JPH0990987A - Method and device for voice synthesis - Google Patents

Abstract

PROBLEM TO BE SOLVED: To generate smooth pitch patterns without performing a wasteful point pitch setting. SOLUTION: A morpheme analysis is conducted against an inputted KANJI (Chinese characters) and KANA (Japanese alphabets) mixed sentence in a language analysis processing section 11 and reading and accent information is generated. In a voice synthesis section 2, a phoneme continuation time computation processing section 21 decides the phoneme continuation time based on the reading information. An every phoneme point pitch setting processing section 22 sets point pitch positions at one or two points in four equally divided continuation time of each phoneme based on the accent information and the height of the pitch is set at that location. The setting time of the point pitch and the frequency are decided by the phoneme position which sets the accent type and the pitch and the kind of the phoneme or phoneme environment. The obtained phoneme information, the phoneme continuation time and the point pitch setting position are written into a file 31 as a voice symbol column by a voice symbol column generation processing section 23. Then, the pitch pattern of the voice to be synthesized based on the phoneme continuation time of the voice symbol column and the point pitch position and phoneme parameters are generated based on the phoneme information and finally a voice is synthesized.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice synthesizing method and apparatus suitable for obtaining a smooth intonation similar to a human voice.

[0002]

2. Description of the Related Art As a speech synthesizer of this type, a text-to-speech synthesizer (Text) that converts a text (sentence) into a symbol string (speech symbol string) composed of phonemes and prosody and generates speech from the symbol string.
There is known a speech rule synthesizing device that performs -to-Speech (TTS) processing (sentence speech conversion processing). The sentence speech synthesis processing in this speech rule synthesis apparatus is roughly divided into a language processing unit and a speech synthesis unit. Taking Japanese rule synthesis as an example, it is generally performed as follows. .

First, in the language processing unit, linguistic processing such as morphological analysis and syntactic analysis is performed on the input text (Kanji / Kana mixed sentence) to perform processing such as decomposition into morphemes and estimation of dependency relationships. At the same time, the reading and accent types are given to each morpheme. After that, the language processing unit determines the accent type for each phrase (hereinafter, referred to as an accent phrase) that is a delimiter at the time of reading, by using the accent movement rule such as a compound word for the accent. In the language processing unit of the normal speech rule synthesizing device, the reading and accent type for each accent phrase thus obtained can be output as a symbol string (speech symbol string).

Next, the voice synthesis unit determines the duration of each phoneme included in the obtained reading according to a predetermined rule based on the phoneme environment of the phoneme. Then in speech synthesis,
According to the "reading" and "phoneme duration" obtained as described above, a speech unit file prepared in advance,
That is, the speech feature files are sequentially read from a speech unit file in which the characteristic parameters of the speech are stored in a predetermined synthesis unit, for example, a unit of consonant + vowel (hereinafter, referred to as CV), and the speech units are connected and synthesized. Generate a characteristic parameter sequence of the power speech.

Further, in the voice synthesizing unit, a point pitch is set based on the accent type at the time when the pitch changes in height, and linearly interpolates between a plurality of set point pitches to generate a pitch accent component. Then, an intonation component (usually a monotonically decreasing straight line on the frequency logarithmic axis) is superimposed on this to generate a pitch pattern. Then, in the voice synthesis unit, the periodic pulse obtained from the prosody parameter in the voiced section and the random noise in the unvoiced section are used as the sound source signals, respectively, to calculate the filter coefficient from the feature parameter series of the voice, and apply it to the synthesizer filter to obtain a desired value. Synthesize voice.

Further, as described above, the language processing unit can output the reading and accent type for each accent phrase as a symbol string. Therefore, if the linguistic analysis result is incorrect, the outputted symbol string is edited and then speech synthesis is performed. A voice can be synthesized by inputting it to the processing unit.

[0007]

Conventionally, the above-described speech synthesis technique has been available, but the following problems occur in the pitch pattern generated by the speech synthesis device to which the speech synthesis technique is applied. was there.

That is, in the conventional speech synthesis technique, when the pitch accent component is generated based on the accent type, as shown in FIG. 16, the point pitch is set to only one pitch for each syllable (one syllable). Therefore, there is a problem that a fine pitch control cannot be performed and a smooth pitch pattern cannot be obtained.

As a method for improving this, Japanese Patent Laid-Open No.
As in the invention described in Japanese Patent No. 164800, it is sufficient to increase the number of point pitches set per syllable, but in many cases it is necessary to set a point pitch that is not necessary.

Further, a finer control of the pitch is made possible by a designation method such as "setting the point pitch setting position after several ms (milliseconds) from the time point of the beginning of what syllable from the beginning". A method is conceivable, but if the accent type is the same, the point pitch is constantly set at the same time after the same phoneme from the beginning, so a smooth pitch pattern cannot be obtained and the synthesized speech It is possible that intonation becomes unnatural.

Further, the prior art has the following problems. That is, in a conventional speech synthesizer, speech reading and accent type to be synthesized are once output as a symbol string (speech symbol string) by language processing, and this is input to the speech synthesis processing to synthesize speech. However, since only the accent type is given to the symbol string as a parameter for giving the intonation to the synthesized speech, the fine control of pitch pattern generation is entrusted to the speech synthesis processing, and the user can make the smoothness independently. It was impossible to create a pitch pattern.

The present invention has been made in consideration of such circumstances, and an object thereof is to realize a voice synthesizing method which enables fine pitch control without setting an unnecessary point pitch and can generate a smooth pitch pattern. And to provide a device.

Another object of the present invention is that the user can control the pitch pattern by changing the phonetic symbol string, the pitch pattern can be easily controlled by this phonetic symbol string, and a smooth pitch pattern can be obtained by only a small number of designations. It is to provide a voice synthesizing method and device capable of generating a voice.

Still another object of the present invention is that a user who is not familiar with voice synthesis technology can easily edit a voice symbol string and can create a pitch pattern that is the same as a voice uttered by a human. An object is to provide an editing device.

[0015]

In order to solve the above-mentioned problems, the speech synthesis method according to the first aspect of the present invention provides a plurality of point pitches on the basis of accent information of a word or a phrase. A voice synthesis method for synthesizing a voice based on a pitch pattern generated by interpolating between a plurality of point pitches, wherein a position on the time axis at which the point pitch is given is a phoneme of a phoneme to give the point pitch. It is characterized in that it is determined based on at least one of the environment and the phoneme type.

Next, the speech synthesis apparatus according to the second aspect of the present invention is input with a speech unit storage medium for accumulating speech characteristic parameters by speech units consisting of predetermined synthesis units. Characteristic parameter generation processing means for generating a characteristic parameter of a speech to be synthesized by connecting the read speech units from the speech unit storage medium based on the phoneme information to be synthesized; It is a phoneme-based point pitch setting processing means for setting a point pitch for each phoneme using accent information input corresponding to phoneme information, and at least the phoneme environment or phoneme type of the phoneme to which the set point pitch belongs. The point pitch setting processing means for each phoneme that determines the position of the point pitch on the time axis based on one of the points and the adjacent point pitch set by the point pitch setting processing means for each phoneme are complemented. The pitch pattern generation processing means for generating a pitch pattern of the voice, the characteristic parameter of the voice to be synthesized generated by the characteristic parameter generation processing means, and the pitch pattern generated by the pitch pattern generation processing means And a synthesizing filter processing means for synthesizing.

Next, the voice synthesizing method according to the third aspect of the present invention is different from the voice synthesizing method according to the first aspect, in that the position on the time axis where the point pitch is given is not a phoneme, It is characterized in that the point pitch is determined based on at least one of the syllable environment and the syllable type of the syllable to be given.

Next, in the voice synthesizing apparatus according to the fourth aspect of the present invention, instead of the phoneme per-point pitch setting processing means in the voice synthesizing apparatus according to the second aspect, input is made corresponding to phonological information. Syllabic point pitch setting processing means for performing point pitch setting for each syllable using accent information,
A syllable-by-syllable pitch setting processing means is provided for determining the position of the point-pitch on the time axis based on at least one of the syllable environment of the syllable to which the set point pitch belongs and the syllable type. It is characterized in that a pitch pattern of a voice is generated by interpolating between the set adjacent point pitches.

Next, the voice synthesizing method according to the fifth aspect of the present invention and the voice synthesizing apparatus according to the sixth aspect of the present invention set at least 4 points within the duration of one phoneme in order to set the point pitch for each phoneme. It is characterized in that the point pitch specifiable time points of points are set, and the point pitch is designated for two or less points.

Next, a voice synthesizing method according to a seventh aspect of the present invention and a voice synthesizing apparatus according to the eighth aspect are a voice synthesizing method according to the fifth aspect and a voice synthesizing apparatus according to the sixth aspect. Unlike the point pitch setting for each phoneme,
To set the point pitch for each syllable, set at least four points that can be specified for the point pitch within the duration of one syllable.
It is characterized in that the point pitch is designated for two or less points among them.

Next, the speech synthesizer according to the ninth aspect of the present invention analyzes the text to be synthesized by the speech synthesizer according to the second, fourth, sixth or eighth aspect. It is characterized in that a language analysis processing means for generating phonological information and accent information is added to provide a text-to-speech conversion (TTS) processing (sentence-speech synthesis processing) function.

