CN1019233B - Chinese characters transforming mode - Google Patents
Chinese characters transforming modeInfo
- Publication number
- CN1019233B CN1019233B CN 89102915 CN89102915A CN1019233B CN 1019233 B CN1019233 B CN 1019233B CN 89102915 CN89102915 CN 89102915 CN 89102915 A CN89102915 A CN 89102915A CN 1019233 B CN1019233 B CN 1019233B
- Authority
- CN
- China
- Prior art keywords
- sound
- speech
- mentioned
- splittable
- speech sound
- 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.)
- Expired
Links
Images
Landscapes
- Document Processing Apparatus (AREA)
Abstract
To obtain a phonetic character KANJI(Chinese character) conversion system suitable for Chinese language by performing word tone KANJI conversion based on a word tone separating system in which a tone frequency method is applied. When a syllable is inputted from a syllable input means 11, a word tone and tone frequency retrieval means 2 retrieves data from a dictionary 1, and sends it to a storage means 3. A syllable judging means 4 judges a node where a tone frequency phrase ends. An optimum word tone separation type generating means 5 generates a word tone separation type setting the tone frequency phrase as an object, and sends it to a separation type storage means 6, and also, generates the optimum word tone separation type, and sends it to a KANJI string conversion means 7. The KANJI string conversion means 7 receives data of the optimum word tone separation type from the optimum word tone separation type generating means 5, and retrieves the KANJI vocabulary of the dictionary 1 sequentially, setting each separated word tone as a header, and selects a KANJI word judged as the word nearest to a targeted word out of homophony words, and outputs a KANJI word string by connecting the KANJI words. The KANJI word string is sent to a KANJI word string storage means 8 and a document storage means 12.
Description
The present invention is a kind of Chinese character mapping mode that can be applied to aspects such as Chinese robot calculator or Chinese text processor, and particularly a kind of syllable string that Chinese pinyin sound sign indicating number is imported is transformed into the Chinese character mapping mode of Chinese character speech string automatically.
As the input in Chinese mode, existing a kind of technology of importing the syllable string with the phonetic sound sign indicating number that omits tone.For example, the cooperative development of Nippon Electric Company, Ltd. and Guangzhou Electronics Technology Inst., Chinese Academy of Sciences of Inst. of Computing Techn. Academia Sinica, on January 29th, 1988 " MC formula input in Chinese system " (the seeing the Peoples Daily on February 1st, 1988) by the national technical appraisement of Chinese Government be like this.This input in Chinese system adopts and replaces the two-combats mode about the initial and the final Two bors d's oeuveres, is unit input Chinese word sound with the syllable, can import at a high speed.
What this input in Chinese system adopted is the Chinese character mapping mode that the syllable string of input can be transformed into automatically the Chinese character speech.This Chinese character mapping mode comprises two parts, and the one, be the dictionary of the Chinese character speech of each speech sound of index stores with the Chinese word sound, the 2nd, can respond input and from the Chinese character speech of dictionaries store, access the treating apparatus of one of them.The speech sound here is meant " pronunciation " of each word in the Chinese.What this treating apparatus used is a kind of identical mode of using with the japanese type processor of the longest consensus method.The longest so-called consensus method is exactly that the feature of " agglutinative language " of the Japanese that utilizes on the linguistics to be said is carried out the method that suitable speech cent cuts automatically to the syllable string.Thereby in the japanese type processor, the longest consensus method has been made very big contribution, and it promptly greatly reduces the access times of transfer key by the number of times that manpower is cut apart the syllable string.
But " isolating language " that Chinese belongs on the linguistics to be said adopts the longest consensus method or other similar fashion, will produce many syllable string speech cents and cut improper problem.So the Chinese character conversion accuracy in the above-mentioned Chinese character mapping mode can not be satisfactory.
Perhaps can frequently use transfer key to carry out the speech cent when the input operation cuts, thereby to the purpose that improves Chinese character conversion accuracy, but Chinese is the form of enumerating with Chinese character to be constituted, and the cut-point between speech and the speech is difficult to clear and definite, cuts apart still and has any problem so carry out the syllable string by manpower.
The objective of the invention is to provide a kind of Chinese character mapping mode that can automatically the syllable string of Chinese pinyin sound sign indicating number form input be transformed into Chinese character string with quite high accuracy.
Another object of the present invention is that a kind of Chinese character mapping mode that just can automatically the syllable string of Chinese pinyin sound sign indicating number form input be transformed into Chinese character string of handling through the short time will be provided.
The present invention is a kind of Chinese character mapping mode, wherein comprises two parts, the one, and be the dictionary of the Chinese character speech of each speech sound of index stores with the speech sound of Chinese, the 2nd, can access the treating apparatus of one of the Chinese character speech of this dictionaries store.
Feature of the present invention is as follows:
It is index with the Chinese word sound that above-mentioned dictionary contains an energy, according to the speech sound of above-mentioned Chinese character speech, and the frequency level memory storage of storage monosyllabic word sound and disyllabic word voice grade.Above-mentioned treating apparatus comprises that then speech voice grade indexing unit, speech voice grade memory storage, optimum speech cent cut generating apparatus and Chinese character word and search device.
Last predicate voice grade indexing unit can be according to the monosyllabic word sound of just having imported in the syllable string of having imported already, retrieves the frequency level memory storage of above-mentioned dictionary and accesses the frequency level of this monosyllabic word sound; Simultaneously can also be according to this monosyllabic word sound disyllabic word sound that these two monosyllabic word sound syllables of monosyllabic word sound that it imports previously constitute that coexists, retrieve above-mentioned dictionary and access the frequency level of this disyllabic word sound.
Last predicate voice grade memory storage can be stored above-mentioned two monosyllabic word sounds speech sound separately and frequency level, and the speech sound of above-mentioned disyllabic word sound and frequency level.
When can not retrieving above-mentioned disyllabic word sound, just the front with the above-mentioned monosyllabic word sound of just having imported is considered as node.In the syllable string that the above-mentioned monosyllabic word sound of just having imported is imported previously, this node just constitutes an information processing sentence before with the node that the front has obtained.The effect that last predicate voice grade accesses device is to access the speech sound of monosyllabic word sound in this information processing sentence and disyllabic word sound and level frequently from last predicate voice grade memory storage.
Each Splittable that above-mentioned optimum speech sound Splittable generating apparatus can combine according to monosyllabic word sound and disyllabic word sound in the above-mentioned information processing sentence, obtain the monosyllabic word sound of each Splittable of formation and the frequency level summation of disyllabic word sound, and generate the optimum speech sound Splittable of above-mentioned information processing sentence according to this summation.
Each speech sound that above-mentioned Chinese character word and search device can split according to above-mentioned optimum speech sound Splittable is retrieved above-mentioned dictionary, one of Chinese character speech that to access with this speech sound be index.
Adopt the present invention can obtain a kind of can pin-point accuracy ground, high response speed ground syllable string that Chinese pinyin sound sign indicating number form is imported is transformed into the Chinese character mapping mode of Chinese character automatically.
Fig. 1 is the time diagram of the explanation principle of the invention.
Fig. 2 represents each speech sound and the quantity of information thereof on the example sentence speech sound string.
The network (1) (2) of the speech sound string in Fig. 3 presentation graphs 2 and Chinese character transformation results (3).
Fig. 4 represents the calcspar of Chinese character mapping mode of the present invention.
The memory contents complete list of each memory storage that Fig. 5 is comprised for the Chinese character mapping mode among Fig. 4.
Fig. 6 represents the speech sound Splittable that syllable input sequence number may occur at from 1 to 7 o'clock.
