Classifications

• • G—PHYSICS
  • G10—MUSICAL INSTRUMENTS; ACOUSTICS
  • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
  • G10L13/00—Speech synthesis; Text to speech systems
  • G10L13/08—Text analysis or generation of parameters for speech synthesis out of text, e.g. grapheme to phoneme translation, prosody generation or stress or intonation determination

Definitions

• the present invention generally relates to the automatic extraction of linguistic knowledge from a corpus of transcriptions of graphic chains into phonetic chains. More particularly, it relates to the transcription of typographic elements such as characters in a predetermined language into phonetic elements.
• each word of a language constitutes a graphic chain which is transcribed phonetically into a chain of phonemes by a phonetician.
• the phonetician For any new word to add to a learning corpus, the phonetician must intervene to transcribe this new word phonetically.
• the learning corpus thus provides only global grapheme / phoneme transcriptions. For example in the global transcription "lane" / [ry ⁇ 1], the corpus indicates that overall, the graphic chain "lane” translates into phonetic chain. However, it is not explained that in any way, unitarily, the typographic element "r" is transcribed phonetically.
• the global transcription does not also indicate the syllables or graphemes making up the graphic chain and the phonetic elements making up the phonetic chain.
• the invention aims to automatically deduce raw transcriptions of graphic strings, such as words and surnames, for example, into phonetic strings, transcriptions of graphic elements, such as characters, into phonetic elements composing the phonetic strings in order to automatically segment any graphic chain into graphemes and any phonetic chain into phonemes.
• the elementary transcriptions graphic element by graphic element that is to say character by character, then facilitate the automatic global transcription of any additional graphic chain brought to the corpus of graphic chains, on the basis in particular of a concatenation of phonetic elements unequivocally corresponding to the characters of the additional graphic string.
• a method according to the invention automatically corresponds graphic elements making up given graphic strings to phonetic elements making up corresponding phonetic chains, after having initially entered global transcriptions of graphic chains into phonetic chains in a database accessible by the computer and having estimated and recorded in the base of the first probabilities of elementary transcriptions of the graphic elements respectively into the phonetic elements.
• the method is characterized by the following stages: for each transcription of a graphic chain given to M graphic elements into a phonetic chain corresponding to N phonetic elements, determine second probabilities of MN second transcriptions of M graphic chains successively concatenating the M graphic elements in N phonetic chains successively concatenating the N phonetic elements, each according to a respective first probability and the greatest of three respective second probabilities determined previously, and establishing and memorizing a link between the last elements of the graphic and phonetic chains of each second transcription and the last elements of the graphic and phonetic chains of the transcription relating to the largest of the three respective second probabilities so that links established in a matrix of size MN relating to the second probability és constitutes a unique path between the last and first pairs of graphical and phonetic elements of the matrix to segment the given graphic chain into graphemes corresponding respectively to phonemes segmenting the corresponding phonetic chain and to record the correspondences between the graphemes and phonemes in basically, the number of graphic elements in a grapheme being identical to the number of phonetic elements in the
• the respective first probability for determining a second probability relating to a second transcription of a graphic chain concatenating m graphic elements into a phonetic chain concatenating n phonetic elements is relative to the last elements in the graphic chain with m graphic elements and the phonetic chain with n phonetic elements.
• the three respective second probabilities previously determined for the second transcription of the graphic chain with m graphic elements in the phonetic chain with n phonetic elements preferably relate respectively to a second transcription of a graphic chain with m-1 graphic elements in the chain phonetic with n phonetic elements, a second transcription of the graphic chain with m graphic elements into a phonetic chain with n-1 phonetic elements and a second transcription of the graphic chain with m-1 graphic elements into the phonetic chain with n-1 elements phonetic.
• the invention transcribes phonetically from the corpus of global transcriptions such as "lane”
• the invention can be likened to a syllabation which makes it possible by analysis to decompose a global transcription into elementary transcriptions, and to locally match grapheme / phoneme sub-transcriptions.
