JP2002156990A - Method and apparatus for pause duration processing in chinese voice synthesis - Google Patents

Abstract

PROBLEM TO BE SOLVED: To synthesize a voice of Chinese having a natural pause by reproducing a pause for breathing in a human speech. SOLUTION: A word inspection part 12 and a part-of-speech/extension syntax inspection part 13 inspect semantic units in an object sentence of voice synthesis by referring to a dictionary memory 11 stored with information for extending words into a phrase. The positions before and after a semantic unit are provided as pause position candidates. A pause position calculation part 18 calculates weighted values of pauses as respective pause position candidates through a sequence based upon respective semantic units in the sentence and their positions. When a pause position candidate has a weighted value of a pause larger than a predetermined value, it is determined as a pause position and pause duration is specified. Through this calculation of the weighted value of the pause, the pause position in the sentence and its pause duration can accurately be determined.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a pause duration for processing a pause duration of a speech in Chinese speech synthesis for determining a pause position and a pause duration of a synthesized speech and obtaining a natural voice. The present invention relates to a processing method and apparatus.

[0002]

2. Description of the Related Art In spontaneously uttered speech, a pause exists at an appropriate position in a sentence due to a breathing motion of a speaker. As a general rule, periodic breathing in human speech is
Occurs every six to eight Chinese characters (ie, Chinese syllables) and occurs before or after the semantic units that make up the phrase or word. During the breath, the duration of the voice is slightly paused. For speech synthesis of long sentences, properly determining the positions to pause and the durations to pause has a great influence on the naturalness of the synthesized speech.

[0003] Conventional devices for speech synthesis of Chinese text only consider processing the duration of speech of the syllable of each character in the sentence. These devices do not study the processing of pause position and pause duration. See, for example, "Speech Duration Processing Method and Apparatus (Speech Dur
ation Processing Method and Apparatus for Synthesi
FIG. 13 herein relates to Republic of China Patent Application No. 82103222 entitled "Zed Speech System".
FIG. 2 is a block diagram illustrating a processing device for determining the duration of each syllable of a Pinyin sentence of an arbitrary length by using a rhythm (rhyme) as a unit for processing a voice duration. In FIG. 13, reference numeral 110 denotes a Pinyin sentence input unit for inputting a Pinyin sentence using a device such as a keyboard. Reference numeral 111 denotes a phrase inspection unit for inspecting the positions of a plurality of syllables in a phrase according to a plurality of phrase markers. Reference numeral 112 denotes a tone / syllable inspection unit for inspecting a plurality of syllables in a Pinyin sentence using a plurality of tone markers. 113
Represents a syllable / rhythm matching unit for storing the pinyin marker of each single syllable and the specified number of rhythms constituting the syllable. Reference numeral 114 denotes a rhythm inspection unit for inspecting a plurality of rhythms constituting the inspected syllable by using a syllable / rhythm matching unit and storing rhythm data. Reference numeral 115 denotes a basic voice duration storage unit for storing the duration of the basic voice of each rhythm. Reference numeral 116 denotes a basic voice duration determining unit for testing a plurality of rhythms from the basic voice duration storage unit according to the inspected rhythm. A voice duration parameter storage unit 117 is established using the tone of each rhythm, the position of the rhythm in the sentence, the plurality of connected rhythm classes, and the position of the rhyme in the phrase to which the rhythm belongs. To represent. 118
Is the rhythm tone, the rhythm position in the sentence, the connected rhythm classes, and the rhyme position in the phrase to which the rhythm belongs. Represents a syllable voice duration calculator for calculating the duration of a syllable voice by recording the total voice duration of a plurality of rhythms and using it as a plurality of indexing keys to retrieve from storage. . 119 represents a plurality of different registers that the device must use in processing the duration.

[0004]

However, the above-mentioned conventional speech duration processing apparatus described above considers only processing of speech duration of a plurality of syllables of a plurality of characters in a sentence,
No more thorough processing of pause positions and pause durations in sentences is performed. As an example, synthesized Chinese speech "

[Outside 1] The most effective way to overcome the disease is "The most effective way to overcome the disease." In the text waveform illustrated in FIG. 7 synthesized using the conventional speech duration processing device described above, the sentence sounds as if it were spoken in a single breath, which is not a natural sound. Pause position is "strong recruitment will"
And "I know

[Outside 2] The first part and the second part when the example sentence is divided into three parts as described above, between the "victory sickness" and the "most effective method".
And between the second part and the third part, and when the pause duration is determined to be 120 ms and 80 ms, respectively, the synthesized voice has a waveform as shown in FIG. Having. Since there are a plurality of poses that mimic a change in breath, the naturalness of the rhythm of the synthesized voice and the clarity of the meaning are clearly improved. Therefore, only the duration of the speech of each syllable should not be considered for a sentence to be synthesized. More importantly, in order to obtain a very natural and very understandable synthesized language, the pause duration must be added at the appropriate place in the text. It is an object of the present invention to solve a method of determining appropriate positions and durations of a plurality of poses in a sentence.

SUMMARY OF THE INVENTION It is therefore a primary object of the present invention to provide a method and apparatus for processing pause duration in Chinese speech synthesis so that a system for converting Chinese text to speech can overcome the aforementioned problems. .

[0006]

According to a first aspect of the present invention, there is provided a pause duration processing method in Chinese speech synthesis.
Adapted for use in a system for converting Chinese text to speech in a computer system having an input portion and a memory buffer portion, for storing a Chinese vocabulary and corresponding information including part of speech and extended syntax. Configuring a dictionary memory and configuring a table memory of front-paused characters for storing a plurality of previously-paused characters and corresponding information including a plurality of previous-pause parameters;
Forming a table memory of later-paused characters for storing a plurality of later-paused characters and corresponding information including a plurality of later-paused parameters, wherein the semantic includes one of a word and a phrase. Configuring a pose parameter storage unit for storing a plurality of general pose parameters according to the length of the unit, and comparing the input sentence with a plurality of words of a Chinese vocabulary stored in the dictionary memory. Determining the position of the boundary of each word in the input sentence input using the input unit; and referencing the dictionary memory and searching for the part of speech and the extended syntax of each word whose boundary has been determined. , A plurality of words in a sentence in a plurality of frames according to the extended syntax and a part-of-speech relationship of adjacent ones of the plurality of words in the sentence. And combining the plurality of words into the plurality of phrases, so that the positions before and after these semantic units are candidate pose positions. Calculating a weighted value of the pose of the semantic unit, wherein the step retrieves a plurality of previously paused characters and calculates a weighted value for them; Retrieving the characters to be paused and calculating weighted values for them; calculating the weighted values of the poses of the plurality of pose position candidates to obtain a plurality of pose positions and a plurality of poses. Determining a duration, wherein each of the pose position candidates having a pose weighted value greater than a predetermined value is determined to be a pose position. And given a pause level, the pause duration is being determined in accordance with the pause level.

