RU2113726C1 - Computer equipment for reading of printed text - Google Patents

Abstract

FIELD: computer engineering, in particular, information source for blinds and Russian language education. SUBSTANCE: device has unit 1 for optical input of printed text, optical text recognition unit 2, voice synthesis unit 5 from orthography text, speech signal generator (it is not shown on the drawing) and terminal audio unit 6. Goal of invention is achieved by design of unit 1 as scanner, design of unit 5 as unit which generates Russian speech from orthography text. In addition device has text file unification unit 3, text database unit 4, touch-sensitive display 8, interface 7 between touch-sensitive display and personal computer, interface unit 9. EFFECT: increased quality of Russian speech synthesis. 29 cl, 14 dwg

Description

 The invention relates to computer technology and can be used as a source of information for blind and visually impaired people, and also as a means for teaching the Russian language.

 For people who have lost their sight or with a lack of vision, one of the important problems is the task of independently obtaining information, since such ordinary sources of information as books, magazines, etc. are not available to them, there is no special information at all, and periodic sources of information due to irregular transfers become non-periodic for them.

 A well-known tool for obtaining blind information in the form of printed products using a braille-relief system [1] has certain disadvantages. Such books occupy a large volume due to embossed printing, are easily damaged during storage and when reading due to mechanical contact, in addition, their nomenclature is minimal compared to conventional printed products, not to mention the speed of obtaining information. Finally, the number of blind people who own a Braille system tends to decrease.

 Another well-known means for obtaining information for the blind is the so-called "talking book", which is a magnetic tape on which the texts of books, magazines, newspapers, etc. are recorded [2].

 The advantages of this tool is that when you perceive speech information, a person gets tired less, responds faster, and the speed of information exchange is much higher than with the tactile method. With the undoubted advantages of this information exchange device, it is distinguished by low efficiency, which is why the individual requests of a blind user are poorly satisfied, especially in the field of professional activity, since there are only mass, widely used products and there is no special information, while the latest and latest information.

 Closest to the invention is a device using a personal computer with an optical-mechanical block for reading flat-printed text and a programmable speech synthesizer [3].

 One can note a rather high naturalness and intelligibility of synthesized speech at the level of individual words and a significant deterioration of these indicators in the synthesis of continuous speech, when there are noticeable spurious sound effects. In addition, synthesis is carried out along lines of text, and with this approach it is impossible to adequately form a melodic contour, since it determines the intonation of phrases and syntagmas, and a line of prosaic text contains fragments of one or more syntagmas, i.e. synthesized speech sounds “mechanically”.

 Regardless of the quality, foreign-language speech synthesizers are unsuitable for reading Russian-language texts. When synthesizing speech in Russian, many problems arise that are associated with the complexity of the grammar and phonetics of the Russian language, derived from the Cyrillic alphabet, and are the most difficult in world practice, since they require taking into account a very large number of factors, like in no other language. To date, attempts to create such devices have not solved this problem, since these devices did not meet the requirements for the quality of the perceived signal, and all created options were rejected by potential blind users.

 An object of the invention is the development of a computer device for reading flat-printed text: reading and sound reproduction of printed text in real time with high quality synthesized Russian "Slavic" speech by ensuring its naturalness and intelligibility of the phoneme, syllable and verbal.

 The problem is solved in a device for reading flat-typed text, containing a series-connected optical input unit for flat-printed text and an optical text recognition unit, a speech synthesis unit for spelling text, an audio signal generating unit and a terminal unit, in which, according to the invention, an optical input unit for flat-printed text is made in the form scanner, the audio signal generating unit is made in the form of a sound card, the speech synthesis unit for spelling text is made in the form of a synthesizer in Russian speech in spelling text, as well as additionally introduced a text file unification unit, a text database unit, a tactile display, a tactile display interface to a personal computer and an interface unit, the output of the optical text recognition unit through a series-connected text file unification unit and the text database block is connected to the information input of the speech synthesis block according to the spelling text, and through the pairing block, to the input of the tactile display, the output of the sync block A spelling of Russian speech in spelling text is connected through a sound card to a terminal audio block, control inputs of the interface units of a tactile display and scanner, an optical text recognition unit, a text file unification unit, a block for synthesizing Russian speech from spelling text, and a text database block are connected to the corresponding outputs of the block interface.

 At the same time, the unit of unification of the text file can be made in the form of series-connected recognition blocks for multi-column text, blocks for reformatting a text file into one columns, a block for recognizing left and right borders of a text, a block for recognizing red lines, a block for selecting paragraphs, a block for excluding metalanguage characters that are not included in valid alphabet, word wrap block, and text file reformat block.

 The block for synthesizing Russian speech from a spelling text can be made in the form of a block for compiling speech synthesis and containing a series-connected block for selecting the current paragraph, a block for reading the current paragraph, a block for preliminary parsing of the paragraph, a block for selecting words in a paragraph, a subword selector on subsets of the metalanguage alphabet, a block recognition of words in the Russian alphabet, the first block OR, block contextual decryption of abbreviations of words, block matching, block selection of phrases and syntagm, block definitions of communication vnogo type phrases and syntagmas determining unit logical stress, phonetic transcription unit of text, the time processor melodic processor compilation unit, forming unit, and an audio file output unit of the sound file.

 The block of preliminary parsing of the paragraph may contain sequentially connected punctuation mark selector, a switch, a punctuation mark verbalization block and a second OR block, while the second output of the switch through sequentially connected punctuation mark context analysis block, a punctuation mark elimination block for spoken words, a elimination block graphic punctuation fictions and the block of verbalization of the signs of a point, comma and dash in the record of numbers is connected to the second input of the second block OR, and the control input of the switch is the control input of the preliminary parsing unit.

 The block for recognizing words in the Russian alphabet contains a series-connected block for converting symbol codes to one register, a selector for words including punctuation characters, a first word-abbreviation selector and a third OR block, a second output of the first word-abbreviation selector through a fourth OR block connected in series, a morphological block analysis, grapheme recovery unit "yo", abbreviation selector and abbreviation decryption unit connected to the second input of the third OR block, the second output of the abbreviation selector Without a serially connected unit for determining a part of speech and an accentuation unit is connected to the third input of the third OR block, the second output of the word selector including punctuation marks, through the second selector of abbreviations is connected to the fourth input of the third OR unit, the second output of the second selector of abbreviations the hyphen exclusion block is connected to the second input of the fourth OR block, the third output of the word selector including punctuation characters, through the apostrophe exclusion block is connected to the third input of the fourth OR unit, a fourth input coupled to a fourth output selector words containing punctuation marks, the third block is the output of OR output word recognition unit to the Russian alphabet.

 The morphological analysis block contains memory blocks for prefixes, foundations, and inflections, implemented in the form of a corresponding database, while the morphological analysis block is implemented that implements an algorithm for solving the corresponding equation.

 The stress arrangement unit contains a successively connected selector of recognized words, a stressed syllable definition unit, and a fifth OR block.

 The translation unit of metalanguage words into words in the Russian alphabet contains a sequentially connected selector of subsets of the alphabet of the metalanguage, the sixth block OR, the verbalization block of the Arabic digital notation of numbers and the seventh block OR, the second output of the selector of the subsets of the alphabet of the metalanguage through the sequentially connected selector of Latin digits and the conversion unit of Latin numbers in Arabic is connected to the second input of the sixth block OR, the second output of the block for converting Latin numbers into Arabic through series-connected the eighth OR block and the Russification block of Latin alphabet words is connected to the second input of the seventh OR block, the second output of the Latin digit selector is connected to the second input of the eighth OR block, the third input of which is connected to the third output of the metalanguage subset selector, the fourth output of which is through the verbalization block of alphabet characters metalanguage is connected to the third input of the seventh OR block, the output of which is the output of the translation unit of metalanguage words into words of the Russian alphabet.

 The phase and syntagma selection block is implemented that implements the phrase extraction algorithm by breaking the paragraph into lexemes separated by {.?} And the syntax highlighting algorithm in phrases by breaking the phrase into lexemes separated by {,:; -}.

 The communicative type definition block for phrases and syntagms contains a sequentially connected question selector phraser, the syntagma definition block containing a question, the ninth OR block and the communicative type determination syntagm block, the second output of the interrogative phrase selector is connected to the second input of the ninth OR block.

 The unit for determining the communicative type of syntagma is implemented that implements an algorithm for determining the following communicative types: completeness corresponding to syntagms ending with signs {.:;}, First and second types of incompleteness corresponding to syntagms ending with respectively signs {, -}, general and particular questions for syntagms, ending with a {?} and correspondingly not containing or containing a question word, as well as accent syntagmas containing a logical highlight sign on the highlighted word.

 The logical stress determination unit contains a sequentially connected selector of logically distinguished words and a tenth OR block, a second output of a selector of logically distinguished words through a series-connected word selector containing a particular question, and the eleventh OR block is connected to the second input of the tenth OR block, the second output of a word selector containing a particular question, through the selector of semantically significant words connected to the second input of the eleventh block OR, and the second output of the selector of semantically significant words through the block EFINITIONS last notional word syntagm is connected to the third input of the OR block eleventh, tenth block output is the output of OR logical stress determination unit.

 The phonetic transcription block of the text contains a series-connected block for eliminating spelling fictions, a block for converting letter recording into phonetic, and a unit for forming continuous speech.

 The spelling fiction elimination unit comprises a spelling fiction elimination unit at the word ends, an unpronounceable consonant elimination unit, a consonant change replacement unit with an equivalent alphabetic entry, and a word recording unit with a hard pronunciation “e”.

 The block for converting an alphabetic entry into a phonetic one contains a serially connected selector of service words, a unit for eliminating stresses in service words, a unit for assimilating service words, a twelfth OR block, a block for transcribing contexts with a soft sign, a block for transcribing contexts with a hard sign, a block for determining the softness of consonants in front of marked vowels , block of transcription of iotated vowels at the beginning of the syntagma, block of transcription of iotated vowels in the position after the vowel, block of transcription unpaired soft consonants, a unit for determining voiced-deafness of consonants in combinations of consonants, a unit for stunning noisy voiced consonants at the end of a syntagm, a unit for assimilating hard consonants to soft ones, a unit for transcribing double consonants, a selector for weakly stressed words, a unit for determining the degree of reduction of vowels in accented words, unit for replacing vowels of the first degree of reduction, the thirteenth block OR and unit of replacing vowels of the second degree of reduction, second output of the service word selector through the unit for replacing stresses in weakly stressed words vah is connected to the second input of the twelfth OR block and the third output - to the third input unit of the twelfth or the second output selector slaboudarnyh words through the block determining the degree of reduction in slaboudarnyh words is coupled to a second input of the thirteenth OR block.

 The unit for the formation of continuous speech is made implementing the corresponding algorithm.

 The unit for determining the degree of reduction of vowels is implemented by implementing an algorithm in which the first degree of reduction or the second degree of emphasis corresponds to vowels in the positions of the first pre-stressed syllable, in unstressed positions at the end of the syntagma, in the open end of the word, if the next word begins with the stressed syllable and the position after the vowel , and the second degree of reduction, or the first degree of highlighting - to the vowel in all other unstressed positions, the third degree of highlighting corresponds to the vowels in the stressed positions except the last stressed a syllable of a syntagma, the vowels of which corresponds to the fourth power of selection.

 The unit for replacing vowels of the first degree of reduction is implemented that implements an algorithm for the contextual replacement of reduced unstressed vowels of the first degree of reduction by appropriate substitutions.

 The unit for determining the degree of reduction of vowels in weakly stressed words is implemented that implements the algorithm for reducing a pseudo-stressed vowel in weakly stressed words to the first degree of reduction.

 The unit for replacing vowels of the second degree of reduction is implemented that implements an algorithm for contextually replacing reduced unstressed vowels of the second degree of reduction by appropriate substitutions.

 The time processor is implemented implementing an algorithm that sets each symbolic element of the phonetic record in accordance with a time interval that determines the duration of the corresponding sound of the compilation element, while the vowels are divided into three classes, reflecting differences in their phonetic quality.

 The melodic processor contains sequentially connected syntagma segmentation unit for syllabic fragments, a class unit for determining a syllable fragment and a unit for determining the parameters of the law of changing the frequency of the fundamental tone for syllable segments.

 The syntagma segmentation block for syllabic fragments is implemented implementing an iterative algorithm for solving the corresponding equations.

