[background technology]
Prior art has multiple to hand-held mobile terminal or computing machine input Chinese text mode, these class input modes are constantly being improved and development, and still, these input methods are in use generally quick inadequately.At present the mode of importing Chinese information to hand-held mobile terminal or computing machine has following several mode.
The first kind be system based on Chinese character being resolved into the figure composition, different figure compositions is assigned on the corresponding button, by key tap once or several times figure composition input system is realized the input of Chinese character like this.The input of Chinese character is to need our that the font structure of Chinese character is understood, simultaneously could be faster under the prerequisite that will understand very much the layout of button and corresponding figure composition to hand-held mobile terminal or computing machine input Chinese character.The shortcoming of this input mode is: the combination of this Chinese character image will be by manually realizing, brain wants the figure composition of font and the position that this word composition phase time is answered button simultaneously in the input Chinese character, so the efficient of its input is very low concerning most domestic consumer.
Second class is the rule of system based on Chinese character phonetic, this is by the phonetic to system's input Chinese character, this is the normal mode that adopts of typist of specialty, and this spelling input method is the mode of now commonplace input Chinese character, prerequisite also be requirement we the phonetic of each Chinese character is understood.The problem that this mode exists: our brain when the input Chinese character wants the phonetic and the corresponding phonetic place button of Chinese character, and this in addition input is sent out and also will be leaned on artificial key tap to realize operation.So the efficient of its input is dissatisfactory concerning most domestic consumer.
The 3rd class input method that to be system combine voice-character input and the non-voice character that adds, be added to non-voice on the phonic symbol since the identical character of difference pronunciation.Example includes the Chinese phonetic spelling (BrP 2,158,776 that on November 20th, 1985 authorized) of radical sign and has the Chinese phonetic spelling (on November 25th, 1992 authorized, Chinese patent 1,066,518) of stroke number.
These methods will be remembered the rule of thinking or require to calculate stroke number that the latter has obviously reduced the speed of input.
The 4th class is the IBM of U.S. International Business Machine Corporation (IBM) in Chinese patents number is 96105876.5, patent " handling the system and method for Chinese text " by name.This input method is with voice Chinese (phonetic and BPMF) input computing machine and converts hanzi form to.This system has the novel keyboard of diacritic key (and corresponding ASCII coding), thereby allows the user to utilize each speech text syllable of being imported of diacritic note of indication syllable tone.The input mode that this system combines based on Chinese character pronunciation and Chinese-character pronunciation tone, the mode that he provides a kind of more detailed solution Chinese character to import, see it is more effective in theory than several input modes in front, if according to this input mode input Chinese character one by one, accuracy rate can be higher so.But in fact this input mode is to lack the understanding of Chinese people being used the custom of Chinese character, because the most people of China seldom consider the tone of Chinese words in daily use mandarin process, concerning most domestic consumer, be difficult to deacclimatize this use-pattern.The input mode of the Chinese popular use habit of so a kind of incompatibility its input efficiency concerning Chinese is very low.
Convert the mode of Chinese text to relevant for Chinese speech input, it is 02106017.7 that Patent Office of the People's Republic of China authorizes Nokia Oy's patent No., and denomination of invention is " short message message service method, an Apparatus and system ".The problem that this invention mainly solves is the solution that sends short speech sound message.Mention and to be stored in the phrase tone signal in the mobile phone in this part patent document the inside, by speech recognition software the voice signal of oral account is converted to text, only be to mention in the content, do not announce out technical scheme how to discern translation with the speech recognition software translation.
[summary of the invention]
In order to overcome the prior art speech conversion is the inefficient technical matters of text, the invention provides a kind of can be efficient realization voice be changed to text provide the basis voice dictionary formation method.
In order to overcome the prior art speech conversion is the inefficient technical matters of text, and the invention provides a kind of speech conversion that can efficiently realize is the audio recognition method of text.
In order to overcome the prior art speech conversion is the inefficient technical matters of text, and the invention provides a kind of speech conversion that can efficiently realize is the speech recognition system of text.
The present invention solves the technical scheme that above-mentioned first technical matters adopted: a kind of voice dictionary formation method is provided, comprises: the literal that regular meeting is used in combination in A, the arrangement daily life; B, the standard pronunciation of described literal is stored by syllable code.
The present invention solves the technical scheme that above-mentioned second technical matters adopted: a kind of audio recognition method is provided, comprises: A, input voice; B, carry out speech analysis, code that comparison input voice form and the phonetic code in the voice dictionary, the phonetic code in the described voice dictionary are that the standard pronunciation of the literal that is used in combination by regular meeting in the storage daily life obtains; The corresponding text of C, the resultant phonetic code of output comparison.
The present invention solves the technical scheme that above-mentioned the 3rd technical matters adopted: a kind of speech recognition system is provided; comprise voice input module, sound identification module and output module that is linked in sequence and the received pronunciation module that voice dictionary is installed; sound identification module is connected with described received pronunciation module; its pronunciation according to each word or several words provides corresponding alternative text, and the phonetic code in the described voice dictionary is to obtain by the standard pronunciation of storing the literal that regular meeting is used in combination in the daily life.
