JP3575919B2 - Text-to-speech converter - Google Patents

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a text-to-speech converter that synthesizes and outputs speech based on input character string information.
[0002]
[Prior art]
Text-to-speech converters that take text information such as text data as input, convert it into speech, and output it have no restrictions on the output vocabulary. The application in the field can be expected.
[0003]
For example, text data created by a word processor or the like can be converted into voice and output, and used for proofreading of sentences. Also, by utilizing the feature that a response message can be easily created and changed only by editing the text, it can be used in communication services such as telephones.
[0004]
FIG. 2 shows the configuration of a conventional text-to-speech conversion device (Japanese text-to-speech conversion) that inputs Japanese (kanji and kana mixed sentences). Hereinafter, an outline of the conventional apparatus will be described with reference to FIG.
[0005]
In FIG. 2, a text analysis unit 101 uses a pronunciation dictionary 102 to generate a phonemic prosody symbol string from a sentence mixed with Kanji or Kana input from a character information input unit 100. Here, the phonemic prosody symbol string is a description of the input sentence reading, accent, intonation, and the like as a character string, and is called an intermediate language. The pronunciation and accent of each word are registered in the pronunciation dictionary 102, and the text analysis unit 101 generates a phonemic prosody symbol string while referring to the pronunciation dictionary 102.
[0006]
The synthesis parameter generation unit 103 extracts a speech unit (sound type) based on the phoneme prosody symbol string, and obtains a phoneme duration (sound length) and a fundamental frequency (sound pitch) according to a predetermined rule. A parameter for speech synthesis such as a pattern (hereinafter, referred to as a synthesis parameter) is generated.
[0007]
Of these, speech units are analyzed and generated from utterance data when words or the like are uttered, are basic units of speech for synthesis, and a synthesized waveform is generated by superimposing these. . In the following, a combination of basic elements of speech such as CV (consonant-vowel) and VCV (vowel-consonant-vowel) is called a speech unit, and an element that realizes a waveform of the speech unit is called a speech unit. Each speech unit corresponds to, for example, a set of a plurality of speech units. The speech unit data is stored in a speech unit data storage unit 104 such as a ROM, and the synthesis parameter generation unit 103 recognizes a speech unit from a phoneme prosody symbol string and extracts corresponding speech unit data.
[0008]
The speech synthesis unit 105 generates a synthesized waveform (speech signal) based on the synthesis parameters generated by the synthesis parameter generation unit 103. Such a synthesized voice signal is output as voice through a speaker, or transmitted to another device via a communication line.
[0009]
As a second conventional example, in the above-described conventional example (referred to as a first conventional example), synthesis parameters such as a fundamental frequency pattern, a pitch of a sound, a length of a pause, and a phoneme duration are determined according to a predetermined rule. In order to enhance the naturalness of the determined speech, there is a method of providing a synthetic speech parameter using a prosodic feature parameter extracted by statistically processing the result of analyzing the prosodic feature of the real speech.
[0010]
[Problems to be solved by the invention]
In the first conventional example, the synthesis parameter is determined according to a predetermined rule according to a phoneme symbol string obtained by converting an input text, and is monotonous as compared with natural speech.
[0011]
Also in the second conventional example, natural speech is analyzed to extract and use prosodic feature parameters such as a fundamental frequency pattern, a pitch, a pause length, power, and a phoneme duration. However, there is a problem that various utterance styles such as papers, novels, and conversations cannot be freely expressed by using only a set of prosodic feature parameters of one utterance style.
[0012]
Therefore, it has already been proposed to prepare prosodic feature parameters for a plurality of utterance styles and use them when switching the synthesis parameters, but the relationship between the utterance styles was not clear. Could not be adjusted by the user to emphasize or weaken.
[0013]
Therefore, in a text-to-speech conversion device having means for generating a prosodic feature of a synthesized speech using a predetermined criterion, a difference in the prosodic feature amount caused by a difference in speech style is emphasized or weakened, and various utterances are emphasized. There is a need for a text-to-speech converter that can generate synthesized speech in a style and can read aloud in a prosodic pattern that suits the user's preference.
[0018]
[Means for Solving the Problems]
To solve the above SL problems, the text-to-speech conversion system of the second invention, the text-to-speech converter for converting the audio signal character information inputted, the plurality of speech style comprising at least normal style, the recitation tone A prosodic parameter table holding characteristics, an utterance style designating unit for selecting an utterance style, an emphasis degree designating unit for designating the degree of emphasis of the utterance style, and each of the utterance style and the reference utterance style selected by the utterance style designating unit And a prosody parameter adjustment unit that corrects the prosody parameter in accordance with the emphasis degree and the difference designated by the emphasis degree designation section.
[0019]
Here, it is desirable that the prosody parameter corrected by the prosody parameter adjusting means be at least a phoneme duration or a pitch pattern.
[0020]
As described above, in the text-to-speech conversion apparatus of the present invention, the prosodic parameters can be changed while adjusting the degree of change according to the user's preference, and a synthesized speech more suited to the user's preference can be obtained.
[0021]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, a first embodiment in which a text-to-speech conversion device according to the present invention is applied to a device for Japanese sentences will be described in detail with reference to the drawings. Here, FIG. 1 is a block diagram showing the entire configuration of the text-to-speech converter of the first embodiment.
[0022]
In FIG. 1, a text-to-speech conversion apparatus according to a first embodiment includes a character information input unit 10, a text analysis unit 11, a pronunciation dictionary 12, a synthesis parameter generation unit 13, a speech unit data storage unit 14, a speech synthesis unit 15, The apparatus includes a synthesis parameter changing unit 16 as an utterance style changing unit and an utterance style designating unit 17.
[0023]
Here, the character information input unit 10, the text analysis unit 11, the pronunciation dictionary 12, the synthesis parameter generation unit 13, the speech unit data storage unit 14, and the speech synthesis unit 15 perform the same operations as the conventional text-to-speech conversion device. Therefore, detailed description is omitted.
[0024]
In this embodiment, a case where the reading tone is changed from the reading tone to the conversation tone will be described as an example. The utterance style includes a normal style, a reading style, a conversation style, an announcer style, and the like.
[0025]
The synthesis parameter generation unit 13 extracts the corresponding speech unit data from the speech unit data storage unit 14 based on the phoneme symbol string, and converts the speech synthesis prosody parameters such as the duration of the phoneme, the pause length, the power, and the fundamental frequency pattern. Generate.
[0026]
The utterance style specifying unit 17 is provided with a switch capable of specifying one utterance style to be used from a plurality of utterance styles ranging from a reading style to a conversation style degree.
[0027]
FIG. 4 shows an example in which the utterance style designating unit 17 is formed by software. The left end of the scroll bar indicates the style 1 indicating the most utterance-style utterance style. And the right end indicates the style 10 indicating the utterance style closest to the conversation style. By sliding the bar of the 10-step scroll bar, a desired utterance style can be selected. In FIG. 4, the bar indicates the style 6.
[0028]
The synthesis parameter changing means 16 modifies the speech synthesis prosody parameters in accordance with the user's designation in the speech style designating section 17. In the case of the first embodiment, the synthesis parameter to be changed is the average length per mora. In addition, the synthesis parameter changing means 16 can modify prosodic features such as phoneme duration, fundamental frequency pattern, pitch and power.
[0029]
Next, the detailed operation of the text-to-speech conversion apparatus of the first embodiment will be described with reference to the flowchart of FIG.
[0030]
First, character information (text data such as kanji and kana mixed sentences) is fetched (step 201), and the character information is analyzed and converted into a phonological prosody symbol string for each phrase (step 202).
[0031]
Next, speech unit data to be used sequentially is extracted from the speech unit data storage unit 14 in accordance with the phoneme prosody symbol string (step 203). Then, for each phrase, a prosody parameter (a parameter defining a phoneme duration, a fundamental frequency pattern, power, etc.) is generated based on the phoneme prosody symbol string (step 204). Next, the synthesizing parameter changing means 16 changes the synthesizing parameter generated in step 204 in accordance with the specification of the speech style specifying unit 17 (step 205).
[0032]
A method for changing the synthesis parameters will be described. When comparing spoken and spoken voices, there are various prosodic features between them. First, of the reading voice and the conversation voice, the conversation voice has larger fluctuations in the prosodic parameters. For example, the pitch, the power, the average duration per mora, and the pause length fluctuate more in conversational speech than in reading speech.
