CN106294310B - A kind of Tibetan language tone prediction technique and system - Google Patents
A kind of Tibetan language tone prediction technique and system Download PDFInfo
- Publication number
- CN106294310B CN106294310B CN201510325742.6A CN201510325742A CN106294310B CN 106294310 B CN106294310 B CN 106294310B CN 201510325742 A CN201510325742 A CN 201510325742A CN 106294310 B CN106294310 B CN 106294310B
- Authority
- CN
- China
- Prior art keywords
- word
- unit
- speech
- boundary
- tibetan language
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Machine Translation (AREA)
Abstract
The invention discloses a kind of Tibetan language tone prediction technique and systems, comprising: receives Tibetan language text to be processed；Word segmentation processing is carried out to the Tibetan language text to be processed, obtains each word unit；According to context information of institute's predicate unit in the Tibetan language text to be processed, the part of speech of institute's predicate unit is determined；It predicts the rhythm boundary of the Tibetan language text to be processed, and according to the part of speech of rhythm boundary word unit, adjusts the word elementary boundary of rhythm boundary；According to the part of speech of each word unit, tone prediction is carried out to the syllable unit of the Tibetan language text to be processed after adjustment word elementary boundary, obtains the tone information of Tibetan language text to be processed.Using the present invention, it can solve modified tone problem of more tone recognition words in different rhythm boundaries, effectively improve Tibetan voice systematic difference effect.
Description
Technical field
The present invention relates to Tibetan language field of information processing, and in particular to a kind of Tibetan language tone prediction technique and system.
Background technique
Speech synthesis is the important component in language information processing, defeated after referring to text through certain conversion
The process of voice out, and make the voice of synthesis that there is good naturalness and intelligibility as far as possible, text-processing is speech synthesis system
The front end text analyzing of system is handled.Wherein, the uniqueness that Tibetan language tone prediction technique is pronounced due to Tibetan language is Tibetan language text-processing
One emphasis of middle research.
Tibetan language includes Lhasa words, health bar words, Anduo County's words etc., wherein based on being talked about with Lhasa.One prominent voice of Lhasa words
Feature is exactly tone, and Tibetan language as described below refers mainly to Lhasa words.Tibetan voice synthetic method mainly include making character fonts and
Tone prediction etc..Briefly, tone is exactly the height of sound.Due to particularity existing for Tibetan language itself, so that Tibetan language
More tone recognition words are different in the tone information of different rhythm boundaries, and the variation of tone can seriously affect semantic understanding.
If the tone tune type of Tibetan language cannot be predicted accurately, it will reduce the application effect of Tibetan language making character fonts.Existing Tibetan language sound
Adjusting prediction technique is typically all rule-based method, i.e., classifies to the initial consonant of syllable, simple or compound vowel of a Chinese syllable, according to sorted initial consonant
With the combined situation of simple or compound vowel of a Chinese syllable, the tone tune type of syllable is obtained by looking into tone tune type table.But existing method becomes tone tune type
The analysis of change mechanism is sufficiently complete, does not consider the characteristic of Tibetan language itself, causes existing Tibetan language tone prediction technique quasi-
More tone recognition words are really predicted in the tone of different rhythm boundaries, so that the naturalness of Tibetan voice synthesis reduces or even shadow
Ring intelligibility.
Summary of the invention
The embodiment of the present invention provides a kind of Tibetan language tone prediction technique and system, solves in Tibetan language more tone recognition words not
With the tone different problems of rhythm boundary, so that Tibetan voice synthesis is more natural.
For this purpose, the embodiment of the present invention provides the following technical solutions:
A kind of Tibetan language tone prediction technique, comprising:
Receive Tibetan language text to be processed；
Word segmentation processing is carried out to the Tibetan language text to be processed, obtains each word unit；
According to context information of institute's predicate unit in the Tibetan language text to be processed, institute's predicate unit is determined
Part of speech；
It predicts the rhythm boundary of the Tibetan language text to be processed, and according to the part of speech of rhythm boundary word unit, adjusts rhythm
Restrain the word elementary boundary of boundary；
According to the part of speech of each word unit, to the syllable unit carry out sound of the Tibetan language text to be processed after adjustment word elementary boundary
Prediction is adjusted, the tone information of Tibetan language text to be processed is obtained.
Preferably, the context information according to each word unit in the Tibetan language text to be processed, really
The part of speech of each word unit includes: calmly
To this progress of Tibetan language and literature subordinate sentence to be processed；
Predict part of speech of institute's predicate unit in sentence；
Determine the type of institute's predicate unit；
According to the type of institute's predicate unit, the part of speech of institute's predicate unit is adjusted.
Preferably, described to include: to this progress of Tibetan language and literature subordinate sentence to be processed
Predict the level-one part of speech of each word unit, the level-one part of speech include: verb, notional word, pronoun, function word, general affixe,
Verb configuration affixe；
If Dan ChuifuThe level-one part of speech of previous word unit is verb or verb configuration affixe, then at Dan Chuifu
For sentence boundary；
If Dan ChuifuThe level-one part of speech of previous word unit is not verb or verb configuration affixe, then passes through system
The method of meter modeling predicts sentence boundary.