A pause duration processing apparatus in Chinese speech synthesis according to a second aspect of the present invention is used in a system for converting Chinese text to speech having a system which is a sentence input section and a memory buffer section. An adapted dictionary memory for storing Chinese vocabulary and corresponding information including part-of-speech and extended syntax, storing a plurality of previously paused characters, and corresponding information including a previous pause parameter. A table memory of characters to be paused before, a table memory of characters to be paused to store a plurality of characters to be paused later, and corresponding information including a post-pause parameter; A pose parameter storage unit for storing a general pose parameter in accordance with the length of a semantic unit including any one of the above, and the dictionary memory Boundary position determining means for determining a position of a boundary of each word in the input sentence input by using the input unit by comparing a plurality of words in the stored Chinese vocabulary with the input sentence, A search unit for referring to the dictionary memory and searching for the part of speech and the extended syntax of each word whose boundary position has been determined; and the extended syntax and a plurality of words in the sentence that are adjacent to each other. A combining means for combining a plurality of words in a sentence into a plurality of phrases according to the part of speech, and a position before and after these semantic units after a combination of the plurality of words into a plurality of phrases. Calculating means for calculating a weighted value of the pose of each of the plurality of semantic units divided from the input sentence so as to be a candidate, wherein the means searches for a plurality of characters to be paused before, It Means for calculating a weighted value for the plurality of pose position candidates, and means for searching for a plurality of characters to be paused and calculating a weighted value for them. A plurality of pose positions and a plurality of pause durations by calculating the calculated pose values, and each of the pose position candidates having a pose weight value greater than a predetermined value is provided. Is determined to be a pause position and given a pause level,
The pause duration is determined according to the pause level.

According to the data configuration and processing steps of the pause duration processing method in Chinese speech synthesis according to the first aspect of the present invention, a Chinese language of an arbitrary length waiting for speech synthesis is provided. The sentence first undergoes a word checking step in which the syllable position of each word in the sentence is checked by comparing it with words stored in a pre-configured dictionary memory.
Next, in the part of speech / extended syntax checking step, the part of speech and the extended syntax of each word are checked with reference to the dictionary memory. Further, in the phrase expansion step, adjacent ones of the plurality of words in the sentence are combined into a plurality of phrases according to the expanded syntax and the relationship between the plurality of parts of speech. At this time, all words or phrases in the sentence are regarded as basic semantic units. In general, the more characters in a semantic unit, the deeper the meaning is implied thereby. Positions before and after these semantic units are candidates for pause positions. Then, in the step of calculating the weighted value of the pose,
The weighted values of the examined or expanded semantic unit poses are calculated according to the order of the semantic units in the sentence. The weighted values are calculated in three ways: before pose, after pose, and general pose. The calculation of the weight value of the previous pose is performed via a comparison of the pre-configured previously paused character with the previously paused character in the table memory. Once the previously paused character has been examined, a weighted value of the pause is calculated for it. The calculation of the weight value of the later pose is performed via a comparison of the pre-configured later-paused character with the later-paused character in the table memory. If the character to be paused later is examined, a weighted value of the pause is calculated for it. The calculation of the weight value of the general pose is based on the number of characters in the semantic unit. By finding a pose parameter from a pre-configured pose parameter storage, a weighted value is calculated for it. Finally, in the step of determining the pause position and the pause duration, whether a pause is required before and after the plurality of semantic units according to the weighted value of the pose of the plurality of semantic units, The time is determined. From the above results, a plurality of pause positions and a plurality of pause durations corresponding to natural speech can be acquired for a Chinese sentence waiting to be synthesized.

According to the configuration of the pause duration processing apparatus for Chinese speech synthesis according to the second aspect of the present invention, first, the word checker compares the word with a word stored in a preconfigured dictionary memory. Check the location of the syllables of each word in a Chinese sentence of any length waiting to be synthesized. Next, the part-of-speech / extended syntax checking unit checks the part of speech and the extended syntax of each word with reference to the dictionary memory. Further, a phrase expansion unit is used to combine adjacent ones of a plurality of words in a sentence into a plurality of phrases in accordance with the expanded syntax and the relationship between the plurality of parts of speech. At this time, a plurality of words or phrases in the sentence are regarded as basic semantic units. The greater the number of characters in a semantic unit, the deeper the meaning it implies. Positions before and after these semantic units are candidates for pause positions. Then, via the pose weighted value calculator, the weighted values of the poses of the inspected or expanded semantic units are calculated according to the order of the semantic units in the sentence. . The weighted values are calculated in three ways: before pose, after pose, and general pose. The calculation of the weight value of the previous pose is performed via a comparison of the pre-configured previously paused character with the previously paused character in the table memory. Once the previously paused character has been examined, a weighted value of the pause is calculated for it. The calculation of the weight value of the later pose is performed via a comparison of the pre-configured later-paused character with the later-paused character in the table memory. If the character to be paused later is examined, a weighted value of the pause is calculated for it. The calculation of the weight value of the general pose is based on the number of characters in the semantic unit. By finding a pose parameter from a pre-configured pose parameter storage, a weighted value is calculated for it. Finally, via a pause position and pause duration determination unit, according to the weighted values of the poses of the plurality of semantic units, whether or not a pause is required before and after the semantic unit, Is determined. From the above results, a plurality of pause positions and a plurality of pause durations corresponding to natural speech can be obtained for a Chinese sentence that is synthesized and is waiting to be output for use.

[0010] Other features and advantages of the present invention will become apparent in the following detailed description of the preferred embodiments, taken in conjunction with the accompanying drawings.

[0011]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a system block diagram illustrating a preferred embodiment of a pause duration processing apparatus for a system for converting Chinese text to speech according to the present invention. As illustrated in FIG. 1, the system includes the following components:

Reference numeral 10 denotes a sentence input unit, such as a keyboard, for inputting text of a sentence to be synthesized.

Reference numeral 11 denotes a dictionary memory for storing a plurality of Chinese words and corresponding information such as part of speech and extended syntax. A schematic diagram illustrating the configuration of the dictionary memory 11 is shown in FIG. The parts of speech include nouns (N), verbs (V), adjectives (J), adverbs (A), and the like. The expansion syntax for determining whether to expand from a word to a phrase is φ φ indicating that there is no expansion.
And AN indicating that a noun is connected later, BN indicating that a noun is connected before, AV indicating that a verb is connected later, and B indicating that a verb is connected before.
V. Further, the dictionary memory 11 may include other information such as a voice marker. The phonetic marker can also be called a phonetic symbol (pronunciation symbol). For example, “qiu2 sheng1 yi4 zhi4” (representing “Chinese willingness” in Chinese) shown in FIG. 9 is a phonetic symbol string combined from the phonetic symbol It is. The numbers “2, 1, 4, 4” here represent tones that are accents unique to Chinese.

Reference numeral 12 denotes a word inspection unit that determines the position of a syllable of each word in the input sentence by comparing the word with the word stored in the dictionary memory 11.

Reference numeral 13 denotes a part-of-speech / extended syntax check unit for searching the dictionary memory 11 for a part-of-speech and an extended syntax corresponding to each inspected word.

The phrase expansion unit 14 expands a word adjacent to each other into a unit (or a plurality of phrases) having a more complete meaning by using the part of speech and the expansion syntax of each word. It represents.

Reference numeral 15 denotes a “character to be paused before” (hereinafter, “character to be paused”).
(Referred to as the previous pause character) and a table memory of characters to pause before for storing corresponding information. The preceding pause character is a character that requires a pause before the character when speaking Chinese. The corresponding information is, for example, a parameter (fp) of the weight value of the previous pose for calculating the weight value of the previous pose of the previous pose character. A schematic diagram of the structure of the table memory 15 for the previous pause character is shown in FIG.

16 is a "character to be paused later" (hereinafter referred to as "character").
Character table memory for storing later-paused characters) and corresponding information. The post-pause character is a character that requires a pause after the character when speaking Chinese. The corresponding information is, for example, a parameter (bp) of the weight value of the post-pose for calculating the weight value of the post-pose of the post-pose character. A schematic diagram of the structure of the post-pause character table memory 16 is shown in FIG.