 The unit for determining the class of a syllable fragment is implemented that implements the corresponding algorithm, in which syllable fragments are divided into classes depending on the communicative type of syntagma, the degree of emphasis of the vowel syllable, and on the coincidence of the syllable with the melodic center of the syntagma, which coincides with the logical stress of the syntagma, while for completeness and two types nine classes of syllables are distinguished: incomplete syllables, stressed syllables with a vowel degree of three equal to those that do not coincide with the melodic center, after a stressed syllable that does not coincide with the melodic center, the melodic center for syntagmas with the type of completeness respectively with the degree of vowel highlighting n = 4 and n = 3, the melodic center for syntagmas with the first and second types of incompleteness and the degree of vowel highlighting respectively n = 3 and n = 4, for syntagmas with the communicative type “general question” four classes of syllables are distinguished, respectively unstressed and stressed syllables are not melodic centers with n = 3, the last stressed syllable is not a melodic center, melodic center with degree E saliency vowel respectively n = 3 and n = 4, and for syntagmas communicative type "special issue" isolated syllables two classes - coincide and not coincide with the center of the melody.

 The unit for determining the parameters of the law of changing the frequency of the fundamental tone for segments of syllables is made implementing the corresponding algorithm.

 The compilation unit contains a sequentially connected context extraction unit, a compiler element code encoder, an acoustic segment synthesis unit, a time and frequency response unit for compilation elements, and a composition unit.

 The block of the acoustic segment synthesis base is made in the form of a database block, the acoustic segment synthesis base contains the basic compilation elements in the form of digitized segments of the natural phonemic speech wave — allophones, which are acoustically and perceptually distinguishable contextual implementations of phonemes.

 The compiler element code encoder is implemented implementing the compilation element code generation algorithm.

 The unit for changing the time and frequency characteristics of the compilation element is implemented implementing the algorithm of the corresponding functional transformation.

 The implementation of a computer device for reading flat-printed text in the form of a series-connected optical input unit for flat-printed text, made in the form of a scanner, and an optical text recognition unit, a compilation unit for synthesizing speech from spelling text, an audio signal generating unit in the form of a sound card, and an end signal, as well as containing a text database unit, a tactile display, a tactile display interface unit with a personal computer, and an interface unit and associated connections between them allows reading and sound reproduction with high quality synthesized speech of the text made in any printed font in Russian.

 The proposed implementation of a unit for unifying a text file, a unit for synthesizing Russian speech from a spelling text, a unit for preliminary parsing of a paragraph, a unit for recognizing words in the Russian alphabet, a unit for morphological analysis, an unit for placing stress, a unit for translating metalanguage words into words in the Russian alphabet, a unit for selecting phrases and syntagm, block for determining the communicative type of phrases and syntagm, block for determining the communicative type of syntagm, block for determining logical stress, phonetic block text scribing, spelling removal unit, letter to phonetic conversion unit, unit speech formation unit, vowel reduction degree determination unit, vowel replacement degree determination unit, vowel reduction degree determination unit in weakly pronounced words, vowel replacement unit of the second reduction degree, temporary processor, melodic processor, syntagma segmentation block for syllabic fragments, syllabic fragment class definition block, acoustic-comp block compilation unit synthesis-element basis, the encoder element compilation code and the changes block of time and frequency characteristics compilation element allows to increase the quality of sound reproduction Russian synthesized speech by providing respectively its phoneme, syllable verbal and intelligibility and naturalness it.

 In FIG. 1-14 are structural electrical diagrams of the following devices and blocks: in FIG. 1 - a computer device for reading flat-printed text; in FIG. 2 - block unification of a text file; in FIG. 3 - block synthesis of Russian speech according to the spelling text; in FIG. 4 - block preliminary parsing of the paragraph; in FIG. 5 - block recognition of words in the Russian alphabet; in FIG. 6 - stress placement unit; in FIG. 7 - block translation of metalanguage words into words in the Russian alphabet; in FIG. 8 - block definition of a communicative type of phrases and syntagmas; in FIG. 9 - logical stress determination unit; in FIG. 10 - block phonetic transcription of the text; in FIG. 11 - block elimination of spelling fiction; in FIG. 12 - block conversion of letter recording into phonetic; in FIG. 13 - block melodic processor; in FIG. 14 - compilation unit.

 The device (Fig. 1) contains a block 1 for optical input of flat-printed text, a block 2 for optical text recognition, a block 3 for unifying a text file, a block 4 for a text database, and a block 5 for synthesizing Russian speech from spelling text, a terminal audio block 6, a pairing unit 7, tactile display 8 and unit 9 of the interface.

 Block 3 unification of the text file (figure 2) contains a block 10 recognition of multicolumn text, block 11 reformatting the text file into a single column, block 12 recognition of left and right borders of the text, block 13 recognition of red lines, block 14 paragraph selection, block 15 exclude characters metalanguage not included in the valid alphabet, block 16 eliminating word wraps and block 17 reformatting the text file.

 Block 5 of the synthesis of Russian speech according to the orthographic text (Fig. 3) includes a block 18 for selecting the current paragraph, a block 19 for reading the current paragraph, a block 20 for preliminary parsing of the paragraph, a block 21 for highlighting words in the paragraph, a selector 22 for subwords on subsets of the metalanguage alphabet, block 23 recognition of words in the Russian alphabet, block 24 translation of words metalanguage into words in the Russian alphabet, first block OR 25, block 26 of the contextual decoding of abbreviations of words, block 27 matching, block 28 of the selection of phrases and syntagm, block 29 of determining communicative ty and phrases and syntagmas determination logic block 30 syntagmas stress in block 31 the phonetic transcription of the text, the time processor 32, processor 33 melodic, the compilation unit 34, the unit 35 forming the sound file and the block 36 output sound file.

 Block 20 of the preliminary parsing of the paragraph (figure 4) contains a selector 37 punctuation marks, a switch 38, a block 39 of verbalization of punctuation marks, a second block OR 40, a block 41 of the contextual analysis of punctuation marks, a block 42 of eliminating punctuation fictions, a selector 43 of punctuation marks in the record numbers and block 44 verbalization of punctuation marks in the record numbers.

 Block 23 recognition of words in the Russian alphabet (figure 5) includes a block 45 for converting capital grapheme codes, a selector 46 of words including punctuation marks, a first word selector 47, a third OR block 48, a fourth OR block 49, a morphological analysis block 50, a block 51 grapheme recovery “yo”, abbreviation selector 52, abbreviation translation unit 53, part of speech determination unit 54, stress accentuation unit 55, second abbreviation selector 56, hyphen hyphenation unit 57 and apostrophe exclusion unit 58.

 Block 55 accent (6) contains a selector 59 recognized words, block 60 determine the stressed syllable, the fifth block OR 61 and block 62 determine the stressed syllable according to empirical rules.

 Block 24 of the translation of metalanguage words into words in the Russian alphabet (Fig. 7) includes a selector 63 of the subsets of the metalanguage alphabet, a sixth block OR 64, a block 65 for verbalization of the Arabic numeric recording of numbers, a seventh block OR 66, a selector 67 of Latin digits, a block for converting Latin letters 68 numbers in Arabic, the eighth block OR 69, block 70 of the Russification of words of the Latin alphabet and block 71 of the verbalization of metalic alphabet characters.

 The unit 29 for determining the communicative type of phrases and syntagmas (Fig. 8) contains a selector 72 for interrogative phrases, the unit 73 for determining the syntagma containing the question, the ninth block OR 74 and the unit 75 for determining the communicative type of syntagm.

 The logical stress determination unit 30 (Fig. 9) includes a selector 76 of logically allocated words, a tenth block OR 77, a selector 78 of words containing a particular question, a selector 79 of semantically significant words, a block 80 for determining the last significant word of the syntagma, an eleventh block OR 81, and a block 82 logical allocation.

 Block 31 phonetic transcription of text (figure 10) includes a block 83 to eliminate spelling functions, block 84 converting the letter recording into phonetic and block 85 forming a continuous speech.

 Block 83 of the elimination of spelling fiction (11) includes a block 86 of the elimination of spelling fiction at the ends of words, block 87 of the elimination of unpronounceable consonants, block 88 replacement of combinations of consonants with equivalent letter writing and block 89 converting the recording of words with a hard pronunciation of "e".

 Block 84 converting the letter recording into phonetic (Fig. 12) includes a selector 90 service words, a unit 91 for eliminating stresses in the service words, a unit 92 for assimilating service words, a twelfth block OR 93, a unit 94 for replacing stresses in low-impact words, a unit 95 for transcribing contexts with soft sign, hard sign context transcribing unit 96, consonant softness determining unit 97 for marked vowels, iotated vowel transcribing unit 98 at the beginning of the syntagma, iotated vowel transcribing unit 99 in pos after the vowel, block 100 transcribing unpaired soft consonants, block 101 for determining the voiced-deafness of consonants in consonant combinations, block 102 for stunning ringer consonants at the end of the syntagma, block 103 for assimilating hard consonants to soft ones, block 104 for transcribing hard consonants, soft selector 105, weak selector 105 block 106 for determining the degree of reduction of vowels in words with accent, block 107 for replacing vowels of the first degree of reduction, block 108 for determining the degree of reduction of vowels in low-impact words, twelfth block OR 109 and block 110 s vowel amenas of the second degree of reduction.

 Block 33 of the melodic processor (Fig. 13) contains a block 111 for segmenting the syntagma into syllable fragments, a block 112 for determining the class of the syllable fragment and a block 113 for determining the parameters of the law of changing the frequency of the fundamental tone for segments of syllables.

 The compilation unit 34 (FIG. 14) includes a context extraction unit 114, a compilation element code encoder 115, an acoustic segment synthesis unit 116, a time and frequency response unit 117 of the compilation elements, and a composition unit 118.

 The device operates as follows.

 According to the start signal from the interface unit 9 to the optical input unit 1 of the flat-printed text, the unit 1 starts inputting graphical information made in any printed font. As an optical input unit for flat-printed text, commercially available hand-held or flatbed scanners, for example, Hewlett Packard scanners, which are more automated and therefore more convenient for a blind user, can be used. To enter book text using a flatbed scanner, preliminary book flashing is necessary; all other input operations are carried out automatically. A hand-held scanner requires pre-adjustment by the sighted operator (mainly brightness), has a narrower capture band and is more sensitive to skew.

 At the output of block 1, an image of the input text appears in one of the graphic formats (usually in TIFF format). By the control signal from block 9, the optical text recognition unit 2 starts recognition of graphic symbols of the alphabet for converting the image of the text into a text file. A graphic file, as you know, is a matrix of an image stored in operational or long-term memory by resolution elements - pixels. For portability of images, as well as for their compression in order to save memory, various standard graphic file formats are used - PCX, GIF, TIFF, etc. A text file, as you know, is a familiarity matrix, where each familiarity corresponds to a code of a certain alphabet symbol (space, letter, punctuation mark, various special characters).

 Of the developed optical recognition systems, CuneiForm by Cognitive Technologies Ltd. is the most efficient. and FineReader, development of Dialog-MEPhI.

 Both systems are characterized by high recognition efficiency (no more than 1-3 errors per 1 page for typographic text, text printed on a laser or matrix 24-needle printers, the first copy of typewritten text (CuneiForm). In addition, it provides recognition of mixed texts (Cyrillic and Latin alphabet), and FineReader also effectively recognizes low-quality texts (for example, photocopies, texts printed on a 9-needle matrix printer). Interface unit 9 provides the integration of these optical systems spoznavaniya claimed in device-specific user blind.

 From the output of block 2, the text file is transferred to block 3 of unification of the text file. This block brings the resulting text file in accordance with the capabilities of the speech synthesizer, which are more limited than the capabilities of a living subject reading flat-printed text.

 In addition to the characters of the alphabet of the synthesized language, the text may include various metalanguage characters and non-textual inserts. Each text has a certain graphic structure (headings, paragraphing, etc.), fragmenting the text into fragments that are finished in a meaningful sense, and the techniques for such fragmentation are quite diverse. The text can have more or less standard typographic design: alignment of lines to the left and right, highlighting of paragraphs with a red line, lack of non-text inserts. However, if there are non-textual inserts, this structure is violated: several left or right borders may appear. Typewritten text typically has fuzzy alignment on the right border. Sometimes paragraphs are not highlighted with a red line, etc. The graphic structure of the text may turn out to be violated at the output of optical recognition unit 2, if the typographic text is typed in a non-width font, alignment along the right border of the text may be violated (lines will turn out to be of unequal length), the beginning of lines may be shifted , the text in general may turn out to consist of several columns, due to recognition errors in the text file, characters that are not included in the valid alphabet may appear.