Next, in the speech synthesis method according to the tenth aspect of the present invention, a plurality of speech features prepared in advance are input with a speech symbol string describing prosody information including phonological information and pitch information of the speech as an input. From the group of speech units consisting of parameters, a plurality of speech units are selected and connected according to the phoneme information in the speech symbol string to generate parameters for expressing the phoneme of the speech, and in the speech symbol string. A pitch synthesizing method for generating a pitch pattern of a voice according to the prosody information of the voice, and synthesizing a voice based on a pitch pattern and a parameter expressing the phonology of the voice, wherein the pitch information in the voice symbol string is a point pitch. It is characterized by having been given.

Next, a speech synthesizer according to an eleventh aspect of the present invention comprises a speech unit storage medium for accumulating a group of speech units, each of which is prepared in advance and composed of a plurality of speech characteristic parameters. Inputting at least phonological information in a phonetic symbol sequence describing prosodic information including phonological information of speech to be synthesized and pitch information given by a point pitch, and inputting at least phonological information from the speech element storage medium according to the phonological information. Phonological parameter generating means for generating a parameter expressing a phoneme of a voice by selecting and connecting phonemes, and at least prosodic information in the phonetic symbol string are input, and a pitch pattern of a voice to be synthesized according to the prosodic information is input. Prosody parameter generating means for generating, parameters expressing the phoneme generated by the phonological parameter generating means, and pitch generated by the prosody parameter generating means Characterized in that a synthesizing filter processing means for synthesizing a speech from the turn.

Next, in the voice synthesizing method according to the twelfth aspect of the present invention and the voice synthesizing apparatus according to the thirteenth aspect, the pitch synthesizing method according to the tenth aspect provides the pitch information in the voice symbol sequence. Different from the pitch information applied in the speech synthesizer according to the eleventh aspect, not only is it given as a point pitch, but the position on the time axis of that point pitch is also a point pitch based on the starting point of each phoneme. It is characterized in that it is specified.

Next, in the voice synthesizing method according to the fourteenth aspect of the present invention and the voice synthesizing apparatus according to the fifteenth aspect, not only the pitch information in the voice symbol string is given as a point pitch, but also the point pitch. The position on the time axis is different from the pitch information applied in the speech synthesis method according to the twelfth aspect and the speech synthesis device according to the thirteenth aspect, and is not the start point of each phoneme but the syllable of each syllable. It is characterized in that the point pitch is specified based on the starting point.

Next, the speech synthesis method according to the sixteenth aspect of the present invention and the speech synthesis device according to the seventeenth aspect include at least four points within the duration of each phoneme included in the phoneme information in the phonetic symbol sequence. Is set, and the prosodic information in the phonetic symbol string is described by setting the point pitches for the time points of two points or less among them.

Next, the speech synthesizing method according to the eighteenth aspect of the present invention and the speech synthesizing apparatus according to the nineteenth aspect include at least four points within the duration of each phoneme included in the phoneme information in the phonetic symbol sequence. By setting the point pitch specifiable time points of, and specifying the position on the time axis of the point pitch based on the starting point of each phoneme for the time points of 2 points or less, the prosodic information in the phonetic symbol string is It is characterized by being described.

Next, the speech synthesis method according to the twentieth aspect of the present invention and the speech synthesis apparatus according to the twenty-first aspect include at least four points within the duration of each syllable included in the phoneme information in the phonetic symbol string. Is set, and the prosodic information in the phonetic symbol string is described by setting the point pitches for the time points of two points or less among them.

Next, the speech synthesis method according to the twenty-second aspect of the present invention and the speech synthesis device according to the twenty-third aspect include at least four points within the duration of each syllable included in the phoneme information in the phonetic symbol sequence. The point probable time in the phonetic symbol sequence is set by setting the points on the time axis that can be specified as the point pitch, and the points on the time axis based on the start point of each syllable are specified for the points less than 2 points. It is characterized by being described.

Next, a phonetic symbol string editing apparatus according to a twenty-fourth aspect of the present invention inputs a phonetic symbol string in which prosodic information including phonological information of a voice and pitch information given by point pitch is described as an input. From a group of prepared speech units consisting of a plurality of characteristic parameters of speech, a plurality of speech units are selected and connected in accordance with the phoneme information in the speech symbol sequence to generate a parameter expressing a phoneme of speech. At the same time, a voice symbol string edit is applied to a voice synthesizer that generates a voice pitch pattern according to the prosodic information in the voice symbol string and synthesizes the voice based on the parameters and pitch pattern expressing the phoneme of the voice. A device for indicating a plurality of point pitches according to the pitch information in the above-mentioned phonetic symbol sequence, and indicating a change of the point pitch displayed by this display means. And a voice symbol string correction processing unit for changing the corresponding point pitch in the display in response to a change instruction from the input unit and reflecting the change result in the voice symbol string. And

Next, a phonetic symbol string editing apparatus according to a twenty-fifth aspect of the present invention is replaced with the display means, the input means and the phonetic symbol string correction processing means in the phonetic symbol string editing apparatus according to the twenty-fourth aspect. Display means for graphically displaying a plurality of point pitches according to the pitch information in the phonetic symbol sequence, input means for inputting an instruction to move the position of the point pitch displayed by the display means, and this input means A voice symbol string correction processing means for moving the position of the corresponding point pitch being displayed on the screen in response to a movement instruction by the user and reflecting the new position of the new point pitch after the movement in the voice symbol string. Is characterized by.

Next, a phonetic symbol string editing apparatus according to a twenty-sixth aspect of the present invention uses the pitch information in the phonetic symbol string in place of the display means in the phonetic symbol string editing apparatus according to the twenty-fifth aspect. A display means for displaying a plurality of point pitches and a pitch contour obtained by interpolating these point pitches in a graph is provided.

Next, in the phonetic symbol string editing device according to the twenty-seventh aspect of the present invention, the phoneme information in the phonetic symbol string is displayed on the display means of the phonetic symbol string editing device according to the twenty-fifth or twenty-sixth aspect. The pitch contour obtained by analyzing the reference voice corresponding to is displayed as a reference pattern in a graph.

With the above-described structure, in the invention according to the first to fourth aspects, a pitch pattern that is smoother and closer to the intonation of the human voice is generated from the phonological information and the accent type as compared with the prior art. You can

Further, in the inventions according to the fifth to eighth aspects, extremely fine pitch control is enabled while the number of set point pitches is minimized, and by extension, a voice closer to the suppression of human voice is provided. Can be synthesized.

In the invention according to the ninth aspect, sentence-speech conversion (TT) for synthesizing speech from text.
While realizing the S) function, it is possible to obtain the same effect as that of the invention according to the second, fourth, sixth or eighth aspect.

Further, in the inventions according to the tenth to twenty-eighth aspects, it becomes possible for a user to easily provide a user's favorite pitch pattern for voice synthesis with a minimum of labor.

[0038]

Embodiments of the present invention will be described below with reference to the drawings. [First Embodiment] FIG. 1 is a block diagram showing the schematic arrangement of a speech synthesis apparatus according to the first embodiment of the present invention.

This speech synthesizer is realized by executing dedicated software (sentence / speech conversion software) on an information processing apparatus such as a personal computer, and has a speech / speech conversion (TTS) processing function, that is, a text. It has a sentence-to-speech conversion processing (sentence-to-speech synthesis processing) function for generating speech from, and its functional configuration is roughly divided into a language processing section 1, a speech synthesis section 2, and a speech symbol string editing section 3.

The language processing unit 1 manages a process of analyzing an input sentence, for example, a kanji / kana mixed sentence, to generate reading information and accent information. The speech synthesizing unit 2 generates a speech symbol string in which the phoneme information, the duration of each phoneme, and the point pitch position for each phoneme are described based on the sentence analysis result in the language processing unit 1. It controls the process of generating speech based on the symbol string. The voice symbol sequence editing unit 3 controls the process of editing the voice symbol sequence converted and generated by the voice synthesizing unit 2 or created by the user according to the user operation.

In the speech synthesizer shown in FIG. 1, a document (here, Japanese document) to be subjected to sentence-to-speech conversion (speech) is stored as a text file (not shown). In this device, according to the sentence voice conversion software,
Sentences containing kanji and kana are read one by one from the file, and the language processing unit 1 and the speech synthesis unit 2 perform sentence-speech conversion processing described below to synthesize speech.

First, the kana-kana mixed sentence read from the text file is processed by the language analysis processing unit 1 in the language processing unit 1.
Input to 1. The linguistic analysis processing unit 11 performs morphological analysis of an input kanji / kana mixed sentence and generates reading information and accent information. Morphological analysis is an operation of analyzing which character string forms a phrase in a given sentence, and what the structure of the word is.

For that purpose, the language analysis processing unit 11 uses a morpheme dictionary 12 having a morpheme, which is the minimum constituent element of a sentence, as an entry word, and a connection rule file 13 in which connection rules between morphemes are registered. . That is, the language analysis processing unit 11
Finds all the morpheme sequence candidates obtained by matching the input sentence with the morpheme dictionary 12 (brute force method), and from among them, refers to the connection rule file 13 to find a combination that can be connected grammatically before and after. Output. In the morpheme dictionary 12, grammatical information used at the time of analysis and morpheme reading and accent types are registered. For this reason, if a morpheme is determined by morphological analysis, reading and accent type can be given at the same time.