Reference numeral 17 denotes a pose parameter storage unit for storing information of a parameter (pp) of a weight value of a general pose corresponding to the number of a plurality of characters forming the phrase of the semantic word and according to the number. Represents A weighted value of the general pose of the semantic word is calculated according to the number of characters forming the phrase of the semantic word. The parameter of the weight value of the general pose is a parameter related to a situation that requires a pause before a certain character and a situation that requires a pause after a certain character. . A schematic diagram of the structure of the pause parameter storage unit 17 is shown in FIG.

Reference numeral 18 denotes a table memory 1 for the previous pause character.
5, the post-pose character table memory 16 and the parameter of the weight value of the pose obtained from the pose parameter storage unit 17, and according to the position of the semantic word in the sentence, 5 illustrates a pose weighted value calculator for calculating weighted values of word poses. The calculation is performed by multiplying the position of the semantic word by the parameter, so that the weighted value of the pose of the semantic word located later in the input sentence has a larger value. Have.

19 determines whether a pause should be inserted before or after the semantic word, and determines the pause duration based on the magnitude of the calculated weighted value of the semantic word pose. This represents a pause position and pause duration determination unit for determination. Next, the text data of the sentence into which the pause position and the pause duration are inserted, which is output from the pause position and pause duration determination unit 19, is subjected to voice synthesis by a known voice synthesis unit.

Reference numeral 20 denotes a register and a memory buffer unit which are used by the apparatus when processing a plurality of pause positions and a plurality of pause durations, the register including a plurality of different registers and a plurality of memory buffer areas.
The sentence input to the sentence input unit 10 is subsequently sent to the word inspection unit 1
2. The part-of-speech / extended syntax checking unit 13, the phrase expanding unit 14, the pause weighted value calculating unit 18, the pause position and the pause duration calculating unit 19 are sequentially processed, and each block is a register and a memory buffer. Write data using different parts in the unit 20,
Then read it.

When the apparatus processes multiple pause positions and multiple pause durations, different registers and memory buffer areas must be used. In FIG. 1, they are omitted and not shown, but are necessary in actual implementation. The register and the memory buffer area are, as constituent elements, a TextBuffer memory buffer area for storing text data of an input sentence, and a Wd_i register (numbers 1, 2, 3,...) For storing a number specifying a word in the sentence. For example, 1 designates the first word in a sentence) and a Wd array register for storing the value of each inspected word (word start position, word length) in the input sentence. In the Wd array register, for example, the following expression indicates that the fourth word in the sentence starts from the fifth syllable and has a word length of two syllables.

[0025]

## EQU1 ## Wd [4] = (5,2)

The register and memory buffer area further includes a Wd_type array register for storing the part of speech of each checked word in the input sentence. For example, the following expression indicates that the part of speech of the second word in the sentence is a noun.

[0027]

[Equation 2] Wd_type [2] = N

The register and memory buffer area further includes a Wd_expand array register for storing expanded syntax of each examined word in the input sentence. For example, the following expression is an extended syntax of the first word in the sentence, and indicates that a noun is connected after the first word.

[0029]

## EQU3 ## Wd_expand [1] = AN

The register and memory buffer area further include a Wd_phr array register for storing syllable values (phrase length, phrase position) forming each phrase (phrase) in the input sentence. For example, the following expression indicates that the fourth syllable in the sentence forms the first syllable of a three-syllable phrase.

[0031]

[Equation 4] Wd_phr [4] = (3,1)

[0032] The register and the memory buffer area further include Phr_ for storing a start position of a phrase in a sentence.
A start register, a Phr_end register for storing the end position of the phrase in the sentence, a Phr_length register for storing the length of the phrase in units of syllables, and a value of each phrase in the input sentence (phrase start syllable) ,
And a Phrase array register for storing the length of the phrase. In the Phrase array register, for example, the following expression indicates that the fourth phrase in the sentence starts from the ninth syllable and the phrase length of two syllables is To have

[0033]

[Equation 5] Phrase [4] = (9, 2)

The register and the memory buffer area further include P_ for storing a number designating a phrase in a sentence.
An i register, a P_begin register, and a P_end register (using Equations 1, 2, 3,..., for example, 1 designates the first phrase in a sentence). Here, the P_begin register and the P_end register correspond to a start pointer and an end pointer of a processing section for determining whether or not it is necessary to insert a pause duration. Further, in order to execute the above determination, an S_range register for storing a value calculated using the values of the P_begin register and the P_end register is included.

The register and the memory buffer area further include f for storing the pause parameter of the phrase.
It includes a p register, a bp register, and a pp register.
Here, the value of the fp register represents the before pause parameter, the value of the bp register represents the after pause parameter, and p
The value of the p register represents a general pause parameter.

The register and memory buffer area further includes a W_minor register and a W_major register for storing weighted values for determining a pause. 80 m when the weighted value of the pose is greater than the value of the W_minor register but less than the value of the W_major register
A shorter pose of s is determined. When the weighted value of the pose is greater than or equal to W_major, a longer pause of 120 ms is determined. In this embodiment, the value of the W_minor register indicating “weighted value of shorter pause” is set to 20, and the value of the W_major register indicating “weighted value of longer pause” is set to 40. ing.

The register and memory buffer areas further include a Wfp register, a Wbp register, and a Wpp register for storing weighted values of the pause of the phrase. The value of the Wfp register indicates the weight value of the previous pause. The value of the Wbp register indicates the weight value of the post-pause. The value of the Wpp register indicates the weight value of the general pose.

The register and the memory buffer area further include P_ for storing a number specifying a pause phrase.
Includes pause register. For example, when the value of the P_pause register is equal to 2, it indicates that a pause is required after the second phrase.

The register and the memory buffer area further store a number specifying a syllable at a pause position.
Includes S_pause register. For example, when the value of the S_pause register is equal to 7, it indicates that a pause is required after the seventh syllable.

The register and the memory buffer area further include a D_pause register for storing a pause duration in units of milliseconds. In this embodiment, the longer pose is 120 ms and the shorter pose is 80 ms.

The register and memory buffer area further include a pause position and a pause duration (S_paus
e, D_pause) for storing a List_pause array register. For example, the following expression indicates that the first pause position has a duration of 120 ms after the seventh syllable.

[0042]

## EQU6 ## List_pause [1] = (7,120 ms)

FIGS. 2 to 6 show flowcharts of the operation of the pause duration processing apparatus for the system for converting Chinese text to speech according to the embodiment of FIG. A semantic word is used as a unit. The following steps are described in FIGS.

In step S1 of FIG. 2, the text of the sentence is input to the TextBuffer memory buffer area.

Then, in step S2, it is checked whether the key of the currently input text is an end key in the text. If it is the end key, the flow proceeds to step S3. Otherwise, the flow returns to step S1.

In step S3, the text in the sentence is examined, and each word in the sentence is found by comparing the text with a plurality of words in the dictionary memory 11. The position of the word in the sentence and the length of the word are stored in the Wd array register.

Next, in step S4, the part of speech and the extended syntax corresponding to each word are searched from the dictionary memory 11 according to each inspected word in the Wd array register, and these are stored in the Wd_type array register and the Wd_expand array register, respectively. It is memorized.

Next, in step S5, according to each examined word in the Wd array register, the constituent data of each syllable corresponding to the word is stored in the Wd_phr array register.