Fig. 7 represents the speech sound string of Fig. 2 is carried out the example explanation that the speech cent cuts processing.
Fig. 8 represents the tree structure of speech sound Splittable.
Fig. 9 (1) expression is carried out another example explanation that the speech cent cuts processing to speech sound string.
The speech cent of Fig. 9 (2) expression and Fig. 9 (1) cuts the explanation of the speech sound Chinese character conversion process that processing carries out synchronously.
Figure 10 is a speech sound Splittable complete list.
The explanation that node is aimed in Figure 11 (1) (2) expression.
Figure 12 is positive Splittable complete list.
Splittable generative process key diagram when Figure 13 imports for syllable.
Figure 14 is positive Splittable generative process key diagram.
The key diagram that Figure 15 imports the preceding of sequence number and then changes along with syllable for speech sound Splittable.
Figure 16 represents to handle the Chinese character conversion process key diagram that carries out synchronously with the speech message breath of positive Splittable.
Figure 17 (a) (b) represents Chinese character mapping mode process flow diagram by other examples of the present invention embodied.
Inventive principle at first is described.Here " the speech sound syllable frequency of occurrences in the language statistics data " abbreviated as frequency.If establishing the syllable number of certain speech sound is m, the frequency in the language statistics data is n, and when being N, then frequency P is total syllable number of objects of statistics (total Chinese character number):
P＝（m×n）/N
Below, the objects of statistics syllable number of " modern Chinese frequency dictionary " that N employing Beijing Language Institute writes, with N=1,807,398 are illustrated.
Speech sound quantity of information I can use following formulate:
I＝-log
aP
Here, the end of logarithm is any arithmetic number.Below, the end of establishing logarithm is 2, the logarithm brief note of frequency P is logP.So draw:
I＝-logP
When reality was used I, radix point can be cast out later, makes it to become integer, and this is level frequently.
Lift the object lesson explanation below.
According to the speech occurrence frequency data in Beijing Language Institute's " modern Chinese frequency dictionary " table one " press alphabetical sequence arrange frequency vocabulary ", the disyllabic word and the frequency thereof that belong to speech sound jishu are:
The speech technology is recorded and narrated odd number radix product progression and is added up to
Frequency 509 21000 602 times
P＝（m×n）/N＝（2×602）/1807398＝0.0006662
I=-log P=10,608 → 10(level)
The frequency level that is speech sound jishu is 10 grades.
Use same calculating, the frequency level that can draw speech sound ji is 8 grades, and the frequency level of speech sound shu is 9 grades, and the frequency level of speech sound xing is 9 grades, and the frequency level of speech sound jishu is 10 grades, and the frequency level of speech sound shuxing is 18 grades.
The speech voice grade network that utilizes these data can draw shown in Figure 1.In Fig. 1, the numerical value below the speech sound is frequency level separately.
Path from start to end has following three, the frequency level on the path and as follows respectively:
Path speech sound string frequently level and calculate frequently level and
（1） ji/shuxing 8+18＝26
（2） ji/shu/xing 8+9+9＝26
（3） jishu/xing 10+9＝19
Wherein, the frequency level and the minimum of path (3).That is to say that the speech cent of this speech sound string cuts, is that the highest speech cent of this syllable string probability is cut.This result is in full accord with the situation of Modern Chinese.
Fig. 2 is representing the speech sound, frequency of single disyllabic word sound that may exist on the example sentence syllable string, level and Chinese vocabulary frequently respectively.
This input in Chinese of example sentence system has many advantages
Syllable string zhe zhong zhong wen shu ru xi tong ju you xu duo you dian
Go on to say below with reference to Fig. 3.Alphabetical sequence a represents monosyllabic word sound to n.The disyllabic word sound is represented in the combination of two alphabetical sequences of center horizontal line bottom and upper segment.In addition, the following added numeral frequency of these monosyllabic word sounds and disyllabic word sound level.
So-called " speech message breath is handled sentence " refers to and might carry out the syllable string that speech message breath is cut apart.Here, " non-speech message breath is handled sentence " refers to " the non-conversion sentence " that there is no need to change into from the speech change of tune Chinese character speech, refers to arabic numeral, Roman capitals, pure Chinese numbers and punctuation mark or the like.Give an example, the pure Chinese numbers here refers to " 29 " in " 20 No. nine " and so on." one " in " generally " " necessarily " " single-minded " then is considered as the part of speech, does not make pure Chinese numbers and handles.Then cease the sentence separate processes as for three speech sounds more than the syllable, use well-known mode conversion to become Chinese character, no longer specify here with the speech message.
Simultaneously, the each point that meets following (1) to (4) definite condition is called " node " respectively.
(1) the tie point position of information processing sentence and non-information processing sentence
(2) beginning of the sentence position
When (3) certain the monosyllabic word sound syllable in the information sentence can not constitute the disyllabic word sound with the monosyllabic word sound syllable of its front, the position before these two monosyllabic word sounds
Consider that from these nodes the speech sound string among Fig. 3 (1) just can be divided into four speech message breaths and handle sentence.
The speech voice grade network that Fig. 3 (2) expression is done according to the speech sound string of Fig. 3 (1).In this speech voice grade network, black line is represented level and minimum path frequently.
Fig. 3 (3) expression carries out to example sentence that the speech cent cuts according to the black line on the speech voice grade network of Fig. 3 (2) and the Chinese character transformation results of homonym after selecting.Select by homonym, selected the maximum speech of level frequently in the single syllable from Fig. 2 and disyllabic each homonym vocabulary (by frequency level series arrangement).
From Fig. 3 (3), be that the Chinese character conversion of zhong is not transformed into straight correct " kind " to speech sound b only, be transformed into " in ", that is to say homonym erroneous conversions phenomenon has taken place.But, but obtained right-on result aspect cutting at the speech cent.
Fig. 4 represents the calcspar of the Chinese character mapping mode that first example according to the present invention carries out.This Chinese character mapping mode has been used above-mentioned principle, comprises dictionary 1.Dictionary 1 is the Chinese character speech that index stores each speech sound with the speech sound of Chinese.Dictionary 1 contains a level memory storage frequently.Frequently level memory storage part is monosyllabic word sound in the index stores Chinese character speech speech sound and disyllabic word sound and level frequently thereof with the speech sound of Chinese.Specifically, dictionary 1 has the data of following (1) to (3):
(1) with the syllable is the speech sound index that unit is recorded and narrated
(2) the frequency level of each speech sound
(3) the corresponding unisonance Chinese character word of each speech sound (is that kanji code is recorded and narrated with Chinese character coding set with the GB message exchange)
Example:
(1) speech sound index ji shu
(2) the frequency level 10
(3) vocabulary (position) 28284285(technology) 28394286(records and narrates) the 38704293(odd number)
In addition, this Chinese character mapping mode also comprises a treating apparatus that can access one of Chinese character speech that dictionary 1 stored.As the following explanation of being done successively, this treating apparatus comprises speech voice grade indexing unit 2, speech voice grade memory storage 3, speech voice grade and accesses device 4, optimum speech sound Splittable generating apparatus 6 and Chinese character word and search device 7.
Speech voice grade indexing unit 2 during from syllable input medias such as keyboard 11 input, can retrieve the data of following (1) to (6) at syllable from dictionary 1, and it is sent into speech voice grade memory storage.