 The optical recognition unit allows you to select columns of text, however, this is done manually in interactive mode, which is unacceptable to a potential user. Block 10 automatically checks for more than one column in the text. A sign of multi-column text is the presence of spaces in the same line positions. We will consider each line as a vector, and transform the vector lines according to the following rule: the i-th component corresponding to the i-th position in the line is 0 if there is a space in this position, and 1 if a character other than a space in this position . The vector summation of the obtained row vectors and the comparison of the components of the resulting vector with the threshold allows us to select the connected areas corresponding to the text columns in the line; the multi-column text file in block 11 is reformatted to single-column.

In order to highlight the headings and paragraphs of the text, you must first check the alignment of the text on the left and right borders. Recognition of the left and right borders of the text is carried out in block 12. Let l _i be the position in the i-th line corresponding to the first character other than the space character, and r _{i the} position corresponding to the last character other than the space and the line feed character. Further, let L be the set of values of l _i and R the set of values of r _i for a given text.

 On the sets L and R, the corresponding frequency distributions of the values of the left and right boundaries of the lines in the text are determined, as well as the distribution of the right borders of the lines ending with a hyphen and the right boundaries of the lines not ending with the characters of the end of the phrase.

 The right borders of the text are recognized by their correlation with the boundaries of lines ending with hyphens, or, if there are no word hyphens in the text, with the boundaries of "incomplete" lines (lines that do not end with punctuation marks at the end of phrases). For fuzzy right borders, their statistical characteristics (mathematical expectation and variance) are determined. Left borders are defined on a subset of the lines following the above.

 Indentation of red lines is determined on a subset of lines whose beginning does not coincide with the selected left borders, by their correlation with the beginning of phrases and with the “incompleteness” of the previous line (that is, the punctuation mark of the end of the phrase is at the end of the line, and the end of the line does not reach the right border of the text or deviation from the fuzzy border exceeds the tolerance interval). From the found values of the indentation of the red lines and the left borders, the starting positions for the red lines are determined.

 Red lines (if they are in the text) are recognized in block 13 by the corresponding position of the beginning of the line and provided that the beginning of the line corresponds to the beginning of the phrase (to exclude accidental matches).

 In block 14, paragraphs are highlighted. Typically, in the text, the beginning of a paragraph is highlighted in red, but not always. If no red lines are found in this text, then the end of the paragraph is determined by the "incompleteness" of the line.

 In block 15, the presence of invalid characters in the text is checked, detected invalid characters are replaced by spaces.

 In block 16, hyphenation characters in words are eliminated (to reduce the ambiguity of the "-" character), and block 17 reformatts the text file in accordance with the accepted standard. The text in one column corresponds to this standard with paragraphs highlighted in red lines and with a fuzzy right border. If the next word does not fit on the current line, it is transferred to the next, while the right border is not aligned due to spaces.

 Based on the usual everyday experience, it can be said that a person is not characteristic of a single linear reading of textual information from beginning to end. People usually come back to what they have read before for a more adequate understanding of certain provisions, taking into account what is stated below or simply to refresh some of what they have read before. Naturally, entering and recognizing once already read text each time is a waste of time. A text file takes up a memory space that is 2–3 orders of magnitude smaller than a graphic or sound file, so it makes sense to store the text file once received in long-term memory. 250 megabytes of memory on your hard drive allows you to store up to 100 thousand pages of text (or about 100 pages of voiced text in the form of an audio file). In order to be able to quickly find the desired text file, stored text files must be organized into a certain database (DB), which is already becoming a subject of collective use. The specific construction of the database will be described below.

 Depending on the control signal at the input of block 4, the output of the latter receives the current text file, or a file selected using the database interface. This file from the output of block 4 goes to the input of synthesis block 5 and through the pairing block 7 to the input of the tactile display 8. Synthesis block 5 generates a sound file in one of the audio formats based on the text file, the latter is converted by the sound output system of the Sound Blaster type into analog sound frequency signal. From the linear output, this signal is fed to the terminal audio unit 6, where it is converted into the form required by the user. Passive or active (with volume and timbre control) acoustic systems that convert an analog signal into an acoustic wave (speech stream) and / or an analog tape recorder can be used as a terminal audio block. In the latter case, a magnetic recording of the speech signal is carried out in parallel with the audio output. This magnetic record can later be used as an ordinary “talking book”. The synthesized “talking book” is a by-product and naturally inferior in quality to the “real talking book”, which is a bit of a “theater at the microphone” show. However, it can be quickly obtained at a lower cost (it does not require a qualified announcer and recording studio, the rental price of which is quite high). In addition, the synthesized and the usual "talking book" have different purposes. The usual “talking book” mainly satisfies the aesthetic needs of the user, and the synthesized one only informative.

 The tactile display duplicates the speech output, allowing you to read words that are incomprehensible or simply poorly perceived by hearing, in addition, some specific graphic means of the text, such as tables, generally do not lend themselves well to adequate linear verbalization.

 Synthesis block 5 implements a compilation method for synthesizing speech using spelling text, while segments of the phoneme dimension — allophones — are selected as the basic compilation elements.

 Since in alphabetical writing the basic elements of written speech are letters, which in spoken language correspond to elements of the phoneme dimension, the basic compilation elements naturally choose the same dimension. In this case, the speech synthesizer, regarded as a kind of intellectual solver of a certain task, turns out to be transparent, or articulating. The latter means that such a solver allows not only to obtain the final solution, but also to trace the entire path of its receipt in the form familiar to the given subject area. In this case, this means that translating a written speech fragment into the alphabet of basic compilation elements leads to the problem of phonetic transcription of text familiar to a given subject area, and phonetic values structured and formalized using artificial intelligence technologies can form the basis of the synthesizer's knowledge base.

 In Russian, 10 vowels and 37 consonant phonemes can be distinguished. By themselves, phonemes do not exhaust the whole variety of sounds of Russian speech.

 The articulation of each sound, both vowel and consonant, consists of three phases - the initial, when the active organ of speech of their initial position moves towards the corresponding passive (excursion), middle, or shutter speed, when the active organ of speech is in relation to the passive organ in the position necessary for the production of a given sound, and final, when the active organ of speech returns to its original position (recursion). In the speech stream, the articulations of different sounds seem to overlap: the recursion of this sound in time coincides with the excursion of the next, and the excursion with the recursion of the previous one.

 Therefore, in a continuous speech stream, the acoustic realization of a particular phoneme will depend on the contextual environment. These contextual implementations of phonemes - allophones - are chosen as the basic compilation elements. The total volume of the acoustic-segment synthesis base with this approach is about 100 thousand, which requires enormous labor costs for compiling such a dictionary. The necessary set can be reduced by summarizing the identical contextual influences. With a successful generalization, the problem of adequate matching of the basic compilation elements in speech synthesis is practically solved. The search for possible generalizations and thereby the optimal set of allophones can be solved only taking into account the knowledge of acoustic reflexes of co-articulation processes. This approach can be considered based on phonetic knowledge in the sense that is accepted in research on artificial intelligence.

 For the formation of the allophone base, a special dictionary is selected, consisting of words containing the necessary allophone in the required contexts. The words of this dictionary are pronounced by the announcer, recorded and entered into the machine in digital form. Then, using the software package (for example, Tool Kit), the acoustic segments corresponding to the given allophones are extracted, and an acoustic-segment synthesis base is formed from these segments.

 The resulting acoustic-segment synthesis base contains 687 basic compilation elements, mainly representing segments of a phonemic speech wave, although in some cases this correspondence is violated. For the synthesis of closures and vibrants, more than one acoustic segment is used, and for the synthesis of two-character sequences, for example, shock inflections, one acoustic segment is used.

Vowels are represented by six phonemes in a strong position (in the stressed syllable): {a ^, o ^, y ^, and ^, s ^, э ^} and eight unstressed, including:
- the first degree of reduction {a, y2, u, s, o},
- second degree of reduction {b, b, y1}.

The following are highlighted as left contexts:
- anterior lingual solid {d, t, s, s, z, q, dz, w, x, a ^, a, b, e ^, e},
- labial hard {b, n, c, f, l, y ^, y1, y2, o ^, o},
- anterior lingual nasal {n},
- labial nasal {m},
- vibrant solid {p},
- soft non-nasal: all consonants placed with a soft sign ('), except {m', n '}, as well as {u ^, u, b, s ^, s},
- soft nasal anterior lingual {n '},
- soft nasal labial {m '},
- beginning of syntagma (initial).

The following context classes are highlighted as right:
- front-lingual solid {d, t, s, s, z, q, dz, w, x, n, k, r, x, a ^, a, b, e ^, e, s ^, s} except for the consonant {k , r, x} before {y ^, y1, y2, o ^, o},
- labial firm {b, n, c, φ,,, m, y ^, y1, y2, o ^, o}, as well as {k, r, x}, if {y ^, y1, y2, o ^ o},
- vibrant solid {p},
- soft, i.e. all consonants marked with a soft sign, as well as {u ^, u, b},
- end of syntagma (final).

Consonants are divided into 7 classes:
1) voiced and deaf closing (hard and soft) {b, d, d, p, t, k, b ', d', g ', p', t ', k'},
2) solid fricatives (except [x]) {c, s, w, w, q, f, dz},
3) soft fricatives (except [x ']) {c', z'shch, ch'f ', zh, j'}
4) nasal {m, n, m ', n'},
5) smooth sonants {l, h ^, l ', d'} and {c, b ', x, x'},
6) vibrants {p, p '},
7) iot {j}.

For classes 1) and 2), only the right contexts are allocated, 4 classes in total:
- {y ^, y, o ^},
- final
- for soft sounds, any context except the final one,
- all other contexts.

For class 3), 2 classes of left contexts are distinguished:
- initial
- other; and 2 classes of right contexts:
- final
- any, except the final one, for soft sounds.

 For nasal (class 4) contexts are not allocated, i.e. the implementation of these sounds is independent of the contextual environment.

For class 5), 4 classes of left contexts are allocated:
- {y ^, y, o ^},
- initial
- {u ^, s ^, u, s, b, e ^},
- other,
and 4 classes of right contexts:
- {y ^, y, o ^},
- final
- any other than the final one for soft sounds,
- other.

For vibrant (class 6), left, right, and connected contexts are distinguished, where connected contexts are understood as cases of the interdependent existence of left and right contexts. For this class, this is a condition for the simultaneous presence of a vowel on the left and on the right, i.e. intervocal position. For the remaining cases, 4 classes of left contexts are distinguished:
- {y ^, y, o ^, b, n, φ, c, l, m},
- initial
- {and ^, s ^, and, s, b, e ^} and all soft consonants,
- all other contexts;
and 4 classes of right contexts:
- {y ^, y, o ^, b, n, φ, c, l, m},
- final
- any, except the final, for [p '],
- all other contexts.

 When forming allophone for vibrant, always, except in cases of final position, allophone for the intervocal position is used, combined with contextually determined allophones, the latter are always glued from the consonant or the beginning. In the case of the environment of two consonants, two identical contextually determined allophones are glued together and an allophone for the intervocal position is inserted between them.

For class 7), 4 classes of left contexts are selected that coincide with classes of left contexts for vibrants, and 3 classes of right contexts:
- final
- vowels,
- consonants.

 Block synthesis 5 operates as follows. Sounding of a text file is carried out cyclically - at the end of sounding of the next fragment, sounding of the next begins, etc. The interface provides the ability to return to the read fragment and re-reading it in the selected mode. This technique seems to be justified, since the text file can be quite large and there is no point in waiting until it is completely processed from beginning to end. A fragment of the text is processed, converted into a sound file, and while this sound file is output through the sound input-output system, the next fragment is processed. At the same time, the reaction time is reduced (it is equal to the processing time of one fragment, and not the entire text) and the volume of the output sound file is reduced.

 The smallest unambiguously allocated fragment, syntactically independent of neighboring fragments, is a paragraph. In the proposed synthesizer, reading is carried out in paragraphs; for this, in block 3, paragraphs were recognized and selected.

 In block 18, the current paragraph is selected for reading. A sound text file is delivered to the information input, and the address of the paragraph from which the reading begins begins at the first control input from the interface unit. Subsequently, the address of the current paragraph goes to the second control input from the audio file output block 35 (next paragraph, return to the previous one, repeated output, etc.). Block 19 reads the current paragraph (translation from long-term memory into operational memory, and block 20 performs preliminary syntactic analysis of the paragraph (preliminary processing of punctuation marks).