Next, in step S6 of FIG. 3, Wd_i
The value in the register is set to 1, and the phrase expansion process starts from the first word.

Next, in step S7, it is determined whether or not the (Wd_i) -th word has the extended syntax. (When the value is φ, this indicates that the word has no extended syntax). When the (Wd_i) -th word has the extended syntax, it is determined at the same time which extended syntax the word has, and the flow proceeds to step S8. Otherwise, the flow is step S
Proceed to 11.

In step S8, according to the extended syntax determined in step S7, it is determined whether or not the part of speech of an adjacent word before or after the (Wd_i) -th word corresponds to the extended syntax. If there is such a response, the flow proceeds to step S9. Otherwise, the flow goes to step S11.

In step S9, the phrase expansion process is started. As the expansion proceeds, the word specified by the value of the Wd_i-1 register is selected as the word to be expanded. As the expansion proceeds, the word specified by the value of the Wd_i + 1 register is selected as the word to be expanded. When multiple words are expanded into a phrase, the phrase is then considered a new word to be expanded to perform the expansion process. The adjacent expanding word and the word to be expanded are combined to form an expanded phrase. The start position Phr_start and the end position Phr_end for the expanded phrase are found, and the length of the expanded phrase is calculated by the following equation.

[0053]

[Equation 7] Phr_length = Phr_end−Phr_start + 1

Subsequently, the start position Phr_start and the end position P
hr_end and extended phrase length Phr_length are
These are stored in the Phr_start register, Phr_end register, and Phr_length register, respectively.

Next, in step S10, the values of the corresponding syllables in the Wd_phr array register are updated according to the expanded phrase. Individually, the update proceeds as follows.

[0056]

[Equation 8] Wd_phr [Phr_start] = (Phr_length, 1)

## EQU9 ## Wd_phr [Phr_start + 1] = (Phr_length, 2)

The following is continued in the same manner, and finally the following equation is obtained.

[0058]

[Expression 10] Wd_phr [Phr_end] = (Phr_length, Phr_length)

Next, in step S11, it is determined whether or not the value of the Wd_i register has reached a value designating the last word. When the last word has been reached, the flow proceeds to step S13. Otherwise, the flow goes to step S12.

In step S12, the value in the Wd_i register is incremented by one, and then the flow returns to step S7 to continue the phrase expansion process.

In step S13, the values specifying a plurality of syllables forming a phrase in the Wd_phr array register are converted into values in the Phrase array register in which the phrase is a basic unit. At this point, the sentence was broken down into multiple phrases in semantic units. In a following step, these values are taken into account in the calculation of the weighted values of the poses, and whether a pause is required before or after them and an appropriate pause duration are determined.

Next, in step S14 of FIG. 4, the values of the P_begin and P_end registers, which are phrase pointers, are initialized to 1.

Next, in step S15, P_begin
The number of syllables from the register to the phrase specified by the P_end register is counted and stored in the S_range register. The calculation step is performed as follows.

[0064]

S_range = Phrase [P_end]. syllable ＋ Phra
se [P_end]. length-Phrase [P_begin]. syllable

Next, in step S16, S_range
It is determined whether the value of the register is still less than 13 syllables. In a pause (breathing) operation when a human speaks, one cycle corresponds to approximately 6 to 8 Chinese characters (ie, Chinese syllables), and a pause is a semantic unit of a phrase or word. Occurs before or after. The average person needs at least one pause within a section consisting of about 13 syllables. Therefore, in the present embodiment, the threshold value of “13” is adopted, and when the threshold value is smaller than 13 syllables, the flow proceeds to step S17. Otherwise, the flow proceeds to step S19, and determines a pause position and a pause duration from the P_begin register to the phrase specified by the P_end register.

In step S17, it is determined whether the phrase specified by the P_end register is the last phrase of the sentence. When the phrase specified by the P_end register is the last phrase, this indicates that all the phrases in the sentence have been checked. Subsequently, the flow proceeds to step S22, and outputs all pause positions and pause durations. Otherwise, the process proceeds to step S18.

In step S18, the value in the P_end register is incremented by 1, and the flow returns to step S15.

In step S19, the weighted values of the poses of all the phrases from the P_begin register to the phrase specified by the P_end register are calculated,
Pause position S_pause and pause duration D_p set based on a phrase specified by the value of the P_pause register
ause is determined, and the values of the S_pause and D_pause registers are stored in a pause array List_pause. Here, two pose levels, which are set based on experience,
The pause duration is determined according to the values of the W_major and W_minor registers. This step S19 further includes sub steps S19-1 to S19-16 as a subroutine. In sub-step S19-1 of FIG.
The pointer P_i specifies the value (start pointer) of the P_begin register. Next, in sub-step S19-2,
The "previous pause character table memory" 15 (FIG. 10) is searched to determine whether the phrase represented by the value of the P_i register is the previous pause character. If the character is the previous pause character, the flow proceeds to substep S19-3. Otherwise, the flow goes to sub-step S19-5. In sub-step S19-3, in order to determine whether or not the character of the phrase specified by the P_i register is a character to be paused before, the previous pause parameter fp of the character is stored in the table memory 15 of the previous pause character.
Is found, and the weight value Wfp of the previous pose for the character is calculated as follows:

[0069]

Wfp = (Phrase [P_i] .syllable-Phras
e [P_begin]. syllable + Phrase [P_i]. length) ×
fp

Next, in substep S19-4,
Wfp indicates “weighted value of shorter pause”
It is determined whether the value is equal to or more than the value of the W_minor register (the value of the W_minor register in this embodiment is set to 20). If Wfp is greater than or equal to the value of the W_minor register, it is determined that a pause is required before the phrase, and then the flow proceeds to sub-step S19-1 in FIG.
Proceed to 6. Otherwise, the flow proceeds to substep S
Proceed to 19-5. In sub-step S19-5,
The "post-pause character table memory 16" (FIG. 11) is searched to determine whether the phrase represented by the value of the P_i register is a post-pause character. If it is a post-pause character, the flow proceeds to sub-step S19-6. Otherwise, the flow goes to sub-step S19-8. In sub-step S19-6, the "post-pause character table memory" is used to check the post-pause parameter bp of the character of the phrase specified by the P_i register.
16 is retrieved, and a weight value Wbp of the post-pose for the character is calculated as follows.

[0071]

Wbp = (Phrase [P_i] .syllable-Phras
e [P_begin]. syllable + Phrase [P_i]. length) ×
bp

Next, in substep S19-7,
Wbp indicates “shorter pose weighted value”
It is determined whether the value is equal to or more than the value of the W_minor register.
If Wbp is greater than or equal to the value of the W_minor register, it is determined that a pause is required after the phrase, and then the flow proceeds to substep S19-16. Otherwise, the flow goes to sub-step S19-8. In sub-step S19-8, it is determined whether the value of the P_i register has reached a value designating the P_end register. P_i
When the value of the register reaches the value designating the P_end register, the flow proceeds to substep S19-10 in FIG. Otherwise, the flow proceeds to sub-step S19-
Go to 9. In sub-step S19-9, the value of the P_i register is incremented by one. Next, the flow returns to sub-step S19-2. Sub-step S19
At -10, the value of the P_i register again specifies the phrase specified by the P_begin register, and the determination of the general pose is started. Next, in sub-step S19-11, the parameter pp of the general pose weighted value is found from FIG. 12 according to the number of characters in the phrase represented by the value of the P_i register, and the general pose weighted value is found. Wpp is calculated as follows:

[0073]