Preferably, the level-one part of speech of each word unit of prediction includes:
Obtain the candidate level-one part of speech of each word unit；
Extract the context-sensitive feature of current word unit；
According to the context-sensitive feature of current word unit, by the method for statistical modeling from the candidate one of current word unit
The level-one part of speech of current word unit is determined in grade part of speech.
Preferably, the type of institute's predicate unit includes any of the following or a variety of: more tone recognition words, function word, affixe,
Conventional word.
Preferably, the part of speech according to rhythm boundary word unit, the word elementary boundary for adjusting rhythm boundary include:
When the word unit of rhythm boundary is more tone recognition word units, and part of speech is verb or adjective, with syllable
More tone recognition word units are split for unit, carry out subsequent tone prediction using syllable unit after fractionation.
A kind of Tibetan language tone forecasting system, comprising:
Receiving module, for receiving Tibetan language text；
Word segmentation module obtains each word unit for carrying out word segmentation processing to the Tibetan language text to be processed；
Part of speech determining module, for being believed according to context environmental of institute's predicate unit in the Tibetan language text to be processed
Breath, determines the part of speech of institute's predicate unit；
Word elementary boundary adjusts module, for predicting the rhythm boundary of the Tibetan language text to be processed, and according to rhythm side
The part of speech of word unit at boundary adjusts the word elementary boundary of rhythm boundary；
Tone prediction module, for the part of speech according to each word unit, to the Tibetan language and literature to be processed after adjustment word elementary boundary
This syllable unit carries out tone prediction, obtains the tone information of Tibetan language text to be processed.
Preferably, the part of speech determining module includes:
Clause unit, for this progress of Tibetan language and literature subordinate sentence to be processed；
Part of speech predicting unit, for predicting part of speech of each word unit in sentence；
Word type determining units, for determining the type of each word unit；
Part of speech adjustment unit adjusts the part of speech of word unit for the type according to word unit.
Preferably, the clause unit includes:
Level-one part of speech predicts subelement, for predicting the level-one part of speech of each word unit, the level-one part of speech include: verb,
Notional word, pronoun, function word, general affixe, verb configuration affixe；
First boundary determines subelement, if being used for Dan ChuifuThe level-one part of speech of previous word unit is verb
Or verb configuration affixe, then it is sentence boundary at Dan Chuifu；
Second boundary determines subelement, if being used for Dan ChuifuThe level-one part of speech of previous word unit is not
Word or verb configuration affixe then predict sentence boundary by the method for statistical modeling.
Preferably, first boundary determines that subelement is specifically used for: the candidate one of each word unit is obtained by tabling look-up
Grade part of speech；Extract the context-sensitive feature of current word unit；According to the context-sensitive feature of current word unit, pass through statistics
The method of modeling determines the level-one part of speech of current word unit from the candidate level-one part of speech of current word unit.
Preferably, institute's predicate elementary boundary adjustment module is specifically used for: when the word unit of rhythm boundary is more tone moulds
Formula word unit, and part of speech be verb or adjective when, more tone recognition word units are split as unit of syllable, utilize sound after fractionation
It saves unit and carries out subsequent tone prediction.
A kind of Tibetan language tone prediction technique and system provided in an embodiment of the present invention, by by received Tibetan language and literature to be processed
This progress word segmentation processing obtains each word unit, and determines the type of each word unit, then according to the type of word unit to acquisition
Rhythm boundary word elementary boundary is adjusted, and predicts the word unit sound of the Tibetan language text to be processed after adjustment word elementary boundary
It adjusts.Due to having adjusted the word list of rhythm boundary in Tibetan language text according to word unit part of speech during predicting word unit tone
The tone of the word unit of rhythm boundary is predicted on first boundary, and the part of speech of the word unit according to rhythm boundary adjusted, from
And it solves more tone recognition words in Tibetan language and improves Tibetan voice synthesis oneself in the tone different problems of different rhythm boundaries
So degree, and ensure its intelligibility.
Detailed description of the invention
In order to illustrate the technical solutions in the embodiments of the present application or in the prior art more clearly, below will be to institute in embodiment
Attached drawing to be used is needed to be briefly described, it should be apparent that, the accompanying drawings in the following description is only one recorded in the present invention
A little embodiments are also possible to obtain other drawings based on these drawings for those of ordinary skill in the art.
Fig. 1 is the flow chart of Tibetan language tone prediction technique provided in an embodiment of the present invention；
Fig. 2 is a kind of structural schematic diagram of Tibetan language tone forecasting system provided in an embodiment of the present invention；
Fig. 3 is a kind of structural schematic diagram of Tibetan language text processing system provided in an embodiment of the present invention.
Specific embodiment
The scheme of embodiment in order to enable those skilled in the art to better understand the present invention with reference to the accompanying drawing and is implemented
The present invention is described in further detail for mode.Following embodiment is exemplary, and for explaining only the invention, and cannot be solved
It is interpreted as limitation of the present invention.