For example, "Go to the park and read a book."
If you perform a morphological analysis on the sentence, / park / go / go / book / read / read /. And morphemes. At the same time, a reading and an accent type are given to each morpheme, and / ko: eN / e / iqte / ho ^ N / o / yomi
/ Ma ^ su /. Here, the symbol “N” is the sound repellency (“n”),
"Q" is a consonant sound ("tsu"), and ":" is a long sound ("-")
Respectively are represented. In addition, a morpheme containing "^" means that the pitch is high in the phoneme immediately before that and the pitch is dropped in the phoneme immediately after that. When there is no "^", it means that the accent is a flat plate type.

By the way, when humans read a sentence, they do not read by adding accents in units of such morphemes, but by reading several morphemes together and adding accents to the unit.

Therefore, in consideration of such a situation, the language analysis processing unit 11 further collects morphemes in units of giving one accent (hereinafter referred to as an accent phrase) and, at the same time, moves the accents due to the collection. Also estimate. In addition to this, the language analysis processing unit 11 also adds information such as vowel devoicing and pauses (breathing) during reading, and in the above example, finally generates the following reading information.

/ Ko: eNe / iqte. / Ho ^ No
/ Yomima ^ s (u) / Here, the period "." Represents breathing, and "()" represents unvoiced vowels. Now, when the reading information is generated by the language analysis processing unit 11 in the language processing unit 1 as described above, the phoneme duration calculation processing unit 21 in the speech synthesis unit 2 is activated.

The phoneme duration calculation processing unit 21 determines the duration (unit is ms) of each phoneme included in the input sentence according to the reading information generated by the language analysis processing unit 11. The determination process of the duration in the phoneme duration calculation processing unit 21 is realized by an extremely simple algorithm in which the positions of the boundary (CV crossing) between the consonant (C) and the vowel (V) are arranged at equal intervals. Has been done. A specific example of this is shown in FIG. 2 for the accent phrase “yomimasu”.

When the phoneme duration calculation processing unit 21 determines the duration of each phoneme included in the input sentence, the phoneme point pitch setting processing unit 22 in the same speech synthesis unit 2 is activated. The phoneme-based point pitch setting processing unit 22 sets the point pitch position based on the phoneme duration determined by the phoneme duration calculation processing unit 21 and the accent information determined by the language analysis processing unit 11. Here, as the time for setting the point pitch, one point or two points of the duration of each phoneme divided into four at equal intervals (that is, equally divided into four) is determined. A specific example of this point pitch setting is shown in FIG. 3 for the case of the accent phrase “yomimasu” whose phoneme duration is determined as shown in FIG. In FIG. 3, the horizontal axis represents time and the vertical axis represents pitch (unit is octave oct),
The position of the white circle is the point pitch position. Here, the point pitch position (x, y) is 0 to 3 when the phoneme duration is divided into four equal parts.
(However, the final time point of the phoneme at the end of the phrase is 4) and y indicating the pitch (unit oct).

In the present embodiment, the time and frequency (pitch frequency) of the point pitch setting are the accent type, the position of the phoneme for which the pitch is set in the accent phrase, the type of the phoneme or the phoneme environment. Depends on This will be described in detail below with reference to FIGS. 4 (a) and 4 (b).

First, in FIG. 4A, the accent phrase is set to the first vowel counting from the beginning (phrase) of the accent phrase of the flat plate type or undulating accent starting from a consonant, that is, the vowel immediately after the first consonant of the accent phrase. The following is a rule of assigning a point pitch position to be used.

As is clear from FIG. 4 (a), when the accent phrase starts with a consonant, at the point in time when the point pitch is specified in the vowel located immediately after it, the immediately preceding consonant is a voiced consonant or unvoiced. It depends on whether it is a consonant and whether or not the phoneme immediately after the vowel is a syllable, that is, the type of phoneme before and after, in other words, the phoneme environment. There are four combinations (phoneme environments) of phoneme types before and after this (vowel to be set for point pitch) (phoneme environment), and point pitch position assignment rules (rule Nos. (1) to (4)) corresponding to the number of combinations. Is prepared.

Next, FIG. 4B shows the point pitch set for the second (second syllable) vowel or vowel when only the vowel or vowel is counted from the beginning of the accent phrase of the flat plate type or undulating accent. The position assignment rule is shown.

As is apparent from FIG. 4 (b), the point pitch designation time point in the second vowel or vowel sound is whether or not it is syllable, that is, the type of the phoneme itself (phoneme type). ). Here, a point pitch position assignment rule (rule No.
(1)) and a point pitch position assignment rule (rule No. (2)) when the phoneme type is other than sound repellency are prepared.

The point pitch positions determined based on the point pitch position assignment rules shown in FIGS. 4A and 4B are shown in FIGS.
It shows in (d). In FIGS. 5A to 5D, as in FIG. 3, the horizontal axis represents time, the vertical axis represents pitch (unit is octave oct), and the positions of white circles are point pitch positions.

First, in the case where the phoneme at the beginning of the accent phrase is a voiced consonant and the second syllable is not a sound utterance, for example, in the case of the accent phrase "yomimasu (read)", it is shown in FIG. Rule No. (4) and the rule No. in FIG. (2) and are applied, and FIG.
A point pitch position as shown in (a) is generated (set).

Next, in the case of an accent phrase in which the phoneme at the beginning of the accent phrase is a voiced consonant and the second syllable is a syllable, for example, in the case of the accent phrase "roNzimasu (discussed)", FIG. Rule No. (3)
And the rule No. in FIG. (1) and are applied to generate a point pitch position as shown in FIG.

Next, in the case where the phoneme at the beginning of the accent phrase is an unvoiced consonant and the second syllable is an unaccented accent phrase, for example, in the case of the accent phrase "trimasu", in FIG. 4A. Rule No. Rule No. 2 in (2) and FIG. (2) and are applied, and FIG.
A point pitch position as shown in (c) is generated.

Next, in the case of an accent phrase in which the phoneme at the beginning of the accent phrase is a voiceless consonant and the second syllable is a syllable, for example, in the case of the accent phrase "kaNzimasu (feel)", FIG. Rule No. (1)
And the rule No. in FIG. (1) is applied to generate a point pitch position as shown in FIG.

A pitch pattern obtained by interpolating a plurality of point pitch positions generated in this way with a straight line (see FIG. 5).
(Thick line portions in (a) to (d)) closely approximate that of a voice uttered by an actual human.

In a pitch pattern of an actual flat or undulating accent of a voice, a voiced consonant tends to start at a lower pitch than an unvoiced consonant as a leading consonant (FIG. 5 (a)). 5 (c) or FIG. 5 (b) and FIG. 5 (d), all the pitches at the beginning of the consonant part start from 0 [oct], but in the unvoiced consonant part, the pitch is not used during voice synthesis. Since it is not present, the pitch is given from the beginning 0.3 [oct] of the vowel part immediately after, and as a result, the pitch at the beginning of the accent is higher than that of the accent phrase starting with the voiced consonant. Further, it is known that when the second syllable is sound repellant, the start time and the end time of the pitch rise tend to be earlier than when it is not.

Conventionally, the pitch has been given regardless of the phonological species and the phonological environment without considering the actual pitch characteristics of the voice. That is, in the conventional technique as shown in FIG. 16, since only one point pitch can be set at the center of the syllable, the setting of the start pitch can be barely changed depending on the point pitch of the first syllable. Although possible, the start time of the pitch rise and the end time of the pitch rise can be set only at the center time of the first syllable and the center time of the second syllable, respectively. Therefore, in the conventional technology,
It was difficult to approximate the pitch pattern of the actual human voice.

By the way, in order to carry out precise pitch control, it is necessary to set the point pitch at intervals such that the duration of one syllable is divided into four equal parts, as pointed out in Japanese Patent Laid-Open No. 3-164800. is there. Even when considered in terms of phonemes, since vowels without consonant parts have the same phonemes and syllables, it is appropriate to give a point pitch to each phoneme when it is divided into, for example, four equal parts. However, even if it is divided into four parts, it is inefficient to give the pitch frequency at all points, and one or two points are sufficient, and there is a phoneme that does not need to give the point pitch at all.

Of course, as can be seen from the example of FIG. 3, since the consonant part has a relatively short phoneme duration, the division is fine and the interval at the point pitch designation is narrow, but the point pitch can be finely designated. Does not adversely affect pitch control, and it is convenient for processing to divide the same number of vowels. It is also possible to change the number of divisions when the phoneme is a consonant, but this may cause confusion when the user rewrites the phonetic symbol string generated thereafter.

Now, when the point pitch position for each phoneme is determined by the phoneme-based point pitch setting processing section 22, the phonetic symbol string generation processing section 23 in the same speech synthesis section 2 is activated. The phonetic symbol string generation processing unit 23 includes the phoneme information (phoneme sequence) obtained by the language analysis processing unit 11, the phoneme durations obtained by the phoneme duration calculation processing unit 21, and the phoneme per-point pitch setting processing unit. From the point pitch position obtained by No. 22,
A phonetic symbol string describing these is generated.