Wpp = (Phrase [P_i] .syllable-Phras
e [P_begin]. syllable + Phrase [P_i]. length) ×
pp

Next, in sub-step S19-12, it is determined whether or not Wpp is equal to or greater than a value of a W_minor register indicating a "weight value of a shorter pause". When Wpp is equal to or greater than the value of the W_minor register,
It is determined that a pause is required after the phrase, and then the flow proceeds to sub-step S19-16. Otherwise, the flow goes to sub-step S19-13. In sub-step S19-13, the value of the P_i register is P
It is determined whether or not the value specified by the _end register has been reached. When the value of the P_i register has reached the value designating the P_end register, the flow proceeds to substep S19-15. Otherwise, the flow proceeds to sub-step S19-
Proceed to 14. In sub-step S19-14, P_i
The value of the register is incremented by 1, and the flow proceeds to sub-step S19-11. Sub-step S19-
At 15, no pause has been set at this time
There is a syllable consisting of at least 13 characters from the phrase specified by the P_begin register to the phrase specified by the P_end register. Therefore, a forced pose is set and the value of Wpp is made equal to the value of the W_major register, which indicates "weighted value of the longer pause" (the value of the W_major register in this embodiment is set to 40). ing). Next, the flow proceeds to substep S19.
Proceed to -16. In sub-step S19-16, a pause phrase P_pause, a pause position S_pause, and a pause duration D_pause are determined in this step. The values of the S_pause and D_pause registers above are the pause array List_p
Added to ause.

(1) The pause phrase P_pause is determined as follows. When in the front pose,

P_pause = P_i-1 Otherwise,

[Equation 16] P_pause = P_i

(2) The pause position S_pause is determined as follows.

[0077]

S_pause = Phrase [P_pause]. syllable + Ph
rase [P_pause]. length-1

(3) The pause duration D_pause is determined as follows. When Wfp, Wbp or Wpp is greater than W_major

D_pause = 120 ms Otherwise,

[Equation 19] D_pause = 80 ms

Next, in step S20 of FIG.
It is determined whether the phrase specified by the value of the pause register has reached the last phrase of the sentence. When the last phrase has been reached, the flow proceeds to step S22. Otherwise, the flow goes to step S21.

In step S21, the value of the P_begin register (start pointer) and the value of the P_end register (end pointer) specify a phrase immediately following the pause phrase P_pause, and the flow proceeds to step S15.

In step S22, all values of the pause position S_pause and the pause duration D_pause of the whole sentence are set in the pause array List_pa for use in the speech synthesis system.
Output from use, the operation of the device of the present invention ends. Next, the text data of the sentence into which the pause position and the pause duration are inserted, which is output from the pause position and pause duration determination unit 19, is subjected to voice synthesis by a known voice synthesis unit.

[0082]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Referring mainly to the flowcharts of FIGS. 2 to 6, the operation of the pause duration processing apparatus for a system for converting Chinese text to speech of the preferred embodiment constructed above will be described. To illustrate, in the example below, the sentence “Robust

[Outside 3] Enter the most effective method of war sickness.

The process flow of the embodiment proceeds as follows. In step S1 of FIG. 2, a sentence is input using a sentence input unit 10 such as a keyboard. In step S2, the end is detected by detecting the end key in the text. At this time, the text data of the sentence "

[Outside 4] The most effective method of the war is stored in the TextBuffer [] memory buffer area.

Then, in step S 3, by comparing the words with a plurality of words in the dictionary memory 11, the word checking unit 12 determines whether each word in the sentence is “strong”, “target”, or “willing to seek”
`` Yes '' `` Noh

[Outside 5] "" Sengoku """Sick""" Most """Effective""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" V e"""""
Is checked, and the start position of each word in the sentence and the number of characters of the word are stored in the array register as a sequence of pairs of numbers (word start position, word length) in Wd []. That is, it is as illustrated below.

[0085]

[Table 1] ―――――――――――――――――――――――――――――――――― Wd [1] = (1,2) ,… “Strength” Wd [2] = (3,1),… “Target” Wd [3] = (4,4),… “Want to volunteer” Wd [4] = (8,1),… Wd [5] = (9,2),… “Noh

[Outside 6] Wd [6] = (11, 2),… “War Victory” Wd [7] = (13, 2),… “Sick” Wd [8] = (15, 1),… “Most” Wd [9] = (16, 2), ... "Effective" Wd [10] = (18, 1), ... "target" Wd [11] = (19, 2), ... "Method" ----------- ―――――――――――――――――――――――――――

Next, in step S5, in accordance with each word stored in Wd [], the part-of-speech / extended syntax checking unit 13 reads the word from the dictionary memory 11 (the contents of which are shown in FIG. 9). The part-of-speech and the extended syntax are found and stored in the Wd_type array register and the Wd_expand array register, respectively. That is, it is as illustrated below.

[0087]

[Table 2] ――――――――――――――――――----- ] = Φ… “robust” Wd_type [2] = J, Wd_expand [2] = BN… “target” Wd_type [3] = N, Wd_expand [3] = φ… “Want to volunteer” Wd_type [4] = V, Wd_expand [4] = φ… “Yes” Wd_type [5] = A, Wd_expand [5] = φ…

[Outside 7] Wd_type [6] = V, Wd_expand [6] = AN… “War Victory” Wd_type [7] = N, Wd_expand [7] = φ… “Sick” Wd_type [8] = A, Wd_expand [8] = φ ... “ Wd_type [9] = N, J, Wd_expand [9] = φ ... "effective" Wd_type [10] = J, Wd_expand [10] = BN ... "target" Wd_type [11] = N, Wd_expand [11] = φ… “Method” ――――――――――――――――――――――――――――――――――――

Next, the phrase expansion unit 14 is used to start the phrase expansion processing. First, in step S5, according to each examined word in the Wd array register, the configuration information of each syllable that together forms a word is stored in the Wd_phr array register.

[Mathematical formula-see original document] Wd_phr [syllable position] = (length of phrase,
(Position in the phrase). That is, it is as illustrated below.

[0089]

[Table 3] ―――――――――――――――――――――――――――――――――― Wd [1] = (1,2) → Wd_phr [1] = (2,1)… “strong” → Wd_phr [2] = (2,2) Wd [2] = (3,1) → Wd_phr [3] = (1,1)… Wd [3] = (4, 4) → Wd_phr [4] = (4, 1)… “willing to seek” → Wd_phr [5] = (4, 2) → Wd_phr [6] = (4, 3) → Wd_phr [7] = (4,4) Wd [4] = (8,1) → Wd_phr [8] = (1,1)… “Zero” Wd [5] = (9,2) → Wd_phr [9] = (2,1)… "Noh

[Outside 8] → Wd_phr [10] = (2,2) Wd [6] = (11,2) → Wd_phr [11] = (2,1)… "War" → Wd_phr [12] = (2,2) Wd [ 7] = (13, 2) → Wd_phr [13] = (2, 1)… “Sick” → Wd_phr [14] = (2, 2) Wd [8] = (15, 1) → Wd_phr [15] = (1,1)… “most” Wd [9] = (16,2) → Wd_phr [16] = (2,1)… “valid” → Wd_phr [17] = (2,2) Wd [10] = (18, 1) → Wd_phr [18] = (1, 1)… “target” Wd [11] = (19, 2) → Wd_phr [19] = (2, 1)… “Method” → Wd_phr [20] = (2,2) ――――――――――――――――――――――――――――――――――