(1) the monosyllabic word sound RS(u of this syllable)
(2) frequency level IS(u RS(u))
(3) when u be 2 whether exist when above connect RS(u-1) and RS(u) disyllabic word sound RD(u) when (but, not having RD(u), no RD(u) sign send into the speech voice grade and access device 4)
(4) have RD(u) time RD(u) and level ID(u frequently)
(5) when u be 4 when above, whether exist to connect RS(u-2), RS(u-1) and trisyllable sound RT(u RS(u)), the RT(u when existing)
(6) when u be 4 when above, whether exist to connect RS(u-3), RS(U-2), RS(U-1) and quadrisyllable sound RQ(u RS(u)), the RQ(u when existing)
(prerequisite of the explanation of doing here is not store u in the dictionary 1 at the speech more than 5.In Chinese, the speech that pentasyllable is above mostly is synthetic by the speech sound below four syllables.For example, technical demonstration match=technology+performance+match, perhaps ,=technology+demonstration match.So the speech sound of u＞4 needn't deposit dictionary in)
When there not being RT(u) and RQ(u) time, sign is sent into the speech voice grade and is accessed device 4; When existing, speech sound RT(u) or RQ(u) send into speech voice grade memory storage 3.When there being RQ(u) time, RT(u), RD(u), RS(u), ID(u) and IS(u) can not send into speech voice grade memory storage 3.When there being RT(u) time, RD(u), RS(u), ID(u) and IS(u) can not send into speech voice grade memory storage 3.
The memory contents of speech voice grade memory storage 3 is as follows, specifically numbers 3 hurdles referring to the calcspar of Fig. 5.
Monosyllabic word sound RS(u) and frequently level IS(u)
Disyllabic word sound RD(u) and frequently level ID(u)
Trisyllable sound RT(u)
Quadrisyllable sound RQ(u)
The speech voice grade accesses device 4 and also can be considered the node judgment means, the node that will mention below it can be judged, and carry out following processing:
(1) in the time of through the retrieval of speech voice grade indexing unit 2, if there is no speech sound RQ(u) or RT(u), just carry out following processing:
1) do not have RD(u) time: with RS(u) point of speech sound syllable front is judged to be node, and sequence number imported in syllable be reset to 1.And then implement RS(u) become RS(1), with IS(u) become IS(1) processing.
2) have RD(u) time: with RS(u), IS(u), RD(u) and content ID(u) send into optimum speech sound Splittable generating apparatus 6 from speech voice grade memory storage 3.
(2) through the retrieval of speech voice grade indexing unit 2, if there is speech sound RQ(u) or RT(u) (RQ(u) preferential retrieval come out) time, just carry out following processing:
1) have RQ(u) time: will import sequence number is RS(u-3) the syllable front handle as the node position.That is to say that assert that this (u-4) the individual syllable string from input sequence number 1 to (u-4) only is made of monosyllabic and disyllabic word, four syllables from sequence number (u-3) to u then are speech sound RQ(u) speech, and carry out later processing according to this.
2) do not have RQ(u) but have a RT(u) time: will import sequence number and handle as the node position for the syllable front of (u-2).That is to say, the syllable string of identification from input sequence number 1 to (u-3) the individual syllable of (u-3) only is made of monosyllabic and disyllabic word, from input sequence number (u-2) to three syllables of u are speech sound RT(u) speech, and carry out later processing with this.
To sum up, exactly when can not retrieving the disyllabic word sound, just assert that the front of the monosyllabic word sound of just having imported is a node.So in the syllable string of input before this monosyllabic word sound of just having imported, above-mentioned node and the node that has obtained before that just constitute one before and handle sentence.The effect that the speech voice grade accesses device 4 is to access the speech sound of monosyllabic word sound and disyllabic word sound in the above-mentioned processing sentence and level frequently from speech voice grade memory storage 3.6 of optimum speech sound Splittable generating apparatus are object with specific " an information processing sentence ", optimum speech sound Splittable concatenated in the syllable of this information processing sentence, and send in Chinese character word and search device 7 and the optimum speech sound Splittable memory storage 8 and go.Optimum speech sound Splittable generation method treats that the back explains again.
Search device of Chinese character 7 can be accepted optimum speech sound Splittable data from optimum speech sound Splittable generating apparatus 6, and is index with each separated speech sound, the Chinese character word of retrieve stored in dictionary 1.And can index homonym in the Chinese character word, selected certificate is recognized now the most near actual Chinese character speech, and each Chinese character speech that will select couples together and sends into next section treating apparatus with the Chinese character string form.Chinese character string is sent into Chinese character string memory storage and file storage device 12.
In addition, the technology of selected optimum Chinese character speech also has other modes from homonym, for example utilizes the method for statistics frequency of speech and what is called " arrangement of self-study formula speech " method or the like.These modes are the well-known technology of widespread use, no longer explanations here on the Japanese word processor.
In addition, judge when speech voice grade indexing unit 2 to have RQ(u) or RT(u) time, the input sequence number refund 4 or the speech sound string of 3 o'clock optimum speech sound Splittable must be with quadrisyllable sound RQ(u) or trisyllable sound RT(u) couple together.So the optimum segmentation type when being necessary input sequence number from (u-1) to (u-4) is preserved.Optimum speech sound Splittable memory storage 8 just has this function.
The memory contents of optimum speech sound Splittable memory storage 8 and Chinese character string memory storage 9 sees square sequence number 8 and 9 hurdles among Fig. 5 for details.
The function and the effect of the optimum speech sound Splittable generating apparatus 6 that explanation now is shown in Figure 4.
As previously mentioned, so-called optimum speech sound Splittable is meant speech sound string is cut apart and the frequency level of each speech sound of obtaining and be the type of minimum value.
Finding a kind of reasonable and succinct method that can find out the optimum speech sound Splittable of any syllable string, is the key of specific implementation speech message breath facture.Explain two kinds of methods below.First method is called " registration computing method ", exactly whole speech sound Splittables that may occur are quantized, register in advance, each input syllable all comply with each Splittable the speech of a level substitution frequently sound calculate frequently level with, and to each level and comparing frequently, therefrom find out have minimum level frequently and the optimum segmentation type.Second method is called " progressively method of formation ", and the input one by one according to syllable generates the optimum segmentation type continuously exactly.
" registration computing method "
Fig. 6 represents whole speech sound Splittables that syllable input sequence number u may occur at from 1 to 7 o'clock.But, be that the suppositive sound has only single syllable and double-tone to save two kinds.Among the figure, each syllable represents with Roman capitals such as abc that by input sequence the speech cent cuts usefulness "/" expression.But the Roman capitals here do not have direct relation with the Roman capitals of each syllable of expression example sentence among Fig. 2 and Fig. 3.T(u) be the speech sound Splittable of binary format.About T(u) implication, for instance, if the speech sound Splittable ab/cd/e of the 1st row then be binary one 0101 during u=5 among Fig. 5.Can be listed as into following table contrasts:
Speech sound Splittable ab/cd/e
Binary digit T 10101
The beginning of T 1 expression syllable string foremost.Below, do not exist the speech cent to cut between the binary digit 0 expression syllable, binary digit 1 expression exists the speech cent to cut.The end of T had not both had 1 not have 0 yet, and e can constitute speech sound ef or e/f after this expression can't be determined next syllable f input now.
This T(u) is called " Splittable variable ".Use T(u), can record and narrate the Splittable of information sentence compactly, simultaneously can be each T(u of each u) be registered in the speech sound Splittable memory storage 8 shown in Figure 4 and go.When the syllable input is arranged, optimum speech sound Splittable generating apparatus 6 shown in Figure 4 just can be as level and calculation element are started working frequently, calculate the frequency level and the ∑ I of each speech sound of each speech sound string of each Splittable that constitutes present u respectively, and then start working as the Splittable determination device again, the type that has minimum value among each ∑ I is obtained as the optimum segmentation type.