FIG. 6 shows the accent phrase "yomimasu".
An example of a phonetic symbol string for "(read)" is shown. As is clear from the figure, the phonetic symbol string includes phoneme information (phoneme sequence) 610 and prosody information 620. This prosody information 620 includes intonation component 621 and phonological information 6
10 phoneme durations and point pitch position information (phoneme duration / pitch information) 622 determined for each phoneme. The intonation component 621 indicates the pitch spontaneous drop component, and the description example of “166, 0.2” in FIG. 6 is 0.2 from 166 Hz (hertz) toward the end of the phrase.
[Oct] Shows the natural falling component of the falling pitch. Further, “;” in the phonetic symbol string in FIG. 6 indicates the end of the accent phrase.

The speech symbol string generation processing section 23 passes the generated speech symbol string to the pitch pattern generation processing section 24 in the speech synthesis section 2 (through a memory not shown), and the pitch pattern generation processing section 24 processes it. Alternatively, the phonetic symbol string editing unit 3 can edit the phonetic symbol string once and write the phonetic symbol string in the phonetic symbol string file 31 in the phonetic symbol string editing unit 3. The edit processing of the voice symbol sequence in the voice symbol sequence file 31 by the voice symbol sequence editing unit 3 will be described in detail in a second embodiment described later, and the description thereof is omitted here. However, in the second embodiment, the phonetic symbol string editing unit 6 is used instead of the phonetic symbol string editing unit 3.

Now, the pitch pattern generation processing unit 24 in the voice synthesis unit 2 is based on the voice symbol string passed from the voice symbol string generation processing unit 23 or stored in the voice symbol string file 31. A process of generating (prosodic parameter) is performed. That is, the pitch pattern generation processing unit 24 linearly interpolates between the point pitches based on the phoneme duration and the point pitch position for each phoneme set in the target phonetic symbol string to perform pitch accent components. For example 1
It is generated every 0 ms (millisecond). The pitch pattern generation processing unit 24 further adds an intonation component (spontaneous drop component of pitch) set in the target speech symbol sequence to obtain a pitch pattern for every 10 ms of speech to be synthesized.

On the other hand, the phoneme parameter generation processing unit 25 in the speech synthesis unit 2 receives the phoneme information (phoneme) in the phonetic symbol string passed from the phonetic symbol string generation processing unit 23 or stored in the phonetic symbol string file 31. The process of generating a phoneme parameter based on the sequence) is performed in parallel with the pitch pattern generation process by the pitch pattern generation processing unit 24 as follows, for example.

First, in the present embodiment, a total of 137 voices obtained by cutting out a 19th-order cepstrum obtained by analyzing an actual voice with a window length of 20 ms and a window period of 10 ms by a consonant + vowel (CV) unit are analyzed. A speech unit file (not shown) in which the units are accumulated is prepared. The contents of this speech unit file are stored in a speech unit region (hereinafter, referred to as a speech unit memory) 26 secured in, for example, a main memory (not shown) at the start of the sentence-speech conversion process according to the sentence-speech conversion software. It is assumed to be loaded.

The phonological parameter generation processing unit 25, in accordance with the phonological information in the speech symbol string passed from the speech symbol string generation processing unit 23 or stored in the speech symbol string file 31, has the above-mentioned CV-based speech unit. Are sequentially read from the voice unit memory 26, and the phoneme parameters (feature parameters) are generated by connecting the read voice units.

A pitch pattern is generated by the pitch pattern generation processing unit 24, and a phoneme parameter generation processing unit 25 is generated.
When the phoneme parameter is generated by the above, the synthesis filter processing unit 27 in the voice synthesis unit 2 is activated. As shown in FIG. 7, the synthesis filter processing unit 27 includes a white noise generation unit 271, an impulse generation unit 272, a driving sound source switching unit 273, and an LMA (Log Magnitude Approximation).
) Filter (logarithmic amplitude approximation filter) 274, and synthesizes speech as follows from the generated pitch pattern (prosodic pattern) and phonological parameters.

First, in the voiced part (U) of the voice, the driving sound source switching part 273 switches it to the impulse generating part 272 side. The impulse generation unit 272 generates impulses at intervals according to the pitch pattern generated by the pitch pattern generation processing unit 24, and drives the LMA filter 274 by using this impulse as a sound source. On the other hand, in the unvoiced part (V) of the voice, the driving sound source switching part 273 switches it to the white noise generating part 271 side. The white noise generation unit 271 generates white noise, and drives the LMA filter 274 using this white noise as a sound source.

The LMA filter 274 uses the speech cepstrum (cepstral parameter) directly as a filter coefficient. In the present embodiment, since the phoneme parameter generated by the phoneme parameter generation processing unit 25 is the cepstrum as described above, this phoneme parameter becomes the filter coefficient of the LMA filter 274 and is driven by the sound source switched by the drive sound source switching unit 273. As a result, a synthetic voice is output.

The voice synthesized by the synthesis filter processing unit 27 (the LMA filter 274 therein) is a D / A (not shown).
A (digital / analog) converter converts it into an analog signal and outputs it to a speaker or the like through an amplifier so that it can be heard as sound. [Second Embodiment] FIG. 8 is a block diagram showing the schematic arrangement of a speech synthesis apparatus according to the second embodiment of the present invention.

The point that the speech synthesizer shown in FIG. 8 differs most from the speech synthesizer shown in FIG. 1 is that instead of the point pitch setting processing unit 22 for each phoneme, the point pitch setting processing unit for each syllable that controls the point pitch setting for each syllable. 42 is provided.

This speech synthesizer has a sentence-speech conversion processing function realized by executing dedicated software (sentence-speech conversion software), that is, a sentence-speech conversion processing (sentence-speech synthesis processing) function for generating speech from text. Things
The functional configuration is roughly divided into a language processing unit 4 and a voice synthesis unit 5.
And a phonetic symbol string editing unit 6.

The language processing section 4 controls the processing of analyzing an input sentence, for example, a kanji / kana mixed sentence, to generate reading information and accent information. The speech synthesis unit 5 describes, based on the result of sentence analysis by the language processing unit 4, a phonetic symbol string in which phonological information, duration of each syllable (of consonant parts and vowel parts), and point pitch position of each syllable are described. And a process of generating a voice based on this voice symbol string. The voice symbol string editing unit 6 controls the process of editing the voice symbol string converted and generated by the voice synthesizing unit 5 or created by the user in accordance with a user operation.

In the speech synthesizer of FIG. 8, a document to be subjected to sentence-to-speech conversion (speech) (Japanese document here).
Is stored as a text file (not shown). In this device, according to the sentence-speech conversion software, the kanji and kana mixed sentences are read from the file one by one,
The language processing unit 4 and the speech synthesis unit 5 perform the sentence-speech conversion processing described below to synthesize speech.

First, the kana-kana mixed sentence read from the text file is processed by the language analysis processing unit 4 in the language processing unit 4.
Input to 1. The language analysis processing unit 41 performs morphological analysis of an input kanji / kana mixed sentence and generates reading information and accent information.

For this purpose, the language analysis processing unit 41 uses a morpheme dictionary 42 having a morpheme, which is the minimum constituent element of a sentence, as an entry word, and a connection rule file 43 in which connection rules between morphemes are registered. . That is, the language analysis processing unit 41
Finds all the morpheme sequence candidates obtained by matching the input sentence with the morpheme dictionary 42 (brute force method), and refers to the connection rule file 43 from among them to find a combination that can be connected grammatically before and after. Output. In the morpheme dictionary 42, morpheme readings and accent types are registered together with grammatical information used during analysis. For this reason, if a morpheme is determined by morphological analysis, reading and accent type can be given at the same time.

For example, as in the first embodiment,
Morphological analysis is performed for the sentence "Go to the park and read the book." / Park / Go / Go / Book / Read / Read /. And morphemes. At the same time, each morpheme is given a reading and an accent type, and becomes / kouen / e / itte / ho / n / wo / yomi / mass /. Here, the morpheme with "^" is an accent in which the pitch is high in the syllable immediately before that (note that it is not a phoneme unlike the first embodiment), and the pitch is dropped in the syllable immediately after that. Means that. Also,
When there is no "^", it means a flat accent.

By the way, when humans read a sentence, they do not read by adding accents in units of such morphemes, but make a group of several morphemes and read by adding an accent to the unit.

Therefore, in consideration of the above, the language analysis processing unit 41 further collects the morphemes by one accent phrase (a unit giving an accent) and at the same time estimates the movement of the accent due to the collection. In addition to this, the language analysis processing unit 41 also adds information such as vowel devoicing and pause (breathing) during reading, and in the above example, finally generates the following reading information.

/ Koenye / Itte. / Ho ^ o / Yomima ^ (s) / Here, the period "." Represents breath, and "()" represents a syllable in which a vowel is devoiced.

When the reading information is generated by the language analysis processing unit 41 in the language processing unit 4 as described above, the phoneme duration calculation processing unit 51 in the speech synthesis unit 5 is activated. The phoneme duration calculation processing unit 51 includes a language analysis processing unit 4
According to the reading information generated in 1, the durations (unit: ms) of the consonant part and the vowel part of each syllable included in the input sentence are determined. In the phoneme duration calculation processing unit 51, the duration determination processing is performed so that the positions of the boundaries (CV crossings) between consonants (C) and vowels (V) are arranged at equal intervals, as in the first embodiment. Is realized (see FIG. 2) by an extremely simple algorithm.

