JPH05165486A - Text voice transforming device - Google Patents

Abstract

PURPOSE:To prevent a broadcast forbidden word itself from being sounded at the text voice transforming device to transform an input sentence to audio signals and to sound them by tabling read forbidden words and judging whether it is the read forbidden word or not while successively segmenting input sentences for the unit of a word. CONSTITUTION:A read forbidden word table 22 is provided and this read forbidden word table 22 is composed of a storing means for a system user to freely add or erase contents. Thus, at a Japanese text voice transforming device 20, the system user can freely set the read forbidden words and replaced expressions. In this case, a text analysis part 24 judges whether it is the read forbidden word or not by successively segmenting the input sentences for the unit of the word, the read forbidden word is stopped sounding based on the judged result, and it is sounded while replacing the expression of the read forbidden word.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a text-to-speech converter for converting an input sentence composed of text data into a voice signal and outputting the pronunciation, for example, applied to a Japanese text-to-speech converter for a Japanese sentence. You can

[0002]

2. Description of the Related Art A Japanese text-to-speech converter converts a kanji-kana mixed sentence (text data) created by a word processor or obtained by reading from an OCR into a voice signal and outputs the voice signal.

FIG. 2 shows the configuration of a conventional Japanese text-to-speech converter. In FIG. 2, a Japanese text-to-speech converter 1 is input with a kanji / kana mixed sentence D1 represented by text data such as JIS code, and the text analysis unit 2 converts the kanji / kana mixed sentence D1 into a pronunciation dictionary 4. The stored contents are used to convert to a phonological prosodic symbol string D2.

Here, the phonological prosody symbol string is an intermediate language in which the reading, accent, intonation, etc. of the input sentence are described as a character string. As shown in FIG. 3, the pronunciation dictionary 4 stores information such as notations, readings, and accents for generating the phonological prosodic symbol string D2 in units of words. The above-described text analysis unit 2 analyzes the input sentence by sequentially performing the matching between the notation of the pronunciation dictionary 4 and the input sentence, and determines the reading, accent, and intonation of the input sentence, thereby phonological prosodic symbol string. Generate D2.
This phonological prosody symbol string D2 is given to the synthesis parameter generation unit 6.

The synthesis parameter generation unit 6 uses the stored contents of the speech unit data unit 8 to generate a phonological prosodic symbol string D2,
It is converted into a synthesis parameter DP for synthesizing an audio signal. In speech synthesis, a speech unit (basic unit of speech) having a type of sound to be generated, a phoneme duration (sound length) that represents the duration of the sound, and a fundamental frequency pattern (pitch) of the sound. Will need to be selected. Therefore, the synthesis parameter generation unit 6 selects the phoneme duration and the fundamental frequency pattern based on the accent and intonation information included in the phoneme prosodic symbol string D2. Further, the speech unit data portion 8 is searched based on the reading information included in the phonological prosody symbol string D2, and the speech unit is selected by this. The synthesis parameter DP formed in this way is given to the speech synthesis unit 10.

The voice synthesizing section 10 synthesizes a voice signal based on the synthesis parameter DP and gives it to the speaker 12 to output a sound. At this time, the speech synthesis unit 10 connects the speech units and synthesizes the speech signal so that the synthesized sound approaches the natural sound while using the phoneme duration and the fundamental frequency pattern as a reference.

As a result, the Japanese text-to-speech conversion apparatus 1 can sequentially read aloud the Japanese sentences (kanji / kana mixed sentences) D1 that have not been separated.

[0008]

However, in the conventional Japanese text-to-speech conversion device 1, the input kanji-kana mixed sentence D1 includes slang words, slang words, discriminative words, abusive words, etc. ) Is included, there is a problem of reading these terms without exception. Such a term is a term that should normally not be spoken, and is unpleasant to the listener.

As a method of solving this problem, a method of not registering this type of term in the pronunciation dictionary 4 can be considered. However, in a system that processes unknown words,
There is a problem that this term is eventually read aloud by guessing from other registered notations.

On the other hand, a method is also conceivable in which the operator checks the input sentence in advance, or the operator responds to it when the system is requested to process it as an unknown word. However, there is a problem that the work takes time and cannot be completed completely, and the work itself becomes complicated when the amount of texts increases.

The present invention has been made in consideration of the above points, and when an input sentence contains a term whose pronunciation output is not appropriate, a text that can prohibit the pronunciation of this type of term. It is intended to provide a voice conversion device.