 Block 21 highlights in the paragraph the words - tokens, separated by a space or a line break character (word breaks are eliminated in block 3). The sequence of codes corresponding to the Selected paragraph word stream arrives at selector 22. Selector 22 separates words consisting of Russian alphabet codes and metalanguage codes (letters of the Latin alphabet, numbers, special characters). If the word consists of codes of letters of the Russian alphabet, then it is processed in block 23 for recognizing words in the Russian alphabet, where on the basis of morphological analysis the words of the Russian language are recognized, part of speech is determined and the stress is automatically placed in words, otherwise the word is processed in block 34 where it is translated into words in the Russian alphabet. From the output of the first block OR 25, combining the results of these two branches of processing, the stream of paragraph word codes is input to block 36, where contextual decryption of word reductions not recognized in block 23 is performed. Rules for contextual decoding of abbreviations of words are presented in the corresponding paragraphs of the synthesizer knowledge base.

 As a result, the ambiguity of using a point is eliminated in block 26 — only the end of the sentence is determined at the output of the current block. In block 27, the verbal equivalents of the digital record obtained in block 24 and the decoding of abbreviations by gender, number and case based on the analysis of the context are coordinated.

 In block 28, phrases and syntagmas are highlighted in the paragraph as a sequence of tokens separated by punctuation marks. By analyzing punctuation marks and words of phrases and sitagms in block 29, the communicative type of the latter is determined, and in block 30 - logical stress. Based on information about accents in words and structured phonetic knowledge of the rules of the transition "letter-phoneme" in words with known accent in block 31, automatic phonetic transcription of the text of the paragraph is carried out. The time processor 32 and the melodic processor 33 form parameters for controlling the prosody of the synthesized speech: with the help of the time processor 32 the required durations of sound elements are determined, and with the help of the melodic processor 33 - melodic contour (approximation of the law of variation of the frequency of the fundamental tone).

 In the compilation unit 34, the actual assembly of the speech message is carried out: the phonetic record is converted into a record in the alphabetical language of the compilation elements, the necessary compilation elements in the form of digitized segments of the natural speech wave are extracted from the database, they are modified in accordance with the parameters defined in blocks 32 and 33 , and assembling a voice message by composing modified fragments. The file at the output of block 34 is a digitized audio signal.

 In block 35, this file is converted to a format compatible with the sound input-output system used (for example, wav-, vjc-, or snd-format). In block 36, the received sound file is transmitted for output through the sound card. Based on the analysis of interruptions from the keyboard, a control signal is generated at the second output of the block to go to the next paragraph, to re-display the current in the same mode, to re-form the current in another mode, or to return to the previous paragraph.

 Punctuation is characterized by the fact that it has a very small amount of funds. Two features of punctuation marks are connected in this: the breadth of their meaning and polysemy. To reduce this ambiguity in block 20 is a preliminary parsing of the paragraph. The block works as follows.

 The selector 37 selects punctuation character codes in the paragraph, which, depending on the control signal supplied to the switch 38, are processed either in block 39 or in blocks 41,42,43,44.

 In block 39, punctuation marks are verbalized. The user selects this mode if he wants to have a complete picture of the punctuation in the source text of the paragraph, including those that do not receive intonation in oral speech. The voice message in this case resembles a dictation for a typist.

 In another mode, the paragraph syntax is reflected only in intonation.

 To reduce the aforementioned ambiguity of punctuation marks in block 41, these characters are analyzed in context. Dash and hyphen characters are separated (the dash has spaces on the left and right). The brackets are separated as a punctuation mark (the presence of opening and closing brackets, as in this fragment) and the bracket as a graphic tool. The apostrophe is divided into the function of quotation marks (the presence of a space before or after the apostrophe) and the apostrophe used in the spelling of words ("entrance", "a'nature"). The point and the ellipsis, the point as a punctuation mark and the decimal point or comma in the decimal fraction record) are separated. In block 42, punctuation fictitious, i.e. punctuation means of graphic speech that are not reflected in oral speech. A purely graphic trick is that after the closing parenthesis a punctuation mark is placed, which should end part of the sentence before the parentheses. In block 42, this character is placed before the opening bracket, and the brackets themselves are replaced by periods if the closing bracket is followed by a terminating phrase, or by commas if there is a space. The point is replaced by an exclamation mark and an ellipsis if it is followed by a new paragraph or a word with a capital letter (otherwise, a space). Replaced by spaces are dashes before the quotation marks at the beginning of the paragraph (a purely graphic technique denoting a dialogue). Quotes are replaced by spaces. If there are several punctuation marks in a row (as a result of the above transformations or errors of the optical recognition system), weaker characters are absorbed by stronger ones.

In block 43, punctuation marks are selected based on the digital context. If the context is digital (first exit), punctuation marks are verbalized in block 44, for example:
5.6 - "5, point 6,
5.6 - "5, comma 6",
fax machine. 276-43-12 - "Fax. 246, Dash 43, Dash 12".

 Block 23 word recognition in Russian works as follows. In block 45 codes are analyzed codes of the characters that make up this word. Uppercase codes are converted to lowercase codes, and information about the presence of capitalization codes in a word is stored. Further, depending on the presence of a punctuation mark in the word and the type of this mark, the address of the word is transmitted for further processing. If the word ends with a period (the first output of the selector 46), processing is then carried out in the first selector 47 of the abbreviations. The latter checks for the presence of a given word in the database (in the "list") of commonly used word abbreviations. If the word is found there, it is not processed in block 23 and enters its output from the first output of the first selector 47 through the third block OR 48.

 If the word is not an abbreviation6, i.e., the dot is the end of sentence, then it is treated as an ordinary word of the Russian language without any punctuation marks (for the latter, the fourth output of selector 46 is the fourth block OR 49). This processing is carried out in block 50 morphological analysis. Morphological analysis is carried out by cutting off inflections and prefixes and highlighting the basis of the lexical meaning of the word.

Morphological analysis is equivalent to solving the equation
W = X _p • X ⁽¹⁾ • ... X ^(n-1) • X ⁽ⁿ⁾ ,
where W is the analyzed word,
X _p is the prefix or empty word,
X ⁽¹⁾ , ..., X ^(n-1) - one or more stems (possibly with a connecting vowel at the end),
X ⁽ⁿ⁾ - inflection or empty words.

 The equation is solved as follows. In each cycle, m symbols are cut off to the right, and the left part of the word is considered as a potential basis and its presence is checked in the basis database. If this part of the word is not found in the database, m increases by one (initial value m = 0), and the cycle repeats. If no corresponding basis is found in any of the cycles, an attempt is made to select one of the list prefixes at the beginning of the word, if this succeeds, the prefix is cut off and the process is repeated for the rest of the word. If the prefix cannot be selected, the word is considered unrecognized.

 If the basis is found in the database, then many inflections for the given basis are selected from the database, and the part of the word separated on the right is compared with them. If it coincides with one of the inflections, the word is recognized, for this word form information is extracted from the speech part, grammatical form and stress (in Russian in polybasic words the stress is always determined by the last basis). This information is associated with this word form. If none of the inflections coincides with the right part of the word, and one of them, including the inflection of zero length, is its initial occurrence, the corresponding word on the left, including the connecting vowel, if it is detected is cut off, and the process is repeated for the remaining part on the right. Thus, words without endings, flexibly mutable words, words formed using prefixes, polybasic words are recognized.

 In the Russian letter quite often not all letters of the Russian alphabet are used, in particular, grapheme "e" is often used instead of grapheme "yo", and an apostrophe is sometimes used instead of a solid dividing sign. The letter "yo" is missing in many text editors, including the OSR CuneiForm editor. For the correct phonetic transcription, automatic restoration, where required, of the grapheme "yo" is necessary. The database contains both spellings of the words with "yo", and information about the presence of the word "yo" together with other grammatical information is sent to the output of block 50. According to this information, in block 51, the desired letter "e" is replaced by "yo".

In block 52, unrecognized words associated with capital letters are perceived as abbreviations. In block 53, the abbreviations are transcribed according to the following rules:
- two-letter abbreviations are read "by letter" (composition of the names of the letters) with emphasis on the last vowel (RF - er ^ f, KA - kaa ^);
- three-letter and four-letter abbreviations are read “like a word” in the absence of a combination of two or more consonants with an accent on the first consonant vowel, which is covered in front, in other cases - “spell” with an accent on the last syllable. For example, VOS - vo ^ c, EMP - emi ^, UFO - enelo ^, NATO - on ^ that, OPEC - ope ^ k, ODMO - odeemo ^;
- all the others are read “like a word” if in the abbreviation there are no more than two consonants in a row, or “spell” otherwise, the emphasis falls on the last vowel covered on the right by the consonant (ASALM - asa ^ lm, UNESCO - unesco, OBXSS - obehaese ^ s).

 If the word is not taken as an abbreviation, then in block 54, based on the grammatical information associated with the word, the code for the part of speech is determined and in block 55 the stressed syllable is indicated.

Codes of words containing a hyphen character from the second output of the selector 46 go to the input of the second selector 56. The latter, similarly to the first selector 47, recognizes abbreviations written through a hyphen (for example, gr-citizen ^ n, instit-institute ^ t ) These abbreviations in block 23 are not processed. If the word is not an abbreviation, then in block 57 the hyphen is excluded, and the following rules are used:
- in complex words formed with the help of part of the stem - “half-”, in complex adjectives (dark blue) and in words with particles something, either, something, something, so, so, two parts of the word are glued together, and if the part of the word after the hyphen begins with an iotated vowel, then the hyphen is replaced by a solid sign (dark blue - deep blue, half apple - half apple);
- in other cases, the hyphen is replaced by a space.

 Next, the word goes to the fourth block OR 49 and is processed according to general rules.

 In block 58, the apostrophe sign is excluded, after which the word is processed according to general rules. The latter is found in words of foreign origin (a'la, a'naturel) specified by the list, in proper names of Irish, French and sometimes Spanish origin after "o" and "d" (O'Higgins, d'Artagnan, d'Alameda) . In these cases, the parts of the words separated by the apostrophe are glued together. In Russian words, an apostrophe is sometimes written instead of a dividing solid sign after prefixes in front of soft vowels, in these cases the solid sign is restored.

 Unit 56 accents works as follows.

 Based on the information associated with the given word, the recognized and unrecognized words are separated in the selector 59.

 For recognized words, the stressed syllable is denoted, and for unrecognized words, an attempt is made to determine the stress using some rules of thumb. So, for example, in Russian, masculine nouns, in the nominative case, ending in "ism", in all word forms have an accent on this syllable (empirio-criticism ^ Зм, behaviorior ^ Зм, etc.), masculine nouns with a second base "logs" have an accent on the connective "o" (filo ^ logo, dento ^ log).

 Block 24 translation of metalanguage words into words in the Russian alphabet works as follows. The selector 63 separates the words consisting of various characters of various subsets of the metalanguage alphabet.

 If the word consists of Arabic numbers, then the processing is carried out according to the chain: the first output of the selector 63, the sixth block OR 64, block 65. In the last, the digital record of the number is replaced by its verbal equivalent. At the second output of the selector 63, words consisting of characters of the Latin alphabet are highlighted, of which at the first output of the selector 67 words are selected consisting of characters of Latin digits - {C, L, X, V, I}. In block 68, numbers recorded in Latin digits are recognized. Since these symbols denote not only Latin numbers, but can also be found in the words of languages with Latin script, in block 68 an attempt is made to convert the sequence of these symbols into an Arabic digital notation. If the combination of these characters does not satisfy the rules for writing numbers in Latin digits, then the sequence of characters is considered as an ordinary word in the Latin alphabet, transmitted to the second output of block 68 and then processed in block 70, as well as words written in Latin letters, but not containing characters that Latin numbers are indicated (second output of block 67).

 If the combination of characters satisfies the rules for writing Latin digits, the number is converted in block 68 to the Arabic notation and through the sixth block OR 64 goes to block 65, where it is verbalized.

 In block 70, words written in Latin letters are translated into the Russian alphabet. Since the language is unknown, translation is carried out according to the relevant rules for Latin words. It turns out a little funny, but understandable (windows - windows, "made in USA" - "made in usa").

 All other characters of the metalanguage {%, /, +, \} are verbalized in block 71 (% - "percent", / - "fraction", + - "plus", \ - "slash", etc.).

 The melodic outline of a phrase depends on its communicative type. The following types of sentences are distinguished in the Russian language: narrative, interrogative, incentive, and exclamation. Narrative sentences are most common in speech, often significant passages of works consist of sentences of this type only. Exclamation sentences have an emotional tone. The many forms of the contours of the fundamental tone are very diverse, and a significant part of the rules for controlling intonation are not only not formalized, but even unknown. Since the main purpose of the proposed complex is to transmit information, the number of possible forms of voice communication is more limited than with voice communication between people. The main communicative type of sentence in the transmission of information is the narrative type, which, as noted above, is generally the main type in writing. Therefore, the synthesizer does not reproduce intonation features, reflecting secondary shades of a predominantly expressive nature, and the many communicative types of sentences come down to two: narrative and interrogative, punctuation marks clearly reflected in graphic speech.