When the duration of each syllable (consonant portion and vowel portion) included in the input sentence is determined by the phoneme duration calculation processing unit 51, the syllable-to-syllable point pitch setting processing unit 52 in the same speech synthesis unit 5 is determined. Is started. The syllable-based point pitch setting processing unit 52 sets the point pitch position based on the duration determined by the phoneme duration calculation processing unit 51 and the accent information determined by the language analysis processing unit 41. Here, as the time for setting the point pitch, one point or two points of the duration of each syllable divided into four at equal intervals (that is, equally divided into four) are determined. A concrete example of this point pitch setting is shown in FIG.
FIG. 9 shows the case of the accent phrase “Yomima (su)” in which the duration of the syllable (of the consonant part and the vowel part) is determined as described above. In FIG. 9, the horizontal axis represents time, the vertical axis represents pitch (unit is octave oct), and the positions of white circles are point pitch positions. Where the point pitch position (x,
y) is represented by x indicating the time points 0 to 3 (where the last time of the syllable at the end of the phrase is 4) obtained by dividing the duration of the syllable into four equal parts, and y indicating the pitch (unit oct).

In this embodiment, the point pitch setting time and frequency (pitch frequency) are the accent type, the position of the syllable for which the pitch is set in the accent phrase, the type of the syllable or the syllable environment. Depends on This will be described in detail below with reference to FIGS.

First, FIG. 10 (a) shows a rule for assigning a pitch position which is set to a head syllable (first syllable) of an accent phrase having a flat type or an up-and-down type accent, starting from a syllable including a consonant.

As is apparent from FIG. 10 (a), at the point in time when the point pitch in the head syllable of the flat or undulating accent phrase is specified, the consonant contained in the head syllable is a voiced consonant or an unvoiced consonant. And whether or not the syllable immediately after the first syllable (second syllable) is syllable, that is, the type of the first syllable (the type of the target syllable itself for the point pitch setting) and the type of the syllable immediately after that, in other words, the syllable. It depends on the environment. There are four types of combinations (syllabic environment) of the type of the leading syllable and the type of the syllable immediately after which the point pitch is set, and the point pitch position assignment rules (rule No.
(1) to (4)) are prepared.

Next, FIG. 10B shows a rule for assigning a pitch position set in the second syllable of an accent phrase having a flat plate type or a relief type accent. As is apparent from FIG. 10 (b), at the point in time when the point pitch in the second syllable of the flat type or undulating accent phrase is designated, whether or not the syllable is syllable, that is, the type of the syllable itself ( Syllable type). Here, a point pitch position assignment rule (rule No. (1)) when the syllable type is sound repellency and a point pitch position assignment rule (rule No. (2)) when the syllable type is other than sound repellency. Is prepared.

The point pitch positions determined on the basis of the point pitch position assignment rules shown in FIGS. 10A and 10B are shown in FIGS.
It shows in (d). In FIGS. 11A to 11D, as in FIG. 9, the horizontal axis represents time, the vertical axis represents pitch (unit is octave oct), and the positions of white circles are point pitch positions.

First, in a case where the consonant in the first syllable of the accent phrase is a voiced consonant and the second syllable is a non-syllable accent phrase,
For example, in the case of the accent phrase "Yomimasu (read)", the rule No. in FIG. (4) and the rule No. in FIG. (2) is applied to generate (set) the point pitch position as shown in FIG.

Next, when the consonant in the first syllable of the accent phrase is a voiced consonant and the second syllable is a syllable,
For example, in the case of the accent phrase “longjimas (discussed)”, the rule No. in FIG. (3)
And the rule No. in FIG. (1) and are applied,
Point pitch positions as shown in FIG. 11B are generated.

Next, when the consonant in the first syllable of the accent phrase is an unvoiced consonant and the second syllable is an accent phrase which is not a syllable,
For example, in the case of the accent phrase "trimas (take)", the rule number in FIG. Rule No. 2 in (2) and FIG. (2) is applied to generate a point pitch position as shown in FIG. 11 (c).

Next, in the case of an accent phrase in which the consonant in the first syllable of the accent phrase is a voiceless consonant and the second syllable is a syllable,
For example, in the case of the accent phrase “Kanjimasu (feels)”, the rule No. in FIG. (1)
And the rule No. in FIG. (1) and are applied,
Point pitch positions as shown in FIG. 11D are generated.

A pitch pattern obtained by interpolating a plurality of point pitch positions generated in this way with a straight line (see FIG. 11).
Similar to the pitch pattern obtained in the first embodiment, the thick line portions in (a) to (d)) closely approximate that of an actual human voice.

In a pitch pattern of an actual flat or undulating accent, a voiced consonant tends to start at a lower pitch than a voiced consonant as a leading consonant of an accent block. Further, it is known that when the second syllable is sound repellant, the start time and the end time of the pitch rise tend to be earlier than when it is not.

To reiterate what has already been described in the first embodiment, conventionally, the pitch is given irrespective of the syllable type and the syllable environment without considering such actual pitch characteristics of the voice. It was Changing the pitch pattern in consideration of the syllable type and the syllable environment is possible with the conventional technique as shown in FIG. However, in the prior art,
Since only one point pitch can be set at the center of a syllable, it is possible to barely change the setting of the start pitch by changing the point pitch of the first syllable. Can be set only to the central time point of the first syllable and the central time point of the second syllable, respectively. Therefore, it has been difficult for the conventional technique to approximate the pitch pattern of the actual human voice.

By the way, in order to perform precise pitch control, it is necessary to set the point pitch at intervals of about four equal parts of the duration of one syllable, as pointed out in Japanese Patent Laid-Open No. 3-164800. is there. However, even if it is divided into four, it is inefficient to give the pitch frequency at all time points, and one or two points are sufficient, and in some syllables, it is not necessary to give the point pitch at all.

When the point pitch setting processing unit 52 for each syllable determines the point pitch position for each syllable, the speech symbol string generation processing unit 53 in the same speech synthesis unit 5 is activated. The phonetic symbol string generation processing unit 53 includes the phonological information obtained by the language analysis processing unit 41, the duration of each syllable (of the consonant portion and the vowel portion) obtained by the phonological duration calculation processing unit 51, and each syllable point. From the point pitch positions obtained by the pitch setting processing unit 52, a phonetic symbol string describing these is generated.

FIG. 12 shows an example of a phonetic symbol string for the accent phrase "Yomimas (read)". As is clear from the figure, the phonetic symbol string is composed of phoneme information 1210 and prosody information 1220. This prosody information 1220 includes intonation component 1221, syllable duration (vowel duration, consonant duration and vowel duration) determined for each syllable of phonological information 1210, and point pitch information (syllable). Duration / pitch information) 1222. The intonation component 1221 represents a spontaneous pitch decrease component, like the intonation component 621 in the first embodiment. Further, “;” in the phonetic symbol string in FIG. 12 indicates the end of the accent phrase. Note that “(s)” in the phoneme information 1210 is a “s” in which the vowel is devoiced.
Represents

The voice symbol string generation processing unit 53 passes the generated voice symbol string to the pitch pattern generation processing unit 54 in the same voice synthesis unit 5 (through a memory (not shown)), and the pitch pattern generation processing unit 54 performs the processing. It is possible to proceed with the processing or to write the phonetic symbol string in the phonetic symbol string file 61 in the phonetic symbol string editing unit 6 once in order to edit the phonetic symbol string. The process of editing the phonetic symbol string in the phonetic symbol string file 61 by the phonetic symbol string editing unit 6 will be described later.

The pitch pattern generation processing unit 54 in the voice synthesis unit 5 is based on the voice symbol string passed from the voice symbol string generation processing unit 53 or stored in the voice symbol string file 61. A process of generating (prosodic parameter) is performed. That is, the pitch pattern generation processing unit 54 also sets the duration (vowel duration or consonant duration and vowel duration) and point pitch position for each syllable set in the target phonetic symbol sequence. Then, linear interpolation is performed between the respective point pitches to generate a pitch accent component, for example, every 10 ms. The pitch pattern generation processing unit 54 further adds an intonation component (spontaneous drop component of pitch) set in the target speech symbol sequence to obtain a pitch pattern for every 10 ms of speech to be synthesized.

On the other hand, the phoneme parameter generation processing unit 55 in the speech synthesis unit 5 also includes the phoneme information in the phonetic symbol string passed from the phonetic symbol string generation processing unit 53 or stored in the phonetic symbol string file 61. The process of generating the phoneme parameters is performed in parallel with the pitch pattern generation process by the pitch pattern generation processing unit 54, for example. The phonological parameter generation processing by the phonological parameter generation processing unit 55 is performed in the same manner as the phonological parameter generation processing unit 25 in the first embodiment, and the description thereof is omitted.

The pitch pattern generation processing unit 54 generates a pitch pattern, and the phoneme parameter generation processing unit 55.
When the phoneme parameter is generated by the above, the synthesis filter processing unit 57 (having the same configuration as the synthesis filter processing unit 27 as shown in FIG. 7) is activated, and the synthesis filter processing unit 27 in the first embodiment is used. Similarly, speech is synthesized from the generated pitch pattern and phoneme parameter.

