KR102417806B1 - Voice synthesis apparatus which processes spacing on reading for sentences and the operating method thereof - Google Patents

Abstract

Disclosed are a speech synthesizing apparatus for performing spacing processing on sentences and an operating method thereof. According to the present invention, when speech synthesis is performed for a predetermined sentence, based on the average pronunciation required time for each phoneme, the average duration of silence, and the appearance ratio of silence for each word obtained from predetermined training sentences, the A technology for supporting a more natural synthetic voice to be generated by selecting a word predicted to have a silent section among words and inserting a silence section at the end of the selected word.

Description

Speech synthesizing apparatus for space-reading processing for sentences and operating method thereof

The present invention relates to a speech synthesizing apparatus for performing space-reading processing on sentences and to a method of operating the same.

Recently, as text-to-speech (TTS) technology for converting text into speech develops, various speech processing-based services using this technology have been released.

In particular, since the text-to-speech technology is a speech synthesis technology that converts text into speech, it has a very high application value in the field of automatic speech response where information must be delivered only by voice without visual information or in the field of assistive tools for the visually impaired.

In general, speech synthesis technology models a speech waveform from speech data as a parameter, and then synthesizes a speech waveform corresponding to a desired text from the acoustic model. Unit selection speech synthesis, in which small pieces (speech waveform fragments) are selected from the corpus to correspond to the text to be synthesized and to be uttered most naturally It is divided into technology or a combination of these two methods (hybrid speech synthesis).

Either way, in order for the synthesized speech waveform to become natural and clear high-quality speech, the speech units generated from phonological phenomena according to various contexts appearing in human voice must be well modeled as an acoustic model or effectively used as a sound piece within the corpus. should be included In addition, spacing occurs frequently between words and phrases in Korean voices, and through this spacing, the linguistic meaning of the voice is better conveyed, and the naturalness of the voice itself is improved in terms of listening. . Here, spacing in speech is interpreted as a short pause in which the amplitude of the waveform is 0 or small enough to be inaudible in the actual physical meaning.

For this reason, if silence, which means spacing, is appropriately added according to the context in the synthesized speech generated by the speech synthesizer, the meaning can be transmitted more clearly and naturally as if uttered by a human.

Accordingly, there is a need for research on a speech synthesis technology that supports a more natural synthesis of a synthesized voice by adding silence to an appropriate position of the synthesized voice in the process of synthesizing a predetermined sentence into a voice.

According to the present invention, when speech synthesis is performed for a predetermined sentence, based on the average pronunciation required time for each phoneme, the average silence interval time and the appearance ratio of silence for each word obtained from predetermined training sentences, the It is intended to support the generation of a more natural synthetic voice by selecting a word predicted to have a silence section among words and inserting a silence section at the end of the selected word.

In the speech synthesizing apparatus for performing spacing processing on a sentence according to an embodiment of the present invention, when each of a plurality of preset training sentences is pronounced as a voice, each of a plurality of phonemes included in the plurality of training sentences A pronunciation required time information storage unit in which a measurement value of the average pronunciation required to be pronounced is stored; A silence time information storage unit in which a measurement value for When a text-to-speech (TTS) command for the first sentence is applied by the user, the ratio storage unit in which the measured values for the ratio at which the silent section appears immediately after being matched with each other are stored, the first sentence An information extractor for extracting a plurality of first phonemes constituting the first sentence and a plurality of first words constituting the first sentence from the After checking the measurement value of the average pronunciation required time corresponding to each word, for each of the plurality of first words, the measurement values for the average pronunciation time of the phonemes included in each word are added to generate a sum value, Measurement of the average interval time with reference to a designation unit for designating the sum value generated for each of the plurality of first words as a word pronunciation time corresponding to each of the plurality of first words, and the silence time information storage unit a value is checked, and a measurement value of a ratio corresponding to each of the plurality of first words is checked with reference to the ratio storage unit, and the plurality of words are pronounced based on a word pronunciation time corresponding to each of the plurality of first words. After calculating the pronunciation duration between the first words of A word selector that selects at least one second word predicted to have a silent section in the first sentence based on time. When the at least one second word is selected, the at least one second word in the first sentence 2 Generates a synthesized voice for the first sentence based on a silence symbol insertion unit that inserts a preset silence symbol at the end of a word and a preset speech synthesis model, wherein the silence symbol is inserted in the first sentence and a voice synthesizer for generating the synthesized voice including a silence section corresponding to the preset silence time.