Accordingly, six communicative types of syntagmas are distinguished:
- completeness (with possible logical stress) for syntagms limited by the signs {.:;},
- the first type of incompleteness for syntagms limited by a comma,
- the second type of incompleteness for syntagms limited by a dash,
- a general question for syntagmas limited by a question mark and not containing a question word,
- a particular question for syntagmas that are limited by a question mark and contain a question word, which falls under logical stress,
- with logical highlighting for syntagmas containing a logical highlighting sign.

 A logical highlight mark may be present in a text file if manual markup of the text has been previously performed, as is done, for example, in fiction books.

 Block 29 determining the communicative type of phrases and syntagm works as follows. In selector 72, interrogative phrases are highlighted. If the phrase ends with a [?], Then control is transferred to block 73, where interrogative words are searched for in the phrase. If such a word is found, at the end of the syntagma containing it, a question mark with a sign of a special question is put, the word is marked, and at the end of the phrase, if this syntagma is not the last, a dot is put. The ninth OR block 74 combines both processing branches. In block 75, the communicative type of syntagma is determined by the punctuation marks separating the syntagmas and the presence of a question word or a logical highlight in the syntagma.

 The logical stress determination unit 30 operates as follows.

 Logical stress, as already noted, does not find formal reflection in graphic speech, but is determined by the semantics of the text. Well-known methods of machine semantic analysis are very far from perfect and work only in an interactive mode, since the concept of meaning is intuitive and poorly formalized. Therefore, the synthesizer uses simplified rules for determining logical stress.

 The selector 76 selects words with a logical highlight in the syntagma. If such a word is found, then no additional actions are performed in the block; if it is not found, then the interrogator words marked in block 73 are highlighted in selector 78. If such a word is found, in block 82, the logical stress sign is assigned to this word instead of a simple accent mark. If a question word is not found, in block 79, cases where semantically significant words are determined by context without a complete semantic analysis of the sentence are checked. Such, for example, are the cases when comparative particles are used: “as if”, “as if”, “as if”, “for sure”, “not so much”, or the amplifying particle “this”, as well as “means” in combination with the infinitive. The significant word in the phrase following it logically stands out.

 If such special cases are not found, in block 80 the last significant word of the syntagma is determined, to which the logical stress falls.

 Block 31 of phonetic transcription of the text functionally solves the following three tasks: eliminating spelling fiction (block 83), converting the letter recording to phonetic (block 84) and forming continuous speech (block 85).

 Translation is carried out by the substitution operation Sb (W, X, Y), where the occurrence of X in the word W is replaced by the subword Y.

 In block 86 of the elimination of spelling fictions, word substitutions are implemented in accordance with the rules of clauses 2.3 of the knowledge base: spelling fonts in the endings of the genitive case of the singular masculine and neuter for the "-th" and "-th" and a soft sign at the end words after hissing in adverbs, feminine nouns and the infinitive.

 At block 87, unpronounceable consonants are excluded.

 In block 88, transcription of cases where the phoneme on the letter is indicated by a combination of consonants is carried out.

 In block 89, in words with strong pronunciation of consonants before "e" (mainly in words of foreign origin), "e" is replaced by "e".

To convert alphabetic character codes into phoneme character codes (block 84), we divide the set of consonants S into the following disjoint classes:
S ₁ = {n, b, t, d, k, r, c, f, s, s, p, m, n, l, x} - paired in hardness and softness;
S ₂ = {C, G, W} - unpaired solid;
S ₃ = {h, uh} - unpaired soft;
S ₄ = {st}.

We divide vowels into two disjoint classes:
G ₁ = {a, o, y, e, s} - "solid",
G ₂ = {i, e, u, e, and} - "soft."

The following subsets are distinguished on the set of phonemes:
f ₁ - solid consonants;
f ₂ - soft consonants;
φ ₃ = {a, o, y, e, s, and} are vowels;
f ₄ = {t, s, n, f, k, w, x, t, t ', c', n ', f', k ', h', u '} - noisy deaf;
f ₅ = {d, s, b, c, d, f, h, z, d ', z', b ', c', r ', j', f '} - noisy voiced;
f ₆ = {t, d, s, s, n} - solid dental;
f ₇ = {t ', d', s', s', n '} - soft tooth;
Φ ₈ = {k, r, x} - back-lingual solid;
f ₉ = {k ', r', x '} - posterior soft
and set the correspondence by deafness-voicing by displaying
φ ₁ : f ₄ -> f ₅ ,
and the correspondence in terms of softness-hardness - by mappings
φ ₂ : f ₆ -> f ₇ , φ ₃ : f ₈ -> f ₉ .

 Block 84 converting the letter recording into phonetic works as follows.

 The selector 90 distinguishes service and weak-impact words defined by "lists". Service words in block 91 lose emphasis and in block 92 are assimilated with the significant word. In this case, auxiliary proclitic words (prepositions) are “stuck” to the next word (if this word begins with a “soft” vowel, then “b” is put instead of a space). Service words-enclitics (particles) "stick" to the previous word. The word resulting from the composition, through the eleventh block OR 93 enters block 95 and then transcribed according to general rules.

 At a block 94, the stress mark in low impact words is replaced by the mark “low stress”. All other words without changes through the eleventh block OR 93 go to block 95 for processing.

' - знак мягкости,
* - знак, разделяющий синтагмы.Contexts containing the soft sign code in block 95 are transcribed using the sequence of operations of context-specific substitutions:
- Sb (W, s * *, s'), s ∈ S ₁ ;
- Sb (W, s ₁ s ₂ , s ₁ 's ₂ ), s ₁ ∈ S ₁ ∪ S ₃ , s ₂ ∈ S ₁ ;
- Sb (W, s ₂ bs, s ₂ s), s ₂ ∈ S ₂ , s ∈ S;
- Sb (W, ssg, s'y '(g)), s ∈ S ₁ ∪ S ₃ , g ∈ G;
- Sb (W, sьg, sй '(g)), s ∈ S ₂ , g ∈ G ₂ ;
where the map α: G ₂ -> ф ₃ is given by the correspondence table

'- a sign of softness,
* - a sign separating syntagmas.

Contexts containing the solid sign code in block 91 are transcribed by substitution
- Sb (W, s'g, s'y 'α (g)), s ∈ S ₁ \ {d, t, s, s, h}, g∈G ₂ .

а отображения β и ζ заданы соответственно таблицами

Транскрибирование гласных в начале синтагмы осуществляется в блоке 97 путем подстановки
- Sb(W, *g, й′α(g)), g ∈ G₂\{и}.Transcription of consonant-vowel combinations (SG) is carried out in block 96 by substitutions

and the mappings β and ζ are given by the tables, respectively

Transcription of vowels at the beginning of syntagma is performed in block 97 by substitution
- Sb (W, * g, ′α (g)), g ∈ G ₂ \ {and}.

Vowel-Vowel (GG) collisions are transcribed in block 98 by substitution
- Sb (W, gg ₂ , gйα (g ₂ )), g ∈ G, g ₂ ∈ G ₂ \ {and}.

Soft consonants {h, uh} in the absence of a soft sign are transcribed in block 99 by substitution
-Sb (W, s, s'), s ∈ S ₃ .

При сочетании более двух согласных два последних перехода осуществляются последовательно в несколько шагов.Block 100 takes into account the tendency characteristic of modern Russian language for voicing noisy deaf consonants in front of noisy voiced and stunning noisy voiced consonants in front of noisy deaf people. This is implemented by substitutions.

With a combination of more than two consonants, the last two transitions are carried out sequentially in several steps.

.In Russian, in the place of voiced consonants at the end of a word (syntagma), the corresponding deaf are pronounced. This is taken into account in block 101 by appropriate substitutions:

.

In block 102, double consonants in the hard-soft combination are transcribed by substitution
-Sb (W, ss ′, s′s ′), s ∈ Φ ₁ .

.Block 103 takes into account the tendency to soften dental hard to dental soft and unpaired soft and to soften posterior hard to posterior soft Corresponding substitutions:

.

In block 104, double consonants are transcribed by substitutions:
Sb (W, ss, s :), s ∈ Φ ₁ ;
Sb (W, s's', s "), s′∈ Φ ₂ .

In Russian, vowel phonemes highlighted in a subset of f ₃ can be pronounced only in stressed syllable. In unstressed syllables in the Russian literary language, vowels are pronounced with varying degrees of reduction, i.e. abbreviations are pronounced less clearly, and most importantly, in unstressed syllables, some vowels do not differ, coinciding with each other.

 Vowels are reduced in different ways in ordinary and weakly stressed words. These words are separated in the selector 105 according to the results of the operation of block 94.

 The stressed vowel is the most powerful in duration and intensity (from the stressed vowels, the vowels in the last stressed syllable of the syntagma stand out as the most powerful among stressed vowels).

 The first degree of reduction is one step lower in duration and intensity and corresponds to the vowel position in the first pre-stressed syllable, unstressed positions at the end of the syntagma, at the open end of the word, if the next word begins with the stressed syllable, and the position after the vowel.

 The second stage of reduction is even lower in duration and intensity and corresponds to all other unstressed positions.

 These rules for determining the degree of reduction are implemented in block 106.

 In weakly stressed words in block 108, a pseudo-vowel is reduced to the first degree of reduction, and all other vowels, if any, are reduced as a general rule.

 The indistinguishability of some vowels in unstressed syllables is taken into account in block 107, where the reduced vowels of the first degree of reduction are replaced in accordance with the following rules.

 In the positions corresponding to the first degree of reduction, [a] and [o] are pronounced as the corresponding reduced [and] in the position after the soft consonant, in all other positions as pronounced [a].

 The vowel [e] of the first degree of reduction is replaced by [s] or [and] in the positions respectively after hard and soft consonants.

 In weakly stressed words, a pseudo-shock vowel reduced to the first degree of reduction preserves quality.

 In block 110, the following rules are implemented for replacing reduced vowels of the second degree of reduction.

 In the positions corresponding to the second degree of reduction, the vowels [o, a, e] do not differ and are replaced by [b] in the position after the hard consonant and by [b] in the position after the soft consonant. In Russian, [s] can only be in position after a solid consonant, and [and] after a soft one. Replacement - according to the same rule.

 The pronunciation of sounds at the word boundaries is different depending on whether pauses are maintained between the words or the words form a continuous speech stream. Transcription of word joints for continuous speech is carried out in block 85 in accordance with the pronunciation standards of the Russian literary language. The pronunciation of concatenated voiced-deaf consonants at the end-beginning of a word in continuous speech is similar to the pronunciation of such conjunctions at the junctions of the morphological parts of the word, "and" at the beginning of a word after a solid consonant at the end of the previous word is replaced by "s".

 In the block of the temporary processor 32, the required compilation element durations are determined. The temporary processor 32 implements the following algorithm.

.The duration of the vowels is determined by the degree of reduction and the phonetic quality of the vowel. We divide the many vowel sounds into three classes, reflecting the differences in their phonetic quality:

.

For any vowel in a phonetic record of text, we define the degree of highlighting n. The first, lowest degree of emphasis (n = 1) corresponds to the vowels of the second degree of reduction, the second (n = 2) - to the vowels of the first degree of reduction, the third (n = 3) - to the stressed vowel with the exception of the last stressed word in the syntagma, this case corresponds to fourth degree of isolation (n = 4). Then for ∀g∈Φ _i , i∈ {1,2,3} the duration of the vowel, depending on the degree of highlighting, is determined by the relation
τ _i (n) = τ _{i, 0} + (n-1 + [n / 4]) / 2
where [] is the integer part of the number, and τ _{i, 0 /} is a constant fixed for the ith class.

For stressed vowels in real speech, there is a dependence of the duration on the sequence number in the syntagm of the word containing them (q) and the number of syllables (m) in this word. For vowels with a degree of isolation 3, this dependence is small and can be neglected, for vowels with a degree of isolation 4, this dependence is significant and approximated by the relation
τ _i ′ (4) = τ _{i, 1} + 0.74 (τ _i (4) -τ _{i, 1} ) (m-1) (q-1),
where τ _{i, 1} is the duration fixed for the ith class.

где δ - фиксированная для всех гласных величина,
κ - длительность инерционного хвоста.At the end of the syntagma, vowels are extended regardless of their degree of reduction and phonetic quality. The effect of consonants on the duration of vowels is taken into account only in the most striking cases, namely, in the position in front of the intervocal vibrants. In the position in front of the intervocal vibrants and at the end of the syntagma, the duration of any vowel is determined by additional ratios, respectively

where δ is a fixed value for all vowels,
κ is the duration of the inertial tail.

 When determining the temporal characteristics of consonants, the following factors were taken into account, the positions of the consonant with respect to the boundaries of the syntagma and the phonetic word, the intervocal-non-intervocal position, the position in the cluster (consonant clusters), the simple-complex composition of the basic compilation elements necessary for sound synthesis of consonants.

The consonants have the maximum duration in the intervocal position (in the position between the vowels). This duration is taken as a reference. The dependence of the duration of the consonants on the segmented environment is determined by the ratio
τ = β _i τ ₀ , i∈ {1,2,3,4}
where τ ₀ is the reference duration,
i = 1 for the position in front of the vowel (β ₁ = 0.8),
i = 2 for the position before the consonant (β ₂ = 0.6),
i = 3 for compilation elements corresponding to the explosion phase of the consonant consonants and vibrants (β ₃ = 1,0),
i = 4 for long consonants (β ₄ = 1.3).

The melodic processor 33 determines the law of a temporary change in the frequency of the fundamental tone within the syntagma. This law is approximated by a piecewise linear dependence, and for each compilation element the values of the fundamental frequency are determined for as its initial and final melodic characteristics. The values of these characteristics are quantized on a scale of musical tones and are formed from left to right by syllabic cycles, i.e. within the sequence s _n g, where g is a vowel, s _n , (n≥0) is a sequence of n consonants preceding a vowel.

The syntagma is segmented into syllable fragments in block 111. This segmentation is equivalent to an iterative solution of the equation
С ⁱ = S ⁱ g ^{i, 1} C ^{i + 1} , ∀g∈Ψ _g with initial conditions С ⁰ = С,
where g ^{i, 1} is the first occurrence of any vowel symbol in the phonetic notation g∈Ψ _g in the word С ⁱ ,
C is a syntagma considered as words in the alphabet Ψ = Ψ _s ∪Ψ _g ,,
Ψ _g = Ψ ₁ ∪ Ψ ₂ ∪ Ψ ₃ is a subset of vowels,
Ψ _s - a subset of consonants, the i-th syllable fragment refers to the composition S ⁱ g ^{i, 1} .

 When determining the melodic characteristics of compilation elements included in the current syllable, the following factors are taken into account: the communicative type of syntagma, the position of the syllable relative to the melodic center of the contour (main syllable syllable) - coincidence, left, right, position of the syllable relative to the boundaries of the syntagma, the degree to which the vowel is highlighted in the current syllable, the degree to which the vowel is highlighted in the previous syllable, the number of symbolic elements in the syllable, the type of symbolic element (vowel-consonant) and its position relative to the beginning of the syllable (ne first-time).

It is accepted that the fundamental frequency variations are one octave relative to the base level F _b .

For communicative types of syntagma - completeness and two types of incompleteness - seven classes of syllables are distinguished :,
K ₁ - unstressed syllables (n≤2),
K ₂ - simple shock layers (n = 3),
K ₃ - the last stressed syllable of the syntagma, which is not a melodic center (n = 4),
K ₄ - the last stressed syllable of the syntagma, which is the melodic center (n = 4) for a syntagma with a type of completeness,
K ₅ - melodic center that does not coincide with the last stressed syllable (n = 3) for a syntagma with a type of completeness,
K ₆ is the melodic center for syntagmas with the first type of incompleteness (n = 4),
K ₇ is the melodic center for syntagmas with the first type of incompleteness (n = 3),
K ₈ is the melodic center for syntagmas with the second type of incompleteness (n = 4),
K ₉ is the melodic center for syntagmas with the second type of incompleteness (n = 3).

 For each class of syllables, a set of rules is specified by which the initial and final melodic characteristics of the symbolic elements of the syllable are determined.

For class K _1, 5 rules are set:
1) F _n = F _to = F _b for all elements of the syllable.

 2) Decrease by half a tone for the first element of the syllable, then ΔF = 0.

3) The frequency of the fundamental tone decreases uniformly (in a grayscale scale), F _k for the last syllable element is lower than F _{n of the} first syllable element by half an octave if the current syllable is the first of the syntagma stressed syllables, and by 3 half tones if it is not the first one.

4) For consonants F _n = F _k = F _{to the} previous vowel, for the vowel F _n = F _{to the} previous vowel, F _k three semitones below F _b .

5) Same as 4), but for the vowel F _k = F _b .

For class K _2, 5 rules are set:
1) The frequency of the fundamental tone increases uniformly (in a grayscale scale), F _k for the last syllable element is higher than F _{n the} first syllable element by half an octave.

2) Similar to rule 1) for K ₁ .

3) The frequency of the fundamental tone increases uniformly (in the grayscale scale), F _k for the last syllable element is higher than F _{n the} first syllable element by half an octave if the current syllable is the first of the syntagma stressed syllables, and by 3 half tones if it is not the first one.

4) Similar to rule 2) for K ₁ .

5) Similarly to rule 5) for K ₁ .

For class K _3, 2 rules are set:
1) Similar to rule 4) for K ₁ , but for vowels F _k half an octave below F _b .

2) Similar to rule 1) for K ₁ .

For class K _4, 2 rules are set:
1) For consonants F _n = F _k = F _b ,
For a vowel, F _n = F _b , F _k half an octave below F _b .

2) The frequency of the fundamental tone is uniformly (in a grayscale scale) reduced, and for the last element of the syllable F _k , half an octave below F _b .

For class K _5, 2 rules are set:
1) For consonants
for the first F _n = F _k and half an octave above F _b ,
further F _n = F _to
For a vowel, F _n is half an octave higher than F _b , F _k = F _b .

2) For consonants F _n = F _k = F _{k of the} preceding vowel,
for the vowel F _n = F _{to the} preceding element, F _to = F _b .

For the classes K _h , h ≥ 6, 2 rules are given:
1) For the first element of the syllable F _n = F _b , then the frequency of the fundamental tone evenly (in the grayscale scale) increases, and for the last element of the syllable F _k a sixth higher than F _b .

2) For the first element of the syllable F _n = F _{to the} previous vowel, then the frequency of the fundamental tone increases uniformly (in a grayscale scale), the total increase is sixth.

The application of the above rules is contextually determined. We define a map ξ: L -> R ¹ on the set L of syntagma syllables, and ξ (l) = m if l∈ K _m , ξ (*) = 0.

, где i = ξ(l) для слога - левого контекста, j - для правого, а значение элемента а_i,j соответствует номеру правила.Then the conditions for applying the above rules can be defined by the matrix

, where i = ξ (l) for the syllable is the left context, j is for the right, and the value of the element a _{i, j} corresponds to the rule number.

For class K ₁ :
and _{0, j} = 1 for ∀j,
a _2.6 = 3,
a _{8, j} = 4, a _{9, j} = 5 for ∀j,
a _{i, j} = 2 of all other i and j.

For class K ₂ :
a _{0, j} = 1 for j ∈ {1,2,4,5} and a _{0, j} = 2 for all other j,
a _{1, j} = 3 for j ∈ {1,2,4,5},
and _{9, j} = 5 for ∀j,
and _{i, j} = 4 of all other i and j.

For class K ₃ :
and _{7, j} = 2 for ∀j,
and _{i, j} = 1 of all the remaining i and j.

For class K ₄ :
and _{0, j} = 1 for ∀j,
and _{i, j} = 2 of all the remaining i and j.

For class K ₅ :
and _{0, j} = 1 for ∀j,
and _{i, j} = 2 of all the remaining i and j.

For classes K _h , h ≥ 6:
and _{0, j} = 1 for ∀j,
and _{i, j} = 2 of all the remaining i and j.

For syntagms with the communicative type “general question”, we distinguish the following classes of syllables:
Q ₁ - unstressed and simple stressed syllables (n≤3),
Q ₂ - the last stressed syllable - not a melodic center (n = 4),
Q ₃ - melodic center (n = 3),
Q ₄ - melodic center (n = 4).

There are 4 rules for class Q ₁ :
1) For consonants F _n = F _k = F _b , for a vowel F _n = F _{k of the} preceding element, and F _k 1 semitone higher than F _b .

2) For the first element of the syllable, F _n = F _{k of the} previous vowel, and F _k one semitone lower, for subsequent elements ΔF = 0.

3) For consonants, rule 2) applies, for a vowel F _n = F _{k of the} previous sound, F _k = F _b .

4) Similar to rule 3), but for a vowel F _k half an octave below F _b .

Accordingly, the elements of the matrix determining the application of these rules are equal to:
and _{0, j} = 1 for ∀j,
a _{1, j} = 2 for j ≤ 2 and a _{1, j} = 1 for j ≥ 3,
and _{2, j} = 2 for ∀j,
and _{3, j} = 3 for ∀j,
a _{4, j} = 4 for ∀j.

For class Q ₂ , a contextless rule is applied, similar to rule 1) for Q ₁ , but F _k is half an octave below F _b .

For classes Q ₃ and Q _4, 2 rules are specified:
1) If the syllable is the first in the syntagma, then for the first element F _n = F _b , then the frequency of the fundamental tone evenly (in a grayscale scale) increases for the last element of the syllable F _k an octave above F _b .

2) If the syllable is not the first in the syntagma, then a rule similar to 1) applies, but for the first element F _n = F _{to the} previous vowel, and for the last - F _k an octave higher than F _{n of the} first element.

For a particular question, two classes of syllables P ₁ - non-melodic center and P ₂ - melodic center are distinguished.

For class P _1, 5 rules are set:
1) F _n = F _to = F _b ,
2) Similar to rule 1) for K ₂ ,
3) Similar to rule 2) for Q ₁ ,
4) For the first element F _n = F _{k of the} previous vowel, then the frequency of the fundamental tone uniformly (in a grayscale scale) decreases and for the last element F _k a sixth below the F _{n of the} first element,
5) For the first element, F _n = F _{k of the} preceding vowel, F _k = F _b , for all other F _n = F _k = F _b .

 Rule 1) applies to syllables located to the left of the melodic center, if the next syllable is also not a melodic center, rule 2) - for syllables immediately preceding the melodic center, rule 3) - for syllables to the right of the melodic center (but not immediately after it), rule 4) - for the syllable following the melodic center, if this is the last syllable or last stressed syllable, rule 5) - for the syllable following the melodic center, if there is at least one stressed syllable on the right.

For class P _2, 2 rules are set:
1) For the first syllable of the syntagma, for consonants F _n = F _k = F _b , for a vowel F _n sixths higher than F _b , and F _k three semitones above F _b .

2) For other contexts, for consonants F _n = F _k = F _k , the preceding element, for a vowel F _n = F _{k the} preceding element, and F _k three semitones above F _b .

For consonants at the end of the syntagma, F _n = F _k = F _{to the} previous vowel.

The compilation unit 34 operates as follows. In block 114, the syntagms are processed from left to right, and left and right contexts (preceding and following characters) are allocated for each current character. The encoder 115 generates compilation element codes corresponding to the phonetic symbols in the corresponding context. The compilation element code is a tuple k ₁ , k ₂ , k ₃ , k ₄ >, where k ₁ is the code of the current character element, k ₂ is the code class of the character element, k ₃ is the class code of the left context, and k ₄ is the code of the class of the right context.

 According to these codes, the corresponding synthesis elements in the form of segments of a digitized speech wave are extracted from block 116 of the acoustic-segment synthesis base. In block 117, these sequences of samples are converted in accordance with the time and frequency parameters of the compilation elements defined in the time processor 32 and the melodic processor 33.

For quasiperiodic segments of the speech wave, the time and frequency characteristics of the corresponding compilation element are changed by functional transformation
φ (t) = Σφ _i (t) _ → φ ′ (t) = Σφ _j (t), i≤n ₀ , j≤n
where φ (t) is the segment of the natural speech wave corresponding to the current compilation element;
φ _i (t) is the orthogonal sequence of functions coinciding with φ (t) on the interval of the i-th quasiperiod;
n ₀ is the number of quasiperiods in the segment of the natural speech wave,
n = max (2, [τ (F (t _to ) + F (t _n )) / 2] is the required number of quasiperiods,
τ is the required duration of the compilation element defined in the block of the temporary processor;
F (t _n ) and F (t _k ) - respectively, the initial and final values of the frequency of the fundamental tone defined in the block of the melodic processor.

- результат функционального преобразования φ_i(t), где индекс i для ∀j определяется областью истинности предиката

Функциональное преобразование

определено следующим образом

где

- требуемая длительность j-го квазипериода.

is the result of the functional transformation φ _i (t), where the index i for ∀j is determined by the region of truth of the predicate

Functional conversion

defined as follows

Where

- the required duration of the j-th quasiperiod.

 For noise segments of a speech wave, the frequency of the fundamental tone naturally does not change (for a noise signal, such a parameter is completely absent), only the duration changes by repeating or trimming the base element.

 As can be seen from the above relations, when the duration of the sound changes, the duration of the exposure phase first of all changes, and the durations of the excursion and recursion phases are preserved if possible.

 In composition block 118, the converted sequences of samples corresponding to compilation elements are recorded sequentially one after another in the order determined by the symbolic phonetic record.

 In block 35, an audio file of the selected format is formed, i.e. Before the obtained sequence of samples, the header of the sound file is formed.

 Block 36 outputs the sound file through the sound card.

 When implementing a text database, an approach similar to that used in the Xanadu system is implemented. With this approach, relevant documents (books, articles, etc.) are taken, translated into electronic form without semantic changes and docked to each other, forming one "big book". Specialized add-ons allow you to identify any book as a whole and each of its components separately. To identify information in the database, add-ons are used such as a list of article titles in alphabetical order, a list of authors in alphabetical order, a hierarchical index index for available information in the form of a "Contents" section. This approach guarantees comprehensive access, but tedious to browse.

 For faster retrieval of information relevant to the request, access to information through the mechanism of keywords is used. The filter mechanism provides selectivity, while filters can be specified through keywords or filtering can be performed with document attributes.

 The database interface provides a dialogue between the user and the computer at the acoustic and tactile levels. Computer messages (menu items and other messages) are duplicated by a speech output through a sound card and displayed on a tactile display. The user enters information into the computer through the PC keyboard in the "voice acting" mode (when you press a key, the typed character is called), the typed message is voiced through the synthesizer with confirmation or non-confirmation of the input. The information entered can be the names of articles, a list of keywords, filter attributes (for example, the range of publication years - see all the publications in a given author’s database from such-and-such to such-and-such year), or a comment made by the user on the document (abstract, subject area , a list of keywords, etc.). This information can be either independently typed by the user or typed by the sighted operator on an individual user card on a flexible magnetic medium. The user only selects a menu item: keyboard input, input from a floppy disk. On an individual card, all user communication sessions with the system are also recorded, so the user can get quick access to the information that he worked with earlier, but whose attributes he forgot. From a floppy disk, a document can also be entered and entered into the database by someone previously converted into electronic form.

 The interface unit 9 provides for the automatic start of all device blocks in the necessary sequence, at the end of each step, the user is given a voice message and a confirmation is requested for the next step (pressing one of the two keys is perceived by the computer as the answer “yes” or “no”). Upon completion of work with the document, control is transferred to the database interface so that the user can associate the necessary information with the document.

 The proposed device allows a blind or visually impaired user to independently and independently receive education, both general and professional, independently and independently carry out intellectual and professional activities using hardware and software systems, which allows to solve the social problem of the professional and social adaptation of people with disabilities in a real social environment. With the proposed device, a blind person can compile an information database of his choice and use for this purpose any kind of literature intended for sighted users, including unique ones, for example, on the history of music, codes of laws, history, science, etc. d. Using this device, the user can practically independently or with minimal use of unauthorized persons receive intellectual development, get a higher education and become a high-class professional.

 This device can also be used for independent and independent study and improvement of the Russian language using live and / or synthesized Russian speech directly from a printed or other type of information carrier by both sighted and blind users, as well as people with disabilities with a violation of the musculoskeletal system.

 The device can be used for collective use in libraries for the blind, in schools for the blind in lessons and when conducting independent study of the material.

 The benefits of using this device to learn a language are obvious. In addition to the above features and advantages, the user simultaneously perceives the visual reproduction of the text and the simultaneous speech sound of the text. The simultaneous use of two analyzers - auditory and visual, can dramatically increase the perception of any information in Russian.

Claims (29)

 1. A computer device for reading flat-printed text, containing a series-connected optical input unit for flat-printed text and an optical text recognition unit, a speech synthesis unit for spelling text, an audio signal generating unit, and an audio terminal unit, characterized in that the optical flat input unit for text printing is made in the form of a scanner , the block for generating an audio signal is made in the form of a sound card, the block for synthesizing speech according to the spelling text is made in the form of a block for synthesizing Russian speech according to To the graphic text, a text file unification unit, a text database unit, a tactile display, a tactile display interface unit to a personal computer and an interface unit are additionally introduced into the device, while the output of the optical text recognition unit through a series-connected text file unification unit and a text database unit connected to the information input of the block of the synthesis of Russian speech according to the spelling text, and through the pairing block - with the input of the tactile display, the output of the block of the synthesis of speech and orthographic text through a sound card connected to the terminal audio unit, the control inputs interface unit tactile display and scanner unit OCR unit unification text file, block the synthesis of Russian speech orthographic text and text base unit are connected to the corresponding interface unit outputs.  2. The device according to claim 1, characterized in that the unit of unification of the text file contains sequentially connected block recognition multicolumn text, block reformatting the text file into one-column, block recognition of left and right borders of the text, block recognition of red lines, paragraph selection block, block exclusion characters of metalanguage that are not included in the valid alphabet, block for word wraps, and block for reformatting a text file.  3. The device according to claim 1, characterized in that the block for the synthesis of Russian speech according to the spelling text is made in the form of a block for compilation speech synthesis and contains a series-connected block for selecting the current paragraph, a block for reading the current paragraph, a block for preliminary parsing of the paragraph, a block for selecting words in paragraph, subword selector on subsets of the metalanguage alphabet, block for recognizing words in the Russian alphabet, first block OR, block for contextual decoding of abbreviations of words, matching block, block for selecting phrases and syntagm, bl ok determination of the communicative type of phrases and syntagmas, a logical stress determination unit, a phonetic text transcription unit, a time processor, a methodological processor, a compilation unit, an audio file generation unit and an audio file output unit, the first output of which is the output of a Russian speech synthesis unit from a spelling text, and the second output is connected to the second control input of the current paragraph selection block, the first control input of which is the first control input of the Russian language synthesis block by a graphic text, the second control input of which is the control input of the preliminary parser of the paragraph, the second output of the subword selector on the subsets of the metalanguage alphabet through the block of translation of metalanguage words into words in the Russian alphabet is connected to the second input of the first OR block.  4. The device according to claim 3, characterized in that the block of preliminary parsing of the paragraph contains sequentially connected punctuation mark selector, a switch, a punctuation mark verbalization block and a second OR block, the second output of the switch through sequentially connected punctuation mark context analysis block, the elimination block is not punctuation marks that are reflected in spoken language, the block for eliminating graphic punctuation frictions, and the block for verbalizing the signs of a point, comma, and dash in the record of numbers is connected the second input of the second OR block and the control input of the switch control input is pre parsing unit.  5. The device according to claim 3, characterized in that the block for recognizing words in the Russian alphabet contains a series-connected block for converting symbol codes to one register, a selector for words including punctuation characters, a first abbreviation selector and a third OR block, the second output of the first selector abbreviations through a fourth block OR connected in series, a morphological analysis block, a grapheme recovery block “yo”, an abbreviation selector and an abbreviation decryption unit are connected to the second input of the third IL block And, the second output of the abbreviation selector through a series-connected speech part determination unit and the accentuation unit is connected to the third input of the third OR block, the second output of the word selector including punctuation marks is connected to the fourth input of the third OR block through the second abbreviation selector, the second output the second selector of abbreviation words through the hyphen block is connected to the second input of the fourth OR block, the third output of the word selector including punctuation characters is through the apostr exclusion block F is connected to the third input of the fourth OR unit, a fourth input coupled to a fourth output selector words containing punctuation marks, the third block is the output of OR output word recognition unit to the Russian alphabet. 6. The device according to claim 5, characterized in that the morphological analysis unit contains memory blocks for prefixes, bases and inflections, implemented in the form of a database, the relationship between which is specified by the display
F: M _ → S,
where M is the set of bases defining the lexical meaning of the words of the Russian language;
S - many classes of inflections,
morphological analysis unit is implemented implementing the algorithm for solving the equation
W = X _p • X ⁽¹⁾ • ... • X ^(n-1) • X ⁽ⁿ⁾ ,
where W is the analyzed word;
X _p is the prefix or empty word;
X ⁽¹⁾ ,. . ., X ^(n-1) - one or more stems (possibly a connecting vowel at the end);
X ⁽ⁿ⁾ - inflection or empty word.  7. The device according to claim 5, characterized in that the stress allocation unit comprises sequentially connected a recognized word selector, a stressed syllable determination unit and a fifth OR block, the second output of the recognized word selector through an empirical rule stress allocation unit is connected to the second input of the fifth OR block .  8. The device according to claim 3, characterized in that the unit for translating metalanguage words into words in the Russian alphabet contains a sequentially connected selector of subsets of the metalanguage alphabet, the sixth OR block, the verbalization block of the Arabic digital number record and the seventh OR block, the second output of the selector of the subsets of the metalanguage alphabet through a series-connected selector of latin digits and a block for converting latin numbers to Arabic, connected to the second input of the sixth block OR, the second output of the block for converting latin numbers to ar through the eighth OR block and the Russification block of Latin alphabet words connected in series with the second input of the seventh OR block, the second output of the Latin digit selector is connected to the second input of the eighth OR block, the third input of which is connected to the third output of the subset of the metalanguage alphabet, the fourth output of which the metalanguage alphabet verbalization block is connected to the third input of the seventh OR block, the output of which is the output of the metalanguage word translation block into the words of the Russian alphabet.  9. The device according to claim 3, characterized in that the block for selecting phrases and syntagma is implemented that implements the algorithm for selecting phrases by breaking a paragraph into lexemes separated by {.?} Symbols, and the algorithm for highlighting syntagms in phrases by breaking a phrase into lexemes separated by characters ( ;; :-).  10. The device according to claim 3, characterized in that the unit for determining the communicative type of phrases and syntagm contains sequentially connected selector of interrogative phrases, the unit for determining syntagma containing a question, the ninth block OR and the unit for determining communicative type of syntagm, the second output of the selector of interrogative phrases is connected to the second input of the ninth block OR.  11. The device according to claim 10, characterized in that the communicative type determination unit of the syntagma is implemented that implements an algorithm for determining the following communicative types: completeness corresponding to syntagms ending with signs {.:;), The first and second types of incompleteness corresponding to syntagms ending with signs respectively (, -), general and particular questions for syntagmas ending with {?} and respectively not containing or containing a question word, as well as accented syntagmas containing a logical emphasis and the highlighted word.  12. The device according to claim 3, characterized in that the logical stress determination unit comprises a sequentially connected selector of logically allocated words and a tenth OR block, a second output of a selector of logically distinguished words through a series-connected selector of words containing a particular question, and the eleventh OR block is connected to the second input of the tenth block OR, the second output of the selector of words containing a particular question, through the selector of semantically significant words connected to the second input of the eleventh block OR, and the second output of the selector and semantically meaningful words through the last determination unit notional word syntagm is connected to the third input of the OR block eleventh, tenth block output is the output of OR logical stress determination unit.  13. The device according to claim 3, characterized in that the phonetic transcription block of the text contains sequentially connected spelling elimination block, a letter to phonetic conversion unit, and a unit speech formation unit.  14. The device according to p. 13, characterized in that the spelling removal unit contains spelling units in the word endings, an unpronounceable consonant elimination unit, a consonant combination replacing unit with an equivalent alphabetic entry, and a solid pronunciation pronunciation unit “e” .  15. The device according to p. 13, characterized in that the unit for converting the letter recording into the phonetic contains a serially connected selector of service words, a unit for eliminating stresses in service words, a unit for assimilating service words, the twelfth OR block, a transcription block of contexts with a soft sign, a transcription block solid-sign contexts, a unit for determining the softness of consonants before marked vowels, a unit for transcribing iotated vowels at the beginning of a syntagma, a unit for transcribing iotated vowels in positions after the vowel, unit for transcribing unpaired soft consonants, unit for determining the voiced-deafness of consonants in combinations of consonants, unit for stunning noisy voiced consonants at the end of the syntagma, unit for assimilation of solid consonants before soft, unit for transcribing double consonants, selector of weakly stressed words, unit for determining the degree of reduction of vowels in stressed words, a vowel replacement unit of the first degree of reduction, the thirteenth OR block and a vowel replacement unit of the second degree of reduction, the second output of the service word selector Without the accent replacement block in low impact words, it is connected to the second input of the twelfth OR block, and the third output is connected to the third input of the twelfth OR block, the second output of the low impact word selector is connected to the second input of the thirteenth OR block through the block for determining the degree of reduction in low impact words.  16. The device according to item 13, wherein the unit for the formation of continuous speech is implemented that implements the algorithm for phonetic transcription of word joints in the syntagma in accordance with Russian orthoepic norms.  17. The device according to p. 15, characterized in that the unit for determining the degree of reduction of vowels is made implementing an algorithm in which the first degree of reduction or the second degree of emphasis corresponds to vowels in the positions of the first pre-stressed syllable, in unstressed positions at the end of the syntagma, in the open end of the word if the next word begins with the stressed syllable, and the position after the vowel, and the second degree of reduction, or the first degree of emphasis, with the vowel in all other unstressed positions, the third degree of emphasis corresponds to the vowels in stressed positions, except for the last stressed syllable of the syntagma, the vowel of which corresponds to the fourth degree of emphasis. 18. The device according to p. 15, characterized in that the unit for replacing vowels of the first degree of reduction is implemented that implements an algorithm for the contextual replacement of reduced unstressed vowels of the first degree of reduction by the following substitutions:
Sb (W, y, y2) - for the vowel "y" in any context;
Sb (W, s'g, s'и), g∈ {o, a, э}, Sb (W, sе, sy) - for consonant-vowel contexts ("sg" or "s'g");
Sb (W, о, а) - for all other contexts.  19. The device according to p. 15, characterized in that the unit for determining the degree of reduction of vowels in low impact words is implemented that implements an algorithm for reducing the pseudo-impact vowel in weak impact words to the first degree of reduction. 20. The device according to p. 15, characterized in that the unit for replacing vowels of the second degree of reduction is implemented that implements an algorithm for the contextual replacement of reduced unstressed vowels of the second degree of reduction by the following substitutions:
Sb (W, y, y1) - for the vowel "y" in any context,
Sb (W, sg, sb), g∈ {o, a, e, s} and Sb (W, s'g, s'b), g∈ {o, a, e, and} - for contexts of type " consonant-vowel. " 21. The device according to claim 3, characterized in that the time processor is implemented implementing an algorithm that sets each symbolic element of the phonetic record in accordance with a time interval that determines the duration of the corresponding sound - compilation element, while the vowels are divided into three classes that reflect differences in their phonetic quality
Ψ ₁ = {b, a, a ^∧ };
Ψ ₂ = {b, u, and ^∧ , s, s ^∧ , y1, y2, y ^∧ };
Ψ ₃ = {e, e ^∧ , o, o ^∧ },
where ^ is the stress sign, the duration of the vowels depending on the degree of emphasis n is determined by the relation
τ _i (n) = τ _{i, 0} + (n-1 + [n / 4]) / 2, i∈ {1,2,3},
where i is the vowel class index;
[a] is the integer part,
τ _{i, 0} is a constant fixed for each class, the duration of the vowels depending on the number containing their words in the syntagma (q), and the number of syllables in the word (m) are determined for the vowels of the fourth degree of emphasis by the additional relation