Next, the process of editing the phonetic symbol string in the phonetic symbol string file 61 by the phonetic symbol string editing unit 6 will be described with reference to the flowchart of FIG. First, the phonetic symbol string editing unit 6 controls the phonetic symbol string correction processing unit 6 in addition to the phonetic symbol string file 61 and the process of correcting the phonetic symbol string according to the phonetic symbol string editing tool (phonetic symbol string editing software).
2, a display device 63 such as a CRT display and a liquid crystal display, and an input device 64 such as a keyboard and a mouse. The display device 63 and the input device 64 may be the display device and the input device included in an information processing device such as a personal computer that realizes the voice synthesizer of FIG. The phonetic symbol string file 61 includes a phonetic symbol string generated as described above by the phonetic symbol string generation processing unit 53 in the language processing unit 4,
Alternatively, a voice symbol string created by the user using a personal computer or the like is stored.

When the voice symbol string correction processing unit 62 in the voice symbol string editing unit 6 is activated through the input device 64 according to the user's operation, it is generated by the voice symbol string generation processing unit 53 or created by the user. Then, the phonetic symbol string having the structure shown in FIG. 12 is read from the phonetic symbol string file 61 (step S1).

Next, the phonetic symbol string correction processing unit 62, in accordance with the designation of the syllable duration time and the point pitch position in the read phonetic symbol string, in the same manner as the pitch pattern generation processing unit 54 in the phonetic synthesis unit 5, for example, A pitch pattern whose horizontal axis is the time axis is generated (step S2), and the pitch pattern is displayed together with the phoneme symbol indicated by the phoneme information in the read voice symbol string and the range of the phoneme on the time axis. Is displayed as a graph on the display screen of (step S
3). This pitch pattern display example is indicated by reference numeral 141 in FIG. 14 for the case of the phonetic symbol string in FIG.
The pitch pattern (pitch pattern 141) is generated (drawn) by linearly interpolating between the point pitch positions set in the phonetic symbol string.

In the pitch pattern display in this embodiment, as shown in FIG. 14, the point pitch positions already designated in the phonetic symbol string are displayed as white circles and are not designated in the phonetic symbol string. However, the point pitch positions that the user can specify are displayed as black circles. The positions of these white circles and black circles can be moved by the user operating the mouse or the like of the input device 64.

For example, if a mouse is used,
First, the user operates the mouse to move the mouse cursor to the circle with the desired point pitch and presses the mouse button in that state. In this state, the user moves the mouse to move the mouse cursor to a desired position, and then releases the mouse button.

The voice symbol string correction processing unit 62 moves the circle position (point pitch position) designated for movement in response to the user's mouse operation (step S4). Then, the phonetic symbol string correction processing unit 62 newly interpolates a straight line between the moved circle and the circles before and after the circle. Further, the phonetic symbol string correction processing unit 62 changes (corrects) the phonetic symbol string read from the phonetic symbol string file 61 according to the movement (change) of the circle position (step S5).

Thus, for example, the pitch pattern displayed as shown in FIG. 15B according to the phonetic symbol sequence shown in FIG. 15A is moved to the position shown in FIG.
In the case of the change as shown in (c), the phonetic symbol string is changed as shown in FIG.

Further, the phonetic symbol string correction processing unit 62 is shown in FIG.
As indicated by the broken line in FIG. 3, the pitch pattern obtained by analyzing the actual voice is displayed as the reference pattern 142 together with the pitch pattern 141. This allows the user to
While looking at the reference pattern 142, if the position of the circle (dotted position) on the pitch pattern 141, which is graphed based on the phonetic symbol string, is corrected so as to come closest to the reference pattern, it will be very similar to the voice uttered by a human. It is possible to easily create (edit) a speech symbol string capable of synthesizing a speech having a close intonation.

Now, the voice symbol string corrected according to the pitch pattern change according to the user's operation, when a save instruction is given from the user through the input device 64 (step S6), the voice symbol string correction processing is performed. It is saved in the phonetic symbol string file 61 by the unit 62 (step S7).

When the phonetic symbol string modification processing unit 62 saves the phonetic symbol string, if the user has designated reading of the next phonetic symbol string instead of ending the phonetic symbol string editing process (step (S8, S9), the process returns to step S1, the next phonetic symbol string is read from the phonetic symbol string file 61, and the process from step S2 is performed on the phonetic symbol string.

Such a voice symbol sequence editing process is similarly performed in the voice symbol sequence editing unit 3 (in the voice symbol sequence correction processing unit 32) in the first embodiment. Now, the voice symbol string editing unit 6 (the voice symbol string correction processing unit 6 in
The voice symbol sequence in the voice symbol sequence file 61 corrected by the voice symbol sequence edit process according to 2), that is, the voice symbol sequence is automatically generated by the voice symbol sequence generation processing unit 53 and then corrected according to the user operation, or created by the user. After that, the voice symbol string corrected according to the user operation is provided for voice synthesis by the voice synthesizer 5. As a result, it is possible to synthesize a voice having intonation extremely close to the voice uttered by a human.

[0118]

As described above, according to the present invention, the number of set point pitches can be minimized, but finer pitch control can be performed as compared with the conventional method, so that a smoother and smoother human voice can be suppressed. A pitch pattern close to can be generated from a text or phonetic symbol string. Further, according to the present invention, there is also an effect that the user can easily provide a user's favorite pitch pattern for voice synthesis with a minimum of labor.

[Brief description of drawings]

FIG. 1 is a block diagram showing a schematic configuration of a speech synthesizer according to a first embodiment of the present invention.

FIG. 2 is a diagram for explaining a method of determining a phoneme duration in the first embodiment.

FIG. 3 is a diagram for explaining a pitch setting method according to the first embodiment.

FIG. 4 is a diagram showing an arrangement of point-pitch position assignment rules (point-pitch setting rules) in consideration of phoneme types or phoneme environments according to the first embodiment.

FIG. 5 is a view showing a specific example of a point pitch position determined according to the rule of FIG. 4 for four accent phrases.

FIG. 6 is a diagram showing an example of a phonetic symbol string according to the first embodiment.

FIG. 7 is a block diagram showing a configuration of a synthesis filter processing unit 27 in FIG.

FIG. 8 is a block diagram showing a schematic configuration of a speech synthesizer according to a second embodiment of the present invention.

FIG. 9 is a diagram for explaining a pitch setting method according to the second embodiment.

FIG. 10 is a diagram summarizing and showing a point pitch position assigning rule (point pitch setting rule) in consideration of a syllable type or a syllable environment in the second embodiment.

FIG. 11 is a diagram showing a specific example of a point pitch position determined according to the rule of FIG. 10 for four accent phrases.

FIG. 12 is a diagram showing an example of a phonetic symbol string according to the second embodiment.

13 is a flowchart for explaining a phonetic symbol string editing process by a phonetic symbol string editing unit 6 (including a phonetic symbol string correction processing unit 62) in FIG.

FIG. 14 is a diagram showing an example of pitch pattern display in a voice symbol string editing process.

FIG. 15 is a diagram showing an example of changing a pitch pattern and a phonetic symbol string in a phonetic symbol string editing process.

FIG. 16 is a diagram for explaining a conventional pitch setting method.

[Explanation of symbols]

1, 4 ... Language processing unit, 2, 5 ... Speech synthesis unit, 3, 6 ... Speech symbol string editing unit, 11, 41 ... Language analysis processing unit, 21, 51 ... Phonological duration calculation processing unit, 22 ... Phonological units Point pitch setting processing unit, 23, 53 ... Speech symbol string generation processing unit, 24, 54 ... Pitch pattern generation processing unit (prosodic parameter generation processing unit), 25, 55 ... Phonological parameter generation processing unit (feature parameter generation processing unit) 26, 56 ... Speech element memory, 27, 57 ... Synthesis filter processing section, 31, 61 ... Speech symbol string file, 32, 62 ... Speech symbol string correction processing section, 33, 63 ... Display device, 34, 64 ... Input device, 52 ... Pitch setting processing unit for each syllable, 610, 1210 ... Phoneme information, 620, 1220 ... Prosodic information, 622 ... Phoneme duration / pitch information, 1222 ... Syllable duration time -Pitch information.