For a better understanding of the present invention, Tibetan language tone prediction technique in the prior art is carried out briefly first below
It is bright.Existing Tibetan language tone prediction technique is usually used rule-based method and predicts the tone of text to be processed,
Such as: according to the combination of the initial and the final, looks into tone tune type table and obtain the tone tune type of syllable.Single syllable tone is in Tibetan language
It is determined by initial consonant and simple or compound vowel of a Chinese syllable, wherein initial consonant is made of pre-script, upper word adding, base word, down word adding, the sound of initial consonant
Tune is divided into high and low two class, indicates that the starting point of syllable tone is high and low.In general, base word is turbid situation initial consonant to be high, base word
Be clear situation initial consonant be it is low.Pre-script, upper word adding can change base word tone and obtain height.Simple or compound vowel of a Chinese syllable be by vowel character, back word adding,
Back word adding is constituted again, simple or compound vowel of a Chinese syllable can be divided into 3 classes according to the ending of a final of simple or compound vowel of a Chinese syllable, i.e., long simple or compound vowel of a Chinese syllable promotees simple or compound vowel of a Chinese syllable, single vowel simple or compound vowel of a Chinese syllable.It is right
When syllable carries out tone prediction in current word unit, the type of current syllable sound mother is determined according to the initial and the final type list first
Combination, the sound parent type table are generally constructed by domain expert；Then tone tune type table is searched, determines the initial and the final combination
Syllable tone tune type contains initial consonant and the various combined tone tune types of simple or compound vowel of a Chinese syllable in the tone tune type table, is generally basede on rule
Method building, the rule describes the characteristic of speech sounds of Tibetan language.Such as syllableWherein, base word isIt is preceding to add
Word isInitial consonant isBelong to high class；Pre-script isInitial consonant attribute is not changed, vowel sign isAfterwards plus
Word isSimple or compound vowel of a Chinese syllable isBelong to rush simple or compound vowel of a Chinese syllable, so that it is determined that sound mother's group of syllable is combined into the simple or compound vowel of a Chinese syllable of high initial consonant and rush, passes through
Tone tune type table is looked into, syllable is obtainedTone be falling tone (f).However, the prior art is not associated with existing for Tibetan language itself
Characteristic, such as do not consider more tone recognition words different rhythm boundaries tone with conjugations different problems, to make
Decline at the application effect of voice system.
Tibetan language tone prediction technique and system provided by the invention, by being obtained to this progress of Tibetan language and literature word segmentation processing to be processed
Each word unit is obtained, and predicts the part of speech of each word unit, then obtains the rhythm boundary of Tibetan language text to be processed, and according to rhythm side
The part of speech of boundary's word unit adjusts its word boundary, solves more tone recognition words in different rhythm boundaries, tone different problems,
Tone prediction then is carried out to the Tibetan language text to be processed after adjustment word elementary boundary, obtains the tone letter of Tibetan language text to be processed
Breath.Due to having adjusted the boundary of rhythm boundary word unit, and according to the part of speech of word unit behind adjustment word boundary, its sound is predicted
It adjusts, solves the problems, such as modified tone of more tone recognition words in different rhythm boundaries, so that the tone of prediction is more acurrate, effectively mention
The application effect of voice system is risen.
Technical solution and technical effect in order to better understand the present invention, below with reference to flow chart and specific implementation
Example is described in detail.
As shown in Figure 1, being the flow chart of Tibetan language tone prediction technique provided in an embodiment of the present invention, comprising the following steps:
Step 101, Tibetan language text to be processed is received.
Step 102, to this progress of Tibetan language and literature word segmentation processing to be processed, each word unit is obtained.
In the present embodiment, principle can be segmented according to Tibetan language to be segmented.
Specifically, first with double vertical symbolsIt is segmented for label, then with double vertical symbolsOr Dan ChuifuTo be segmented in the segmentation of label, obtain with double vertical symbolsOr Dan ChuifuFor the text phase to be processed on boundary
String should be segmented.
Further, it obtains after text to be processed accordingly segments string, it can also be to can in the text to be processed participle string
It can be marked to stick together the case adverbial verb of case adverbial verb, the case adverbial verb that sticks together refers to that word unit is followed by having the lattice of adhesive properties to help
Word, the case adverbial verb can be combined into a syllable with the word unit ultima that gets adhered in the form of back word adding, and
Become the ending of a final of this syllable on pronunciation.
Step 103, the context information according to each word unit in the Tibetan language text to be processed, determines institute
State the part of speech of each word unit.
In practical applications, can by tabling look-up or determining using method of statistical modeling etc. the part of speech of each word unit,
In the present embodiment, by context-sensitive feature of each word unit in sentence, each word unit is predicted by the method for statistical modeling
Part of speech adjust the part of speech of each word unit then further according to the type of word unit, with ensure obtain word unit part of speech standard
True property, can specifically include:
To this progress of Tibetan language and literature subordinate sentence to be processed；
Predict part of speech of each word unit in sentence；
According to the type of word unit, the part of speech of word unit is adjusted.