Strong / target / recruitment will / requirement / noh

[Outside 9] / War / sickness / most / effective / effective / method

Thereafter, in step S6 of FIG. 3, Wd_i
The value in the register is set to 1, and the expansion processing of the first word “robust” is started. In step S7,

## EQU21 ## It is determined whether Wd_expand [1] １φ. Since Wd_expand [1] is φ, no expansion processing is necessary, and the process proceeds to the expansion processing of the next word.
In step S11, the value of the Wd_i register is checked,
It is determined whether the value specifying the last word has been reached. The result is "NO", and the value of the Wd_i register is changed to 2 in step S12. Next, in step S7 again, the expansion processing of the second word is continued,

## EQU22 ## It is determined whether or not Wd_expand [2] ≠ φ. here

## EQU23 ## Since Wd_expand [2] = BN, which represents an extended syntax for generating a phrase in the order of N and BN with nouns connected before, the part of speech of the previous word is checked in step S8. Is done. At this time,

[Expression 24] Wd_type [Wd_i-1] = N represents a noun corresponding to the extended syntax BN. In this way, the second (Wd_i = 2) word “target” and the first (Wd_i−1 = 1) word “robust” are expanded to form the phrase “robust”. In other words, Wd_phr
[1], Wd_phr [2], and Wd_phr [3] are expanded to include a start position Phr_start = 1 and an end position Phr_end = 3.
And the phrase length

Form a new phrase with Phr_length = 3-1 + 1 = 3. In step S9, these values are stored in the Phr_start register, Phr_end register, and Phr_length register, respectively. Subsequently, in step S10, the value associated with this phrase containing three syllables in the Wd_phr array register is updated as follows.

[0092]

[Table 4] ―――――――――――――――――――――――――――――――――― Wd_phr [1] = (2,1) → Wd_phr [1] = (3,1) Wd_phr [2] = (2,2) → Wd_phr [2] = (3,2) Wd_phr [3] = (1,1) → Wd_phr [3] = (3) , 3) ――――――――――――――――――――――――――――――――――――

Next, in step S11, it is determined that the value of the Wd_i register has not yet reached the value designating the last word. Therefore, in step S12, the value of the Wd_i register is incremented by 1 to 3, and Continue the expansion processing of the word "Shouki will". The expanded phrase “robust” is regarded as a word to be expanded in a later expansion process. In step S7

After determining that Wd_expand [3] = φ, since the value of the Wd_i register has not yet reached the value specifying the last word in step S11, the value of the Wd_i register is only 1 in step S12. The value is incremented, and step S7 is executed again. Thus, steps S7, S8, S9, S10,
S11 and S12 are repeated to process the fourth word, the fifth word,... Up to the eleventh word “method”. The phrase expansion process is completed by detecting that the last word in the sentence has been reached in step S11. At this time, the values in the Wd_phr array register are as follows.

[0094]

[Table 5] ―――――――――――――――――――――――――――――――――― Wd_phr [1] = (2,1) → Wd_phr [1] = (3,1)… “Robust” Wd_phr [2] = (2,2) → Wd_phr [2] = (3,2) Wd_phr [3] = (1,1) → Wd_phr [ 3] = (3,3) Wd_phr [4] = (4,1) → Wd_phr [4] = (4,1)… “willingness to seek” Wd_phr [5] = (4,2) → Wd_phr [5] = (4,2) Wd_phr [6] = (4,3) → Wd_phr [6] = (4,3) Wd_phr [7] = (4,4) → Wd_phr [7] = (4,4) Wd_phr [8 ] = (1,1) → Wd_phr [8] = (1,1)… “Yes” Wd_phr [9] = (2,1) → Wd_phr [9] = (2,1)… ”Noh

[Outside 10] Wd_phr [10] = (2,2) → Wd_phr [10] = (2,2) Wd_phr [11] = (2,1) → Wd_phr [11] = (4,1)... 12] = (2,2) → Wd_phr [12] = (4,2) Wd_phr [13] = (2,1) → Wd_phr [13] = (4,3) Wd_phr [14] = (2,2) → Wd_phr [14] = (4,4) Wd_phr [15] = (1,1) → Wd_phr [15] = (1,1) ... “most” Wd_phr [16] = (2,1) → Wd_phr [16 ] = (3,1) "effective" Wd_phr [17] = (2,2) → Wd_phr [17] = (3,2) Wd_phr [18] = (1,1) → Wd_phr [18] = ( 3,3) Wd_phr [19] = (2,1) → Wd_phr [19] = (2,1) ... “Method” Wd_phr [20] = (2,2) → Wd_phr [20] = (2,2) ――――――――――――――――――――――――――――――― ----

From the above, it can be seen that the 11 words in the sentence “

[Outside 11] Eight phrases are "strong" and "willing will"
`` Yes '' `` Noh

[Outside 12] ”, Victory sickness,“ most ”,“ effective ”and“ method ”.

Next, in step 13, Wd_phr [] using syllables as index keys to facilitate the calculation of the weighted value of the pose for each phrase.
The value in the array register is Phr using the phrase (specified by "Phrase_id") as the index key
ase [] Converted to the value in the array register. Phrase
[] The format of the array register is a pair of numbers (syllab
le, length)

[Mathematical formula-see original document] Phrase [Phrase_id] = (syllable, length). After the conversion in step S13, the Phrase
[1] The following values are obtained from the array register to the Phrase [8] array register.

[0097]

[Table 6] ―――――――――――――――――――――――――――――――――― Phrase [1] = (1,3) , Phrase [2] = (4,4), Phrase [3] = (8,1), Phrase [4] = (9,2), Phrase [5] = (11,4), Phrase [6] = (15,1), Phrase [7] = (16,3), Phrase [8] = (19,2). ――――――――――――――――――――――――――――――――――――

When starting from step S14 in FIG.
The phrases in the sentence are considered to be semantic processing units. By calculating the weighted value of the pose of each semantic unit, it is determined whether a pause duration is required before and after the semantic unit. At first,
As the processing section, the number of phrases including at least 13 syllables (including both syllables) is subsequently searched from the sentence. In step S14, the section start pointer P_begin is initially set to 1, and the section end pointer P_end is set to be the same as the start pointer. Subsequently, in step S14, P_begin and P_end
Are calculated and stored in the S_range register. Therefore, the following equation is established.

[0099]

S_range = Phrase [P_end]. syllable ＋ Phra
se [P_end]. length-Phrase [P_begin]. syllable =
1 + 3-1 = 3

Thereafter, in step S16, it is determined whether or not S_range <13 indicating that the section is 13 syllables or less is satisfied. The value of the P_end register then specifies the next phrase. In step S17,
If it is determined that the value of the P_end register has not yet reached the value designating the last phrase, step S18
In, the value of the P_end register is incremented by 1, and the P_begin register (corresponding to the start pointer) and P_e
The calculation of the number of syllables between the phrases respectively designated by the nd register (corresponding to the end pointer) is continued, and it is determined in step S15 whether or not the number of syllables is 13 or more syllables. Steps S15, S16, S to go around
17 and S18 are repeated. P_end register value is 5
In step S15, the value of the S_range register is determined to be 14, and in step S16, the value of the S_range register is determined to be 13 or more syllables. Next, proceeding to step S19, the pause position and the pause duration of the section constituted by the expanded five phrases designated by P_begin = 1 to P_end = 5 are determined as described below.