in addition, the duration of any vowel in the position in front of the intervocal vibrants and the duration of the vowels at the end of the syntagma are determined by additional ratios, respectively

where δ is a fixed value for all vowels, ∂b is the duration of the iterative tail, the duration of the consonants depending on the segment environment and the longitude of the consonant is determined by the relation
τ = β _i τ ₀ , j∈ (1,2,3,4},
where τ ₀ is the reference duration corresponding to the duration in the intervocal position;
β ₁ = 0.8 for the position after the consonant before the vowel and β ₂ = 0.6 for the position between the consonants, while for compilation elements corresponding to the explosion phase of the consonant consonants and vibrants β ₃ = 1, β ₄ = 1.3 for long consonants.  22. The device according to claim 3, characterized in that the melodic processor comprises sequentially connected syntagma segmentation block for syllable fragments, a syllable fragment class determination unit and a fundamental tone law parameter determination unit for syllable segments. 23. The device according to p. 22, characterized in that the syntagma segmentation block for syllabic fragments is made implementing an iterative algorithm for solving equations
C ⁽ⁱ⁾ = S ⁽ⁱ⁾ g ^{(i, l)} C ^{(i + l)} , ∀g∈ψ _g ,
with initial conditions C ^(o) = C,
where g ^{(i, l)} is the first occurrence of any symbolic element of the vowel g∈Ψ _g in the word C ⁽ⁱ⁾ ;
C - syntagma, considered as a word in the alphabet;
Ψ = Ψ _s ∪ Ψ _g , Ψ _q = Ψ ₁ ∪ Ψ ₂ ∪ Ψ ₃ - a subset of vowel symbolic elements;
Ψ _s is a subset of consonant symbolic elements;
By the i-th syllable fragment we mean the composition S ⁽ⁱ⁾ g ^{(i, l)} .  24. The device according to p. 22, characterized in that the unit for determining the class of the syllable fragment is implemented that implements the corresponding algorithm, in which syllable fragments are divided into classes depending on the communicative type of syntagma, the degree of emphasis on the vowel syllable, and on the coincidence of the syllable with the melodic center of the syntagma that matches with the logical emphasis of syntagma, while for completeness and two types of incompleteness, nine classes of syllables are distinguished: unstressed syllables, stressed syllables with a vowel degree of three, not coinciding with the melodic center, the last stressed syllable that does not coincide with the melodic center, the melodic center for syntagmas with the type of completeness, respectively, with the degree of vowel highlighting n = 4 and n = 3, the melodic center for syntagmas with the first and second types of incompleteness and the degree of vowel highlighting, respectively respectively n = 3 and n = 4, for syntagmas with the communicative type “general question” four classes of syllables are distinguished, respectively unstressed and stressed syllables are not melodic centers with n = 3, the last stressed syllable is not melodic a center, a melodic center with vowel degrees of emphasis n = 3 and n = 4, respectively, and for syntagmas with the communicative type "private question" two classes of syllables are distinguished - coinciding and not coinciding with the melodic center. 25. The device according to p. 22, characterized in that the unit for determining the parameters of the law of changing the frequency of the fundamental tone for syllable segments is implemented that implements an algorithm, each symbolic element - a syllable segment associating its initial (F _n ) and final (F _k ) melodic characteristics while a linear approximation of the law of changing the frequency of the fundamental tone on the interval of the duration of each segment of the syllable
F (t) = F _n + (tt _n ) / (t _to -t _n ) ΔF, t∈ [t _n , t _to ];
ΔF = F _to -F _n
where [t _n , t ^k ] is the interval of duration of the ith segment, the values of F _n and F _{k are} quantized on the scale of musical tones, their values are determined by the class of the syllable, the position of the syllable relative to the boundaries of the syntagma (first, non-first, last), the classes of syllables - left and right contexts, the number of segments in a syllable, the type of symbolic element (vowel - consonant), its position relative to the beginning of the syllable (first - nonfirst) and are given tabular on a valid subset of the vector product (K ∪ {*}) ⊗K⊗ (K ∪ {*}), where K is the set of syllable classes, * is the sign, I share In addition, for the deaf slit and deaf afflicants, for the deaf bow and explosion in the deaf cushions, for the segments of vocalization and pause, the vibrant is F (t) = const = F _min , where F _min is the minimum of the frequency range of the main tone of the speaker, for Voiced sonic, a linear change in the frequency of the fundamental tone corresponds to the phase of the bow, and for the phase of the explosion F (t) = F _min .  26. The device according to claim 3, characterized in that the compilation unit comprises sequentially connected context allocation unit, a compiler element code encoder, an acoustic segment synthesis unit, a unit for changing the time and frequency characteristics of compilation elements, and a composition unit. 27. The device according to p. 26, wherein the block of the acoustic segment synthesis base is made in the form of a database block, the acoustic segment synthesis base contains basic compilation elements in the form of digitized segments of the natural phonemic speech wave — allophones that are acoustically and perceptually distinguishable contextual implementations of phonemes, while in order to reduce the volume of the base determined by the complete calculation of compilation elements of a given type for the synthesized language, identical contextual influences are generalized taking into account the acoustic reflexes of co-articulation processes, for which the set of phonemes Ψ of the synthesized language is divided into classes differentiated by contextual influence, for each of which the sets of left and right contexts are divided into classes of left and right contexts that are identical in influence for a given phoneme class, while each phoneme of the ith class ψ∈Ψ is represented in the base N _i = card (L _i ) card (R _i ) by implementations, each of which is a realization ψ _i in the context of λ _j ψ _i ρ _k , where L _i = UL _{i , j} and R _i = UR _{i, k} are respectively but the sets of classes of left and right contexts for the ith class of phonemes, classes of phonemes are represented by a class of vowel phonemes and seven classes of consonant phonemes, classes of consonant phonemes include voiced and deaf connecting, hard and soft fricatives, respectively, except for hard and soft (x), nasal, flowing sonants, including hard and soft (x) and (c), vibrants and iota, while the class of nasal is indifferent to contextual influences, the classes of voiced and deaf cusps are indifferent to the left context, and for the right contexts h three classes of classes - respectively ruined, final, soft, and all the rest, for classes of hard and soft fricatives, two classes of left contexts are distinguished - the initial and all the rest, and the same classes of right contexts as the classes of voiced and deaf closed ones, for the class of smooth sonants - four classes of left contexts, respectively, ruined, initial, {and ^, and, s ^, s, e ^, b} and all the rest, and four classes of right contexts, respectively, ruined, final, any context for soft sounds and all the rest, vibrant class includes hard and soft phonemes (p) in the intervocal position, and for the non-intervocal position four classes of left contexts are distinguished, respectively {y ^, y, o ^, b, n, f, c, l, m}, initial, {u ^, s ^, and, b, e ^} and soft consonants, all other contexts, and four classes of right contexts, respectively, the first class is identical to the first class of the left context, finite, any, except the final, context for soft (p) and all other contexts, for of the iota class, four classes of left contexts are distinguished, respectively, ruined, initial, {and ^, s ^, and, s, b, e ^} and all soft consonants, and all other contexts, and three classes of right contexts, respectively final, vowels and consonants, the class of vowel phonemes respectively includes stressed phonemes {a ^, y ^, and ^, s ^, е ^, о ^}, unstressed phonemes of the first degree of reduction { a, y2, u, s, e, o} and unstressed phonemes of the second degree of reduction {b, b, y1}, ten classes of left contexts are distinguished for vowels, respectively anterior lingual solid, labial hard, posterior lingual solid, anterolingual nasal, labial nasal, postalveolar hard, soft nasal, soft nasal anterior lingual, nachal flaxen and soft nasal labial, and eight classes of right contexts, respectively, anterolingual hard, labial hard, postalveolar hard, soft consonants, as well as {and ^, and, b}, and final, with the initial and final contexts are understood as pauses at the beginning and end of syntagma. 28. The device according to p. 26, characterized in that the compiler element code encoder is implemented implementing the compilation element code generation algorithm in the form of a tuple << K _f , K _{k, f} , K _l , K _r >>, where K _f is the phoneme code , K _{k, f} is the phoneme class code, K _l is the left context class code, K _r is the right context class code. 29. The device according to claim 3, characterized in that the unit for changing the time and frequency characteristics of the compilation element is made implementing a functional transformation algorithm

where φ (t) is the segment of the natural speech wave corresponding to the current compilation element φ _i (t) is the orthogonal sequence of functions coinciding with φ (t) on the interval of the i-th quasiperiod, n _o is the number of quasiperiods in the segment of the natural speech wave, n = max (2, [τ (F (t _to ) + F (t _n ) / 2] is the required number of quasiperiods, τ is the required duration of the compilation element, F (t _n ) and F (t _to ) are the given initial and the final value of the frequency of the fundamental tone, while

is the result of the functional transformation φ _i (t), where the index i for ∀j is determined by the truth domain of the predicate
P (i) = (i = j ^ (n = n _o vn> n _o ^ j≤ [n _o / 2] vn <n _o ^ j≤ [n / 2]) v;
(i = j + n _o -n ^ n> n _o ^ j≥n- [n _o / 2] vn <n _o ^ j≥ [n / 2]) v;
(i = [n _o / 2] ^ n> n _o ^ [n / 2] <j ^ j <n- [n _o / 2]),
and functional transformation

defined as follows

Where

- the required duration of the j-th quasiperiod.

Images