Fig. 7 represents with registering the processing example that computing method is done.Example sentence is the same with Fig. 3, is " zhe zhong zhong wen shu ru xi long " (this input in Chinese system).Optimum segmentation type when the Splittable among the figure shown in the black line is each u.It should be " zhe/zhong/zhongwen/shuru/xitong " that correct speech cent cuts, and obtained right-on result visible above-mentioned cutting apart.The next line of Fig. 7 is represented the speech sound Chinese character transformation results of each syllable input sequence number.The Chinese character conversion has become the Chinese character speech that has the highest statistics frequency in the homonym.From the result that last u equaled at 8 o'clock, zhong does not become " kind ", " in " but become by mistake, in addition, does not have other unisonance mistakes.
" progressively method of formation "
Fig. 8 represent U be 1 to 6 one of any the time, whole Splittables that may occur when allowing till the speech sound syllable length to 6 of speech message breath process object.A pass that amount M imports sequence number U with syllable that needs to handle is M=2
U-1By each U whole speech sounds all be the speech sound string that constitutes by monosyllabic word sound (for example U then was a/b/c/d/e/f at 6 o'clock) be placed on topmost delegation syllable number (equaling syllable input sequence number) for the speech sound of U only one speech sound string (for example U be 6 be abcdef) be placed on delegation bottom, by the speech sound string Splittable in the middle of following rule conversion successively, just can obtain this table again.
In the middle of Fig. 8, it from U 1 o'clock speech sound string " a ", successively from left to right, simultaneously from top to bottom, look along the arrow direction, just be appreciated that when whole speech sounds of the syllable number that may exist existed, how the form of speech sound Splittable in an organized way generated out when each syllable is imported.Just as shown in this figure, all speech sound Splittables constituted one well-ordered " speech sound Splittable tree-structured " (among the figure, mark+d ,+de etc. do not represent to import d, de, represents the process that each Splittable changes successively).
In a Splittable with U syllable, the end is 2 with the quantity q of the type that p syllable speech sound finishes
U-p-1(but q is greater than 0).For example, in the Splittable of six syllable number, the type that finishes with two syllable speech sounds is q=8.In Fig. 8, when U=6, p=2, q=8 type is arranged under the a/b/c/d/ef continuously, and they add at each end of U=4 type that all disyllabic word sound ef draws.So, if found a corresponding optimum type of 8 types during with U=4, just in 8 types that the ef with U=6 finishes, show so minimum level frequently and type have only one, that is exactly to add ef on the optimum type when U=4.Finally can both decide a unique type, this situation is all generally set up any U and any p.For example, in the middle of the type of U=7, be on the optimum type of U=5, to add f with the optimum type in the middle of the type of f end; With the optimum type in the middle of the type of def end is to add def on the optimum type of U=3; With the optimum type in the middle of the type of cdef end is to add cdef on the optimum type of U=2.Therefore, can draw following results to whole " speech sound Splittable tree-structured ":
If U is since 1, the every increase by 1 of U, just progressively remove to ask the optimum segmentation type one time, so as long as from the best candidate type of the speech sound syllable number equivalent amount of necessity, select level and the most small-sized frequently.
Thereby, can draw following very simple clear conclusions, if promptly p=4(is because when carrying out Language Processing to Chinese, the syllable number maximum of a speech is decided to be 4 just enough), U can select level frequently and a minimum type as the optimum segmentation type at 1 to 3 o'clock from U candidate's type so, and if U be not less than can select level frequently and minimum type as the optimum segmentation type at 4 o'clock from 4 candidate's types.
Fig. 9 (1) expression p is 4, N be carried out in 1 to 6 o'clock four select minimum level frequently of the same form and type choice Processing Algorithm.Each array variable is defined as follows among the figure: IS, ID, IT and IQ represent with the syllable of just having imported to be the frequency level of single syllable, double-tone joint, triphone and the quadrisyllable sound at end respectively; PS, PD, PT and PQ represent respectively with single syllable, double-tone joint, triphone and quadrisyllable sound finish have minimum separately frequency level and each speech sound string the frequency level and; P be frequency level minimum in these four speech sound strings and, the numeral in the bracket is syllable input sequence number U.
Fig. 9 (2) represents to press each section of U, carries out the algorithm of speech sound Chinese character conversion in minimum level frequently of carrying out Fig. 9 (1) and calculating.Each array variable is defined as follows among the figure: HS, HD, HT and HQ represent it is believed that in the Chinese character speech that has IS, ID, IT and IQ speech sound the most definite Chinese character speech on the statistics respectively; That KS, KD, KT and KQ then represent respectively to finish with HS, HD, HT and HQ, have minimum separately level frequently and the conversion Chinese character speech string of each speech sound string; K represents and the corresponding Chinese character speech of P string that the numeral in the bracket is U.
So, just can not use Splittable variable T fully, but obtain the speech sound the frequency level and, it is compared, through repeated calculation, the speech sound string of progressively obtaining the optimum segmentation type along with advancing of U, and carry out the conversion of speech sound Chinese character synchronously.
Specify below.All the above-mentioned frequency level and the ∑ I of Splittable are stored among the frequency level and memory storage that comprises in the optimum speech sound Splittable generating apparatus 6 shown in Figure 4.When the syllable input was whenever arranged, optimum speech sound Splittable generating apparatus 6 was just started working as speech sound input detection device, and output speech sound input detecting signal.Receive after this speech sound input detecting signal, optimum speech sound Splittable generating apparatus 6 is just started working as level and comparison means frequently, accesses the frequency level and the ∑ I of whole Splittables from level and memory storage frequently, and these frequencies grade and ∑ I are compared.Frequently the Splittable of level and ∑ I minimum is decided to be optimum speech sound Splittable at last.
" second example of Chinese character transformation mode "
Be illustrated with regard to the employed Chinese character mapping mode of second example of the present invention below.In this Chinese character mapping mode, except that above-mentioned node, also use the notion of " accurate node ".If the frequency level of a disyllabic word sound is not less than the frequency level sum of the monosyllabic word sound of the frequency level of monosyllabic word sound of its first syllable and second syllable, just the position between these two monosyllabic word sounds is called accurate node so.For example, in model sentence, between zhe and the zhong accurate node.
About judging the problem of accurate node, do some supplementary notes referring again to Fig. 4.At first to obtain each self-frequency level sum of monosyllabic word sound of just having imported in the syllable string of syllable input media 11 inputs and these two monosyllabic word sounds of monosyllabic word sound of before it, importing.This and can by the speech voice grade access device 4 as the additive operation device work go to try to achieve.And then, can also access device 4 as output unit work and with this be decided to be by the speech voice grade with reference to level frequently.At this moment, can access device 4 to the speech voice grade and be considered as determining the reference reference frequency level setting device of level frequently.
Access in the device 4 at the speech voice grade, the frequency level of the disyllabic word sound that each syllable by above-mentioned two monosyllabic word sounds is constituted and with reference to level frequently compares mutually.At this moment, to access device 4 be the work of installing as a comparison to the speech voice grade.When being not less than the frequency level of above-mentioned disyllabic word sound with reference to the frequency level, the speech voice grade accesses device 4 and accesses device work again as the speech voice grade, and position before the monosyllabic word sound of just having imported is decided to be accurate node.
During concrete the processing, this accurate node also should equally be treated with node.Now reason is described as follows:
Figure 10 represents that syllable input sequence number is the complete list of 1 to 8 o'clock speech sound Splittable.If establishing the quantity of type is v(u), so following formula is set up:
v（u）＝v（u-2）+v（u-1）
The result is selected in Figure 11 (1) and (2) expression, specific practice is respectively to take out two from the speech sound string colony of the Splittable of certain u, investigate between the speech sounds all on two speech sound strings and whether can satisfy accurate node condition, frequency level and size to two speech sound strings compares mutually simultaneously, finally selects a level and a little side frequently.The arrow direction is the speech sound string of selecting among the figure.