Step is a) to this progress of Tibetan language and literature subordinate sentence to be processed
Since the basic composed structure of Tibetan language sentence is " subject-object-predicate "；Therefore, text to be processed is divided
When sentence, first have to determine predicate verb position.The present embodiment is according to Dan ChuifuThe level-one part of speech of previous word unit is come true
Order, which hangs down, to be accorded withWhether corresponding boundary is a boundary.Wherein, the level-one part of speech includes: verb, notional word, pronoun, void
Word, general affixe, verb configuration affixe, it is described to may include: to this progress of Tibetan language and literature subordinate sentence to be processed
Predict the level-one part of speech of each word unit；
Sentence boundary is determined according to the level-one part of speech of each word unit.
In practical applications, the level-one part of speech that word unit can be carried out by the method for statistical modeling is predicted, specific to predict
When, the candidate level-one part of speech of each word unit is obtained by tabling look-up first；Then the context-sensitive feature of current word unit is extracted,
Wherein, context-sensitive feature can be for current word unit be previous and/or the candidate level-one part of speech of the latter word unit, and works as
Candidate level-one part of speech of preceding word unit etc.；Finally according to the context-sensitive feature of current word unit, pass through the side of statistical modeling
Method determines the level-one part of speech of current word unit from the candidate level-one part of speech of current word unit, wherein statistical model such as decision tree
Model, level-one part of speech can specifically include:
1. notional word specifically includes: noun, adjective, adverbial word, number；
2. pronoun, such as demonstrative pronoun(I),(you),(he)；
3. function word specifically includes: case adverbial verb, conjunction, preposition；
4. verb；
5. general affixe, specifically includes: noun affixe, adjective affixe, adverbial word affixe, verb morphological affix, it is described dynamic
Word morphological affix refers to be added in verb after change the verb affixe of original semanteme；
6. verb configuration affixe, specifically includes: suffix, modal particle, auxiliary verb, verb affixe string, the verb configuration word
Sew generally to be added in and do not change the originally semantic verb affixe of verb after verb, the expression of verb configuration affixe is syntactic function.Institute
Type of attachment when stating the i.e. multiple verb configuration affixes of verb affixe string while occurring.
When determining sentence boundary according to the level-one part of speech of each word unit, it is specifically as follows: judges Dan ChuifuPrevious word
Whether the level-one part of speech of unit is verb or verb configuration affixe, if it is, thinking Dan ChuifuFor sentence boundary；Otherwise,
Method based on statistical modeling predicts sentence boundary, and the statistical model such as decision-tree model obtains the sentence of Tibetan language text to be processed
Boundary.
It should be noted that in the present embodiment, double vertical symbolsCorresponding boundary is both segment boundary and sentence boundary.
And Dan ChuifuCorresponding boundary is not necessarily a boundary, it is also possible to be phrasal boundary or word boundary.
Step b) predicts part of speech of each word unit in sentence
Since level-one part of speech includes a variety of different part-of-speech informations, when such as level-one part of speech being notional word, comprising noun, adjective,
Four kinds of adverbial word, number part-of-speech informations, so that the level-one part of speech harmony tune of word unit is not also one-to-one relationship, due to more tones
When the part of speech difference of mode word, tone may also be different, tone when such as more tone recognition words are as noun and as adjective
When tone may be different.Therefore, the present embodiment, according to contextual feature of the word unit in sentence, predicts word list after subordinate sentence
The specific part of speech of member, i.e. the second level part of speech of word unit；The second level part of speech is identical as the part of speech of ordinary meaning, and such as noun describes
Word etc..The present embodiment carries out the prediction of second level part of speech, specific prediction technique and prior art phase to each word unit in each subordinate sentence
Together, as predicted using the method for statistical modeling, the second level part of speech of each word unit is obtained.
Step c) determines the type of word unit
Since word unit same in Tibetan language text is in different context environmentals, different parts of speech can be taken on, shown
Different tones, the second level part of speech range predicted as unit of sentence is larger, the second level part of speech inaccuracy for being easy to cause prediction to obtain；
In order to more accurately predict the tone information of Tibetan language text to be processed, therefore, to assure that step b) obtains the second level part of speech of each word unit
Accuracy.Further the second level part of speech of the word unit of step b) prediction can be adjusted by the type of word unit, because
This, needs first to determine the type of word unit.
In practical applications, can according to tone different context the form of expression by word dividing elements be 4 seed types, i.e., it is more
Tone recognition word, function word, affixe, conventional word, specific as follows:
1. function word: for the function word of different initial consonants in isolated syllable, tone is different；If high initial consonant is in the effect of different back word addings
Under, tone is h (height), f (drop)；For mother under the action of different back word addings, tone is l (low), r (liter) in a low voice；It is several in flow
All function word tones are all read as l or r, such as case adverbial verb, conjunction, preposition, state, modal particle all read it is lower；
2. tone recognition word more than: for more tone recognition word units when as different parts of speech, tone is different, such as:Latin transliteration is khrom skor, and tone integrated mode is hh when doing noun, and tone combines when doing verb
Mode is fr.
3. affixe: it is softly relatively similar with Mandarin Chinese with the pronunciation of weak read mode, such as ba, wa, bo, pa, po, ma, mo
Deng.
4. conventional word: there is inherent modified tone rule in tone.For example, two syllable lists are all read as the syllable of low-key l when reading
A word is formed, is not to be read as ll, but be often read as lh, i.e., dissyllabic low-key becomes to a high-profile.