In addition, in the method of operating a speech synthesis apparatus for performing spacing processing on a sentence according to an embodiment of the present invention, when each of a plurality of preset training sentences is pronounced by voice, the maintaining a pronunciation duration information storage unit in which a measurement value of an average pronunciation duration required for each of a plurality of phonemes to be pronounced is stored; Maintaining a silence time information storage unit in which a measurement value for an average interval time between non-silent sections is stored, a plurality of words included in the plurality of training sentences, and each of the plurality of training sentences is pronounced by voice When each of the plurality of words is pronounced, the step of maintaining the ratio storage unit in which the measured values for the ratio of the silent section appear in correspondence with each other are stored, the text-to-speech command for the first sentence by the user If approved, extracting a plurality of first phonemes constituting the first sentence and a plurality of first words constituting the first sentence from the first sentence, referring to the pronunciation required time information storage unit After checking the measurement value of the average pronunciation time corresponding to each of the first phonemes of designating the summation value generated for each of the plurality of first words by generating a value as a word pronunciation time corresponding to each of the plurality of first words; the average interval with reference to the silence time information storage unit Check the measured value for time, check the measured value for the ratio corresponding to each of the plurality of first words with reference to the ratio storage unit, and determine the word pronunciation time corresponding to each of the plurality of first words After calculating the pronunciation duration between the plurality of first words based on the average interval time, the ratio corresponding to each of the plurality of first words and the plurality of first words selecting at least one second word predicted to have a silence section in the first sentence based on a pronunciation duration between the words; if the at least one second word is selected, the at least one second word in the first sentence Inserting a preset silence symbol at the end of one second word and generating a synthesized voice for the first sentence based on a preset speech synthesis model, wherein the silence symbol is inserted in the first sentence and generating the synthesized voice including a silence section corresponding to a preset silence time.

According to the present invention, when speech synthesis is performed for a predetermined sentence, based on the average pronunciation required time for each phoneme, the average silence interval time and the appearance ratio of silence for each word obtained from predetermined training sentences, the By selecting a word predicted to have a silence section among words and inserting a silence section at the end of the selected word, it is possible to support the generation of a more natural synthesized voice.

1 is a diagram showing the structure of a speech synthesis apparatus for performing space-reading processing on a sentence according to an embodiment of the present invention.
2 is a flowchart illustrating a method of operating a speech synthesis apparatus for performing space reading processing on a sentence according to an embodiment of the present invention.

Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings. These descriptions are not intended to limit the present invention to specific embodiments, and should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention. While describing each drawing, like reference numerals are used for similar components, and unless otherwise defined, all terms used in this specification, including technical or scientific terms, refer to those of ordinary skill in the art to which the present invention belongs. It has the same meaning as is commonly understood by those who have it.

In this document, when a part "includes" a certain component, it means that other components may be further included, rather than excluding other components, unless otherwise stated. In addition, in various embodiments of the present invention, each of the components, functional blocks or means may be composed of one or more sub-components, and the electrical, electronic, and mechanical functions performed by each component are electronic. A circuit, an integrated circuit, an ASIC (Application Specific Integrated Circuit), etc. may be implemented as various well-known devices or mechanical elements, and may be implemented separately or two or more may be integrated into one.

On the other hand, the blocks in the accompanying block diagram or steps in the flowchart are computer program instructions that are loaded in a processor or memory of equipment capable of data processing, such as a general-purpose computer, a special-purpose computer, a portable notebook computer, and a network computer, and perform specified functions. can be interpreted as meaning Since these computer program instructions may be stored in a memory provided in a computer device or in a memory readable by a computer, the functions described in the blocks of the block diagrams or the steps of the flowcharts are produced as articles of manufacture containing instruction means for performing the same. it might be In addition, each block or each step may represent a module, segment, or portion of code comprising one or more executable instructions for executing the specified logical function(s). It should also be noted that, in some alternative embodiments, it is also possible for the functions recited in blocks or steps to be executed out of the prescribed order. For example, two blocks or steps shown one after another may be performed substantially simultaneously or in the reverse order, and in some cases, some blocks or steps may be omitted.