For example, when u is 4, can be with regard to ab and a/b, bc and three groups of investigation of carrying out frequency level and magnitude relationship such as b/c, cd and c/d.Following example is three groups of situations that all constitute accurate node condition simultaneously.That is to say that from level and last frequently, selecting condition is the (∩ of ab＜the a/b) (∩ of bc＜b/c) (cd＜c/d).
[example: u=4]
In the above-mentioned example, magnitude relationship is still not clear between speech sound string ab/cd and the a/bc/d.Which will be the optimum segmentation type to these two Splittables, have only actual specific more once both the frequency level and could determine.That is to say, as the object that speech message breath is handled, only remaining ab/cd and these two types of a/bc/d.Shown in above-mentioned example, this left Splittable colony of candidate target that satisfies whole selecting conditions simultaneously and handle as speech message breath is called " positive speech sound Splittable ", slightly is called " positive Splittable " (among figure shown in the surplus).Figure 12 then is the complete list of positive Splittable.
Refer again to Figure 11 (1) and (2), illustrate the implication of " accurate node ".Let us has a look that " u is 4 o'clock selecting condition (3), (4), (7) and (8).As can be seen, the common factor of these selecting conditions is bc 〉=b/c, and selecting the result various is ab/cd, ab/c/d, a/b/cd and a/b/c/d, and the back of b necessarily exists the speech cent to cut.Bc 〉=b/c that is to say, if must exist the speech cent to cut so between b and the c.On the other hand, if bc＜b/c just not necessarily exists the speech cent to cut between b and the c so.In other words, bc cuts match to the speech cent of b/c in the middle of, if bc 〉=b/c, bc cold turkey so.That is to say, equal speech sound bc and just do not existed originally.Such point is defined as " accurate node ".Accurate node has equivalent effect aspect the cutting apart of information processing sentence and information processing sentence with node.
Figure 13 is " Splittable generative process " figure that arrangement is come out, and the figure shows in the input sequence number is 1 to 5 o'clock, and along with the input of each syllable of a～e, how Splittable adapts to selecting condition and generate.
Examine " the Splittable generative process " of Figure 13, can know that just selecting condition and positive Splittable system finally can be summed up as Figure 14 " positive Splittable generative process ".Numeral is 1. to 6. representing syllable input point and syllable input sequence number among the figure.1. be the starting position (being node) of speech sound string.Even the syllable input is carried out forward, but as long as selecting condition continues not satisfy accurate node condition (for example ab is not less than a/b), process is just advanced to right-hand along this figure path topmost so, accurate node do not occur on the speech sound string, and the syllable string of positive Splittable continues always.This syllable string renames and is " positive information processing sentence ".In fact, in Chinese, in limited syllable input sequence number, node must occur, " positive information processing sentence " just interrupts at this point.Accurate node coexists and just constitutes an information processing sentence between one of node that this accurate node drew in the past or accurate node, is referred to as the definite message or answer breath here and handles sentence.For convenience of explanation, also abbreviate positive information processing sentence and definite message or answer breath processing sentence as the information processing sentence sometimes.
Let us is imported the u syllable, and it is supposed to be as the criterion node together between the u-1 syllable.For example, work as u=4, during input d, suppose cd 〉=c/d exactly.It is ab/c/d and a/bc/d that two positive Splittable syllable strings might appear in the result, but the structure of both end parts all is "/d ".Not only u=4 one for example this, in syllable string input process, just when occurring accurate node condition at first, if syllable string end syllable is made as x, the syllable string is " .../x " form certainly so.That is to say, begin to the syllable string of the previous syllable position at end from the syllable foremost of syllable string, is one complete " information processing sentence ", and the syllable in end must be first syllable of next " information processing sentence ".
In Figure 13, in order to be distinguished with "/", cut with " | " mark be expressed as being as the criterion speech cent of node.For example, above-mentioned example just is expressed as ab/c|d and a/bc|d.
Therefore, when accurate node occurs, the input sequence number of above-mentioned x is reset to 1, it is altered to the 1st syllable of next information processing sentence, as shown in Figure 14, as 2. processing procedure is poured into sequence number, the overall process of speech message breath processing will be put very succinct form in order so.In the middle of second example, it is that object is carried out with the speech message breath sentence colony on Figure 14 path topmost only that speech message breath is handled.
List the quantity that becomes the speech message breath of speech message breath processing sentence below.As seen in the middle of second example, the number of packages that needs to handle is compared and will be descended greatly with first example.
[number of packages that needs processing]
Syllable number U 123456789 10 11 12 13 14
The first example number of packages v
112358 13 21 34 55 89 144 233 377 610
The second example number of packages v
2112234579 12 16 21 28 37
" more succinct speech message breath is handled "
Figure 15 represents syllable input sequence number u from 1 when 8 advance, and which kind of mutual relationship the speech sound Splittable that speech message breath is handled sentence is keeping how to change forward.
TS(u among the figure) and TD(u) etc. represent speech message breath each the colony's intermediater-frequency stage of sentence and the minimum type that finish with monosyllabic word sound or disyllabic word sound respectively.These types are called TS type or TD type.T(u) the frequency level of expression TS type and TD type information processing sentence and relatively mutual, the result is level and little type frequently.
U=1, T(1)=a, u=2, T(2)=ab, this is the initial value in the processing procedure.
Thereby the end that horizontal dotted line is illustrated in u=u group type among the figure adds a disyllabic word sound and constitutes the relation of u=u+2 group type.The end that oblique solid line is illustrated in u=u group type adds that thereby a monosyllabic word sound constitutes the relation of u=u+1 group type.Should see that tiltedly solid line can not be that monosyllabic type is drawn from end speech sound.
With thin camber line plural type is drawn together together among the figure, and added codes such as TS, TD, T, this is illustrated in and compares level and size frequently on this position, carries out either-or choice.Carefully not with the aid of pictures, be difficult to understand this process, be illustrated for example below.
[example: T(1)～T(6) selection process]
Screening word table among the last figure shows by the relatively choice of level and size frequently, is defined as the type of optimum speech sound Splittable.For example, when u=3, each self-frequency level of the a/bc of the a/bc of TS type and TD type and comparing because condition ab/c＜a/bc, so TS type ab/c win (screening word), so TS(3) become optimum type T(3), thereby be able to determine at last.Secondly, when u=4, from TD(2) and the TD(4 that comes) be defeated by from TD(3) and the TS(4 that comes), so T(4) be defined as TS(4) a/bc/d(screening word).During above-mentioned u=5, should compare by ab/cd/e, ab/c/de and these three types of a/bc/de originally, but both comparisons of back are stage of 3 to have decided victory or defeat already at u, i.e. the frequency level of a/bc/de and be exactly and equate certainly not greatly than the frequency level of ab/c/de.So the a/bc/de that fails when u=3 stamps bracket, be illustrated in the later processing procedure except it.When u=6 is, also needn't compare choice, only to TD(6 to four types) a/bc/d/ef and TS(6) the ab/c/de/f amphitypy selects just enough.So, no matter u how, always minimum level frequently and relatively choice handle and can go on by either-or method.Selecting the result is deciding between a TD type and TS type certainly.When selecting the failure type to be the TD type, as TD(3), its end connects a monosyllabic word sound, can participate in pair elimination serieses when one section later u once more.When the type of failure when being the TS type, its end connects a disyllabic word sound, and when two sections later u, this type is invalid on selection process.As mentioned above, the choice of optimum speech sound Splittable needs only each section along with u, carries out twice level additive operation frequently, and level and size have just disposed more frequently again.