With respect to prior art, the beneficial effect of voice dictionary formation method of the present invention is: because: one, most people are by remembeing that text pronunciation realizes that most people seldom have the notion of tone in daily communication to the memory of literal; Two, big people from portion is more weak to the consciousness of the implication of single word, all realizes linking up by phrase, sentence in daily communication major part.The present invention just is based on the above-mentioned custom that people use language; the standard pronunciation of the literal that regular meeting in the daily life is used in combination is stored as voice dictionary; import pronunciation accurately according to the language use habit in the daily life to system; as long as the voice of input meet the pronunciation of corresponding everyday words in the voice dictionary or sentence; just can identify the voice of input; convert corresponding text to, improve the speed that speech conversion becomes text widely.
With respect to prior art, the beneficial effect of audio recognition method of the present invention is: because: one, most people are by remembeing that text pronunciation realizes that most people seldom have the notion of tone in daily communication to the memory of literal; Two, big people from portion is more weak to the consciousness of the implication of single word, all realizes linking up by phrase, sentence in daily communication major part.The present invention just is based on the above-mentioned custom that people use language; the standard pronunciation of the literal that regular meeting in the daily life is used in combination is stored as voice dictionary; import pronunciation accurately according to the language use habit in the daily life to system; as long as the voice of input meet the pronunciation of corresponding everyday words in the voice dictionary or sentence; just can identify the voice of input; convert corresponding text to; improve speed widely, do not need prior art to cooperate the method for the complexity of keyboard input to hand-held mobile terminal or computing machine input text literal.
With respect to prior art, the beneficial effect of speech recognition system of the present invention is: because: one, most people are by remembeing that text pronunciation realizes that most people seldom have the notion of tone in daily communication to the memory of literal; Two, big people from portion is more weak to the consciousness of the implication of single word, all realizes linking up by phrase, sentence in daily communication major part.The present invention just is based on the above-mentioned custom that people use language; the standard pronunciation of the literal that regular meeting in the daily life is used in combination is stored as voice dictionary and is installed in the received pronunciation module; import pronunciation accurately according to the language use habit in the daily life to system; as long as the voice of input meet the pronunciation of corresponding everyday words in the voice dictionary or sentence; just can identify the voice of input; convert corresponding text to; improve speed widely to hand-held mobile terminal or computing machine input text literal; prior art cooperates the complication system of keyboard input relatively, and speech recognition system of the present invention is simply efficient.
[embodiment]
The present invention is further described below in conjunction with drawings and embodiments.
Referring to Fig. 1, speech recognition system 100 of the present invention comprises that the voice input module 10, speech conversion pre-processing module 20, sound identification module 30, voice post-processing module 40, the speech data that are linked in sequence convert text data library module 60, memory module 70 and output module 90 to.Also comprise received pronunciation module 50 and text-processing module 80, sound identification module 30 converts text data library module 60 to described received pronunciation module 50 and speech data and is connected, and text-processing module 80 connects memory module 70.
Voice input module 10 is voice entry system; Speech conversion pre-processing module 20 carries out speech analysis in order to remove noise; Sound identification module 30 matching standard voice modules 50 provide corresponding alternative text according to the pronunciation of each word or several words; Received pronunciation module 50 is equipped with voice dictionary, and the phonetic code in the described voice dictionary is to obtain by the standard pronunciation of storing the literal that regular meeting is used in combination in the daily life.
Voice post-processing module 40 is carried out grammer, semantic analysis according to the voice contrast of front and back different interval, finds out the most logical speech; Speech data converts text data library module 60 to and by the received pronunciation module 50 that has phonetic dictionary the speech conversion of input is become text, and the text feature data are provided; Memory module 70 becomes speech conversion the data storage of text; The text editing operation is made amendment, moves, duplicates or the like to 70 li texts that provide of 80 pairs of memory modules of text-processing module; 90 of output modules become speech conversion the data of text to export by desired form.
The present invention can be applicable to the equipment that Chinese character input hand-held mobile terminal or computword input etc. need the literal input.Below be example with Chinese character input hand-held mobile terminal, specify the present invention.
At first, form the needed voice dictionary of speech recognition (consulting Fig. 2), comprise step:
The Chinese text 701 that regular meeting is used in combination in A, the arrangement daily life;
B, with the standard mandarin pronunciation of described literal by syllable code storage 702.
Use the custom of Chinese to arrange according to Chinese in the structure layout of this cover voice dictionary.This use habit refer to the present age Chinese Chinese formed the Chinese character that phrase, Chinese idiom, allusion, famous person's name, famous person's well-known phrase that agreement becomes social custom, famous place name, school's name, landscape realm name, sentence or the like every regular meeting in daily life is used in combination.