In the present embodiment, the type that word unit can be obtained by way of looking into various word cell type dictionaries, may be used also
The type of word unit is obtained in a manner of by originally carrying out artificial marking types to Tibetan language and literature to be processed in advance.
Step d) adjusts the part of speech of word unit according to the type of word unit.
In the present embodiment, the part of speech of word unit is adjusted according to the type of each word unit, it is ensured that the second level word of each word unit
The accuracy of property.For example, the part of speech of word unit can be adjusted according to the type and its context-sensitive feature of word unit.In addition, also
The part of speech of word unit can be adjusted by the method for statistical modeling.Specifically, word unit part of speech adjustment can be divided into it is following several
Kind situation:
1. function word unit part of speech adjusts
The effect of function word is the different sentence element of connection, since the function word with different role is identical in form, but
Different parts of speech may be taken in actual context, i.e. function word has ambiguous category part of speech；In the present embodiment, above and below word unit
Literary environment carries out the adjustment of function word part of speech using the method for statistical modeling, to obtain the second level part of speech of accurately each word unit.
For example, case adverbial verbIt is also likely to be conjunction, case adverbial verb can be determined by the connection of ingredient each in sentenceIn sentence
The specific location of son is when case adverbial verb or to work as conjunction.
The adjustment of the word unit part of speech of tone recognition more than 2.
Using function word adjusted as boundary demarcation text fragments to be processed, part of speech tune is carried out to more tone recognition word units
It is whole, the part of speech of more tone recognition words is adjusted with specific reference to context environmental or using the method for statistical modeling.For example, more tones
Mode word unit(between two "/" labels) part of speech is different in following two sentences:
①.
Wherein,Work as verb in end of the sentence, indicates to occur, occur, tone group is combined into rl；
②.
Wherein,When noun indicates that history, biography, tone group are combined into lr in this sentence.
When part of speech adjusts, according toPlace context environmental, adjusts accordingly.
3. affixe unit and the adjustment of conventional word unit part of speech
Word unit and conventional word unit where affixe unit directly use the second level part of speech of each word unit, the present embodiment not into
The adjustment of row part of speech.It should be noted that affixe is divided into independent affixe and fixed two kinds of affixe part for depending on other word units,
In the present embodiment, the part of speech for setting independent affixe and affixe part is identical, for example, other word lists are depended in independent affixe and fixation
The part of speech of the affixe part of member can be affixe.
Step 104, the rhythm boundary of the Tibetan language text to be processed is predicted, and according to the word of rhythm boundary word unit
Property, adjust the word elementary boundary of rhythm boundary.
Rhythm boundary refers in verbal communication, in order to express semantic information, and the pause between the word and word occurred, rhythm side
The part of speech of more tone recognition words at boundary is related with rhythm boundary, and tone can change with the change of part of speech, therefore, it is necessary to
According to the part of speech of rhythm boundary word unit, the word elementary boundary of rhythm boundary is adjusted.
In the present embodiment, firstly, rhythm Boundary Prediction is carried out to text to be processed, for example, the prediction to stall position
The as prediction on rhythm boundary, specific Boundary Prediction process is same as the prior art, for example, can be according to the context phase of word unit
Information is closed, rhythm Boundary Prediction is carried out using statistical modeling method；Then, according to rhythm boundary word unit second level adjusted
Part of speech adjusts the word elementary boundary of rhythm boundary.Specifically, when the type of the word unit of rhythm boundary is more tone modelings
Formula word unit, and its second level part of speech be verb or adjective when, more tone recognition word units are split as unit of syllable, utilize
Syllable unit after fractionation carries out tone prediction.
Step 105, according to the part of speech of each word unit, to the syllable list of the Tibetan language text to be processed after adjustment word elementary boundary
Member carries out tone prediction, obtains the tone information of Tibetan language text to be processed.
In practical applications, can own using syllable unit as the load bearing unit of tone in the Tibetan language text to be processed
Each syllable unit carries out tone prediction in word unit, such as according to tone predicted characteristics, is determined by looking into tone tune type table every
The tone of a syllable unit, wherein tone predicted characteristics can be initial consonant classification, the simple or compound vowel of a Chinese syllable class of current syllable of current syllable
Not, the position before and after current syllable where the initial and the final classification of syllable, current syllable in word unit, word list where current syllable
Length, part of speech of current syllable place word unit of member etc..Furthermore, it is possible to according to the tone rule of Tibetan language pronunciation to word unit
Tone is adjusted, for example, all affixes are set as weak reading, can specifically set all syllable units of affixe as weak reading, often
With affixe such as ba, wa, bo, pa, po, ma, mo.
Further, the tone of each syllable unit obtained through this embodiment can be applied to speech synthesis field, example
Such as, the process that the present embodiment carries out tone prediction can be carried out after completing Tibetan language making character fonts, to make finally to synthesize
Tibetan voice it is more natural.