[0012]

In order to solve such a problem, according to the first aspect of the present invention, a text-to-speech conversion device for converting an input sentence composed of text data into a sound signal and outputting the sound is output. Means were provided. That is, a reading-prohibited term table that stores reading-prohibited terms and an input sentence is cut out in word units, and a reading-prohibition table is searched to determine whether or not a word included in the input sentence is a reading-prohibited term. The judgment means and the pronunciation prohibition means for prohibiting the pronunciation of the word corresponding to the reading-prohibited term based on the judgment result of the reading-prohibited term judgment means are provided.

Further, in the second aspect of the present invention, the following means are provided in the text-to-speech conversion apparatus for converting an input sentence composed of text data into a sound signal and outputting the sound. That is, a reading prohibited term table that stores a pair of the reading prohibited term and a replacement expression that represents this reading prohibited term, the input sentence is cut out in word units, and the reading prohibited table is searched to search for the word included in the input sentence. A reading prohibition term judging means for judging whether or not is a reading prohibition term, and a reading prohibition term replacing means for converting the reading prohibition term into a substitution expression and pronounced based on the judgment result of the reading prohibition term judging means are provided. ..

[0014]

In the first aspect of the present invention, the reading-prohibited term for prohibiting the pronunciation output is stored in advance in the reading-prohibited term table. When a sentence consisting of text data is entered,
The reading-prohibited term determination means cuts out the input sentence on a word-by-word basis, searches the reading-prohibition table, and determines whether or not the word included in the input sentence is a reading-prohibited term.
Then, the pronunciation prohibiting means prohibits the pronunciation of the word corresponding to the reading-prohibited term based on the judgment result of the reading-prohibited term judging means. For example, the pronunciation of a reading-prohibited term is prohibited by making a beep sound be pronounced or silenced instead of the reading-prohibited term.

Further, in the second aspect of the present invention, a pair of a reading prohibited term and a replacement expression expressing the reading prohibited term is stored in advance in the reading prohibited table. When a sentence consisting of text data is input, the reading-prohibited term determination means cuts the input sentence into words and searches the reading-prohibition table to determine whether the word included in the input sentence is a reading-prohibited term. .. Then, the reading-prohibited term replacing means converts the reading-prohibited term into a replacement expression based on the judgment result of the reading-prohibited term judging means and causes the pronunciation.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment in which the present invention is applied to a Japanese text-to-speech converter will be described in detail below with reference to the drawings.
Here, FIG. 1 is a block diagram showing the configuration of this embodiment, and the portions corresponding to those in FIG. 2 described above are designated by the same reference numerals.

In FIG. 1, the Japanese text-to-speech conversion apparatus 20 of this embodiment has a reading-inhibited term table 22.
Is equipped with.

FIG. 4 shows the reading-prohibited term table 2
2 shows the configuration of No. 2. As shown in FIG. 4, the read-aloud term table 22 is read out (sound output).
Is stored with a pair of a term forbidden (hereinafter, referred to as a reading prohibited term) and a replacement expression (hereinafter, referred to as a replacement expression). In this embodiment, replacement expressions such as "blind person" and "early person" are stored for reading-prohibited terms such as "tsubo" and "blind". The read-aloud term table 22 is formed by a storage means that allows the system user to freely add and delete the contents, whereby the system user of the Japanese text-to-speech converter 20 is prohibited to read the content. The terms and replacement expressions can be freely set.

The text analysis unit 24 basically converts the inputted Kanji / Kana mixed sentence D1 into a phonological prosodic symbol string D3, as in the conventional case. However, the text analysis unit 24 of this embodiment also executes the process related to the reading-prohibited term shown in FIG. By executing the processing shown in FIG. 5, it is possible to prevent the pronunciation of the reading-prohibited terms registered in the reading-prohibited term table 22 from occurring.

When the kana-kana mixed sentence D1 is input, the text analysis unit 24 starts the process of FIG. 5 to cut out one word by using the reading of the pronunciation dictionary 4, and further stores the pronunciation dictionary 4. The contents are used to convert into a phonological prosodic symbol D3 (steps 101 and 102). Then, the reading prohibited term table 22 is searched to determine whether or not the cut out word corresponds to the reading prohibited term (step 1).
03).

If the cut-out word is not a reading-prohibited term, the text analysis unit 24 further judges whether it is the end of sentence (step 104). If it is the end of the sentence, that is, if the processing relating to the reading-prohibited term for the last word has been executed, the processing shown in FIG. 5 is ended (step 1).
05). On the other hand, if the cut-out word is neither a reading-prohibited term nor a sentence end, the text analysis unit 24 returns to step 102 described above and cuts out the next word.

That is, while the word cut out from the input sentence is not a reading-prohibited term, the text analysis unit 24 repeats the loop of steps 102-103-104-102 to sequentially convert it into the phonological prosodic symbol D3.

If the input sentence contains a reading-prohibited term and the word is cut out in step 102, a positive result is obtained in step 103. At this time, the text analysis unit 24 determines the processing mode for the reading-prohibited term (step 106).

In the case of this embodiment, the Japanese text-to-speech converter 20 is designed so that the system user can select the processing mode of the reading-prohibited term in advance. As the processing modes, there are a beep sound mode in which the reading prohibited term portion is pronounced with a beep sound, and a paraphrasing mode in which the reading prohibited term portion is pronounced with the replacement expression. Such processing mode determination may be performed, for example, at a position corresponding to the mode of the DIP switch, or by determining the content of the flag holding the mode information input via the keyboard or the like. Is also good.

When the system user has selected the beep sound mode in advance, the text analysis unit 24 uses the phonological prosodic symbol string D3 for the word as a beep sound generation command for the word being processed which corresponds to the reading prohibited term. After the conversion, the process moves to the above-mentioned step 104 (step 10).
7).

On the other hand, if the system user has previously selected the paraphrase mode, the text analysis unit 24 discards the phonological prosodic symbol string D3 for the reading-inhibited term,
The word corresponding to the read-aloud term in the input sentence is replaced with the replacement expression, and the process returns to step 102 (step 108). When the process returns to step 102 in this way, one word is cut out from the replacement expression.

As described above, the text analysis unit 24 of this embodiment cuts out one word from the input sentence and converts it into the phonological prosodic symbol D3, and when the cut out word is a reading-prohibited term. , A beep sound pronunciation command is output instead of the phonological prosodic symbol D3, or the phonological prosodic symbol D3 for the replacement expression of the reading-prohibited term is output.

FIG. 6 shows the output of the text analysis unit 24 when the input sentence "She is blind." Is input. When the beep sound mode is selected, reading and accent are alternately continued, and a beep sound pronunciation command [BEEP] is inserted in the reading prohibited term "blank".
A phonological prosodic symbol string D3 "P] Da" is generated. When the paraphrase mode is selected, the phonological prosodic symbol string D3 for a sentence similar to the sentence in which the input sentence of “She is blind” is replaced with “She is a blind person”, that is, "Kanojowa 1, Meno 1, Fuji Yuna 1, Hitoda 2" is generated.

In the conventional Japanese text-to-speech conversion apparatus 1, in this case, the phoneme prosodic symbol string D2 of "Kanojowa 1, Mekurada 3" in which reading and accent are alternately continuous is generated.

In this embodiment, the reading-prohibited term is replaced with the replacement expression for processing because the usability for the system user can be improved. Hereinafter, the reason for this will be described in detail.

When the phonological prosodic symbol string D3 is generated, even if the reading and accent of the reading prohibited term are replaced, the pronunciation of the reading prohibited term can be prevented in advance. However, in this method, if the system user is allowed to freely set and delete the read-aloud words, the system user must eventually register the reading and accent of the read-aloud words. That is, in this case, the reading prohibited term table 22 is treated in the same manner as the pronunciation dictionary dedicated to the user, and the user himself / herself registers the reading and accent of the reading prohibited term in his / her own user dictionary. However, registering accents actually requires skill,
There is a problem that registration is difficult for system users who are not familiar with it. In particular, when the replacement expression is a phrase composed of a plurality of words, it becomes more difficult to register the accent as compared with the case where the replacement expression is one word. Therefore, in this embodiment, by replacing the notation of the reading-prohibited term, it is possible to easily register, and the replaced notation is analyzed by the text analysis unit 24, and the reading and accent are added.

The text analysis unit 24, after obtaining readings and accents for all the words in the sentence, makes a comprehensive judgment so that the sentence as a whole is formatted and determines the readings and accents of the input sentence. Phonological prosodic symbol string D
It is designed to generate 3.

The synthesis parameter generation unit 26, like the synthesis parameter generation unit 6 (FIG. 2), generates the synthesis parameter DP1 based on the phonological prosodic symbol string D3.
However, the synthesis parameter generation unit 26 uses the phonological prosodic symbol string D.
Beep sound command [BEEP] inserted in 3
When it detects, a synthetic parameter (that is, data such as position and length) for beep sound generation is generated and output.

In the above configuration, the input sentence D1 is converted into the phonological prosody symbol string D3 by the text analysis unit 24, and this phonological prosody symbol string D3 is converted into the synthesis parameter DP1 for speech synthesis by the synthesis parameter generation unit 26. ..

At this time, the input sentence D1 is the text analysis unit 24.
Table 2
Based on 2, it is determined whether or not each word corresponds to a read-aloud term. Here, when the prohibited term is applied, in the beep sound mode, the beep sound pronunciation command is replaced to generate the phonological prosodic symbol string D3, whereas in the paraphrase mode, the replacement expression of the reading prohibited term table 22 is replaced. As a result, a phonological prosodic symbol string D3 is generated.

As a result, in the beep sound mode, the beep sound is output in place of the read-aloud words, whereas in the paraphrase mode, the read-aloud words are replaced with a predetermined expression and read.

Therefore, according to the above embodiment, the input sentence is cut out word by word, the reading prohibited term table 22 is searched, and the reading prohibited term is replaced with a beep sound or a replacement expression, thereby omitting a complicated operation. It is possible to prevent the pronunciation of the reading-prohibited terms from occurring.

In the above embodiment, the case where the reading-prohibited term is replaced with the beep sound or the replacement expression has been described. However, a peculiar sound other than the beep sound is generated,
It is also possible to silence and stop the pronunciation itself.

In the above embodiment, two processing modes are provided as the processing mode for the reading-prohibited term and the user is allowed to select the processing mode. However, the apparatus can realize only one of the processing modes. Is also good.

Further, in the above embodiment, the text-to-speech converter for Japanese sentences is shown, but the present invention can be applied to the text-speech converter for other languages.

[0041]

As described above, according to the present invention, the reading-prohibited terms are made into a table, and the input sentence is sequentially cut out in word units to judge whether or not the reading-prohibited terms are used.
It is possible to easily judge whether it is a reading prohibited term, and based on this judgment result, the pronunciation of the reading prohibited term is prohibited, or the expression of the reading prohibited term is replaced and pronounced. A text-to-speech converter that can prevent pronunciation can be realized.

[Brief description of drawings]

FIG. 1 is a block diagram showing a text-to-speech conversion apparatus according to an embodiment.

FIG. 2 is a block diagram showing a conventional Japanese text-to-speech conversion device.

FIG. 3 is a schematic diagram showing the contents of a pronunciation dictionary.

FIG. 4 is an explanatory diagram showing the contents of a reading-prohibited term table according to the embodiment.

FIG. 5 is a flowchart showing the operation of the text analysis unit of the embodiment.

FIG. 6 is an explanatory diagram showing a phonological prosodic symbol string according to the embodiment.

[Explanation of symbols]

4 ... Pronunciation dictionary, 8 ... Speech element data part, 10 ... Speech synthesis part, 12 ... Speaker, 20 ... Japanese text-to-speech conversion device, 22 ... Read-aloud term table, 24 ... Text analysis part, 26 ... Synthesis parameter generation Department.

Claims (2)

[Claims] 1. A text-to-speech conversion device for converting an input sentence consisting of text data into a voice signal and outputting the pronunciation, and a reading-prohibited term table for storing reading-prohibited terms; Based on the reading-prohibited term determination means that searches the reading-prohibited table and determines whether the word included in the input sentence is a reading-prohibited term, and the determination result of the reading-prohibited term determination means,
A text-to-speech conversion device comprising: a pronunciation prohibition unit that prohibits pronunciation of a word corresponding to the above-mentioned reading-prohibited term. 2. In a text-to-speech conversion device for converting an input sentence composed of text data into a voice signal and outputting the pronunciation, a read-aloud term table storing a pair of a read-prohibited term and a replacement expression expressing the read-prohibited term. And a reading-prohibited term judging means for cutting out the input sentence in units of words, searching the reading-prohibition table and judging whether or not the word included in the input sentence is the reading-prohibited term, and the reading-prohibited term judging means. Based on the judgment result of
A text-to-speech conversion device, comprising: a reading-prohibited term replacement unit that converts the reading-prohibited term into the replacement expression and pronounces it.