The storage mode of described Chinese speech syllable code is all stored in a unit (unit refers to a code string) mode.Example: " China " just is stored as a unit to the pronunciation syllable code of two words of China; " brave the wind and the waves " just the pronunciation syllable of four words that brave the wind and the waves to be stored as a unit; " If three of us are walking together, at least one of the other two is good enough to be my teacher " is stored as a unit to the pronunciation syllable of If three of us are walking together, at least one of the other two is good enough to be my teacher seven words; " will ought be deposited farsighted " ought deposit the farsighted unit that is stored in to will; " Zhuge Liang " is stored as unit or the like to Zhuge Liang.The arrangement of this structure can be increased work efficiency greatly.Then, be divided into different groups according to use habit on dictionary Chinese speech syllable code is arranged, the order of the arrangement of each group is defaulted as according to frequency permutation used in everyday, and frequency of utilization highland more is arranged in the front more.For example: the pronunciation of input " braving the wind and the waves " four words, the selective text that system provides are to select and " braving the wind and the waves " immediate pronunciation in the code string of forming from all four words, the approaching more front that is arranged in more.It is the group of a unit storage with single word all that one group of Chinese character in all dictionaries is also arranged simultaneously, in order to the input of single word sound.
Phonetic dictionary is the spoken voice of the mandarin of standard, different sound ray, the volumes of the compatible men and women, old and young of speech recognition.The voice if the user can not speak standard Chinese pronunciation, system still carries out voiced translation to " dialect phonetic " syllable of importing by Chinese speech dictionary syllable according to Chinese speech dictionary syllable and converts text to, and the people who grasps this " dialect phonetic " can see and understand the text.
Secondly, carry out speech recognition (consulting Fig. 3).Behind phonetic entry hand-held mobile terminal or the computing machine, the back system entered the speech syllable that is defaulted as input automatically and carried out text-converted more than 801 seconds (user is time of being provided with of Adjustment System as required).Also can carry out to system's order of input by one, its method flow is as follows:
A, input voice 802 are pressed acknowledgement key by system validation " quantity " 803 to be processed, are finished by voice input module 10
B, carry out speech analysis, code that comparison input voice form and the phonetic code in the voice dictionary comprise substep:
B1, noise treatment 804 are handled by speech conversion pre-processing module 20;
B2, carry out speech analysis, grammatical analysis and contextual analysis 805~807, handle by sound identification module 30;
Code that b3, comparison input voice form and the phonetic code in the voice dictionary are handled by sound identification module 30;
B4, speech data convert text data to, convert text data library module 60 to by speech data and handle.
B5, judge whether that several voice connect 808, if, carry out dynamic optimization cutting 809, if not, jump to step C;
The corresponding text of C, the resultant phonetic code of output comparison comprises substep:
The corresponding text and the corresponding operational order of the resultant phonetic code of screen display output comparison of c1, output module 90, text puts in order by ordering 810 before and after the frequency of utilization height;
C2, manual confirmation option 811;
C3, utilize text-processing module 80 to carry out Edit Text 812, send and storage text 813, directly store text 814 or directly send the wherein a kind of of text 815,, jump to substep c4, if not, then finish if carry out Edit Text 812;
C4, directly store text 816 or directly send text 817.
The present invention utilizes people to use the custom of language dexterously; the standard pronunciation of the literal that regular meeting in the daily life is used in combination is stored as voice dictionary; import pronunciation accurately according to the language use habit in the daily life to system; as long as the voice of input meet the pronunciation of corresponding everyday words in the voice dictionary or sentence; just can identify the voice of input; convert corresponding text to; improve speed widely, do not need prior art to cooperate the method for the complexity of keyboard input to hand-held mobile terminal or computing machine input text literal.
Prior art cooperates the complication system of keyboard input relatively, and the present invention does not need specific keyboard to cooperate, and speech recognition system is simply efficient.The input method that the non-voice character of importing and adding based on phonetic characters with respect to prior art combines, the present invention does not need to add the combination of non-voice character, thereby it is simple efficient, because prior art will be remembered the rule of thinking or require to calculate stroke number that the latter has obviously reduced the speed of input.Resolve into the method for figure composition with respect to prior art based on the rule of Chinese character phonetic or Chinese character, the present invention uses phonetic entry, greatly improves text input speed, and is simple and convenient.Human when doing the same thing, his notice is limited in the unit interval, the present invention compares with the mode of in the past various input in Chinese system, be to discharge to greatest extent to sell, as long as we concentrate on notice in the pronunciation of face, import pronunciation accurately according to the Chinese use habit in the daily life to system, can improve our speed widely, transmission that especially suitable mobile phone Chinese short message ceases and exchanging that the Chinese short message on the network ceases to hand-held mobile terminal or computing machine input Chinese.
The custom of above-mentioned use Chinese according to Chinese people, be meant: one, most of China popular (except the personage of language specialty) is by to remembeing Chinese pronunciation to the memory of Chinese, most people seldom have the notion of tone in daily Chinese exchanges, this meets our use habit; Two, most of China popular (except the personage of language specialty) is more weak to the consciousness of the implication of single Chinese words, exchanges major part at daily Chinese and all realizes linking up by phrase, sentence, and this also meets our use habit.