Tibetan language tone prediction technique provided in an embodiment of the present invention, by segmenting received Tibetan language text to be processed
Processing obtains each word unit, and determines the type of each word unit, then according to the type of word unit to the rhythm boundary of acquisition
Word elementary boundary is adjusted, so that carrying out tone to the Tibetan language text after adjustment word elementary boundary according to the part of speech of word unit
When prediction, it is contemplated that influence of the part of speech of more tone recognition words to the tone of rhythm boundary word unit can solve more in Tibetan language
Tone recognition word different rhythm boundaries tone different problems so that Tibetan voice synthesis it is more natural.
Correspondingly, the present invention also provides Tibetan language tone forecasting systems, as shown in Figure 2, comprising:
Receiving module 201, for receiving Tibetan language text；
Word segmentation module 202 obtains each word unit for carrying out word segmentation processing to the Tibetan language text to be processed；
Part of speech determining module 203, for the context environmental according to institute's predicate unit in the Tibetan language text to be processed
Information determines the part of speech of institute's predicate unit；
Word elementary boundary adjusts module 204, for predicting the rhythm boundary of the Tibetan language text to be processed, and according to the rhythm
The part of speech of boundary word unit adjusts the word elementary boundary of rhythm boundary；
Tone prediction module 205, for the part of speech according to each word unit, to the Tibetan language to be processed after adjustment word elementary boundary
The syllable unit of text carries out tone prediction, obtains the tone information of Tibetan language text to be processed.
In practical applications, the present embodiment determines it by context-sensitive feature of each word unit where it in sentence
Part of speech, the part of speech determining module 203 include:
Clause unit, for this progress of Tibetan language and literature subordinate sentence to be processed；
Part of speech predicting unit, for predicting part of speech of each word unit in sentence；
Word type determining units, for determining the type of each word unit；
Part of speech adjustment unit adjusts the part of speech of word unit for the type according to word unit.
Wherein, pass through the Dan Chuifu in Tibetan language text to be processedThe level-one part of speech of previous word unit passes through system
The method of meter modeling predicts sentence boundary, and the clause unit includes:
Level-one part of speech predicts subelement, for predicting the level-one part of speech of each word unit, the level-one part of speech include: verb,
Notional word, pronoun, function word, general affixe, verb configuration affixe；
First boundary determines subelement, if being used for Dan ChuifuThe level-one part of speech of previous word unit is verb
Or verb configuration affixe, then it is sentence boundary at Dan Chuifu；
Second boundary determines subelement, if being used for Dan ChuifuThe level-one part of speech of previous word unit is not
Word or verb configuration affixe then predict sentence boundary by the method for statistical modeling.
In practical applications, first boundary determines that subelement is specifically used for: obtaining each word unit by tabling look-up
Candidate level-one part of speech；Extract the context-sensitive feature of current word unit；According to the context-sensitive feature of current word unit, lead to
The method for crossing statistical modeling determines the level-one part of speech of current word unit from the candidate level-one part of speech of current word unit.Wherein, institute
State the position of the front and back word unit level-one part-of-speech information and current word unit of feature such as current word unit in sentence.
In the present embodiment, institute's predicate elementary boundary adjustment module 204 is particularly used in: when the word unit of rhythm boundary
For more tone recognition word units, and when part of speech is verb or adjective, more tone recognition word units are split as unit of syllable, benefit
Tone prediction is carried out with the syllable unit after fractionation.
Finally, it is exchanged by tone prediction module 205 according to the part of speech for each word unit that part of speech determining module 203 obtains
The syllable unit of Tibetan language text to be processed after whole word elementary boundary carries out tone prediction, obtains the tone of Tibetan language text to be processed
Information.
Certainly, in practical applications, which can also further comprise: memory module (not shown), for saving dictionary
Information, tone prediction result etc..In this way, automatically processing to facilitate to this progress of Tibetan language and literature computer to be processed, and store synthesis
Speech related information etc..
In practical applications, the system can be applied in Tibetan voice synthesis field, for example, the system 301 can be with
Making character fonts systems 302 etc. carry out Tibetan language and literature by text processing system collectively as the subsystem of text processing system 400
Present treatment, to improve the naturalness of Tibetan voice synthesis, as shown in Figure 3.
Tibetan language tone forecasting system provided in an embodiment of the present invention, by the received Tibetan language text to be processed of receiving module 201,
Word segmentation processing is carried out by word segmentation module 202, obtains each word unit, each word unit is then determined by part of speech determining module 203
Type, the rhythm boundary word elementary boundary of Tibetan language and literature sheet to be processed is adjusted according to the result of part of speech determining module 203
It is whole, when so that the system carrying out tone prediction to rhythm boundary word unit according to the part of speech of word unit, it is contemplated that more tone moulds
Influence of the part of speech of formula word to the tone of rhythm boundary word unit can solve in Tibetan language more tone recognition words on different rhythm sides
Tone different problems at boundary, so that Tibetan voice synthesis is more natural.
All the embodiments in this specification are described in a progressive manner, same and similar portion between each embodiment
Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for system reality
For applying example, since it is substantially similar to the method embodiment, so describing fairly simple, related place is referring to embodiment of the method
Part explanation.System embodiment described above is only schematical, wherein described be used as separate part description
Unit may or may not be physically separated, component shown as a unit may or may not be
Physical unit, it can it is in one place, or may be distributed over multiple network units.It can be according to the actual needs
Some or all of the modules therein is selected to achieve the purpose of the solution of this embodiment.Those of ordinary skill in the art are not paying
In the case where creative work, it can understand and implement.