[0033]
As an example, when comparing the mora length for each number of mora in a prosodic phrase (phrase) between a reading voice and a conversational or conversational voice, the mora length is almost constant in the reading tone, whereas the mora length in the conversation tone is within one phrase. Tends to become longer as the number of mora becomes smaller.
[0034]
For this, the Transactions of the Acoustical Society of Japan, 1995. Watanabe et al., Which is described in 31-4-6, entitled "Examination of time structure in reading and dialogue speech".
[0035]
Here, the mora is a unit of temporal rhythm such as substantially corresponding to one character of kana.
[0036]
In the first embodiment, based on this feature, the synthesis parameters are changed such that the higher the conversation degree, the longer the average mora length for each mora number in the phrase. For example, when it is desired to provide a difference of ± 1.5 times the mora length duration based on the phrase length of 10 mora, t is the average duration per mora of the reading style, and n is the number of mora of one phrase. Time t 'is
t ′ = − (t / 20) × n + 1. 5t
To convert the duration of each mora. In addition, when the user wants to change the conversation more like conversation, the user can arbitrarily specify the degree by, for example, expanding and contracting by 10% or more.
[0037]
As described above, when the synthesis parameter including the prosodic parameter and the speech unit data is determined, the speech signal is synthesized (step 206) and output (step 207). The output method may be transmission from a speaker or another device via a communication line.
[0038]
As described above, in the text-to-speech conversion apparatus of the first embodiment, in the text-to-speech conversion apparatus having the means for generating the prosodic features of the synthesized speech using the predetermined criteria, Or reading-out) and prosodic features that appear due to differences between other utterance styles, such as conversational style, can be emphasized or weakened, and the synthesis utterance of a certain degree of utterance style can be generated from the normal utterance style, It is possible to realize a text-to-speech conversion device that can read aloud a prosodic pattern that suits the user's preference.
[0039]
Next, a description will be given of a second embodiment in which the text-to-speech conversion device according to the present invention is applied to a device for Japanese sentences.
[0040]
In the second embodiment, the input character information is used to convert natural speech uttered in a plurality of utterance styles into phoneme type durations, pause lengths, power fluctuations, pitch pattern fluctuations (sound pitches). In a text-to-speech conversion device that generates a synthesis parameter using a prosody parameter table created by analyzing each prosody parameter (e.g., difference), and converts the synthesis parameter into a speech signal, the prosody determined according to the utterance style selected by the user. The parameter is compared with the prosodic parameter of the reading style, the difference is obtained, and means for emphasizing or weakening the feature of the prosodic parameter of the speech style is provided.
[0041]
As an example of the prosodic parameters, if it is a phonological duration, it analyzes for each phonological type, the surrounding phonological environment, the phrase position such as the beginning, middle, and end of a sentence, and the mora position, and analyzes each phonological duration. It is an analysis of time.
[0042]
Hereinafter, a text-to-speech converter according to the second embodiment will be described with reference to FIG. Note that the second embodiment will be described by taking as an example a case where the phoneme duration is changed among the prosody parameters such as the duration of each phoneme, the pause length, the power variation, and the pitch pattern. In the second embodiment, a reading-style utterance style is used as a reference utterance style.
[0043]
The text-to-speech conversion device according to the second embodiment includes a character information input unit 10, a text analysis unit 11, a pronunciation dictionary 12, a synthesis parameter generation unit 13, a speech unit data storage unit 14, a speech synthesis unit 15, a plurality of duration tables. 16, an utterance style designation unit 17, an utterance style emphasis unit 20 for changing the phoneme duration, and an utterance style emphasis degree designation unit 19.
[0044]
The character information input unit 10, the text analysis unit 11, the pronunciation dictionary 12, the synthesis parameter generation unit 13, the speech unit data storage unit 14, and the speech synthesis unit 15 perform the same operations as those of the conventional configuration. Description is omitted.
[0045]
The synthesis parameter generation unit 13 extracts the corresponding speech unit data from the speech unit data storage unit 14 based on the phoneme symbol string, and refers to the phoneme duration table of the speech style designated by the speech style designation unit 17. The duration of the phoneme is determined, and prosody parameters for speech synthesis such as pause length, power, and fundamental frequency pattern are generated.