• the division into initial graphemes and phonemes and the one-to-one correspondence of each graphic element to each phonetic element of the cut phonemes is called grapheme
• the invention produces the following alignment: M -. H "u”"e""binds" [r] [y] [ ⁇ ] [1 **].
• the symbol * indicates a silent and meaningless phonetic element.
• FIG. 1 is an algorithm of main steps of the automatic correspondence method according to the invention
• - Figure 2 is a sub-step algorithm of a step of determining first individual probabilities included in the automatic matching method.
• the method of automatic correspondence of graphic elements and phonetic elements comprises main steps El to Eli. These steps are for the most part implemented for example in the form of software i implemented in a terminal, such as a personal computer or a mobile in a cellular radiocommunication network, and linked in particular to a software system for spell checking. of lexical errors which can be integrated into a word processing system or a system of linguistic exercise.
• the terminal contains or can access a database of the type used in artificial intelligence.
• the database stores a corpus C of initial global transcriptions.
• step E1 the global transcriptions (CG
• the segmentation of the chain CG into syllables or graphemes each comprising one or more graphic elements is ignored, and the segmentation of the chain CP into phonemes each comprising one or more phonetic elements.
• the alphabets G and P have around thirty elements. They thus present a possibility of 30 ⁇ 30,900 possible pairs of graphic element and of phonetic element.
• the corpus C contains at least 100,000 global transcriptions of typographic strings CG into phonetic strings CP, which preserves the invention from gross errors in probability estimates, as will be seen below.
• the first elementary transcription probabilities P (gjjp- j ) so that an element graph g ⁇ corresponds to the phonetic element p are first estimated and recorded in the database with the corpus of global transcriptions C.
• the estimated values of the first probabilities are as close as possible respectively to the maximum probability values sought so that the method of the invention operating by iterations converges quickly while avoiding retaining local maxima.
• the initial estimation algorithm E2 of the first probabilities P comprises the following sub-steps E21 to E27.
• IJ contingency numbers ⁇ gi p j 'respectively associated with the elementary transcriptions (jjp j ) of a graphic element of the alphabet G and of a phonetic element of the alphabet P are set to zero .
• the number of contingencies K qj _ p is equal at the end of step E2 to the number of times estimated where the graphic element gj is transcribed in the phonetic element pj in the various global transcriptions of typographic chains CG into phonetic chains CP included in the corpus C.
• the ranks of the graphic elements in the chain CG and the ranks of the phonetic elements in the chain CP are normalized as a function of the respective lengths l g and 1 wearingof the chains CG and CP which may be different.
• the number of contingencies K g ⁇ pj associated with the elementary transcription of the graphic element g. j _ in the phonetic element pj is then incremented by 1 only if the phonetic element p-4 is located at the deduced rank r in the chain CP, as indicated in substeps E24 and E25.
• Sub-steps E22 to E25 are repeated for each global transcription (CG
• each graphic chain CG read in the corpus of the database in order to automatically correspond in a one-to-one manner each segment of the chain CG , called grapheme, comprising one or more one-segment graphic elements, called phoneme, comprising one or more phonetic elements resulting from a segmentation of the corresponding phonetic chain CP.
• a graphic chain CG comprises M consecutive graphic elements g ⁇ to g M and the phonetic chain CP corresponding to the chain CG comprises N consecutive phonetic elements p 1 to p N with the integer N different, or possibly equal to the integer M.
• the probability P (g 1 # ... g m , ... g M ⁇ p lr ...
• P ( ⁇ g 2 - • • g m + ⁇ lP ⁇ P2 • • • - n + l ) only depends on the probabilities of three possible transcriptions: either P (g x g 2 ... g m
• a link is memorized in the computer.
• the links trace a unique path also gradually memorized in the computer and connecting the first couple (g-
• the single path topology in the size matrix MN segments the graphic chains CG into graphemes and the phonetic chains CP into phonemes and aligns the graphic elements and the phonetic elements in one-to-one correspondence.
• a segment of the path follows a portion of a line between two graphic elements
• the concatenation of the graphic elements of the line portion corresponds to the phonetic element of the line supplemented by one or more silent and meaningless phonetic elements in order to form a pair of graphemes and phonemes having the same number of elements, which pair is stored in the computer.
