JPS6161400B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a speech synthesis device that generates arbitrary vocabulary using monosyllables.

Conventional word-registered and sentence-registered voice response devices respond to predetermined sentence patterns, such as "We are closed today. Please call back tomorrow."
We prepared multiple sentences such as, and output the corresponding sentence pattern aloud by specifying the sentence pattern. In addition, when inserting numbers, such as telephone numbers and monetary amounts, into sentences, the numerical information was registered in advance as words, which were then strung together and output as voice.
However, in the bank transfer notification service, the person's name,
In order entry services such as place names, trading companies, and stores, personal names, place names, and product names are essential variable information for voice output services, and preparing this information in advance in word units requires a huge amount of memory capacity. It was hot.

Furthermore, as a voice output method for variable information such as a person's name, place name, product name, etc., there is a monosyllable arbitrary word voice response device. This device is based on the 100 Japanese syllables, and if necessary, adds unit sounds such as consonants and long sounds, and stores approximately 100 to 500 unit sounds, which are output one after another at regular intervals when outputting audio. was outputting audio for each unit sound as is. For this reason, the output speech is equivalent to unit speech uttered word by word at the same intervals, and there is a drawback that interruptions between unit speeches cause deterioration in clarity and naturalness.

On the other hand, a voice response device that uses VCV (vowel-consonant-vowel) phoneme chain requires 700 to 900 phoneme chain information, which is about twice the number of words as the monosyllabic method. In addition to the monosyllabic information expressed in kana characters, accent position information and intonation information are required, and a dedicated VCV synthesis device must be added to the word registration type or sentence registration type voice response device, so the price is low. However, it had the disadvantage that it was difficult to create variable information such as people's names and place names.

In order to solve these drawbacks, this invention improves the clarity and naturalness of output speech by inserting phonological chain information of vowels and vowels or vowels and consonants corresponding between unit speech into the "crossing" part between unit speech. This will be explained in detail below with reference to the drawings.

FIG. 1 shows an embodiment of the speech synthesis device of the present invention, in which a speech storage section 11 stores, for example, fixed information 12 (numerical information may be included as variable information) in the original text shown in FIG. 2A. In the word storage section 13,
Monosyllabic information 14 indicated in kana characters is stored in the monosyllabic storage section 15, and phonological chain information 16 of vowels/vowels or vowels/consonants that occur as a "crossover" of the monosyllabic information 14 is stored in the phonological chain storage section 17. has been done. The output designation information receiving unit 20 receives, for example, a voice output command from a computer center to designate a sentence pattern and receive month/day information 1 which is variable information based on numbers.
8. Receive amount information 19 as well as monosyllabic information 14 of place names and people's names indicated in kana characters.

In response to this, the output designation information receiving unit 20 outputs an audio output start command to the audio output control unit 21. In parallel with this, the phoneme chain information generation unit 22 outputs the monosyllabic information 14 received by the output designation information reception unit 20.
Based on the kana characters, for example, as shown in FIG. 4B, phoneme chain information 16 of vowel/consonant information "in" connecting "mi" and "na" is generated. This can occur, for example, in such a way that the input monosyllabic information 14 is input as a code and its neighbors read out a read-only memory to obtain phoneme chain information. The audio output control unit 21 receives the audio output start command from the output designation information receiving unit 20, reads word groups 12 from the word storage unit 13 in a predetermined order, and supplies them to the audio output unit 23 as output audio. . Furthermore, when inserting monosyllabic information into the output of the corresponding sentence pattern and numeric word, the audio output control section 23 inserts the corresponding The speech information of the monosyllable stored in the syllable storage section 17 is read out and given to the speech output section 23 as output speech, and then the phonological chain information 16 that creates a "crossover" with the next monosyllable information is generated as a phonological chain. The phoneme chain information 16 is received from the phoneme chain storage unit 22 and read out from the phoneme chain storage unit 17 in which the phoneme chain information 16 is stored.
3 as the output audio. Through such a procedure, monosyllabic information and phoneme chain information are combined to arbitrarily edit place names, personal names, etc., and then inserted into word information to create speech information for each sentence. In the text registration type voice response device, voice information is stored in the word storage unit 13 in sentence pattern units and numerical information in word units, and it is clear that the word registration type method can be applied to the reading method. . In other words, fixed information and kana character information are edited in the same manner as in the past, but in this invention, the kana character information is edited using monosyllabic information, and the chain information is incorporated into each monosyllabic information.

As explained above, according to the present invention, any word can be edited by "crossing" between kana characters based on kana character information, that is, by inserting vowel/vowel or vowel/consonant phonological chain information. The syllables are unbroken, the clarity of the word,
The naturalness is easily improved and the phonological chain storage unit 17
This has the advantage that the increase in the amount of memory required is relatively small, and the amount of information supplied to the receiving section 17 is also small.

In the above, this invention was applied when inserting kana characters into an original text containing variable information, but when the output speech is relatively short, this invention can also be applied when editing only monosyllables to create a text. In addition, clarity and naturalness can be obtained by editing the monosyllables by including numbers that are only kana characters, and inserting phonological chain information between adjacent monosyllables.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an example of a speech synthesis device according to the present invention, and FIG. 2 is a diagram showing an example of text outputted from the speech. 11: Speech storage section, 12: Word group, 13: Word storage section, 14: Monosyllabic information, 15: Monosyllabic storage section, 16: Phonological chain information, 17: Phonological chain storage section, 18: Date information, 19 : Amount information, 20: Output specification information receiving section, 21: Audio output control section, 2
2: Phonological chain information generation section, 23: Audio output section.

Claims (1)

[Claims] 1. A storage means for accumulating phonological chains of monosyllables and vowels/vowels and vowels/consonants between adjacent monosyllables that occur, a receiving means for receiving monosyllabic information to be outputted as a voice, and each of the monosyllables from the received monosyllable information. means for generating phonological chain information between adjacent monosyllables or vowels and consonants; and a voice output for sequentially reading the monosyllables and phonological chain information from the storage means in accordance with the received monosyllable information and outputting speech. A speech synthesis device comprising means.