In brief, the above-mentioned alternatively method of selecting can be carried out like this: at first utilize accurate node detection device to detect the accurate node of the monosyllabic word sound front of just having imported on each Splittable.With the Splittable sorter each Splittable is classified again, be divided into the second class Splittable colony that there is the first kind Splittable colony of accurate node in the monosyllabic word sound front of just having imported and does not have accurate node.Use level calculation element frequently again, the whole separately Splittables of first and second Splittable colony are calculated, calculate the summation of level frequently.And then, determine that with first and second Splittable device accesses the Splittable of the frequency level summation minimum separately of above-mentioned first and second Splittable colony again, with it as first and second Splittable.At last, select device that first and second a Splittable intermediater-frequency stage and a little side are elected to be optimum speech sound Splittable by Splittable.
In alternatively's choice method that second example adopts, establishing p and q is two adjacent monosyllabic word sounds, the disyllabic word sound that pq is formed by connecting for both, I
p, I
qAnd I
PqThen be separately the frequency level and, the definition of accurate node at this moment is decided to be 1
PqBe not less than 1
pWith 1
qSum.Why it is not defined as 1
PqLess than 1
pWith 1
qSum, that is because if the frequency level of pq/r and p/q/r and equate, becomes 3 thereby just will comprise p/qr in candidate's number of speech message breath sentence during u=3 so, thus the alternatively of optimum speech sound Splittable is selected become can not.That is to say, the definition of accurate node is decided to be 1
PqBe not less than 1
pWith 1
qSum, purpose are thoroughly to carry out alternatively's choice.Owing to take this measure, making frequently, level has partly obtained solution with identical problem.
But, either-or result, two frequency levels but happen occasionally with identical problem during certain u.Alternatively's method can not prevent level and identical problem frequently fully.At this moment, can adopt one of following three kinds of methods to go to solve, promptly or adopt the end of the sentence tail to be double-tone joint one side perhaps opposite, perhaps lean on people's judgement to carry out the alternatively.
" handling the speech sound Chinese character conversion of carrying out synchronously " with alternatively's formula speech message breath
Utilize above-mentioned alternatively's method, can examine and obtain the optimum segmentation type compactly.Moreover, when speech message breath was handled, the conversion of speech sound Chinese character also can adopt alternatively's mode to carry out.In Figure 16,
U is a syllable input sequence number, since 1.
In the PS type that to be the end finish with monosyllabic word sound minimum level frequently and.
In the PD type that to be the end finish with the disyllabic word sound minimum level frequently and.
P be final minimum level frequently and, PS and PD compare,
If PS is not more than PD, P=PS then.If PS is greater than PD, P=PD then.
1 is the frequency level of speech sound.
1S is the frequency level of monosyllabic word sound for the syllable of just having imported.
1D couples together the frequency level of resulting disyllabic word sound for the syllable of the syllable just imported and its front.
Numeral in top array variable PS, PD, P, 1,1S and the 1D back bracket is u.In addition, in Figure 16, the variable of following relevant Chinese character all is according to the information exchange resulting literal array variable of encode Chinese characters for computer.
KS is the Chinese character speech string that finish with the Chinese word character speech at the end.
KD is the Chinese character speech string that finish with two Chinese character speech at the end.
K is the Chinese character speech string of selecting.
If P=PS, if K=KS then is P=PD, then K=KD.
HS in the Chinese character homonym of the monosyllabic word sound just imported according to recognizing the most definite Chinese character speech.
HD couples together in the Chinese character homonym of the disyllabic word sound that obtains according to recognizing the most definite Chinese character speech with the syllable of its front for the syllable just imported.
The numeral of literal array variable KS, KD, K, HS and HD back bracket the inside is u.
Below, according to Figure 16, explanation obtains the algorithm of the Chinese character speech string K of final goal.
(1) initial value:
P（1）＝IS（1）：PD（2）＝ID（2）：P（2）＝ID（2）
K（1）＝HS（1）：KD（2）＝HD（2）：K（2）＝KD（2）
（2）PS（u）＝PD（u-1）+IS（u），PD（u）＝P（u-2）+ID（u）
KS（u）＝KD（u-1）+HS（u），KD（u）＝K（u-2）+HD（u）
(3) if PS(u) be not more than PD(u), so,
P（u）＝PS（u）：K（u）＝KS（u）
If PS(u) greater than PD(u), so,
P（u）＝PD（u）：K（u）＝KD（u）
Adopt above-mentioned algorithm, first syllable of the information processing sentence during from u=1 begins to handle, and theoretically no matter how big u is, u of every input can both draw according to recognizing the Chinese character speech string that speech cent the most accurately cuts the syllable string of former input.
Above-mentioned situation explanation, adopt this algorithm, syllable of every input, carry out twice integer variable addition and additive operation of twice Chinese character variable in (2), carry out once again in (3) integer variable relatively, promptly only carry out altogether 5 simple calculating just can finish ask for K(u) computing.The work of treatment that need append is the work of indexed search IS and ID and the work of retrieval KS and KD with the speech sound of input just.But, no matter how import sequence number u, always the workload that will pay behind syllable of input is certain, very easy.Another advantage is, this method is not whole speech message breath to be handled sentence carry out the conversion of disposable speech sound Chinese character, but syllable of every input just carries out a suboptimum speech cent to the whole syllable string of having imported and cuts and the Chinese character conversion.This is progressively to handle.As optimum result, Chinese character speech string can be presented on the screen along with each keystroke, so typewriting person's work is very light.
Figure 17 is that the above-mentioned alternatively's formula of explanation is selected the process flow diagram of method.Here, be that the speech sound that unit imports only limits to monosyllabic word sound and disyllabic word sound with the syllable; Speech sound string is two ends formula node then, and may there be accurate node in the centre.With reference to Figure 17 and Fig. 4, handle after beginning on the starting point, after node and accurate node, be reset to 1 at the S1 syllable input sequence number M that the speech message ceases in the sentence on the step.At S2 monosyllabic speech sound input syllable input media 11 on the step.
At S3 on the step, speech voice grade indexing unit 2 is according to the monosyllabic word sound of just having imported in the syllable string of having imported already, the frequency level memory storage of retrieval dictionary 1 accesses the frequency level IS of this monosyllabic word sound, and the Chinese character KS of the most definite Chinese character speech in the homonym of this monosyllabic word sound.Simultaneously, the disyllabic word sound RD that speech voice grade indexing unit 2 also will constitute according to these two monosyllabic word sounds of monosyllabic word sound of above-mentioned monosyllabic word sound and input before it, retrieval dictionary 1 also accesses the frequency level ID of this disyllabic word sound, and the Chinese character KD of the most definite Chinese character speech in the homonym of this disyllabic word sound.
In the S4 step, will judge whether syllable input sequence number M is 1.If M is more than 2 for syllable input sequence number, so just enter the S5 step, whether the existence of judgement disyllabic word sound RD to be.If there is no disyllabic word sound RD just turns to end, promptly handles termination.If there is disyllabic word sound RD, so just enter the S6 step.In S6 step, the frequency level ID of disyllabic word sound, will with the frequency level of monosyllabic word sound and syllable input sequence number be M-1 promptly in the frequency level IS0 sum of the monosyllabic word sound of input the last period, compare.If the frequency level ID of disyllabic word sound is little, so just enter the S7 step, judge whether syllable input sequence number M is 2.When syllable input sequence number 3 when above, just enter S8.