• the graphic element of the column supplemented by one or more graphic elements without meaning corresponds to the concatenation of the phonetic elements of the portion of column in order to form a couple of graphemes and of phoneme having the same number of elements, which couple is memorized in the computer.
• a change of direction of the path towards the horizontal, the vertical or the diagonal in the matrix indicates a segmentation of the chains CG and CP.
• the symbol ⁇ - indicates that the couple (g m , p n ) is linked to the couple (g m -i Pn . ) ' ⁇ - e symbol indicates that the couple (g m , p n ) is linked to the couple (g m , p n _ - [ _); and the symbol indicates that the couple (g m , p n ) is related to the couple (g m -i / Pn- 1 ⁇ •
• bo) indicates that the latter is deduced and therefore linked to the transcription (b
• the symbol indicates a segmentation border between pairs of graphemes and phonemes.
• p j ) transcriptions of each of the graphic elements respectively into the J phonetic elements (step E2) and in particular the contingency numbers ⁇ g l p l to K gI pj (sub-step E25) are again estimated in function in particular of the ranks of elements p honetic placed in the given phonetic chains CG which were segmented into phonemes in the previous step E10.
• second probabilities P (g x , ... g pi • • - n ) of MN second transcriptions of each global transcription of a graphic chain given to M graphic elements (CG) into a corresponding phonetic chain (CP) to N phonetic elements are determined by the execution of steps E3 to E10 so that in the next step E10 links are established between couples (g m / P n ) of a new matrix with M columns and N rows and therefore a corrected path connecting the last couple (9M / PN) to the First couple (g ⁇ , P ⁇ ) in the new matrix of second probabilities of size MN.
• Any new graphic chain added to the corpus can then be automatically transcribed into a phonetic chain segmented into phonemes using in particular the correspondences previously established and recorded according to the invention, which progressively enriches the corpus in the database and increases the accuracy of transcriptions.
• phonetic transcriptions are useful for correcting software systems misspelling to recognize lexical mistakes when entering text on a terminal keyboard.
• the phonetic chain segmented into phonemes by means of the recorded correspondences is used for a spelling correction of the new graphic chain entered.
• the method of the invention can also be used as a tool for automatically generating short SMS messages from text written in everyday language.
• the sentence in French "I look busy" entered in the terminal is automatically transcribed into a short message according to Gl 'ROQP to be transmitted by the terminal, "phonetic strings” [G], [1 '], [R] and [OQP] being phonetically readable by any non-phonetic user.
• the "phonetic chains” [G], [1 '], [R] and [OQP] can be assimilated to phonetic elements to constitute a phonetic chain [Gl'ROQP].
• the steps of the method of the invention are determined by the instructions of a computer program incorporated in a computer such as a terminal, a personal computer, a server or any other computer system.
• the program automatically matches graphic elements composing given graphic chains to phonetic elements making up corresponding phonetic chains, after having initially entered global transcriptions of graphic chains into phonetic chains in a database accessible by the computer and having estimated and recorded in the base of the first probabilities of elementary transcriptions of the graphic elements respectively into the phonetic elements.
• the program includes program instructions which, when said program is loaded and executed in the computer whose operation is then controlled by the execution of the program, carry out the steps of the method according to the invention.
• the invention also applies to a computer program, in particular a computer program on or in an information medium, suitable for implementing the invention.
• This program can use any programming language, and be in the form of source code, object code, or intermediate code between code source and object code such as in a partially compiled form, or in any other form desirable for implementing the method according to the invention.

Landscapes

• Engineering & Computer Science (AREA)
• Computational Linguistics (AREA)
• Health & Medical Sciences (AREA)
• Audiology, Speech & Language Pathology (AREA)
• Human Computer Interaction (AREA)
• Physics & Mathematics (AREA)
• Acoustics & Sound (AREA)
• Multimedia (AREA)
• Machine Translation (AREA)
• Debugging And Monitoring (AREA)

Applications Claiming Priority (2)

Publications (1)

Family

ID=34630305

Family Applications (1)

Country Status (4)

Families Citing this family (6)

Family Cites Families (4)

• 2003
  • 2003-12-18 FR FR0314928A patent/FR2864281A1/fr active Pending
• 2004
  • 2004-12-17 EP EP04816413A patent/EP1711936A2/de not_active Ceased
  • 2004-12-17 WO PCT/FR2004/003278 patent/WO2005062292A2/fr not_active Ceased
  • 2004-12-17 US US10/596,425 patent/US20070055515A1/en not_active Abandoned

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events