Claims (27)

[Claims] 1. A voice synthesizing method for synthesizing a voice based on a pitch pattern generated by giving a plurality of point pitches based on accent information of a word or a phrase and interpolating between the given plurality of point pitches. 2. The speech synthesis method according to claim 1, wherein the position on the time axis at which the point pitch is given is determined based on at least one of the phoneme environment or the phoneme type of the phoneme to give the point pitch. 2. A voice unit storage medium for storing voice feature parameters in the form of a voice unit comprising a predetermined synthesis unit, and a voice unit storage medium for storing voice based on input phoneme information. By using the accent parameter input corresponding to the phoneme information, the feature parameter generation processing means for reading out the voice unit, connecting the read voice units, and generating the feature parameter of the voice to be synthesized. A point pitch setting processing unit for each phoneme that sets a point pitch for each phoneme, and determines the position of the point pitch on the time axis based on at least one of the phoneme environment or the phoneme type of the phoneme to which the set point pitch belongs. A pitch for generating a voice pitch pattern by interpolating between the phoneme-based point pitch setting processing means and the adjacent point pitches set by the phoneme-based point pitch setting processing means. Pattern generation processing means; synthesis filter processing means for synthesizing speech from the characteristic parameter of the speech to be synthesized generated by the characteristic parameter generation processing means and the pitch pattern generated by the pitch pattern generation processing means. A speech synthesizer characterized by: 3. A voice synthesizing method for synthesizing a voice based on a pitch pattern generated by giving a plurality of point pitches based on accent information of a word or a phrase and interpolating between the plurality of given point pitches. 2. The method of synthesizing speech according to claim 1, wherein the position on the time axis at which the point pitch is given is determined based on at least one of the syllable environment and the syllable type of the syllable to which the point pitch is given. 4. A voice unit storage medium for storing voice feature parameters in the form of a voice unit comprising a predetermined synthesis unit, and a voice unit storing medium for storing voice based on input phoneme information. By using the accent parameter input corresponding to the phoneme information, the feature parameter generation processing means for reading out the voice unit, connecting the read voice units, and generating the feature parameter of the voice to be synthesized. A syllable-based point pitch setting processing means for setting a point pitch for each syllable, wherein the position of the point pitch on the time axis is determined based on at least one of the syllable environment or the syllable type of the syllable to which the set point pitch belongs. A pitch for generating a voice pitch pattern by interpolating between the point pitch setting processing unit for each syllable and the adjacent point pitch set by the point pitch setting processing unit for each syllable Pattern generation processing means; synthesis filter processing means for synthesizing speech from the characteristic parameter of the speech to be synthesized generated by the characteristic parameter generation processing means and the pitch pattern generated by the pitch pattern generation processing means. A speech synthesizer characterized by: 5. A voice synthesizing method for synthesizing a voice based on a pitch pattern generated by giving a plurality of point pitches based on accent information of a word or a phrase and interpolating between the plurality of given point pitches. 2. In the speech synthesis method, at least four point pitch specifiable points are set within the duration of one phoneme, and the point pitch is designated for two points or less. 6. A voice unit storage medium for storing voice feature parameters in the form of a voice unit comprising a predetermined synthesis unit, and a voice unit storing medium for storing voice based on input phoneme information. By using the accent parameter input corresponding to the phoneme information, the feature parameter generation processing means for reading out the voice unit, connecting the read voice units, and generating the feature parameter of the voice to be synthesized. It is a point pitch setting processing unit for each phoneme that sets a point pitch for each phoneme, and sets at least four points at which point pitch can be designated within the duration of one phoneme. A pitch pattern of voice is generated by interpolating between the point pitch setting processing unit for each phoneme that specifies the point pitch and the adjacent point pitch set by the point pitch setting processing unit for each phoneme. Pitch pattern generation processing means, and synthesis filter processing means for synthesizing speech from the characteristic parameters of the speech to be synthesized generated by the characteristic parameter generation processing means and the pitch pattern generated by the pitch pattern generation processing means. A speech synthesis apparatus comprising: 7. A voice synthesizing method for synthesizing a voice based on a pitch pattern generated by giving a plurality of point pitches based on accent information of a word or a phrase and interpolating between the given plurality of point pitches. In the speech synthesis method, at least four point pitch specifiable points are set within the duration of one syllable, and the point pitch is designated for two points or less. 8. A voice unit storage medium for storing voice feature parameters as voice units made of a predetermined synthesis unit, and a voice unit storage medium for storing voice based on input phoneme information. By using the accent parameter input corresponding to the phoneme information, the feature parameter generation processing means for reading out the voice unit, connecting the read voice units, and generating the feature parameter of the voice to be synthesized. A syllable-based point-pitch setting processing means for setting a point-pitch for each syllable, wherein at least four point-pitch specifiable times are set within the duration of one syllable, and two or less of these points are specified A pitch pattern of voice is generated by interpolating between the syllable-based point pitch setting processing means for designating the point pitch and the adjacent point pitch set by the syllable-based point pitch setting processing means. Pitch pattern generation processing means, and synthesis filter processing means for synthesizing speech from the characteristic parameters of the speech to be synthesized generated by the characteristic parameter generation processing means and the pitch pattern generated by the pitch pattern generation processing means. A speech synthesis apparatus comprising: 9. The method according to claim 2, further comprising language analysis processing means for analyzing text to be subjected to speech synthesis to generate phoneme information and accent information. The speech synthesizer according to claim 8. 10. A phonetic symbol string, which describes prosodic information including phonological information and pitch information of the phonetic speech, is input, and the phonetic symbol is selected from a group of phonetic pieces composed of a plurality of voice characteristic parameters prepared in advance. A plurality of speech units are selected and connected according to the phoneme information in the sequence to generate parameters for expressing the phoneme of the voice, and a pitch pattern of the voice is generated according to the prosody information in the phonetic symbol sequence, A voice synthesizing method for synthesizing a voice based on a parameter expressing a phoneme and a pitch pattern, wherein the pitch information in the voice symbol string is given in a point pitch. 11. A speech unit storage medium for accumulating a group of speech units composed of a plurality of speech feature parameters prepared in advance, and phonological information of a speech to be synthesized and point pitches. At least the phoneme information in a phonetic symbol sequence describing prosody information including pitch information is input, and a plurality of phonemes are selected from the phoneme storage medium according to the phoneme information and connected to represent a phoneme of a voice. A phonological parameter generating means for generating a parameter, a prosodic parameter generating means for inputting at least prosodic information in the phonetic symbol string, and generating a pitch pattern of speech to be synthesized according to the prosodic information, and the phonological parameter generating means. A synthesis filter process for synthesizing a voice from the parameter expressing the generated phoneme and the pitch pattern generated by the prosody parameter generating means. A speech synthesis apparatus comprising: a processing means. 12. A phonetic symbol string which describes prosodic information including phonological information and pitch information of a voice is input, and the phonetic symbol is selected from a group of phoneme segments prepared from a plurality of voice characteristic parameters prepared in advance. A plurality of speech units are selected and connected according to the phoneme information in the sequence to generate parameters for expressing the phoneme of the voice, and a pitch pattern of the voice is generated according to the prosody information in the phonetic symbol sequence, A voice synthesis method for synthesizing a voice based on a phoneme expressing parameter and a pitch pattern, wherein pitch information in the voice symbol string is given in a point pitch,
A method for synthesizing speech, characterized in that the position of the point pitch on the time axis is specified by the point pitch with reference to the starting point of each phoneme. 13. A speech unit storage medium for accumulating a group of speech units composed of a plurality of speech feature parameters prepared in advance, phonological information of speech to be synthesized, and point pitch. Pitch information, at least the phonological information in the phonetic symbol string describing the prosody information including the pitch information where the position on the time axis of the point pitch is specified by the point pitch with reference to the starting point of each phoneme. Phonological parameter generation means for inputting and selecting a plurality of speech units from the speech unit storage medium according to the phoneme information and connecting them to generate a parameter expressing a phoneme of a voice, and at least a prosody in the speech symbol string. Information is input, and a prosody parameter generating means for generating a pitch pattern of a voice to be synthesized according to the prosody information, and a phoneme generated by the phoneme parameter generating means are expressed. A voice synthesizing apparatus comprising: a parameter and a synthesis filter processing means for synthesizing a voice from the pitch pattern generated by the prosody parameter generating means. 14. A phonetic symbol string which describes prosodic information including phonological information and pitch information of a voice is input, and the phonetic symbol is selected from among a group of phoneme pieces prepared from a plurality of voice characteristic parameters prepared in advance. A plurality of speech units are selected and connected according to the phoneme information in the sequence to generate parameters for expressing the phoneme of the voice, and a pitch pattern of the voice is generated according to the prosody information in the phonetic symbol sequence, A voice synthesis method for synthesizing a voice based on a phoneme expressing parameter and a pitch pattern, wherein pitch information in the voice symbol string is given in a point pitch,
