JPH08328578A - Text voice synthesizer - Google Patents

Abstract

PURPOSE: To obtain voices that are easily understood by pronouncing a nown phrase portion slowly compared to other portions when a speed reading is specified in a text voice synthesizer which synthesizes voices from inputted character information. CONSTITUTION: The synthesizer is provided with a text analysis section 8 which converts inputted character information to phoneme rhythm symbol columns by referring to a pronunciation dictionary 3, a synthesis parameter generating section 4 which converts the columns to synthesis parameters for voice synthesis by referring to a voice piece storage section 5 and a voice synthesis section 6 which synthesizes voice signals based on the parameters. A speed reading processing section 9 is provided in the section 8 and phoneme rhythm symbol columns are generated so that only the nown phrases among the inputted character information are pronounced relatively slowly during a speed voice synthesis.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice synthesizing device for synthesizing and outputting a voice based on input character string information.

[0002]

2. Description of the Related Art A speech synthesizer which receives character information such as text data and converts it into speech and outputs the speech is not limited in output vocabulary. Can be expected to be applied in various fields of use. For example, text data created by a word processor, etc. can be converted into voice and output, or by editing the text, you can easily create and change response messages. But it can be used.

FIG. 6 is a block diagram of a conventional speech synthesizer, which shows an apparatus for performing Japanese text-to-speech conversion using Japanese (Kanji / Kana mixed sentence) as an input. This Figure 6 below
The outline of the conventional device will be described with reference to FIG. In FIG. 6, the conventional text analysis unit 2 uses the pronunciation dictionary 3 to generate a phonological prosodic symbol string from a kanji / kana mixed sentence input from the character information input unit 1. Here, the phonological prosody symbol string describes the reading, accent, intonation, etc. of the input sentence as a character string, and is called an intermediate language. Pronunciation dictionary 3 for reading and accent of each word
The text analysis unit 2 refers to the pronunciation dictionary 3 to generate a phonological prosodic symbol string.

The synthesis parameter generation unit 4 extracts a voice unit (sound type) based on the phonological prosodic symbol string,
According to a predetermined rule, phoneme duration (sound length),
A parameter for voice synthesis (called a synthesis parameter) such as a fundamental frequency (voice pitch) pattern is generated. Of these, speech units are those that are analyzed and generated from pronunciation data when a word or the like is uttered, and are the basic units of speech for synthesis.By superimposing these, a synthesized waveform is generated. It The voice unit data is stored in advance in the voice unit storage unit 5 including, for example, a ROM, and the synthesis parameter generation unit 4 recognizes the voice unit from the phonological prosodic symbol string and stores the voice unit storage unit 5. The corresponding speech unit data is extracted by referring to it.

The voice synthesis unit 6 is a synthesis parameter generation unit 4
A synthesized waveform (voice signal) is generated based on the generated synthesis parameter. Such a synthesized voice signal is output as voice through the speaker 7 or is transmitted to another device via a line. In addition, many voice synthesizers are capable of selecting the speed of voice, a male sound, a female sound, etc. according to the purpose of use.

[0006]

When the reading speed of the text is increased, according to the conventional technique, all the phonemes are shortened almost uniformly and the synthesized speech is monotonous and becomes difficult to hear. Was there. Therefore, the user
I often missed important words, so to avoid this, I had to either read them slowly or repeat them each time.

The present invention provides a speech synthesizer which utters a noun relatively slowly when reading a text at a certain speed or faster than other words, and synthesizes a voice whose content is easy to hear even when reading aloud quickly. It enables you to listen to a lot of sentences in a short time.

