JPH037999A - Voice output device - Google Patents

Abstract

PURPOSE:To improve quality by synthesizing a fixed word by rule synthesis and a variable word by text synthesis. CONSTITUTION:An input document is converted by a document analysis part 1 into a rendering KANA(Japanese syllabary) string, which is divided into the fixed word and variable word by a fixed word detection part 6. Rule synthesis data on the fixed word is read out of a fixed word dictionary 12 by a rule synthesis part 11 and the fixed word is converted into a voice parameter time series. The rendering KANA string of the variable word is converted by a text synthesis part 3 into a voice parameter time series. The voice parameter time series is connected by a switching part 9 and converted by a voice conversion part 5 into a synthetic voice signal. Thus, the fixed word and variable word are synthesized and connected to improve the quality of the fixed word and connect the fixed word and variable word naturally.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a speech output device that outputs a combination of synthesized speech of fixed words and synthesized speech of variable words.

[Prior Art] Conventionally, this type of speech output device includes one that synthesizes fixed words and variable words using a text synthesis method, as shown in FIG. 4, and one that uses a parameter editing synthesis method as shown in FIG. Some synthesized fixed words using a text synthesis method, and synthesized variable words using a text synthesis method.

In the conventional example shown in Fig. 4, a sentence is input to the sentence analysis unit 1, which performs linguistic processing such as syntactic analysis and semantic analysis, and then registers the phonetic table of kanji, the pronunciation of compound words, etc. Create a row of readings while referring to the reading dictionary 2.
From this reading sequence, a text synthesis section 3 generates a speech parameter time series using a speech parameter 4 of the gana, and a seido conversion section 5 converts this speech parameter time series into a synthesized speech signal. In this conventional example, fixed words are also synthesized using the text synthesis method, so fixed words are essentially treated the same as variable words, and an infinite variety of fixed words can be output.

In addition, in the conventional example shown in FIG. 5, after the text analysis unit 1 performs syntactic and semantic analysis on the input sentence, the fixed word detection unit 6 detects whether it is a fixed word or a variable word. If the word is a fixed word, it is converted into speech parameter time series information using the text synthesis section 3 and the speech parameter 4 such as kana.If it is a fixed word, the speech parameter is converted into speech parameter time series information using the fixed word speech parameter dictionary 7 and the parameter editing synthesis section 8. A meter time series is generated, and the speech parameter time series of fixed words and the speech parameter time series of variable words are outputted to the speech conversion section 5 while being switched by a switching section 9.

In addition, in order to eliminate unnaturalness at the connection between fixed words and variable words, the quality of the fixed words is matched by the quality matching unit 10, and the quality of the fixed words is clear when compared with the text synthesis method. Both nature and naturalness will be high.

[Problems to be Solved by the Invention] By the way, in the conventional example shown in FIG. 4 above, the text synthesis rules are insufficient with the current text synthesis technology, so the quality of the synthesized speech, especially the quality regarding naturalness, is satisfactory. There was a problem that it was difficult to describe.

Furthermore, in the conventional example shown in FIG. 5, since the amount of fixed word data to be stored in the fixed word speech parameter dictionary 7 is large,
There was a problem in that the length and type of fixed words were limited due to storage space limitations6.Furthermore, the quality of fixed words and the quality of variable words were significantly different, resulting in unnatural connections between fixed words and variable words. In order to eliminate this problem, it is necessary to provide a quality matching unit 10 that lowers the quality of fixed words to bring them closer to the quality of variable words. For example, among the audio parameters, data regarding the pitch of the audio is stored in the text synthesis unit 3.
There was a way to use the output of

The present invention was made in view of the above-mentioned problems, and its purpose is to provide high quality fixed words, natural connections between fixed words and variable words, and to reduce the amount of data in fixed word dictionaries. There are fewer audio output devices to provide.

In addition, in the invention according to claim 2, the types of variable words are limited,
Moreover, it is an object of the present invention to provide a voice output device that can be used when changes are frequently made, and the invention according to claim 3 further provides a voice output device that can easily change variable words in the invention according to claim 2. It is an object of the invention according to claim 4 to provide a sound output device which can be made smaller and lighter in sound generation according to claim 2.

[Means for Solving the Problems] The present invention provides a speech output device that outputs a combination of fixed words and variable words. As shown in Figure 1, after an input sentence is analyzed by a sentence analysis section 1, a fixed word detection section 6 detects whether it is a fixed word or a variable word. A rule synthesis unit 11 synthesizes fixed words using a rule synthesis method to generate a speech parameter time series, and if it is a variable word, a text synthesis unit 3 synthesizes variable words using a text synthesis method to generate a speech parameter time series. While switching these audio parameter time series by the switching unit 9, the audio converting unit 5
It is designed to be output to and converted into a synthesized audio signal.