1 is a diagram showing the structure of a speech synthesis apparatus for performing space-reading processing on a sentence according to an embodiment of the present invention.

1 , the speech synthesis apparatus 110 according to the present invention includes a pronunciation required time information storage unit 111 , a silence time information storage unit 112 , a ratio storage unit 113 , an information extraction unit 114 , It includes a designation unit 115 , a word selection unit 116 , a silence inserting unit 117 , and a voice synthesis unit 118 .

In the pronunciation required time information storage unit 111, when each of a plurality of preset training sentences is pronounced by voice, the average pronunciation duration required for each of the plurality of phonemes included in the plurality of training sentences is pronounced. The measured values are saved.

For example, when it is said that the plurality of training sentences are '1000 sentences', the pronunciation required time information storage unit 111 contains '1000 sentences' when the '1000 sentences' are actually pronounced by voice. A measurement value of an average pronunciation time required to pronounce each of a plurality of phonemes included in .

In relation to this, if there is a sentence 'the weather is very nice today' among '1000 sentences', the phoneme included in the sentence 'the weather is very nice today' is 'ㅗㄴㅡㄹㅡㄴㄹㅆㅣ' It could be 'Ahhhhhhhh', and if the sentence 'The weather is nice today' is actually pronounced as a voice, the time each phoneme is pronounced within the sentence 'The weather is great today' is '0.500 0.615 0.710 0.754 0.925 1.020 1.356 1.427 1.638 1.996 2.520 2.723 2.856 3.001 3.800 3.873 3.984 4.102 4.310 4.423 4.501'[msec]. In this case, the manager may calculate the duration required for each phoneme to be pronounced based on the time at which each phoneme in the sentence 'The weather is very nice today'. For example, the time it takes to pronounce the second phoneme 'ㄴ' is the difference between '0.710 msec', which is the time for 'ㅡ', and '0.615 msec', which is the time for 'ㄴ', is pronounced in the sentence. It can be calculated as '0.098msec'.

In this way, the manager calculates the total pronunciation time required to pronounce each of the plurality of phonemes included in the '1000' sentences and calculates the average of the pronunciation required time for each phoneme, so that the plurality of phonemes are The average pronunciation required time may be measured and stored in the pronunciation required time information storage unit 111 .

Summarizing the matters described so far, the pronunciation required time information storage unit 111 may store information as shown in Table 1 below.

multiple phonemes Average pronunciation time go 0.078msec you 0.092msec do 0.052msec ... ...

In the silence time information storage unit 112 , when each of the plurality of training sentences is pronounced by voice, a measurement value of an average interval time between silence sections in which no voice is present is stored.

For example, as in the above-mentioned example, if there are '1000' training sentences, and among them there is a sentence 'The weather is very nice today', the sentence 'The weather is very nice today' is voiced. When pronounced, there may be a brief silence between the words 'weather', 'true', and 'like'. In this case, the manager can measure the interval time between two silence sections existing in the sentence 'The weather is really nice today', and in this way, the manager can measure the silence section existing in each of the '1000' training sentences. Interval time can be measured. Then, the administrator can store the measured value in the silence time information storage unit 112 after measuring the average interval time, which is the overall average of the interval time between silence sections in each of the 1000 training sentences. .

That is, the average interval time stored in the silence time information storage unit 112 is an overall average value of the interval time between silence sections appearing in the plurality of training sentences when a plurality of training sentences are pronounced by voice.

The ratio storage unit 113 includes a plurality of words included in the plurality of training sentences, and when each of the plurality of training sentences is pronounced by voice, a silence section is stored immediately after each of the plurality of words is pronounced. Measured values for the appeared ratios are stored in correspondence with each other.

For example, as in the above example, if there are '1000' training sentences, and among them there is a sentence 'The weather is very good today', the sentence 'The weather is very nice today' is 'today'. It can be composed of words such as ', 'weather', 'true', and 'good'. There may be a brief silence between the words 'true' and 'like'. Therefore, in the sentence 'the weather is really nice today', a silent section appears after the word 'weather' and a silent section appears after the word 'true'. At this time, the manager said that from the sentence 'The weather is really nice today', the silence section appears '1 time' for the word 'weather' and the silence section appears '1 time' for the word 'true'. can count

In this way, the manager lists up a plurality of words included in the '1000' training sentences from the '1000' training sentences and counts the number of times each word appears in the '1000' training sentences. By counting and counting the number of times a silent section appears after each word in '1000' training sentences, the ratio of the silent section appearing immediately after each of the plurality of words in '1000' training sentences is pronounced can be calculated and stored in the ratio storage unit 113 .

Summarizing the matters described so far, information may be stored in the ratio storage unit 113 as shown in Table 2 below.

multiple words ratio Today is 10% The weather 23% Oh yeah 85% ... ...

In this way, in a situation in which predetermined information measured based on the plurality of training sentences is stored in the pronunciation required time storage unit 111 , the silence time information storage unit 112 , and the ratio storage unit 113 , the user When a text-to-speech (TTS) command for the first sentence is applied by the One phoneme and a plurality of first words constituting the first sentence are extracted.

For example, if the first sentence is a sentence 'the weather is very nice', the information extracting unit 114 may include a plurality of first sentences 'ㄴ' from the first sentence. It is possible to extract phonemes and a plurality of first words such as 'weather', 'true', and 'good'.

The designation unit 115 refers to the pronunciation required time information storage unit 111 to check the measurement value of the average pronunciation required time corresponding to each of the plurality of first phonemes, and then applies the , by adding up the measurement values for the average pronunciation time of phonemes included in each word to generate a sum value, corresponding to the sum value generated for each of the plurality of first words to each of the plurality of first words It is designated as the pronunciation time of the word to be used.

In relation to this, as in the above-mentioned example, the first phonemes of the plurality are 'ㄴㅆㅣahahahahahahaha,' and the first words of the plurality are 'weather', 'true', 'good'. ', the designation unit 115 refers to the pronunciation required time information storage unit 111 to determine the average pronunciation time of each of the plurality of first phonemes, 'ネㅆㅣahahahahahahaha'. After checking the measurement value for 'weather', by adding up the measured values for the average pronunciation time of the phonemes included in the word 'weather', it is possible to designate the pronunciation time of the word corresponding to the word 'weather' and 'true'. By adding up the measurement values for the average pronunciation time of phonemes included in the word word, it is possible to designate the word pronunciation time corresponding to the word ‘true’, and the average pronunciation time of the phonemes included in the word ‘good’ By summing the measured values, it is possible to designate the pronunciation time of the word corresponding to the word 'good'.

The word selection unit 116 refers to the silence time information storage unit 112 to confirm the measurement value of the average interval time, and refers to the ratio storage unit 113 to determine the value corresponding to each of the plurality of first words. A measurement value of the ratio is checked, and a pronunciation duration between the plurality of first words is calculated based on a word pronunciation time corresponding to each of the plurality of first words, and then the average interval time and the plurality of first words are calculated. At least one second word predicted to have a silent section in the first sentence is selected based on a ratio corresponding to each of the first words and a pronunciation duration between the plurality of first words.

At this time, according to an embodiment of the present invention, the word selector 116 repeats steps (a) to (e) as shown in Table 3 below, thereby selecting the words selected one by one through the process (d). The at least one second word that is predicted to have a silence section in the first sentence may be selected.

(a) determining a word having a first position in the first sentence among the plurality of first words as a reference word
(b) checking the remaining words after the reference word in the first sentence among the plurality of first words
(c) designating a word pronunciation time corresponding to the reference word as a pronunciation duration corresponding to the reference word, and for each of the remaining words, from the reference word to each of the remaining words in the first sentence After calculating the cumulative sum of word pronunciation time, the cumulative sum calculated for each of the remaining words is designated as the pronunciation duration corresponding to each of the remaining words
(d) based on whether the pronunciation duration of each of the reference word and the remaining words is close to the average interval time and the magnitude of the ratio corresponding to each of the reference word and the remaining words, the reference word and the Select any one word predicted to have a silent section in the first sentence among the remaining words
(e) processing so that, among the plurality of first words, a word whose position in the first sentence is immediately following the word selected in step (d) is determined as the reference word in step (a)

At this time, according to an embodiment of the present invention, the word selector 116 calculates an operation value according to the operation of Equation 1 below for each of the reference word and the remaining words in the process (d). A word having the smallest calculation value among the reference word and the remaining words may be selected as any one word predicted to have a silent section in the first sentence.

Here, X _i is the operation value for the reference word and the i-th word among the remaining words, I _i is the pronunciation duration corresponding to the i-th word among the reference word and the remaining words, T is the average The interval time, P _i , is a ratio corresponding to the i-th word among the reference word and the remaining words, and a is a preset ratio control constant and means a positive real value.

For example, as in the above example, when the plurality of first words say 'weather', 'true', or 'good', the word selector 116 refers to the silence time information storage unit 112 and Check the measured value for the average interval time, and check the measured value for the ratio corresponding to each of the plurality of first words 'weather', 'true', and 'good' with reference to the ratio storage unit 113 can

And, the word selector 116 uses 'weather', which is the first word in the first sentence among the plurality of first words, 'weather', 'true', and 'good' as a reference word. can decide

After that, the word selector 116 determines that the position in the first sentence of the plurality of first words, 'weather', 'true', and 'good', is after 'weather' as the reference word. You can check 'true' and 'good' with the remaining words.

Then, the word selector 116 designates a word pronunciation time corresponding to the reference word 'weather' as a pronunciation duration corresponding to the reference word 'weather', and in the first sentence, The cumulative sum of the word pronunciation times up to the word 'weather' and 'true', which is one of the remaining words, is calculated, and the cumulative sum is designated as the pronunciation duration corresponding to the word 'true', and the second By calculating the cumulative sum of word pronunciation times from 'weather', which is the reference word, and 'good', which is another one of the remaining words, in one sentence, the calculated cumulative sum is used to pronounce the word corresponding to the word 'good'. It can be specified as a duration. Here, from the first sentence to the standard words 'weather' and 'good', there are three words 'weather', 'true', and 'good', so the word selector 116 is 'good' In order to calculate the pronunciation duration corresponding to the word ', the word pronunciation time corresponding to 'weather', the word pronunciation time corresponding to 'true', and the word pronunciation time corresponding to 'good' can all be accumulated. have.

In this way, when the pronunciation duration corresponding to the reference word 'weather' and the remaining words 'true' and 'good' is designated, the word selector 116 sets the standard word 'weather' and The standard confirmed from the measurement value and the ratio storage unit 113 for the average interval time checked from the silence time information storage unit 112 and the pronunciation duration time designated for each of the remaining words, 'true' and 'good' By applying the measurement value of the ratio corresponding to each of the word 'weather' and the other words 'true' and 'good' to Equation 1 as an input, the reference word 'weather' and the other words It is possible to calculate an operation value corresponding to each of 'true' and 'good'.

Thereafter, the word selector 116 may select a word having the smallest calculated value from among the reference word 'weather' and the remaining words 'true' and 'good'. In relation, the meaning that the operation value is the minimum means that the pronunciation duration specified for the specific word in Equation 1 approaches the average interval time, and the ratio of silence after the specific word in the plurality of training sentences is large. it means That the pronunciation duration specified for the specific word is close to the average interval time means that the duration of pronunciation from 'weather' as the reference word in the first sentence to the specific word appears in the plurality of training sentences It means that it is close to the average value of the silence interval, and that the ratio corresponding to the specific word is large means that the number of times that silence appears after the specific word in the plurality of training sentences is large. Therefore, when the calculated value in Equation 1 is small, it can be predicted that there is a high probability that silence appears after the word having the corresponding calculated value.

If the word with the smallest calculated value among the reference word 'weather' and the remaining words 'true' and 'good' is called 'true', the word selector 116 is the standard word 'Weather is' and the word 'true' among the remaining words 'true' and 'good' may be selected as a word predicted to have a silent section in the first sentence.

After that, the word selector 116 is a word immediately following the word 'true' in the first sentence among the plurality of first words 'weather', 'true', and 'good'. After confirming 'good', the word 'good' is determined again as a reference word, and the process for selecting a word having a silent section described so far can be repeated.

However, the word 'good' is a word located at the last point in the first sentence, and when 'good' is determined as the reference word, the remaining words after 'good' do not exist, so the word choice is The unit 116 may end the process for selecting a word having a silent section.

In this way, according to the method described so far, the word selector 116 selects the word 'true' among the plurality of first words 'weather', 'true', and 'good' in the silence section in the first sentence. may be selected as the at least one second word predicted to have

When the selection of the at least one second word by the word selector 116 is completed, the silence inserting unit 117 inserts a preset silence at the end of the at least one second word in the first sentence. do.

For example, as in the above example, if the first sentence says 'the weather is great', the at least one second word says 'true', and the preset silence sign is 'pause', a silence symbol The insertion unit 117 inserts a silence sign called 'pause' at the point where the at least one second word 'true' ends in the first sentence 'the weather is very nice', It is possible to generate a corrected first sentence saying '.

The speech synthesis unit 118 generates a synthesized speech for the first sentence based on a preset speech synthesis model, but silences corresponding to a preset silence time for a point where the silence symbol is inserted in the first sentence. The synthesized speech including the section is generated.

For example, when the first sentence in which a silence symbol is inserted says 'the weather is very nice to pause' and the preset silence time is '500 msec', the speech synthesis unit 118 responds to 'the weather is very nice' However, the synthesized voice may be generated by including a silence section lasting for '500 msec' between 'weather' and 'true'.

2 is a flowchart illustrating a method of operating a speech synthesis apparatus for performing space reading processing on a sentence according to an embodiment of the present invention.

In step S210, when each of a plurality of preset training sentences is pronounced by voice, a measurement value of the average pronunciation time required to pronounce each of the plurality of phonemes included in the plurality of training sentences is stored, Maintains a pronunciation required time information storage unit.

In step S220, when each of the plurality of training sentences is pronounced by voice, a silence time information storage unit in which a measurement value of an average interval time between silence sections in which a voice is not present is stored is maintained.

In step S230, when a plurality of words included in the plurality of training sentences and each of the plurality of training sentences are pronounced by voice, a silent section appears immediately after each of the plurality of words is pronounced. It maintains a ratio storage unit in which the measured values are stored in correspondence with each other.

In step S240, when a text-to-speech command for the first sentence is applied by the user, from the first sentence, a plurality of first phonemes constituting the first sentence and a plurality of second phonemes constituting the first sentence 1 Extract words.

In step S250, after checking the measurement value of the average pronunciation required time corresponding to each of the plurality of first phonemes with reference to the pronunciation required time information storage unit, for each of the plurality of first words, each word is Word pronunciation time corresponding to each of the plurality of first words by summing the measured values for the average pronunciation required time of the included phonemes to generate a sum value, so that the sum value generated for each of the plurality of first words designate as

In step S260, the measurement value of the average interval time is checked with reference to the silence time information storage unit, and the measurement value of the ratio corresponding to each of the plurality of first words is checked with reference to the ratio storage unit, , after calculating the pronunciation duration between the plurality of first words based on the word pronunciation time corresponding to each of the plurality of first words, the average interval time and the ratio corresponding to each of the plurality of first words and selecting at least one second word predicted to have a silence section in the first sentence based on a pronunciation duration between the plurality of first words.

In step S270, when the at least one second word is selected, a preset silence symbol is inserted at the point where the at least one second word ends in the first sentence.

In step S280, a synthesized speech for the first sentence is generated based on a preset speech synthesis model, but a silence section corresponding to a preset silence time for a point where the silence symbol is inserted in the first sentence is Generate the synthesized speech included.

At this time, according to an embodiment of the present invention, in step S260, by repeating steps (a) to (e) as shown in Table 3 above, the words selected one by one through step (d) are separated from the first sentence. The at least one second word predicted to have a silent section may be selected.

At this time, according to an embodiment of the present invention, in step S260, after calculating the calculation value according to the operation of Equation 1 for each of the reference word and the remaining words in the process (d), the reference word and a word having the smallest calculation value among the remaining words may be selected as any one word predicted to have a silence section in the first sentence.

In the above, an operation method of the speech synthesis apparatus according to an embodiment of the present invention has been described with reference to FIG. 2 . Here, since the operation method of the speech synthesis apparatus according to an embodiment of the present invention may correspond to the configuration of the operation of the speech synthesis apparatus 110 described with reference to FIG. 1 , a more detailed description thereof will be omitted. .

The method of operating a speech synthesis apparatus according to an embodiment of the present invention may be implemented as a computer program stored in a storage medium for execution through combination with a computer.

In addition, the method of operating a speech synthesis apparatus according to an embodiment of the present invention may be implemented in the form of a computer program command for execution through combination with a computer and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the present invention, or may be known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic such as floppy disks. - includes magneto-optical media, and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like.

As described above, in the present invention, specific matters such as specific components, etc., and limited embodiments and drawings have been described, but these are only provided to help a more general understanding of the present invention, and the present invention is not limited to the above embodiments. , various modifications and variations are possible from these descriptions by those of ordinary skill in the art to which the present invention pertains.

Therefore, the spirit of the present invention should not be limited to the described embodiments, and not only the claims described below, but also all of the claims and all equivalents or equivalent modifications to the claims will be said to belong to the scope of the spirit of the present invention. .

110: Speech synthesizing device for space-reading processing for sentences
111: pronunciation duration information storage unit 112: silence time information storage unit
113: ratio storage unit 114: information extraction unit
115: designation unit 116: interval time generation unit
117: word selection unit 118: silence symbol insertion unit
119: speech synthesis unit

Claims (8)

Pronunciation time information in which a measurement value of the average pronunciation time required to pronounce each of the plurality of phonemes included in the plurality of training sentences is stored when each of the plurality of preset training sentences is pronounced by voice storage;
a silence time information storage unit storing a measurement value of an average interval time between silence sections in which no speech is present when each of the plurality of training sentences is pronounced by speech;
A plurality of words included in the plurality of training sentences, and when each of the plurality of training sentences are voiced, a measurement value of the ratio of the silence section appearing immediately after each of the plurality of words is pronounced a ratio storage unit stored in correspondence with each other;
When a text-to-speech (TTS) command for a first sentence is applied by the user, a plurality of first phonemes constituting the first sentence and a plurality of first sentences constituting the first sentence from the first sentence an information extraction unit for extracting first words of
After checking the measurement value of the average pronunciation required time corresponding to each of the plurality of first phonemes with reference to the pronunciation required time information storage unit, for each of the plurality of first words, the average of the phonemes included in each word A designation unit for generating a sum value by adding up the measurement values for the pronunciation required time, and designating the sum value generated for each of the plurality of first words as the word pronunciation time corresponding to each of the plurality of first words ;
Check the measured value of the average interval time with reference to the silence time information storage unit, check the measured value of the ratio corresponding to each of the plurality of first words with reference to the ratio storage unit, After calculating the pronunciation duration between the plurality of first words based on the word pronunciation time corresponding to each of the first words, the average interval time and the ratio corresponding to each of the plurality of first words and the plurality of first words a word selector configured to select at least one second word predicted to have a silent section in the first sentence based on a pronunciation duration between the first words;
a silence symbol inserting unit that, when the at least one second word is selected, inserts a preset silence symbol at a point where the at least one second word ends in the first sentence; and
A synthesized voice for the first sentence is generated based on a preset voice synthesis model, and the synthesized voice includes a silence section corresponding to a preset silence time for a point where the silence symbol is inserted in the first sentence. speech synthesis unit that generates
including,
The word selector
By repeating steps (a) to (e) below, one word selected through step (d) is selected as the at least one second word predicted to have a silent section in the first sentence. A speech synthesis device that performs space-reading processing on sentences with
(a) determining a word having a first position in the first sentence among the plurality of first words as a reference word
(b) checking the remaining words after the reference word in the first sentence among the plurality of first words
(c) designating a word pronunciation time corresponding to the reference word as a pronunciation duration corresponding to the reference word, and for each of the remaining words, from the reference word to each of the remaining words in the first sentence After calculating the cumulative sum of word pronunciation time, the cumulative sum calculated for each of the remaining words is designated as the pronunciation duration corresponding to each of the remaining words
(d) based on whether the pronunciation duration of each of the reference word and the remaining words is close to the average interval time and the magnitude of the ratio corresponding to each of the reference word and the remaining words, the reference word and the Select any one word predicted to have a silent section in the first sentence among the remaining words
(e) processing so that, among the plurality of first words, a word whose position in the first sentence is immediately following the word selected in step (d) is determined as the reference word in step (a) delete According to claim 1,
The word selector
In the step (d), after calculating an operation value according to the operation of Equation 1 below for each of the reference word and the remaining words, the word having the smallest calculated value among the reference word and the remaining words is said A speech synthesis apparatus for performing spacing processing on a sentence, characterized in that the first sentence is selected as any one word predicted to have a silent section.
[Equation 1]

Here, X _i is the operation value for the reference word and the i-th word among the remaining words, I _i is the pronunciation duration corresponding to the i-th word among the reference word and the remaining words, T is the average Interval time, P _i is a ratio corresponding to the i-th word among the reference word and the remaining words, and a is a preset ratio adjustment constant and is a positive real value. Pronunciation time information in which a measurement value of the average pronunciation time required to pronounce each of the plurality of phonemes included in the plurality of training sentences is stored when each of the plurality of preset training sentences is pronounced by voice maintaining a storage unit;
maintaining a silence time information storage unit in which a measurement value of an average interval time between silence sections in which no speech is present is stored when each of the plurality of training sentences is pronounced by speech;
A plurality of words included in the plurality of training sentences, and when each of the plurality of training sentences are voiced, a measurement value of the ratio of the silence section appearing immediately after each of the plurality of words is pronounced maintaining a ratio storage unit stored in correspondence with each other;
When a text-to-speech (TTS) command for a first sentence is applied by the user, a plurality of first phonemes constituting the first sentence and a plurality of first sentences constituting the first sentence from the first sentence extracting first words of
After checking the measurement value of the average pronunciation required time corresponding to each of the plurality of first phonemes with reference to the pronunciation required time information storage unit, for each of the plurality of first words, the average of the phonemes included in each word designating the sum value generated for each of the plurality of first words as word pronunciation time corresponding to each of the plurality of first words by summing the measurement values for the pronunciation required time to generate a sum value;
Check the measured value of the average interval time with reference to the silence time information storage unit, check the measured value of the ratio corresponding to each of the plurality of first words with reference to the ratio storage unit, After calculating the pronunciation duration between the plurality of first words based on the word pronunciation time corresponding to each of the first words, the average interval time and the ratio corresponding to each of the plurality of first words and the plurality of first words selecting at least one second word predicted to have a silent section in the first sentence based on a pronunciation duration between the first words;
inserting a preset silence symbol at a point where the at least one second word ends in the first sentence when the at least one second word is selected; and
A synthesized voice for the first sentence is generated based on a preset voice synthesis model, and the synthesized voice includes a silence section corresponding to a preset silence time for a point where the silence symbol is inserted in the first sentence. steps to create
including,
The step of selecting the at least one second word
By repeating steps (a) to (e) below, one word selected through step (d) is selected as the at least one second word predicted to have a silent section in the first sentence. A method of operating a speech synthesis device that performs space-reading processing on a sentence of .
(a) determining a word having a first position in the first sentence among the plurality of first words as a reference word
(b) checking the remaining words after the reference word in the first sentence among the plurality of first words
(c) designating a word pronunciation time corresponding to the reference word as a pronunciation duration corresponding to the reference word, and for each of the remaining words, from the reference word to each of the remaining words in the first sentence After calculating the cumulative sum of word pronunciation time, the cumulative sum calculated for each of the remaining words is designated as the pronunciation duration corresponding to each of the remaining words
(d) based on whether the pronunciation duration of each of the reference word and the remaining words is close to the average interval time and the magnitude of the ratio corresponding to each of the reference word and the remaining words, the reference word and the Select any one word predicted to have a silent section in the first sentence among the remaining words
(e) processing so that, among the plurality of first words, a word whose position in the first sentence is immediately following the word selected in step (d) is determined as the reference word in step (a) delete 5. The method of claim 4,
The step of selecting the at least one second word
In the step (d), after calculating an operation value according to the operation of Equation 1 below for each of the reference word and the remaining words, the word having the smallest calculated value among the reference word and the remaining words is said A method of operating a speech synthesizing apparatus for performing spacing processing on a sentence, characterized in that the first sentence is selected as any one word predicted to have a silent section.
[Equation 1]

Here, X _i is the operation value for the reference word and the i-th word among the remaining words, I _i is the pronunciation duration corresponding to the i-th word among the reference word and the remaining words, T is the average Interval time, P _i is a ratio corresponding to the i-th word among the reference word and the remaining words, and a is a preset ratio adjustment constant and is a positive real value. A computer-readable recording medium recording a computer program for executing the method of any one of claims 4 or 6 through combination with a computer. A computer program stored in a storage medium for executing the method of any one of claims 4 or 6 through combination with a computer.

Images