A voice synthesis method characterized in that the position of the point pitch on the time axis is specified by the point pitch with reference to the start point of each syllable. 15. A speech unit storage medium for accumulating a group of speech units composed of a plurality of speech feature parameters prepared in advance, phonological information of speech to be synthesized, and point pitch. Pitch information, the position on the time axis of the point pitch is at least the phonological information in the phonetic symbol sequence describing prosody information including pitch information specified by the point pitch with reference to the starting point of each syllable. Phonological parameter generation means for inputting and selecting a plurality of speech units from the speech unit storage medium according to the phoneme information and connecting them to generate a parameter expressing a phoneme of a voice, and at least a prosody in the speech symbol string. Information is input, and a prosody parameter generating means for generating a pitch pattern of a voice to be synthesized according to the prosody information, and a phoneme generated by the phoneme parameter generating means are expressed. A voice synthesizing apparatus comprising: a parameter and a synthesis filter processing means for synthesizing a voice from the pitch pattern generated by the prosody parameter generating means. 16. A phoneme in a phonetic symbol string from a group of phoneme units prepared from a plurality of voice characteristic parameters prepared in advance, using a phonetic symbol string describing phoneme information and prosodic information of the phonetic as an input. A plurality of speech units are selected and connected according to the information to generate parameters for expressing the phoneme of the voice, and a pitch pattern of the voice is generated according to the prosodic information in the phonetic symbol string to express the phoneme of the voice. A voice synthesizing method for synthesizing a voice based on a parameter and a pitch pattern, wherein at least four point pitch specifiable points are set within the duration of each phoneme included in the phoneme information in the phonetic symbol sequence, A speech synthesis method characterized in that prosodic information in the speech symbol string is described by designating a point pitch for time points of two points or less. 17. A speech unit storage medium for accumulating a group of speech units consisting of a plurality of speech feature parameters prepared in advance, phonological information of speech to be synthesized, and included in the phonological information. At least four point pitch specifiable time points are set within the duration of each phoneme to be recorded, and at least the above-mentioned prosodic information in the phonetic symbol sequence describing prosodic information for which the point pitch is specified for time points of two points or less. Phonological parameter generation means for inputting phonological information, selecting a plurality of speech units from the speech unit storage medium according to the phonological information and connecting them, and generating a parameter expressing a phoneme of a voice; Of at least prosody information, and a prosody parameter generation means for generating a pitch pattern of a voice to be synthesized according to the prosody information, and a phoneme generated by the phonology parameter generation means. And a synthesis filter processing unit for synthesizing a voice from the pitch pattern generated by the prosody parameter generating unit. 18. A phoneme in a phonetic symbol string from a group of phoneme segments made up of a plurality of phonetic characteristic parameters prepared in advance, with a phonetic symbol string describing phoneme information and prosody information of the phoneme as an input. A plurality of speech units are selected and connected according to the information to generate parameters for expressing the phoneme of the voice, and a pitch pattern of the voice is generated according to the prosodic information in the phonetic symbol string to express the phoneme of the voice. A voice synthesizing method for synthesizing a voice based on a parameter and a pitch pattern, wherein at least four point pitch specifiable points are set within the duration of each phoneme included in the phoneme information in the phonetic symbol sequence, It is characterized in that the prosodic information in the phonetic symbol string is described by designating the position on the time axis of the point pitch with respect to the starting point of each phoneme for two or less points among them. Speech synthesis method to be. 19. A speech segment storage medium for accumulating a group of speech segments composed of a plurality of speech feature parameters prepared in advance, phonological information of speech to be synthesized, and included in the phonological information. Within the duration of each phoneme to be set, at least four point pitch specifiable points are set, and for points of less than two points,
At least the phoneme information in the phonetic symbol string describing the prosody information in which the position on the time axis of the point pitch is specified on the basis of the starting point of each phoneme is input, and the phoneme storage medium is input according to the phoneme information. Phonological parameter generation means for selecting a plurality of speech units and connecting them to generate a parameter for expressing a phoneme of a speech, and inputting at least prosody information in the speech symbol string, of a speech to be synthesized according to the prosody information. A prosody parameter generating means for generating a pitch pattern, and a synthesis filter processing means for synthesizing a voice from the parameter expressing the phoneme generated by the phoneme parameter generating means and the pitch pattern generated by the prosody parameter generating means. A speech synthesizer characterized by the above. 20. A phoneme in a voice symbol string is selected from a group of voice units prepared from a plurality of voice characteristic parameters prepared in advance, with a voice symbol string describing phoneme information and prosody information of the voice as an input. A plurality of speech units are selected and connected according to the information to generate parameters for expressing the phoneme of the voice, and a pitch pattern of the voice is generated according to the prosodic information in the phonetic symbol string to express the phoneme of the voice. A voice synthesizing method for synthesizing a voice based on a parameter and a pitch pattern, wherein at least four point pitch specifiable points are set within a duration of each syllable included in the phoneme information in the voice symbol string, A speech synthesis method characterized in that prosodic information in the speech symbol string is described by designating a point pitch for time points of two points or less. 21. A speech unit storage medium for accumulating a group of speech units composed of a plurality of speech feature parameters prepared in advance, phonological information of speech to be synthesized, and included in the phonological information. At least four point pitch specifiable points are set within the duration of each syllable, and at least the above-mentioned prosodic information in the phonetic symbol sequence describing prosodic information for which the point pitch is designated for two points or less. Phonological parameter generation means for inputting phonological information, selecting a plurality of speech units from the speech unit storage medium according to the phonological information and connecting them, and generating a parameter expressing a phoneme of a voice; Of at least prosody information, and a prosody parameter generation means for generating a pitch pattern of a voice to be synthesized according to the prosody information, and a phoneme generated by the phonology parameter generation means. And a synthesis filter processing unit for synthesizing a voice from the pitch pattern generated by the prosody parameter generating unit. 22. A phoneme in a voice symbol string is selected from among a group of voice segments prepared in advance, each of which is prepared by inputting a voice symbol string describing phoneme information and prosody information of the voice. A plurality of speech units are selected and connected according to the information to generate parameters for expressing the phoneme of the voice, and a pitch pattern of the voice is generated according to the prosodic information in the phonetic symbol string to express the phoneme of the voice. A voice synthesizing method for synthesizing a voice based on a parameter and a pitch pattern, wherein at least four point pitch specifiable points are set within a duration of each syllable included in the phoneme information in the voice symbol string, It is characterized in that the prosodic information in the phonetic symbol string is described by designating the position of the point pitch on the time axis with respect to the starting point of each syllable for the time points of two points or less. Speech synthesis method to be. 23. A speech unit storage medium for accumulating a group of speech units composed of a plurality of speech feature parameters prepared in advance, phonological information of speech to be synthesized, and included in the phonological information. Set at least 4 point pitch specifiable time points within the duration of each syllable, and for 2 points or less,
Input at least the phonological unit information in the phonetic symbol sequence describing prosodic information in which the position on the time axis of the point pitch is specified based on the starting point of each syllable, and from the phoneme storage medium according to the phonological unit information. Phonological parameter generation means for selecting a plurality of speech units and connecting them to generate a parameter for expressing a phoneme of a speech, and inputting at least prosody information in the speech symbol string, of a speech to be synthesized according to the prosody information. A prosody parameter generating means for generating a pitch pattern, and a synthesis filter processing means for synthesizing a voice from the parameter expressing the phoneme generated by the phoneme parameter generating means and the pitch pattern generated by the prosody parameter generating means. A speech synthesizer characterized by the above. 24. A group of speech units consisting of a plurality of speech characteristic parameters prepared in advance, using as input a speech symbol string describing prosodic information including phonological information of speech and pitch information given by point pitch. From among them, a plurality of speech units are selected and connected according to the phoneme information in the phonetic symbol string to generate a parameter for expressing the phoneme of the phonetic symbol, and a pitch pattern of the voice is determined according to the prosodic information in the phonetic symbol string. A voice symbol sequence editing device which is applied to a voice synthesizing device for synthesizing a voice based on a parameter expressing a phoneme of the voice and a pitch pattern, the plurality of points according to pitch information in the voice symbol sequence. Display means for displaying the pitch, input means for instructing to change the point pitch displayed by the display means, and response to the change instruction by the input means. Then, the phonetic symbol string editing device further comprises a phonetic symbol string correction processing unit for changing the corresponding point pitch being displayed by the display unit and reflecting the change result in the phonetic symbol string. 25. A group of speech units consisting of a plurality of speech characteristic parameters prepared in advance, using as input a speech symbol string describing prosodic information including phonological information of speech and pitch information given by point pitch. From among them, a plurality of speech units are selected and connected according to the phoneme information in the phonetic symbol string to generate a parameter for expressing the phoneme of the phonetic symbol, and a pitch pattern of the voice is determined according to the prosodic information in the phonetic symbol string. A voice symbol sequence editing device which is applied to a voice synthesizing device for synthesizing a voice based on a parameter expressing a phoneme of the voice and a pitch pattern, the plurality of points according to pitch information in the voice symbol sequence. Display means for displaying the pitch in a graph, input means for inputting an instruction to move the position of the point pitch displayed by the display means, and this input Moving the position of the corresponding point pitch being displayed by the display means on the screen in response to the movement instruction from the means,
A voice symbol sequence editing device, comprising: a voice symbol sequence correction processing means for reflecting the position of the new point pitch after the movement in the voice symbol sequence. 26. A group of speech units consisting of a plurality of speech characteristic parameters prepared in advance, using as input a speech symbol string describing prosodic information including speech phonological information and pitch information given by point pitch. From among them, a plurality of speech units are selected and connected according to the phoneme information in the phonetic symbol string to generate a parameter for expressing the phoneme of the phonetic symbol, and a pitch pattern of the voice is determined according to the prosodic information in the phonetic symbol string. A voice symbol sequence editing device applied to a voice synthesizing device for generating and synthesizing a voice based on a parameter expressing a phoneme of the voice and a pitch pattern, comprising: Display means for displaying the point pitch and the pitch contour obtained by interpolating between these point pitches in a graph, and the position of the point pitch displayed by this display means are moved. That an input means for inputting an instruction, the position of the corresponding point pitch being displayed by the display unit to move on the screen in accordance with the movement instruction by the input means,
A voice symbol sequence editing device, comprising: a voice symbol sequence correction processing means for reflecting the position of the new point pitch after the movement in the voice symbol sequence. 27. The pitch contour obtained by analyzing the reference voice corresponding to the phoneme information in the voice symbol string is graphically displayed as a reference pattern on the display means. Item 26. A voice symbol string editing device according to item 26.