The invention set forth in claim 2 is the invention set forth in claim 1, comprising a variable word registration unit that registers variable words that have been text synthesized, and a combination voice output unit that outputs a combination of variable words and fixed words. The invention according to claim 3 is the invention according to claim 2, in which the variable word registration section and the combination voice output section are included in the same physical component, and the invention according to claim 4 is the invention according to claim 2. In the invention described in item 2, the variable word registration unit and the combination voice output unit are included in different physical components.6 [Operation] According to the present invention, fixed words are synthesized by rule synthesis. Therefore, the quality of fixed words is high, the amount of data in the fixed word dictionary is reduced, and natural synthesized speech can be obtained at the connection between fixed words and variable words.

[Example] The present invention will be explained below with reference to Examples.

Fig. 2 shows an embodiment of the present invention as set forth in claim 1, and this embodiment constitutes a voice output device that reads aloud a sentence that is a combination of fixed words consisting of fixed sentences and variable words consisting of non-fixed sentences. do.

In this embodiment, the sentence analysis unit 1 creates a pronunciation sequence of an input sentence while referring to a pronunciation dictionary 2 in which words that appear in input sentences and their pronunciations are registered in association with each other. The fixed word detection section 6 separates fixed words and variable words while referring to the fixed word dictionary 12, inputs the reading string of the variable word to the text synthesis section 3, and inputs the reading string of the fixed word into the rule synthesis section. 11.

The fixed word dictionary 12 registers the pronunciations of fixed words in association with rule composition data, and the rule composition data includes types of kana, length, strength, height, etc. of kana.

The rule synthesis unit 11 generates a speech parameter time series for a fixed word's pronunciation sequence, and the text synthesis unit 3 generates a speech parameter time series for a variable word from the pronunciation sequence. The generation of these audio parameter time series is performed using audio parameters registered in association with, for example, Japanese kana syllables and PARCOR parameter time series for synthesizing the audio of each kana syllable.

The switching section 9 outputs the regularly synthesized fixed word speech parameter time series and the text synthesized variable word speech parameter time series to the speech conversion section 5 in response to the detection by the fixed word detection section 6. The speech converter 5 converts the input speech parameter time series into a synthesized speech signal.

Next, the operation of this embodiment will be explained.

First, an input sentence is converted into a string of pronunciations in a sentence analysis section 1, and the converted string of pronunciations is separated into fixed words and variable words by a fixed word detection section 6.

For fixed words, the rule synthesis unit 11 reads out the rule synthesis data from the fixed word dictionary 12 and converts them into speech parameter time series. Here, the fixed dictionary 12 has a rate of about 5 times per second.
Approximately 140 bits are required to synthesize one second of Mora's speech, but as in the fixed word speech parameter dictionary 7 of the conventional example shown in Figure 5, approximately 2400 bits are required for one second of synthesized speech. It can be configured very compactly compared to the .

Also, audio day in fixed word dictionary 12! The quality of the synthesized speech produced by the conventional example shown in FIG. 4 is higher than that of the synthesized speech produced by the conventional example shown in FIG. On the other hand, the reading sequence of variable words is converted into a speech parameter time series in the text synthesis section 3.

These two types of speech parameter time series, fixed words and variable words, are connected by the switching unit 9, but since there is no significant difference in quality between them, the connection becomes natural, and as shown in FIG. The quality matching section 10 of the conventional example becomes unnecessary.

In addition, by transforming the speech parameter time series of variable words to match the speech parameter time series of fixed words, such as multiplying the speech height of variable words by a constant to match the height of fixed words, variable words and fixed words can be The connection between the two can be made more natural.

The audio parameter time series connected by the switching unit 9 is converted into a synthesized audio signal by the audio conversion unit 5.

FIG. 3 (a>(b) shows an embodiment according to the invention according to claims 2 to 4, and although the types of variable words are limited to some extent in this embodiment, changes in variable words often occur. This constitutes a speech output device when a text synthesis method is used for synthesis, or because the quality of the synthesized speech of variable words is not required to be high.In the embodiment shown in FIG. This is done at the time of output, but in this embodiment, it is done in the variable word registration section 13 shown in FIG. The word dictionary 14 is shown in Figure 3 (
The single combination voice output unit 15 used in the combination voice output unit 15 in b) is designed not to input sentences, but to input sentence numbers assigned to each sentence of variable words and fixed words. The voice selection unit 16 extracts rule synthesis data from the fixed word dictionary 12 and the variable word dictionary 14 in which variable words have been registered in accordance with the sentence number.

Next, the operation of this embodiment will be explained.