[0046]
The utterance style emphasis degree designation unit 19, reading style and emphasis can switch is provided a speech style degree specified by the speech style specified portion 17, phoneme duration defined with reference to the reading duration table Are compared with each other in the utterance style designated, and the difference is emphasized by the utterance style emphasis unit 20 according to the degree designated by the utterance style emphasis degree designation unit 19.
[0047]
Next, the operation of the text-to-speech converter of the second embodiment will be described with reference to the flowchart of FIG.
[0048]
First, character information (text data such as a sentence mixed with kanji and kana) is fetched (step 601), and the character information is analyzed and converted into a phonological symbol string for each phrase (step 602). Next, speech units to be used sequentially are extracted from the speech unit data storage unit 14 in accordance with the phoneme prosody symbol string (step 603). Then, for each phrase, based on the phonemic prosody symbol string, referring to the duration table of the utterance style specified by the utterance style specification unit 17 and the duration table of the utterance style of the reading style that is the reference utterance style, The phoneme duration is determined, and two types of synthesis parameters (parameters defining the phoneme duration, fundamental frequency pattern, power, etc.) are generated, a designated style and a reading style (step 604). At this time, the duration table is determined by referring to the duration based on factors (such as the type of the accurate phoneme, the surrounding environment, the phrase position, the mora position in the phrase, etc.) obtained by analyzing the natural speech in advance.
[0049]
Next, the utterance style emphasizing unit 20 emphasizes the difference between the designated utterance style continuation time (Tn) and the reading tone continuation time (Ts) according to the degree specified by the utterance style emphasis degree specifying unit 19. To change the phoneme duration. For example, when the emphasis coefficient is α, the final phoneme duration T is T = Ts + α (Tn / Ts−1) Ts
Can be calculated. The emphasis coefficient α may be changed from 0 to several times depending on the degree specified by the emphasis section specification section 19 (step 605).
[0050]
As described above, when the synthesis parameter including the prosodic parameter and the speech unit data is determined, the speech signal is synthesized (step 606) and output (step 607). The output method may be output from a speaker or transmission to another device via a communication line.
[0051]
According to the text-to-speech converter of the second embodiment described above, it is possible to change the utterance style by changing the phoneme duration according to the user's preference.
[0052]
In each of the above embodiments, the text-to-speech conversion apparatus for Japanese sentences has been described, but the present invention can of course be applied to a text-to-speech conversion apparatus for other language sentences.
[0053]
【The invention's effect】
As described above, according to the present invention, it is possible to emphasize or weaken a difference in prosodic feature amount that appears due to a difference in utterance style, to generate a synthetic sound of various utterance styles, and to adjust a rhythm to a user's preference. A text-to-speech converter capable of reading out in a pattern can be realized.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a text-to-speech conversion apparatus according to a first embodiment.
FIG. 2 is a diagram showing a conventional text-to-speech conversion device.
FIG. 3 is a flowchart showing an operation of the text-to-speech conversion apparatus in FIG. 1;
FIG. 4 is an explanatory diagram of a conversation style designating unit 17 of the text-to-speech converter of FIG.
FIG. 5 is a block diagram illustrating a text-to-speech conversion apparatus according to a second embodiment.
FIG. 6 is a flowchart showing an operation of the text-to-speech conversion apparatus in FIG. 5;
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 10 ... Character information input part, 11 ... Text analysis part, 12 ... Speech dictionary, 13 ... Synthesis parameter generation part, 14 ... Speech unit data storage part, 15 ... Speech synthesis part, 16 ... Synthesis parameter change means, 17 ... Speech Style designation unit, 18: prosodic parameter table, 19: speech style emphasis degree designation unit, 20: speech style emphasis unit.

Claims (2)

In a text-to-speech converter for converting input character information into a speech signal,
A prosodic parameter table that retains characteristics in at least a plurality of utterance styles including a normal style and a reading style;
An utterance style designating section for selecting an utterance style;
An emphasis level designating section for designating the emphasis level of the utterance style;
A difference calculation unit that calculates a difference between each prosodic parameter of the utterance style selected by the utterance style designation unit and the reference utterance style,
A text-to-speech conversion apparatus comprising: a prosody parameter adjusting unit that corrects a prosody parameter according to the emphasis degree designated by the emphasis degree designation unit and the difference. 2. The text-to-speech converter according to claim 1, wherein the prosodic parameter corrected by the prosodic parameter adjusting means is at least a phoneme duration or a pitch pattern.

Images