[0101]

[Table 7]

In order to examine the phrases in the section in the sequence and to see if there is a pre-pause character or a post-pause character, first, in sub-step S1 of FIG.
At 9-1, the value of the P_i register is set to the value of the P_begin register (start pointer). At this time, the value of the P_i register is 1, which specifies the phrase “robust”. Next, in sub-step S19-2, "table memory of previous pause character" 15 is searched to find that "robust" is not the previous pause character. Next, in sub-step S19-5, "table memory for post-pause character"
16 is searched to find that "robust" is not a post-pause character. In order to continue checking the next phrase, after confirming in step S19-8 that the value of the P_i register has not yet reached the value specifying the last phrase of the section, P_i is determined in substep S19-9. The value of the register is incremented by one, and the examination of the phrase "request for will" is continued, and these characters (the other 20 phrases) are recognized in the sub-step S19-2 or the sub-step S19-5 by the pre-pause character or the post-pause character. Is not found. Sub-step S19-
At 8, check that the P_i register has not yet reached the last phrase of the section. Next, the value of the P_i register is incremented by 1, and the phrase “ze”
Is continued, and in sub-step S19-2, it is found that the character "ZE" constituting the phrase is a pre-pause character. The parameter fp of the weight value of the pose of “ze” is 5.4, as obtained from the “table memory 15 of the previous pose character”, which is weighted in the sub-step S19-3. The calculated value Wfp is calculated as follows.

[0103]

[Mathematical formula-see original document] Wfp = (Phrase [P_i] .syllable-Phras]
e [P_begin]. syllable + Phrase [P_i]. length) ×
fp = (8-1 + 1) × 5.4 = 43.2

Thereafter, in sub-step S19-4, since Wfp is equal to or greater than the value of the W_minor register (the value of the W_minor register is set to 20 in this embodiment), sub-step S19-16 of FIG. 6 is executed. . Because it is the previous pose, the pose phrase P_pause is

## EQU30 ## P_pause = P_i = 3-1 = 2, and the pause position S_pause is represented by the following equation.

S_pause = Phrase [P_pause]. syllable + Phrase [P_pause]. length-1 = Phrase [2]. syllable + Phrase [2]. length-1 = 4 + 4-1 = 7

Since the value of Wfp is 43.2, W_majo
It is larger than r (the value of the W_major register is set to 40 in this embodiment), and the pause duration follows the following equation.

[0106]

[Formula 32] D_pause = 120 ms

Next, the values of the pause position S_pause (having a value of 7) and the pause duration D_pause (having a value of 120 ms) are stored in the pause array List_pause, respectively, and the values of the phrases specified by the P_begin register and the P_end register are stored. The determination of the pause position and the pause duration is completed. Next, the flow returns to step S20 in FIG.
It is checked whether the phrase specified by the pause register is the last phrase of the sentence. At this time, P_pause = 2
And not the last phrase. Therefore, step S21
Is executed, and the phrase specified by the P_begin and P_end registers is replaced with the phrase specified by the P_pause register.
Specify the next phrase. That is,

P_begin = P_end = P_pause + 1 = 2 + 1 = 3 Steps S15, S16, S17, and S18 to be repeated are repeated until P_end = 8 at which the value of the S_range register becomes 13 in step S16. Step S
Step 19 is executed to prepare for the determination of the pause position and the pause duration of the section composed of the expanded six phrases specified by P_begin = 3 to P_end = 8 as follows.

[0108]

[Table 8]

In order to check in the sequence whether the phrase in the sentence has a pre-pause character or a post-pause character, the value of the P_i register is set to the value of the P_begin register in sub-step S19-1. Circulating sub-steps S19-2, S19-5, S19-8 and 19-
9 is repeated. Finally, the phrase "most" whose value of the P_i register is 6 is examined, and in sub-step S19-2, it is found that the character "most" constituting the phrase is the preceding pause character. The parameter fp of the weighted value of the “most” previous pose is found to be 2.7 from the “previous pose character table memory 15”. In sub-step S19-3, the weight value Wfp of the previous pose is calculated as follows.

[0110]

Wfp = (Phrase [P_i] .syllable-Phras
e [P_begin]. syllable + Phrase [P_i]. length) ×
fp = (15−8 + 1) × 2.7 = 21.6

Next, in sub-step S19-4, W
Since fp is determined to be equal to or greater than the value of the W_minor register (the value of the W_minor register is 20 in this embodiment).
), The sub-step S19-16 is executed. Because the pose is before, the pose phrase P_pause is

P_pause = P_i-1 = 6-1 = 5 Also, the pause position S_pause is

S_pause = Phrase [P_pause]. syllable + Phrase [P_pause]. length-1 = Phrase [5]. syllable + Phrase [5]. length-1 = 11 + 4-1 = 14. Since Wfp = 21.6 and smaller than W_major (the value of the W_major register is set to 40 in this embodiment), the pause duration D_pause is

[Formula 37] D_pause = 80ms

Thereafter, the value (14) of the S_pause register and the value (80 ms) of the D_pause register are set to the pause array Lis.
The determination of the pause position and the pause duration between the phrases stored in t_pause and specified by the P_begin and P_end registers is completed. The flow returns to step S20, and again
Checks whether the phrase specified in the P_pause register is the last phrase in the sentence. At this time, P_pause =
5 is not the last phrase. Therefore, the process proceeds to step S21, where the phrase specified by the P_begin and P_end registers is specified as the next phrase after the phrase specified by the P_pause register. Ie

[Equation 38] P_begin = P_end = P_pause + 1 = 5 + 1 = 6

Steps S15, S16, S17 to go around
And S18 are repeated until P_end = 9. In step S17, the value of the P_end register has reached a value designating the last phrase of the sentence. Then step S22
Is executed, the value of the pause array List_pause is output to the sequence, and the processing operation of the apparatus of the present invention ends. Next, speech synthesis can be performed on the text data into which the pause has been inserted by a known speech synthesis unit.

[0114] The example sentence "

[Outside 13] The pause position and pause duration obtained for the “Strictly effective method” are (7,120 ms) and (14,
80 ms). In other words, there is a pause of 120 ms after the seventh syllable “Shi”, and the 14th syllable “Ma”
Is followed by a pause of 80 ms. FIG. 7 illustrates the waveform of the original synthesized speech according to the prior art speech synthesizer, which waveform is further processed using the pause duration processing device of the present invention, as shown in FIG. Obtain synthesized speech with natural pause durations with complex waveforms.

The present invention should not be limited to the embodiments described above. For example, in the phrase expansion unit 14 of the pause duration processing apparatus of the present invention, a phrase marker can be added at the time of input instead of using phrase expansion syntax in order to expand words adjacent to each other into a phrase. . Put a phrase marker on the input sentence,
A word is detected by referring to the dictionary memory 11 for the input sentence, and then the phrase is automatically extended from the word to generate a phrase.
In addition, whether to perform the judgment process manually in advance, or to automatically execute the judgment process of the phrase using a phrase dictionary,
Alternatively, the position of the phrase can be found either by using a statistical corpus or the like, whereby the pause position and pause duration between the phrases can be determined. Instead, a phrase cache can be constructed to check for phrases in the input sentence in a similar manner.

From the above, in addition to taking into account the duration of the syllable speech of the input sentence, the present invention further inserts the pause duration at semantically appropriate locations in a subtle way. Thus, the problem of prior art speech artifacts due to lack of pause due to breathing is overcome, thereby providing a synthesized speech having a very natural and very understandable natural pause duration.

Although the present invention has been described in what is considered to be the most practical and preferred embodiments, the present invention is not limited to the disclosed embodiments, but encompasses all equivalent arrangements of such modifications. It is to be understood that the intent is to apply to various arrangements within the intent and scope of the broadest interpretation.

[0118]

As described above, according to the apparatus and method for processing a pause duration in Chinese speech synthesis according to the present invention, an apparatus having a system that is a sentence input unit and a memory buffer unit is provided. Memory for storing the vocabulary and corresponding information including part-of-speech and extended syntax, a plurality of characters to be paused before, and a pause before storing corresponding information including a previous pause parameter. A table memory of characters to be paused, a table memory of characters to be paused to store a plurality of characters to be paused later, and corresponding information including a post-pause parameter, and one of words and phrases to be stored. A pose parameter storage unit for storing general pose parameters according to the length of a semantic unit including the word, and a Chinese word stored in the dictionary memory. By comparing the plurality of words with the input sentence, the boundary position determining means for determining the position of the boundary of each word in the input sentence input using the input unit, and referring to the dictionary memory A search unit for searching for the part of speech and the extended syntax of each word whose boundary position has been determined, the extended syntax, and the relationship between the parts of speech of adjacent ones of the plurality of words in the sentence. After combining multiple words into multiple phrases and combining multiple words into multiple phrases,
Calculating means for calculating a weighted value of the pose of each of the plurality of semantic units divided from the input sentence so that positions before and after these semantic units are pose position candidates. The calculating means searches for a plurality of previously paused characters and calculates a weighted value for them, and the means for searching for a plurality of subsequently paused characters and calculates a weighted value for them. The apparatus further comprises a pose determining means for determining a plurality of pose positions and a plurality of pause durations by calculating weighted values of the poses of the plurality of pose position candidates. Each of the candidate pose positions having a weighted value of the pose greater than the given value is determined to be a pose position and given a pose level, wherein the pause duration is It is determined according to the pause level.

Therefore, a voice synthesized using the pause duration processing apparatus according to the present invention has a plurality of poses so as to imitate a change in breath, so that the naturalness and meaning of the rhythm of the synthesized voice are significant. The clarity can be improved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a pause duration processing apparatus for converting a Chinese text into a voice according to an embodiment of the present invention.

FIG. 2 is a first diagram illustrating a pause position and pause duration calculation process executed by the pause duration processing apparatus of FIG. 1;
It is a flowchart which shows the part.

FIG. 3 is a second diagram illustrating a pause position and pause duration calculation process performed by the pause duration processing apparatus of FIG. 1;
It is a flowchart which shows the part.

FIG. 4 is a diagram illustrating a pause position and pause duration calculation process performed by the pause duration processing apparatus of FIG. 1;
It is a flowchart which shows the part.

FIG. 5 is a flowchart showing a first part of a pause position and pause duration calculation subroutine process, which is a subroutine of FIG. 4;

6 is a flowchart showing a second part of a pause position and pause duration calculation subroutine process, which is a subroutine of FIG.

FIG. 7 is a signal waveform diagram of a Chinese text that is speech-synthesized by the speech duration processing apparatus for speech synthesis according to the related art, when no pause is inserted.

FIG. 8 is a signal waveform diagram when a pause is inserted in the waveform of FIG. 7;

9 is a diagram illustrating the structure of the dictionary memory 11 of FIG.

FIG. 10 is a table memory 15 of the previous pause character in FIG. 1;
FIG. 3 is a diagram illustrating the configuration of FIG.

11 is a table memory 16 for a post-pause character in FIG. 1;
FIG. 3 is a diagram illustrating the configuration of FIG.

FIG. 12 is a diagram illustrating a configuration of a pause parameter storage unit 17 of FIG.

FIG. 13 is a block diagram of a speech duration processing apparatus for speech synthesis according to the related art.

[Explanation of symbols]

 10: sentence input unit, 11: dictionary memory, 12: word inspection unit, 13: part-of-speech / extended syntax inspection unit, 14: phrase expansion unit, 15: table memory for front pause character, 16: table memory for post pause character 17: Pause parameter storage unit 18: Pause weighted value calculation unit 19: Pause position and pause duration determination unit 20: Register and memory buffer unit 110: Pinyin sentence input unit 111: Phrase inspection 112, Tone / Syllable Inspection Unit, 113, Syllable / Rhythm Matching Unit, 114, Rhythm Inspection Unit, 115, Basic Voice Duration Storage Unit, 116, Basic Voice Duration Determination Unit, 117, Voice Duration Parameter Storage Unit 118: syllable voice duration calculating unit 119: buffer memory area

Claims (2)

[Claims] 1. A pause duration processing method for a system for converting Chinese text to speech in a computer system having an input unit and a memory buffer unit, comprising: a Chinese vocabulary, a part of speech, and an extended syntax. A dictionary memory for storing corresponding information including a plurality of previously paused characters and a table memory of previously paused characters for storing corresponding information including a plurality of previous pause parameters. A table memory of later-paused characters for storing a plurality of later-paused characters and corresponding information including a plurality of later-paused parameters, wherein one of the words and phrases is stored. Configuring a pose parameter storage for storing a plurality of general pose parameters according to the length of the semantic unit including Determining a position of a boundary of each word in the input sentence input using the input unit by comparing a plurality of words of the Chinese vocabulary stored in the dictionary memory with the input sentence; Referring to the dictionary memory and searching for the part of speech and the extended syntax of each word whose boundary has been determined; and Combining the words in the sentence into the phrases, and after combining the words into the phrases, the positions before and after these semantic units are candidate pose positions, Calculating a weighted value of the pose of each of the plurality of semantic units divided from the input sentence, wherein the step of searching for a plurality of previously paused characters comprises Calculating a weighted value for each of the plurality of pose positions, and searching for a plurality of characters to be paused later, and calculating a weighted value for them. Determining a plurality of pose positions and a plurality of pause durations by calculating a pause position, wherein each of the pose position candidates having a pose weighted value greater than a predetermined value is a pause position. Is determined to be
And a pause level, wherein the pause duration is determined according to the pause level. 2. A pause duration processing device for a system for converting Chinese text to speech having a system which is a sentence input unit and a memory buffer unit, comprising a Chinese vocabulary, a part of speech and an extended syntax. A dictionary memory for storing corresponding information including: a plurality of previously paused characters; a table memory of previously paused characters for storing corresponding information including previous pause parameters; A table memory of later-paused characters for storing later-paused characters and corresponding information including later-pause parameters; and general according to the length of semantic units containing either words or phrases. A pose parameter storage unit for storing a pause parameter; and a plurality of words of a Chinese vocabulary stored in the dictionary memory and an input sentence. By doing so, boundary position determining means for determining the position of the boundary of each word in the input sentence input using the input unit, and each word whose boundary position is determined by referring to the dictionary memory Search means for searching for the part of speech and the extended syntax of the sentence, the plurality of words in the sentence are divided into a plurality of phrases according to the extended syntax, Combining means for combining, after combining a plurality of words into a plurality of phrases, a plurality of each of the plurality of words divided from the input sentence so that positions before and after these semantic units are pose position candidates. Calculating means for calculating a weighted value of the pose of the semantic unit, said means for searching for a plurality of previously paused characters and calculating a weighted value for them; Means for searching for characters to be paused after, and calculating weighted values for them. Pose determining means for determining a plurality of pause durations, wherein each of the pose position candidates having a weight value of the pose greater than a predetermined value is determined to be a pose position, and a pause level is given. Wherein the pause duration is determined according to the pause level.