First, the variable word registration unit 13 registers limited variable words. In this case, the input sentence is read by the text analysis unit 1' and a string of pronunciations is generated while referring to the dictionary 2', and this string of pronunciations is synthesized into text and rule synthesis data for outputting rule synthesis data for variable words is generated. The generation unit 3' converts the data into rule composite data and registers it in the variable word dictionary 14. The reason why the data to be registered in the variable word dictionary 14 is not in the form of a speech parameter time series but in the form of rule synthesis data is that the rule synthesis data is a text synthesis method that generates a speech parameter time series from a reading string. Because the data is obtained in the middle of the process, speech synthesized from the rule synthesis data and speech synthesized from the speech parameter time series are of the same quality, and furthermore, this rule synthesis data is The capacity of the memory constituting the compact name variable word dictionary 14 is small.

Now, in the combination speech output section 15, the speech selection section 16 extracts rule synthesis data from the fixed word dictionary 12 and the variable word dictionary 14 in which variable words have been registered, in accordance with the input sentence number. Here, the method of distinguishing between the variable word sentence number and the fixed word sentence number is to use a specific number or higher as a variable word,
There are methods such as assigning a sentence number smaller than a specific number to a fixed word, and a method of enclosing the sentence number of a variable word in parentheses and inputting the sentence number of a fixed word without parentheses so that it can be identified. Further, as a method for establishing correspondence between sentence numbers and rule synthesis data, there is a method such as providing a correspondence table in the fixed word dictionary 12 and the variable word dictionary 14.

The rule synthesis data read from each dictionary 12, 14 is converted into a speech parameter time series in the rule synthesis section 4, and further converted into a synthesized speech signal by the speech conversion section 5.

By the way, the variable word registration section 13 and the combination voice output section 15
may be provided within the same physical component or may be provided within different physical components; in the former case, registration of variable words and audio output of a combination of variable and fixed words may be performed. Since it is easy to switch and check the audio, it is also easy to change the variable word audio. The generating section 3' can be separated from the combined audio output section 15, which simplifies the configuration of the audio output device and allows it to be made smaller and lighter.

In each of the above embodiments, PARCOR parameters are used as audio parameters, but the present invention is not limited to this. Further, the unit of the voice parameter may be VCV or CVC other than CV (C is a consonant, ■ is a vowel). In addition, the rule synthesis data includes the type of kana, the length, strength, and height of the kana, but it is not limited to this, and the input to the audio output device is the sentence and sentence number in each of the above embodiments. However, the present invention is not limited to these or a combination thereof. For example, in the embodiment shown in FIG. 2, fixed words may be specified by corresponding symbol strings, and variable words may be specified by sentences.

[Effects of the Invention] The present invention provides a speech output device that outputs a combination of fixed words and variable words, which has means for synthesizing fixed words using a rule synthesis method and means for synthesizing variable words using a text synthesis method. Therefore, the quality of fixed words is high, synthetic speech with natural connections between fixed words and variable words can be obtained, and the amount of data in the fixed word dictionary is also reduced.

Furthermore, the invention according to claim 2 includes a variable word registration section that registers variable words synthesized from text, and a combination voice output section that outputs a combination of variable words and fixed words, so that the idea of the variable word can be changed. It is possible to realize a voice output device that can be effectively used in cases where the variable word is limited and changes are often made, and the invention according to claim 3 is the invention according to claim 2, in which the variable word can be easily changed. The invention according to claim 4 is the invention according to claim 2, in which the variable word registration section and the combination voice output section are provided in the same physical component, so that the device can be made smaller and lighter. .

[Brief explanation of drawings]

1 is a basic configuration diagram of the present invention, FIG. 2 is a configuration diagram of one embodiment of the present invention, FIG. 3(a) is a configuration diagram of a variable word registration section of another embodiment of the present invention, FIG. 4B is a block diagram of the combined audio output section same as above, FIG. 4 is a block diagram of a conventional example, and FIG. 5 is a block diagram of another conventional example. 1 is a text analysis section, 3 is a text synthesis section, 5 is a speech conversion section, 6 is a fixed word detection section, 11 is a rule synthesis section, 9 is a switching section, 13 is a variable word registration section, and 15 is a combination speech output section. be.

Claims (4)

[Claims] (1) A speech output device that outputs a combination of fixed words and variable words, characterized by having means for synthesizing fixed words by a rule synthesis method and means for synthesizing variable words by a text synthesis method. Audio output device. (2) The audio output device according to claim 1, further comprising: a variable word registration unit that registers variable words that have been text-synthesized; and a combination audio output unit that outputs a combination of variable words and fixed words. . (3) The audio output device according to claim 2, wherein the variable word registration section and the combination audio output section are provided in the same physical component. (4) The audio output device according to claim 2, wherein the variable word registration section and the combination audio output section are provided in different physical components.