The embodiment of the present invention has been described in detail above, and specific embodiment used herein carries out the present invention
It illustrates, the above description of the embodiments is only used to help understand the method and apparatus of the present invention；Meanwhile for the one of this field
As technical staff, according to the thought of the present invention, there will be changes in the specific implementation manner and application range, to sum up institute
It states, the contents of this specification are not to be construed as limiting the invention.
Claims (11)
1. a kind of Tibetan language tone prediction technique characterized by comprising
Receive Tibetan language text to be processed；
Word segmentation processing is carried out to the Tibetan language text to be processed, obtains each word unit；
According to context information of institute's predicate unit in the Tibetan language text to be processed, the word of institute's predicate unit is determined
Property；
It predicts the rhythm boundary of the Tibetan language text to be processed, and according to the part of speech of rhythm boundary word unit, adjusts rhythm side
Word elementary boundary at boundary；
According to the part of speech of each word unit, it is pre- that tone is carried out to the syllable unit of the Tibetan language text to be processed after adjustment word elementary boundary
It surveys, obtains the tone information of Tibetan language text to be processed.
2. the method according to claim 1, wherein it is described according to institute's predicate unit in the Tibetan language and literature to be processed
Context information in this, determines that the part of speech of institute's predicate unit includes:
To this progress of Tibetan language and literature subordinate sentence to be processed；
Predict part of speech of institute's predicate unit in sentence；
Determine the type of institute's predicate unit；
According to the type of institute's predicate unit, the part of speech of institute's predicate unit is adjusted.
3. according to the method described in claim 2, it is characterized in that, described include: to this progress of Tibetan language and literature subordinate sentence to be processed
Predict that the level-one part of speech of each word unit, the level-one part of speech include: verb, notional word, pronoun, function word, general affixe, verb
Configuration affixe；
It is sentence side at Dan Chuifu if the level-one part of speech of the previous word unit of Dan Chuifu " | " is verb or verb configuration affixe
Boundary；
If the level-one part of speech of the previous word unit of Dan Chuifu " | " is not verb or verb configuration affixe, pass through statistical modeling
Method predict sentence boundary.
4. according to the method described in claim 3, it is characterized in that, the level-one part of speech of each word unit of prediction includes:
Obtain the candidate level-one part of speech of each word unit；
Extract the context-sensitive feature of current word unit；
According to the context-sensitive feature of current word unit, by the method for statistical modeling from the candidate primary word of current word unit
Property in determine current word unit level-one part of speech.
5. according to the method described in claim 2, it is characterized in that, the type of institute's predicate unit includes any of the following or more
Kind: more tone recognition words, function word, affixe, conventional word.
6. the method according to claim 1, wherein the part of speech according to rhythm boundary word unit, adjustment
The word elementary boundary of rhythm boundary includes:
It is single with syllable when the word unit of rhythm boundary is more tone recognition word units, and part of speech is verb or adjective
Position splits more tone recognition word units, carries out subsequent tone prediction using syllable unit after fractionation.
7. a kind of Tibetan language tone forecasting system characterized by comprising
Receiving module, for receiving Tibetan language text to be processed；
Word segmentation module obtains each word unit for carrying out word segmentation processing to the Tibetan language text to be processed；
Part of speech determining module, for the context information according to institute's predicate unit in the Tibetan language text to be processed, really
Determine the part of speech of institute's predicate unit；
Word elementary boundary adjusts module, for predicting the rhythm boundary of the Tibetan language text to be processed, and according to rhythm boundary
The part of speech of word unit adjusts the word elementary boundary of rhythm boundary；
Tone prediction module, for the part of speech according to each word unit, to the Tibetan language text to be processed after adjustment word elementary boundary
Syllable unit carries out tone prediction, obtains the tone information of Tibetan language text to be processed.
8. system according to claim 7, which is characterized in that the part of speech determining module includes:
Clause unit, for this progress of Tibetan language and literature subordinate sentence to be processed；
Part of speech predicting unit, for predicting part of speech of each word unit in sentence；
Word type determining units, for determining the type of each word unit；
Part of speech adjustment unit adjusts the part of speech of word unit for the type according to word unit.
9. system according to claim 8, which is characterized in that the clause unit includes:
Level-one part of speech predicts subelement, for predicting the level-one part of speech of each word unit, the level-one part of speech include: verb, notional word,
Pronoun, function word, general affixe, verb configuration affixe；
First boundary determines subelement, if the level-one part of speech for Dan Chuifu " | " previous word unit is verb or verb
Configuration affixe is then sentence boundary at Dan Chuifu；
Second boundary determines subelement, if the level-one part of speech for Dan Chuifu " | " previous word unit is not verb or moves
Word configuration affixe then predicts sentence boundary by the method for statistical modeling.
10. system according to claim 9, which is characterized in that first boundary determines that subelement is specifically used for: logical
It crosses to table look-up and obtains the candidate level-one part of speech of each word unit；Extract the context-sensitive feature of current word unit；According to current word list
The context-sensitive feature of member determines current word list by the method for statistical modeling from the candidate level-one part of speech of current word unit
The level-one part of speech of member.
11. system according to claim 7, which is characterized in that institute's predicate elementary boundary adjustment module is specifically used for: working as rhythm
Restrain boundary word unit be more tone recognition word units, and part of speech be verb or adjective when, as unit of syllable split it is more
Tone recognition word unit carries out subsequent tone prediction using syllable unit after fractionation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510325742.6A CN106294310B (en) | 2015-06-12 | 2015-06-12 | A kind of Tibetan language tone prediction technique and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510325742.6A CN106294310B (en) | 2015-06-12 | 2015-06-12 | A kind of Tibetan language tone prediction technique and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106294310A CN106294310A (en) | 2017-01-04 |
| CN106294310B true CN106294310B (en) | 2019-05-03 |
Family
ID=57650104
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510325742.6A Active CN106294310B (en) | 2015-06-12 | 2015-06-12 | A kind of Tibetan language tone prediction technique and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106294310B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109523992A (en) * | 2018-11-28 | 2019-03-26 | 鲁东大学 | Tibetan dialect speech processing system |
| CN112735378B (en) * | 2020-12-29 | 2024-05-31 | 科大讯飞股份有限公司 | Thai speech synthesis method, device and equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101650942A (en) * | 2009-08-26 | 2010-02-17 | 北京邮电大学 | Prosodic structure forming method based on prosodic phrase |
| CN103035241A (en) * | 2012-12-07 | 2013-04-10 | 中国科学院自动化研究所 | Model complementary Chinese rhythm interruption recognition system and method |
| CN103165126A (en) * | 2011-12-15 | 2013-06-19 | 无锡中星微电子有限公司 | Method for voice playing of mobile phone text short messages |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9037460B2 (en) * | 2012-03-28 | 2015-05-19 | Microsoft Technology Licensing, Llc | Dynamic long-distance dependency with conditional random fields |
-
2015
- 2015-06-12 CN CN201510325742.6A patent/CN106294310B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101650942A (en) * | 2009-08-26 | 2010-02-17 | 北京邮电大学 | Prosodic structure forming method based on prosodic phrase |
| CN103165126A (en) * | 2011-12-15 | 2013-06-19 | 无锡中星微电子有限公司 | Method for voice playing of mobile phone text short messages |
| CN103035241A (en) * | 2012-12-07 | 2013-04-10 | 中国科学院自动化研究所 | Model complementary Chinese rhythm interruption recognition system and method |
Non-Patent Citations (1)
| Title |
|---|
| 藏语声调形成的过程与社会历史系统状态;江荻;《藏学期刊第二辑》;20031231;第185-190页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106294310A (en) | 2017-01-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107818085B (en) | Answer selection method and system for reading understanding of reading robot | |
| Protopapas et al. | IPLR: An online resource for Greek word-level and sublexical information | |
| CN105989833B (en) | Multilingual mixed this making character fonts of Chinese language method and system | |
| Goslin et al. | PhonItalia: a phonological lexicon for Italian | |
| JP2008225963A (en) | Machine translation device, replacement dictionary creating device, machine translation method, replacement dictionary creating method, and program | |
| CN106294310B (en) | A kind of Tibetan language tone prediction technique and system | |
| Abdul-Mageed et al. | Asma: A system for automatic segmentation and morpho-syntactic disambiguation of modern standard arabic | |
| Bar-Haim et al. | Choosing an optimal architecture for segmentation and POS-tagging of Modern Hebrew | |
| CN105895076B (en) | A kind of phoneme synthesizing method and system | |
| Singha et al. | Part of speech tagging in Manipuri with hidden markov model | |
| Ibrahim et al. | Bel-Arabi: advanced Arabic grammar analyzer | |
| Horváth et al. | Language technology resources and tools for Mansi: an overview | |
| CN106294311B (en) | A kind of Tibetan language tone prediction technique and system | |
| KR102372629B1 (en) | Triple Extraction method using Pointer Network and the extraction apparatus | |
| CN105895075A (en) | Method and system for improving synthetic voice rhythm naturalness | |
| Mamateli et al. | Morphological analysis based part-of-speech tagging for uyghur speech synthesis | |
| Saychum et al. | Efficient Thai Grapheme-to-Phoneme Conversion Using CRF-Based Joint Sequence Modeling. | |
| CN110362803A (en) | A kind of text template generation method based on the combination of domain features morphology | |
| Ahmadi et al. | Providing a suitable method for allophonic labeling of speech corpuses according to the IPA system | |
| Mansour | Morphtagger: Hmm-based arabic segmentation for statistical machine translation | |
| Declerck et al. | How to semantically relate dialectal Dictionaries in the Linked Data Framework | |
| Mahmoud et al. | Uyghur stemming using conditional random fields | |
| Ning | Chinese prosodic phrase prediction based on shallow semantic features | |
| Goyal et al. | Automatic standardization of spelling variations of Hindi text | |
| CN112988965B (en) | Text data processing method and device, storage medium and computer equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |