accent is incorrect, the correct reading or accent is It is input by the operator from the dictionary editing unit 105.

Then, the pronunciation and accent corrected by the operator are stored in the user dictionary 104.

(B) When there is no user dictionary In the word division processing 102, for example, "/"
Is to be processed by attaching "'" to the accent nucleus (the position at which the pitch changes from high to low). The unknown word output unit 111 displays the contents of the unknown word storage unit 110 one by one on the screen of the display, or uses the reading and accent generated by the rule to write a sentence or sentence of its own.
Read aloud with synthetic voice using the voice conversion function. With reference to this, from the user dictionary editing unit 105, the operator into the word string divided by the word division processing unit 102 as follows,
By directly inputting the above symbols "/" and "'", the reading and accent of the incorrect input sentence can be rewritten to the correct reading and accent.

"From Tokyo, I went to / Ikebukuro / by subway, and got on the national train from / Ikebukuro //// Nippori /."<Effect of the invention> As apparent from the above, according to the invention For example, it has a language processing unit and a voice synthesis unit, and the voice processing unit is a Kanji dictionary, an unknown word reading processing unit, an unknown word accent processing unit,
Equipped with an unknown word storage unit and an unknown word output unit, based on rules, it adds readings and accents to unknown words and stores them, and displays the readings and accents of unknown words that are stored according to the operator's request on the display screen. Since it is displayed or uttered as a synthetic voice, it is possible to easily check the correctness of only the reading and accent of the unknown word given based on the above rule collectively, and the wrong reading or accent of the unknown word It becomes easy to fix.

[Brief description of drawings]

FIG. 1 is a block diagram of a Japanese sentence / speech converter according to an embodiment of the present invention. 101 ... Input section, 102 ... Word division processing section, 103 ... Japanese dictionary, 104 ... User dictionary, 105 ... User dictionary editing section, 106 ... Word reading accent processing section, 107 ... Unknown word reading processing section, 108 ... Kanji Dictionary: 109 ... Unknown word accent processing unit, 110 ... Unknown word storage unit, 111 ... Unknown word output unit, 112 ... Prosody processing unit, 113 ... Prosody parameter generation unit, 114 ... Speech parameter generation unit, 115 ... Synthesis unit Voice data dictionary, 116 ... Voice synthesis unit, 117 ... Output unit.

Claims (1)

[Claims] 1. A word processing unit having a word division processing, a word reading accent processing unit, a prosody processing unit and a Japanese dictionary, a prosody parameter generation unit, a voice parameter generation unit, a voice synthesis unit and a voice data dictionary of a unit for synthesis. A Japanese sentence / speech conversion device for converting an input Japanese character string into a synthesized voice, which has a voice synthesis unit having, for obtaining reading and accent of an unknown word not registered in the above Japanese dictionary. , A kanji dictionary that stores readings for each kanji character, and kanji of unknown words contained in the input Japanese character string,
Based on the unknown word reading processing unit for converting into reading using the Kanji dictionary, and the accent rule for obtaining a word accent from the reading of the unknown word obtained by the unknown word reading processing unit,
An unknown word accent processing unit that sets an unknown word accent, an unknown word storage unit that stores the reading and accent of the unknown word processed by the unknown word accent processing unit, and an unknown word stored in the unknown word storage unit Japanese sentence / speech conversion device, which is provided with an unknown word output unit that outputs the reading and accent.