In step, carry out the calculating of following (1) and (2) formula at S8:
PS（M）＝PD（M-1）+IS……（1）
PD（M）＝P（M-2）+ID……（2）
Here, PS(M), PD(M) and P(M) all be meant each the speech sound string of M from 1 to M the frequency level and.And, P(M) and PD(M) when the speech sound that is meant speech sound string end respectively is monosyllabic word sound and disyllabic word sound the frequency level and, P(M) then for from PS(M) and PD(M) the optimum speech message breath sentence speech sound string that chooses of alternatively the frequency level with.
Enter the S9 step, then carry out the calculating of following (3) and (4) formula:
KS（M）＝KD（M-1）+HS……（3）
KD（M）＝K（M-2）+HD……（4）
Here, KS(M), KD(M) and K(M) be meant that all M is 1 Chinese character string to each syllable string of M.KS(M) and the Chinese character speech that KD(M) is meant the end respectively be the character string of Chinese word character speech and two Chinese character speech, the K(M) Chinese character string of then cutting Chinese character speech string for optimum speech cent.
After aforementioned calculation finishes, S10 in the step to level and PD(M frequently) and PS(M) compare mutually.As PD(M) 〉=PS(M) time, change the S11 step over to establish P(M)=PS(M), determine optimum Chinese character string in the S12 step.Perhaps, changing S13 step over to establishes P(M)=PD(M), determine optimum Chinese character string in the S14 step.
If syllable input sequence number M is 1 in S4 step, so just enter the S15 step to establish P(1)=1S, determine optimum Chinese character string in S16 the step.
If the frequency level ID of disyllabic word sound is equal to or greater than monosyllabic audio frequency level IS and IS0 sum in S6 goes on foot, so just enter the S17 step.Syllable input sequence number M enters the S15 step again after the S17 step resets.
If syllable input sequence number M is 2 in S7 step, so just enter the S18 step to establish P(2)=ID, determine optimum Chinese character string in the S19 step.
After S12, S14, S16 or S19 step determine optimum Chinese character string, above-mentioned optimum Chinese character string is exported in output units such as screen display (not expression among the figure) in the step at S20 again, and then, the frequency level IS with present monosyllabic word sound in S21 step again preserves as the frequency level IS0 of the monosyllabic word sound of the last period.At last, syllable input sequence number is marched forward 1, get back to S2 and go in the step in the S22 step.
Claims (7)
1, a kind of Chinese character transform method, this method is by a kind of disposal system, is the dictionary of each speech sound of index stores from the speech sound with Chinese, accesses one of Chinese character speech of this dictionaries store;
It is index with the Chinese word sound that above-mentioned dictionary contains an energy, according to the speech sound of above-mentioned Chinese character dictionary, and the frequency level storage part of storage monosyllabic word sound and disyllabic word voice grade; Above-mentioned disposal system then comprises speech sound, frequency level indexing unit, speech sound.Frequency level memory storage, speech sound.Frequently level accesses device, optimum speech sound Splittable generating apparatus and Chinese character word and search device etc. each installs interaction relationship between each device and the effect among each device is decided by software;
It is characterized in that,
A. the monosyllabic word sound of just having imported in the syllable string of having imported already according to the syllable input media, last predicate sound, a level indexing unit is retrieved the frequency level storage part of above-mentioned dictionary frequently, and accesses the frequency level of this monosyllabic word sound; Simultaneously also according to this monosyllabic word sound disyllabic word sound that these two monosyllabic word sound syllables of monosyllabic word sound that it imports previously constitute that coexists, retrieve above-mentioned dictionary and access the frequency level of this disyllabic word sound;
B. last predicate sound, frequently two monosyllabic word sounds having retrieved of level indexing unit and level frequently, and above-mentioned disyllabic word sound and frequently level deposit predicate sound, grade memory storage frequently in;
When last predicate sound, when frequently a level indexing unit retrieve not go out above-mentioned disyllabic word sound, be considered as node with regard to front with the above-mentioned monosyllabic word sound of just having imported; In the syllable string that the above-mentioned monosyllabic word sound of just having imported is imported previously, just constitute an information processing sentence between the node that this node has obtained with the front;
C. by last predicate sound, frequently level accesses device, from last predicate sound, access the speech sound of monosyllabic word sound in this information processing sentence and disyllabic word sound and level frequently the level memory storage frequently;
D. by above-mentioned optimum speech sound Splittable generating apparatus, according to the speech sound, frequently level accesses each speech sound Splittable that monosyllabic word sound and disyllabic word sound combine in the above-mentioned information processing sentence that device accessed, obtain the monosyllabic word sound of each speech sound Splittable of formation and the frequency level summation of disyllabic word sound, and generate the optimum speech sound Splittable of above-mentioned information processing sentence according to this frequency level summation;
E. according to each speech sound that splits from above-mentioned optimum speech sound Splittable, retrieve above-mentioned dictionary, one of Chinese character speech that to access with this speech sound be index by above-mentioned Chinese character word and search device.
2, the Chinese character transform method of putting down in writing according to claim 1, last predicate sound, frequency level access device and comprise that level setting device, frequency level comparison means and speech sound, frequency level access device etc. more frequently in reference;
It is characterized in that described step C comprises:
C1. determine with reference to the frequency level with reference to a frequency level setting device with above-mentioned;
C2. by being connected the above-mentioned frequency level comparison means on the predicate voice grade memory storage, the frequency level of above-mentioned disyllabic word sound is compared with reference to level frequently with above-mentioned;
In said reference when frequently level is not less than above-mentioned disyllabic word voice grade, just the front with the above-mentioned monosyllabic word sound of just having imported is considered as accurate node, so in the syllable string of input before this monosyllabic word sound of just having imported, between the node or accurate node that the accurate node of the above-mentioned monosyllabic word sound front of just having imported had drawn in the past with this accurate node, just constituted a definite message or answer breath and handled sentence;
C3. by be connected last predicate sound on the above-mentioned frequency level comparison means, level accesses device more frequently, from last predicate sound, access above-mentioned definite message or answer breath the level memory storage frequently and handle sentence and go up the speech sound of monosyllabic word sound and disyllabic word sound and level frequently.
3, the Chinese character transform method of putting down in writing according to claim 2, above-mentioned reference a level setting device frequently comprise devices such as additive operation device and output unit;
It is characterized in that described step C1 comprises:
C11. by being connected speech sound, the above-mentioned additive operation device on the level memory storage frequently, obtain above-mentioned two monosyllabic word sounds the frequency level and;
C12. the frequency level that above-mentioned additive operation is drawn by above-mentioned output unit and as above-mentioned with reference to level output frequently.
4, the Chinese character transform method of putting down in writing according to claim 2 is characterized in that, last predicate sound, frequency level access step C3 again and comprise:
C31. establish above-mentioned each information processing sentence and definite message or answer breath handle the syllable input sequence number of above-mentioned syllable string in the sentence be u(but, u is a nonnegative integer);
C32. be located in each speech sound Splittable of the syllable string that constitutes from the 0th to u number syllable, end speech sound is PS(u for the speech sound that constitutes monosyllabic word sound and the above-mentioned frequency level summation reckling of the whole speech sounds of this syllable string);
Be located in each speech sound Splittable of the syllable string that constitutes from the 0th to u number syllable, end speech sound is PD(u for the speech sound that constitutes the disyllabic word sound and the above-mentioned frequency level summation reckling of the whole speech sounds of this syllable string);
C33. the above-mentioned frequency level of u monosyllabic word sound is 1S(u);
The frequency level of the disyllabic word sound that (u-1) number and u number two monosyllabic word sound constitute is 1D(u);
So the initial value of level summation becomes frequently:
P（1）＝1S（1），P（2）＝1D（2），PD（2）＝1D（2）;
When the u syllable is imported, just carry out following calculating:
PS（u）＝PD（u-1）+1S（u），
PD（u）＝P（u-2）+1D（u）;
Carry out following alternative selection simultaneously:
PS(u) be not more than PD(u) time, P(u)=PS(u);
PS(u) greater than PD(u) time, P(u)=PD(u);
C34. use the P(u that determines with this selection) corresponding speech sound Splittable, carry out the speech cent to the syllable string that is input to the syllable formation till u number and cut.