[0008]

SUMMARY OF THE INVENTION According to the present invention, a character information input section for inputting character information and the character information are read by referring to a pronunciation dictionary in which the reading and accent of words are registered in advance, and the accent and intonation are read. Etc., a text analysis unit for converting into a phonological prosodic symbol string, the phonological prosodic symbol string is accepted, and the phonological prosodic symbol string is referred to by referring to a speech unit storage unit in which a speech unit which is a basic unit of speech is registered in advance. In a text-to-speech synthesizer including a synthesis parameter generation unit that converts a synthesis parameter for speech synthesis and a voice synthesis unit that receives the synthesis parameter and synthesizes a voice signal, the text analysis unit includes a speed reading processing unit, It is characterized by generating a phonological prosodic symbol string that specifies that only noun phrases in input character information should be pronounced relatively late during speed-reading speech synthesis. .

[0009]

The text analysis unit analyzes the character information input to the character information input unit and generates a phonological prosodic symbol string by referring to the pronunciation dictionary. When the speed reading is specified, the speed reading processing unit operates to generate the noun phrase relatively slowly when the phonological prosodic symbol string is generated. The synthesis parameter generation unit refers to the phoneme prosodic symbol string to generate a synthesis parameter by referring to the speech unit storage unit, and the voice synthesis unit converts the synthesis parameter into a voice signal. When the speed reading is designated, only the noun phrase part of the obtained audio signal is sounded later than the other parts.

[0010]

Embodiments of the present invention will be described below with reference to the drawings. In the following explanation, sentences are from the beginning. Up to (period) is shown, and the sentence means one composed of a plurality of sentences. FIG. 1 is a block diagram of a speech synthesizer according to an embodiment. In the figure, reference numeral 1 is a character information input unit for inputting Japanese sentences such as character strings mixed with kanji and kana.

Reference numeral 8 is a text analysis unit of the present invention, which receives a sentence from the character information input unit 1 and analyzes the sentence to generate a phonological prosodic symbol string describing the reading, accent, intonation, etc. of the sentence. The text analysis unit 8 has a speed reading processing unit 9. The speed reading processing unit 9 relatively slows down a phrase including a noun in comparison with other parts when synthesizing a sentence at a fast speed (quick speed). Performs speed-reading processing characterized by specifying to speak. Reference numeral 3 denotes a pronunciation dictionary in which the reading and accent of a word are registered in advance, which is referred to and used in the processing of the text analysis unit 8.

Reference numeral 4 denotes a synthesis parameter generation unit, which recognizes a voice unit based on the phonological prosodic symbol string created by the text analysis unit 8 and extracts the voice unit data registered in the voice unit storage unit 5. To generate synthetic parameters. The speech unit data is previously generated by analysis from the speech data. Reference numeral 6 denotes a voice synthesis unit, which receives the synthesis parameter output from the synthesis parameter generation unit 4 and generates a synthesized waveform (voice signal).

Reference numeral 7 is a speaker for outputting the synthesized waveform synthesized by the present apparatus. FIG. 2 is an operation flowchart of the text analysis unit of the embodiment,
The flow of processing of the text analysis unit of the embodiment will be described with reference to this figure. In the sentence analysis here, the analysis is performed at the syntax level and the semantic level, but in the embodiment shown here, the longest matching method at the word division level (morpheme analysis level) will be described as an example.

When the apparatus starts operating, the text analysis section 8 reads out a text, that is, a Japanese document from the character information input section 1 (S1). The Japanese sentence input from the character information input unit 1 is sentence-parsed into words in order from the beginning and matching with registered words in the pronunciation dictionary 3 is performed. In this sentence analysis, the character strings of the entry words in the pronunciation dictionary 3 are sequentially checked from the beginning of the sentence. If a plurality of headwords can be matched, the character string with the longest word length is selected. In this way, sentence reading and control information such as accent are generated.

In order to perform such sentence analysis, first, a character pointer is set at the beginning (S2). Next, a word (read word) starting with the character pointed by the character pointer is taken out from the pronunciation dictionary 3 (S3). The read word and the long word registered in the pronunciation dictionary 3 are matched with the input character string (S4). In the sentence analysis using the longest matching method, if there is no candidate for the succeeding word after a certain position in the sentence, matching cannot be obtained (S5). In that case, it is necessary to backtrack. Therefore, the character pointer is returned by the length of the previous word (S6), the previous word is traced back, and another candidate is matched again (S4).

If a match is found (S5), the connection relation with the previous word is checked (S7), and the quality of the connection relation is determined (S8). If the connection relationship is negative (NG), the process returns to S4 and the processes up to S8 are repeated. When the connection relation is good (OK) (S8), reading is added to the word (S9). Pronunciation dictionary 3
The phonetic kana, accent type, and part-of-speech information of the word are registered in, and the word information is also obtained at this time.

Next, if the reading speed is set to a certain level or higher and the result of the text analysis is that the word is a noun (S10), an emphasis mark is given to the word (S11). Then, the character pointer is advanced by the length of the word (S12). Then, it is checked whether or not this series of processing has been completed for each sentence (S13).
Returning to the processing of S3, the same processing (S3 to S13) is performed and the processing of one sentence is performed. When processing of one sentence is completed, independent words and independent words, or independent words and adjunct words are combined to generate accent phrases such as compound words and clauses, and one adjacent accent phrase with dependency To form an exhalation paragraph (S14). At this time, the movement, occurrence, and disappearance of accents due to word combination are performed based on the accent combination rule described in the text analysis unit 8. The expiratory paragraph is a unit in which a human speaks in a breath.

Finally, in accordance with the emphasizing mark given in S11, a symbol that reduces the speed of the accent phrase of the word having the emphasizing mark by one step from the surroundings is inserted (S15) to complete the phonological prosodic symbol string. This result is sent to the synthesis parameter generation unit 4 as in the conventional case (S16). Based on the generated phonological prosodic symbol string, the synthesis parameter generation unit 4 produces prosodic parameters such as phoneme duration, fundamental frequency pattern, and power. Is set and the speech unit data to be used is taken out. The voice synthesis unit 6 generates a synthesized waveform based on the synthesis parameter, and outputs a synthesized sound through the speaker 7 or the like.

If there are still sentences to be analyzed (S1
7), returning to the process of S1, the same process (S1 to S16)
Is repeatedly performed, and the processing of the text analysis unit 8 ends. Next, the operation of text analysis will be described with reference to sentence examples. FIG. 3 is an explanatory diagram of an example of processing of the text analysis unit of the embodiment. Taking the sentence “You are good at quick words” in this figure as an example, this is due to the processing of S2 to S13 of FIG. "You are good at-quick words-" is decomposed.

FIG. 4 is an explanatory view of an example of the text analysis result of the embodiment, and shows an example of the text analysis determined by the above processing. From this, it can be seen that this sentence is divided into three accent phrases, "you are", "quick words" and "we are good at." It is. "You"
Is a pronoun, and since this pronoun rarely becomes a keyword in a sentence, an emphasis mark is not attached. Since the emphasis mark is put on the noun "Speaker" by the process of S11, before and after this accent phrase,
Insert {V-} and {V +}. {V-} is a symbol that makes the speed one step slower than before this symbol, and {V +}
Is the opposite.

The above-mentioned symbols {V-} and {V +} are not inserted in the symbol string in the case of specifying the normal speed in the phoneme symbol string output example of FIG. 3, but when the process is performed by designating the speed reading. , {V-}, {V-, as shown in the symbol string in the case of speed reading in FIG.
+} Is inserted. As a result, it is possible to generate a phonological prosodic symbol string for uttering the noun phrase one step later than the surrounding accent phrases. Also, the symbol "P1", "
P3 "and" P0 "are phrase symbols indicating an exhalation paragraph, and"} "and"] "are accent symbols. Since the symbols used in this description are for convenience, what is actually a symbol? You may use.

When there are many accent phrases including nouns in a sentence, the emphasis marks are prioritized so that the reading speed of the entire sentence does not decrease.
FIG. 5 is an explanatory diagram of an example of the priority order of the noun emphasis according to the embodiment.
Accent phrases with the particle "ha" or "ga" attached to the noun are
In many cases, it is easy to become a subject and important, so the priority is raised. In this way, the priority may be determined according to the relationship with the adjacent word or the like, the position in the sentence, and the like, and the emphasis mark may be placed in consideration of the priority. For example, in the sentence, "Handmade cakes made from carefully selected flour, sugar, fresh eggs and milk are very delicious."
The five nouns are "flour", "sugar", "eggs", "milk", "handmade", and "cake", but if you put emphasis marks according to Fig. 5, 0 points are emphasized and 0 points are emphasized. If you add points such as one that may be done, "flour", "sugar", "egg",
"Milk" and "Handmade" are given 1 point, and "cake" is given 5 points (process of S11 in FIG. 2). Figure 2
In S15 of this, 5 points of "cake" according to this emphasis mark
Accent phrases including, here, a speed symbol is added before and after "cake".

By the processing in the above text analysis unit, the noun in the sentence can be uttered more slowly than the surroundings. For this reason, the listener can hear the nouns in the sentence in an emphasized state, and even if they are read aloud quickly, they are less likely to miss the nouns that often become the keywords in the sentence. Further, since the reading speed of the synthetic sound changes, the monotonousness is softened and it becomes easy to hear.

Since the contents are easily heard by the above processing, the speed can be further increased, and a large amount of data can be efficiently heard in a short time. In this embodiment, the noun phrase is uttered more slowly than the surrounding phrases in order to give attention to the noun in order to give a change to the synthetic sound, but in addition, the phonological prosodic symbol string is used to raise the volume by one step. It is also effective to generate them.

[0025]

As described above, when the speed reading process is specified, the text analyzing unit is provided with the speed reading processing unit that processes the noun so that it is spoken more slowly than the surroundings. This has the effect of making it easier to hear the contents even when the speed reading is designated, and further has the effect of further increasing the speaking speed.

[Brief description of drawings]

FIG. 1 is a configuration diagram of a speech synthesizer according to an embodiment.

FIG. 2 is an operation flowchart of a text analysis unit according to the embodiment.

FIG. 3 is an explanatory diagram of an example of processing of a text analysis unit according to the embodiment.

FIG. 4 is an explanatory diagram of a text analysis result example of the embodiment.

FIG. 5 is an explanatory diagram of an example of priority order of noun emphasis according to the embodiment.

FIG. 6 is a block diagram of a conventional speech synthesizer.

[Explanation of symbols]

 1 character information input unit 3 pronunciation dictionary 4 synthesis parameter generation unit 5 speech unit storage unit 6 speech synthesis unit 8 text analysis unit 9 speed reading processing unit

Claims (3)

[Claims] 1. A character information input section for inputting character information,
A text analysis unit that converts the character information into a phonological prosodic symbol string by referring to a pronunciation dictionary in which word readings and accents are registered in advance, and a phoneme unit that is the basic unit of speech that accepts the phonological prosodic symbol string. A speech synthesis unit for converting the phonological prosodic symbol string into a synthesis parameter for speech synthesis by referring to a speech unit storage unit registered in advance, and a speech synthesis unit for accepting the synthesis parameter and synthesizing a speech signal. In a text-to-speech synthesizer provided, a text analysis unit is provided with a speed reading processing unit, and at the time of speed reading speech synthesis, a phonological prosodic symbol string that specifies that only noun phrases in input character information are pronounced relatively late. A text-to-speech synthesizer characterized by generating a. 2. The noun phrase according to claim 1, wherein the noun phrase corresponding to the noun in the input character information is given a priority according to the relationship between the noun in the input character information and the position in the sentence, etc. A text-to-speech synthesis apparatus having a speed reading processing unit for synthesizing a phonological prosodic symbol string designating that only a noun phrase having a high priority is pronounced relatively late when a large number of nouns are included. 3. The speed reading processing unit according to claim 1, further comprising: a phonological prosodic symbol string that specifies that a noun phrase that is pronounced slowly is pronounced at a relatively high volume. And a text-to-speech synthesizer.