5, the Chinese character transform method of putting down in writing according to claim 1, above-mentioned optimum speech sound Splittable generating apparatus comprise each device such as accurate node detection device, speech sound Splittable sorter, the first and second speech voice grades and calculation element, the first and second speech sound Splittable determination devices and speech sound Splittable selecting arrangement;
It is characterized in that described optimum speech sound Splittable generates step D and comprises:
D1. by be connected the predicate sound, frequently level accesses the above-mentioned accurate node detection device on the device again, detects the above-mentioned accurate node of the monosyllabic word sound front that above-mentioned each speech sound Splittable just imported:
D2. cut the type sorter by the last predicate cent that is connected on the above-mentioned accurate node detection device, above-mentioned each speech sound Splittable is divided into the second speech sound Splittable kind that there is the first speech sound Splittable kind of accurate node in the monosyllabic word sound front of just having imported and does not have accurate node;
D3. by above-mentioned first speech voice grade and the calculation element, calculate the last predicate voice grade summation of above-mentioned whole first speech sound Splittable classes;
By above-mentioned second speech voice grade and the calculation element, calculate the last predicate voice grade summation of above-mentioned whole second speech sound Splittable classes;
D4. by being connected the above-mentioned first speech sound Splittable determination device on above-mentioned first speech voice grade and the calculation element, the speech sound Splittable of going up predicate voice grade summation minimum in the above-mentioned first speech sound Splittable kind is defined as the first speech sound Splittable;
By being connected the above-mentioned second speech sound Splittable determination device on above-mentioned second speech voice grade and the calculation element, the speech sound Splittable of going up predicate voice grade summation minimum in the above-mentioned second speech sound Splittable kind is defined as the second speech sound Splittable;
D5. cut the type selecting arrangement by the last predicate cent that is connected on the above-mentioned first and second speech sound Splittable determination devices, elect optimum speech sound Splittable as going up the little side of predicate voice grade summation in the above-mentioned first and second speech sound Splittables.
6, the Chinese character transform method of putting down in writing according to claim 1, above-mentioned optimum speech sound Splittable generating apparatus comprises speech voice grade and each devices such as calculation element and speech sound Splittable determination device;
It is characterized in that described optimum speech sound Splittable generates step D and comprises:
D ' 1. calculates predicate voice grade summation by last predicate voice grade and calculation element according to above-mentioned whole speech sound Splittables;
D ' 2. cuts the type determination device by the last predicate cent that is connected on predicate voice grade and the calculation element, and the speech sound Splittable of last predicate voice grade summation minimum is defined as above-mentioned optimum speech sound Splittable.
7, the Chinese character transform method of putting down in writing according to claim 6, last predicate cent cut the type determination device and comprise the input of speech sound detection device, speech voice grade and memory storage and speech voice grade and comparison means etc.;
It is characterized in that institute's predicate cent cuts type deciding step D ' 2 and comprises:
D ' 21. is by being connected the last predicate sound input detection device on predicate sound, the frequency level indexing unit, and when above-mentioned monosyllabic word sound was imported, output speech sound was imported detecting signal;
D ' 22. stores above-mentioned summation by being connected last predicate voice grade and the memory storage on the predicate sound input detection device according to above-mentioned whole speech sound Splittables;
D ' 23. is by being connected last predicate voice grade and the comparison means on the predicate sound input detection device, from last predicate voice grade and memory storage, access the speech voice grade summation of above-mentioned whole speech sound Splittables according to last predicate sound input detecting signal, simultaneously this summation is compared and the speech sound Splittable of summation minimum is defined as optimum speech sound Splittable.
Applications Claiming Priority (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP63105030A JP2798931B2 (en) | 1988-04-26 | 1988-04-26 | Chinese phonetic delimiter and phonetic kanji conversion |
| JP105030/88 | 1988-04-26 | ||
| JP172163/88 | 1988-07-11 | ||
| JP63172163A JPH0221357A (en) | 1988-07-11 | 1988-07-11 | Word sound dividing system for chinese word |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1037411A CN1037411A (en) | 1989-11-22 |
| CN1019233B true CN1019233B (en) | 1992-11-25 |
Family
ID=26445381
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 89102915 Expired CN1019233B (en) | 1988-04-26 | 1989-04-26 | Chinese characters transforming mode |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN1019233B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH06208559A (en) * | 1993-01-11 | 1994-07-26 | Matsushita Electric Ind Co Ltd | Chinese/kanji converting device |
| CN1107255C (en) * | 1997-08-06 | 2003-04-30 | 顾剑 | Infinite ordered character set Chinese character whole set method and system |
-
1989
- 1989-04-26 CN CN 89102915 patent/CN1019233B/en not_active Expired
Also Published As
| Publication number | Publication date |
|---|---|
| CN1037411A (en) | 1989-11-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN1113305C (en) | Language processing apparatus and method | |
| CN1168068C (en) | Speech synthesizing system and speech synthesizing method | |
| CN1174332C (en) | Method and device for converting expressing mode | |
| CN1096036C (en) | Apparatus and method for retrieving dictionary based on lattice as key | |
| CN1133127C (en) | Document retrieval system | |
| CN1171162C (en) | Apparatus and method for retrieving charater string based on classification of character | |
| CN1281191A (en) | Information retrieval method and information retrieval device | |
| CN1648828A (en) | System and method for disambiguating phonetic input | |
| CN1871597A (en) | System and method for associating documents with contextual advertisements | |
| CN1894688A (en) | Translation determination system, method, and program | |
| CN1542735A (en) | System and method for recognizing a tonal language | |
| CN1652107A (en) | Language conversion rule preparing device, language conversion device and program recording medium | |
| CN1841367A (en) | Communication support apparatus and method for supporting communication by performing translation between languages | |
| CN101034414A (en) | Information processing device, method, and program | |
| CN1910573A (en) | System for identifying and classifying denomination entity | |
| CN1095560C (en) | Kanji conversion result amending system | |
| CN1702650A (en) | Apparatus and method for translating Japanese into Chinese and computer program product | |
| CN1315722A (en) | Continuous speech processing method and apparatus for Chinese language speech recognizing system | |
| CN1120439C (en) | Chinese forming device for machine translation | |
| CN1195293C (en) | Speech recognition method and device, speech synthesis method and device, recording medium | |
| CN1158621C (en) | Information processing device and information processing method, and recording medium | |
| CN1019233B (en) | Chinese characters transforming mode | |
| CN1106619C (en) | Chinese input transition processing device and Chinese input transition processing method | |
| CN1752966A (en) | Method of solving problem using wikipedia and user inquiry treatment technology | |
| CN1753083A (en) | Phonetic symbol method, system reach audio recognition method and system based on phonetic symbol |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C06 | Publication | ||
| PB01 | Publication | ||
| C13 | Decision | ||
| GR02 | Examined patent application | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C15 | Extension of patent right duration from 15 to 20 years for appl. with date before 31.12.1992 and still valid on 11.12.2001 (patent law change 1993) | ||
| OR01 | Other related matters | ||
| C19 | Lapse of patent right due to non-payment of the annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |