CN107248409A - A kind of multi-language translation method of dialect linguistic context - Google Patents
A kind of multi-language translation method of dialect linguistic context Download PDFInfo
- Publication number
- CN107248409A CN107248409A CN201710368832.2A CN201710368832A CN107248409A CN 107248409 A CN107248409 A CN 107248409A CN 201710368832 A CN201710368832 A CN 201710368832A CN 107248409 A CN107248409 A CN 107248409A
- Authority
- CN
- China
- Prior art keywords
- dialect
- code
- feature
- phonetic
- voice
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/02—Feature extraction for speech recognition; Selection of recognition unit
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/40—Processing or translation of natural language
- G06F40/58—Use of machine translation, e.g. for multi-lingual retrieval, for server-side translation for client devices or for real-time translation
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/04—Segmentation; Word boundary detection
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
Abstract
The invention discloses a kind of multi-language translation method of dialect linguistic context, belong to computer language translation technical field, solving language translation instrument of the prior art can not accurately dialect translating, the problem of increasing the mistranslation rate of translation.Condition code storehouse of the present invention including step 1, the condition code composition for setting up various dialect phonetics under mother tongue system；Step 2, the feature compensation code for producing according to the difference of condition code and standard base phonetic feature code dialect, composition characteristic compensation codes storehouse；Step 3, with the compound standard base voice produced under correspondence mother tongue system of the corresponding source voice of feature compensation code；Step 4, standard base voice are converted into the voice or text of object language.Translation for dialect and personalized speech.
Description
Technical field
A kind of multilingual translation system and method for dialect linguistic context, for the translation of dialect and personalized speech, belongs to meter
Calculation machine language translation technical field.
Background technology
In the contacts in each fields such as countries in the world culture, economy, military affairs, the communication of language is particularly important, in order to
Accurately express macaronic meaning main in various foreign affairs contacts for a long time or based on translating with people.In recent years with
The fast development of computer and digital technology, translation is done with computer and digital technology and has made significant headway,
Various outstanding machine translation systems are continued to bring out, more perfect especially in terms of character translation.But active computer is translated
System do in terms of simultaneous interpretation due to the more linguistic context of dialect is different under each mother tongue system or preference of personal pronunciation cause mistranslation rate compared with
It is high, it is impossible to the meaning of accurate expression source voice.
All can only be by the base voice of the relative standard in the mother tongue system of source in numerous language translation systems and interpreting equipment
It is translated as the standard base voice of the target family of languages.But because the source voice that need to be translated is under many circumstances in practical application environment
Non-standard voice, there are some the local dialects in every kind of mother tongue system in major mother tongue systems in the world, and this causes the language of computerization
Speech translation system has very high mistranslation rate because of the difference of linguistic context.
The patent of Patent No. 200820234990.5, is to be used as feature and mark by extracting the word of dialect phonetic, words and phrases
Quasi- base voice carries out directly contrast to find the corresponding standard base voice of dialect phonetic, once the people's pronunciation spoken a dialect is inaccurate,
It is easy for causing translation inaccurate, so that corresponding standard base voice is can not find, and also every dialect being translated is required for
Storage, the problems such as causing carrying cost, high operation hardware cost and the slow operational speed of a computer.
The content of the invention
It is an object of the invention to：Solving language translation instrument of the prior art can not accurately dialect translating, increase
The problem of mistranslation rate of translation, there is provided a kind of multi-language translation method of dialect linguistic context.
The technical solution adopted by the present invention is as follows：
A kind of multi-language translation method of dialect linguistic context, it is characterised in that comprise the following steps：
Step 1, set up various dialect phonetics under mother tongue system condition code composition condition code storehouse；
Step 2, the feature compensation code for producing according to the difference of condition code and standard base phonetic feature code dialect, composition characteristic
Compensation codes storehouse；
Step 3, with the compound standard base voice produced under correspondence mother tongue system of the corresponding source voice of feature compensation code；
Step 4, standard base voice are converted into the voice or text of object language.
Further, the method for building up in condition code storehouse comprises the following steps：
Various dialect phonetic samples under step 1.1, acquisition mother tongue system；
Step 1.2, to dialect phonetic sample carry out pretreatment remove redundancy section, pre-filtering is carried out by bandpass filter
Preemphasis is carried out by a high-pass filter again after processing, to being multiplied by Hamming window after the dialect phonetic sample noise abatement after preemphasis
After carry out end-point detection；
Step 1.3, the pretreated dialect phonetic sample of Hamming window will be multiplied by carry out spectrum analysis, and then carry out feature and carry
Take, the formant of the dialect phonetic sample after feature extraction spectrum analysis, pitch period feature, MFCC and LPCC parameter attributes
Code；
Step 1.4 is by the formant of acquisition, pitch period feature, and it is right after de-redundancy that MFCC and LPCC parameter attributes code is carried out
Multiple dialect phonetic files carry out the probability distribution statistical of condition code, find out condition code of its denominator as the dialect phonetic
Key element；
Step 1.5 by the code with condition code key element use compress mode to recompile for 64 bytes dialect phonetic
Condition code, assigns searching number by this feature code and is incorporated into condition code storehouse.
Further, the extraction step of the formant of the dialect phonetic after spectrum analysis is included in step 1.3：
Smooth spectrum is obtained after homomorphic filtering to the dialect phonetic after spectrum analysis discrete fourier is asked to the spectrum again
Conversion, then extracts the formant parameter of voice signal with DFT spectrums.
Further, the extraction step of the pitch period feature of the dialect phonetic after spectrum analysis is included in step 1.3：
Pitch period feature is extracted using average magnitude difference function method to the dialect phonetic after spectrum analysis.
Further, the extraction step of the MFCC parameters of the dialect phonetic after spectrum analysis is included in step 1.3：
Dialect phonetic after spectrum analysis is subjected to Short Time Fourier Transform and obtains its frequency spectrum, then seeks square of spectrum amplitude
Energy spectrum is obtained, bandpass filtering is carried out with triangle filter equalizer, the number of wave filter is close with critical band number, if wave filter number is
Obtain being output as after M, filtering：X (k), k=l, 2 ..., M, the output to wave filter group are taken the logarithm, and then make 2M points against in Fu
Leaf transformation is that can obtain MFCC parameters.
Further, the extraction step of the LPCC parameters of the dialect phonetic after spectrum analysis is included in step 1.3：
Dialect phonetic after spectrum analysis is carried out to the inverse Z-transform of logarithm modular function after transform, passes through the Fourier of signal
Conversion, the logarithm of modulus, then Fourier transformation of negating obtain LPCC parameters.
Further, the foundation in feature compensation code storehouse uses following steps in step 2：
Step 2.1, the standard base speech samples obtained under mother tongue system, the condition code of extraction standard base voice；
Step 2.2, com-parison and analysis is carried out to dialect phonetic condition code under mother tongue system and standard base phonetic feature code, drawn
The respective probability distribution variances frequency spectrum of general meaning；
Step 2.3, by the difference frequency spectrum carry out radix-minus-one complement superposition draw dialect phonetic feature compensation code；
Step 2.3, feature compensation code and the condition code of dialect phonetic are combined, the condition code after being combined again with standard base voice
Verify and correct by n times, be combined after being corrected to the n times verification of dialect phonetic feature compensation code and draw standard base phonetic feature
Code is in the range of allowable error, and this feature compensation codes are the feature compensation code of the dialect phonetic, and this feature compensation codes are assigned
Searching number is incorporated into feature compensation code storehouse.
Further, the feature compensation code obtaining step in step 3：Pass through the feature compensation code and feature compensation code of source voice
Storehouse carries out similarity retrieval and obtains corresponding feature compensation code.
Further, the feature compensation code in step 3, which is obtained, is set manually by user.
In summary, by adopting the above-described technical solution, the beneficial effects of the invention are as follows：
1st, the present invention can by match source voice and poster base voice feature compensation code, then with source voice and feature benefit
Repay code to be combined, dialect or individualized language can accurately be translated, greatly reduce the mistranslation rate of existing interpreting equipment, it is accurate
True rate may be up to more than 95%；
2nd, the present invention is applicable the dialect phonetic translation under different occasions, significantly improves the language simultaneous interpretation of computerization
Application；
3rd, the present invention can most be represented by extraction each dialect or the feature of personalized speech so that the feature compensation of generation
Code is applied to corresponding dialect or personalized speech, it is to avoid when deviation occurs in pronunciation, causes compound standard base voice to be forbidden
Really the problem of；
4th, requirement of the present invention to storage hardware, operation hardware is low, so as to save hardware cost so that arithmetic speed
It hurry up.
Brief description of the drawings
Fig. 1 sets up the schematic diagram in condition code storehouse and feature compensation code storehouse for system in the present invention；
Fig. 2 is carries out the block schematic illustration of the acquisition of feature compensation code during the translation of unknown languages in the present invention；
The block schematic illustration of the acquisition of feature compensation code Fig. 3 is carries out specifying languages translation in the present invention when；
Fig. 4 is the schematic diagram of the embodiment 1 in condition code storehouse in the present invention, time-domain diagram, sound spectrograph is followed successively by from left to right common
Shake peak, pitch period figure, loudness of a sound figure；
Fig. 5 is the schematic diagram of the embodiment 2 in condition code storehouse in the present invention, time-domain diagram, sound spectrograph is followed successively by from left to right common
Shake peak, pitch period figure, loudness of a sound figure；
Fig. 6 is the schematic diagram of the embodiment 3 in condition code storehouse in the present invention, time-domain diagram, sound spectrograph is followed successively by from left to right common
Shake peak, pitch period figure, loudness of a sound figure；
Fig. 7 is the schematic diagram of the embodiment 4 in condition code storehouse in the present invention, time-domain diagram, sound spectrograph is followed successively by from left to right common
Shake peak, pitch period figure, loudness of a sound figure；
Fig. 8 is the schematic diagram of the embodiment 5 in condition code storehouse in the present invention, time-domain diagram, sound spectrograph is followed successively by from left to right common
Shake peak, pitch period figure, loudness of a sound figure；
Fig. 9 is the schematic diagram of the embodiment 6 in condition code storehouse in the present invention, time-domain diagram, sound spectrograph is followed successively by from left to right common
Shake peak, pitch period figure, loudness of a sound figure.
Embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that specific embodiment described herein is only to explain the present invention, not
For limiting the present invention.
A kind of multi-language translation method of dialect linguistic context, concrete implementation method is as follows：
It (can be each dialect or personalized language under each mother tongue system of the world that each dialect phonetic under mother tongue system should be set up first
Sound or each dialect or personalized speech under the mother tongue system specified) condition code storehouse and feature compensation code storehouse.Condition code
The foundation in storehouse uses following steps：
Massage voice reading is carried out with same section of representational word, declaimer is various dialect men, female voice under certain mother tongue system
Voice, dialect phonetic declaimer, which tackles the dialect, has certain representative and general meaning.
Typing preserves for dialect phonetic file after the text sampling that dialect phonetic declaimer is read aloud, the parameter of voice document
For：Sample rate 11025Hz, monophonic, sampling precision 16,1~30s of file size, every kind of dialect phonetic file is recorded respectively
Each representational text-to-speech file of men and women's sound is at least each 50, and the dialect phonetic file of recording is more, last other side's speech
The probability distribution statistical of sound condition code is more accurate.
Pretreatment to dialect phonetic：Remove redundancy section, pass through again after carrying out pre-filtering processing by bandpass filter
One high-pass filter carries out preemphasis, and end-point detection is carried out to being multiplied by after the signal noise abatement after preemphasis after Hamming window.
The pretreated dialect phonetic of Hamming window will be multiplied by and carry out spectrum analysis, feature extraction, feature emphasis is then carried out
It is to use most represent the formant of dialect phonetic feature, pitch period feature, MFCC and LPCC parameters are analyzed and feature
Extract.
The extraction of formant：Smooth spectrum is obtained after homomorphic filtering to the dialect phonetic after spectrum analysis again to the spectrum
Discrete Fourier transform (DFT) is asked, the formant parameter of voice signal is then extracted with DFT spectrums.
The extraction of pitch period feature：To the dialect phonetic after spectrum analysis using average magnitude difference function (AMDF) method come
Extract pitch period feature.
MFCC extraction：Dialect phonetic after spectrum analysis is subjected to Short Time Fourier Transform and obtains its frequency spectrum, then seeks frequency
Spectral amplitude square energy spectrum, bandpass filtering is carried out with triangle filter equalizer, the number of wave filter is close with critical band number,
If wave filter number is M, obtain being output as after filtering：X (k), k=l, 2 ..., M, the output to wave filter group are taken the logarithm, then
Make 2M point inverse Fourier transforms and can obtain MFCC parameters.
LPCC extraction：Dialect phonetic after spectrum analysis is carried out to the inverse Z-transform of logarithm modular function after transform, passed through
The Fourier transformation of signal, the logarithm of modulus, then Fourier transformation of negating obtain LPCC parameters.
By the formant of acquisition, pitch period feature, MFCC and LPCC parameter attributes code is carried out after de-redundancy to multiple sides
Speech voice document carries out having the probability distribution statistical of dialect phonetic condition code, and emphasis is the pronunciation to vowel, the first formant and
Second formant, pitch curve, LPCC and MFCC probability distribution statistical, find out feature of the denominator as the dialect phonetic
Code key element.
By the code with condition code key element use compress mode to recompile for 64 bytes dialect phonetic condition code,
Searching number, which is assigned, by this feature code is incorporated into condition code storehouse, i.e. condition code storehouse.
The specific foundation in condition code storehouse is as follows：
Embodiment 1
The present embodiment says English model voice " This is a test sample for myself " using American.Adopt
Sample frequency 11025Hz, sampling depth 16bit, monophonic, duration 2.157 seconds.The time-domain diagram of actual analysis, sound spectrograph formant,
Pitch period figure, loudness of a sound figure, as shown in Figure 4；Formant is extracted, pitch period feature, MFCC and LPCC parameter attributes code is carried out
The probability distribution statistical for having dialect phonetic condition code is carried out after de-redundancy to dialect phonetic file, by the generation with condition code key element
Code uses compress mode to recompile for the condition code of the dialect phonetic of 64 bytes, and assigning searching number by this feature code is incorporated into feature
Code storehouse, that is, be incorporated into spectrum signature code storehouse.
Embodiment 2
The present embodiment says English model voice " This is a test sample for myself " using Englishman.Adopt
Sample frequency 11025Hz, sampling depth 16bit, monophonic, duration 2.267 seconds.The time-domain diagram of actual analysis, sound spectrograph formant,
Pitch period figure, loudness of a sound figure, as shown in Figure 5；Formant is extracted, pitch period feature, MFCC and LPCC parameter attributes code is carried out
The probability distribution statistical for having dialect phonetic condition code is carried out after de-redundancy to dialect phonetic file, by the generation with condition code key element
Code uses compress mode to recompile for the condition code of the dialect of 64 bytes, and assigning searching number by this feature code is incorporated into condition code
Storehouse, that is, be incorporated into spectrum signature code storehouse.
Embodiment 3
The present embodiment says English model voice " This is a test sample for myself " using Indian.Adopt
Sample frequency 11025Hz, sampling depth 16bit, monophonic, duration 1.956 seconds.The time-domain diagram of actual analysis, sound spectrograph formant,
Pitch period figure, loudness of a sound figure, as shown in Figure 6；Formant is extracted, pitch period feature, MFCC and LPCC parameter attributes code is carried out
The probability distribution statistical for having dialect phonetic condition code is carried out after de-redundancy to dialect phonetic file, by the generation with condition code key element
Code uses compress mode to recompile for the condition code of the dialect phonetic of 64 bytes, and assigning searching number by this feature code is incorporated into feature
Code storehouse, that is, be incorporated into spectrum signature code storehouse.
Embodiment 4
The present embodiment says that Henan model voice " please input the sample that received pronunciation text makees speech sample using Chinese
Plate ".Sample frequency 11025Hz, sampling depth 16bit, monophonic, duration 4.27 seconds.The time-domain diagram of actual analysis, sound spectrograph is common
Shake peak, pitch period figure, loudness of a sound figure, as shown in Figure 7；Extract formant, pitch period feature, MFCC and LPCC parameter attributes code
The probability distribution statistical for having dialect phonetic condition code after de-redundancy to the progress of dialect phonetic file is carried out, there will be condition code key element
Code use compress mode to recompile for 64 bytes dialect phonetic condition code, by this feature code assign searching number be incorporated into
Condition code storehouse, that is, be incorporated into spectrum signature code storehouse.
Embodiment 5
The present embodiment says that Sichuan model voice " please input the sample that received pronunciation text makees speech sample using Chinese
Plate ".Sample frequency 11025Hz, sampling depth 16bit, monophonic, duration 4.928 seconds.The time-domain diagram of actual analysis, sound spectrograph
Formant, pitch period figure, loudness of a sound figure, as shown in Figure 8；Extract formant, pitch period feature, MFCC and LPCC parameter attributes
Code carries out the probability distribution statistical for having dialect phonetic condition code after de-redundancy to the progress of dialect phonetic file, will be wanted with condition code
The code of element uses compress mode to recompile for the condition code of the dialect phonetic of 64 bytes, and assigning searching number by this feature code compiles
Enter condition code storehouse, that is, be incorporated into spectrum signature code storehouse.
Embodiment 6
The present embodiment " please input the sample that received pronunciation text makees speech sample using Chinese's model voice that speaks standard Chinese pronunciation
Plate ".Sample frequency 11025Hz, sampling depth 16bit, monophonic, duration 3.96 seconds.The time-domain diagram of actual analysis, sound spectrograph is common
Shake peak, pitch period figure, loudness of a sound figure, as shown in Figure 9；Extract formant, pitch period feature, MFCC and LPCC parameter attributes code
The probability distribution statistical for having dialect phonetic condition code after de-redundancy to the progress of dialect phonetic file is carried out, there will be condition code key element
Code use compress mode to recompile for 64 bytes dialect phonetic condition code, by this feature code assign searching number be incorporated into
Condition code storehouse, that is, be incorporated into spectrum signature code storehouse.
It is that source voice can be complex as to the standard base voice of the mother tongue system, Ying Jian after the condition code storehouse of dialect phonetic is set up
Dialect phonetic under vertical mother tongue system is combined required feature compensation code storehouse.Corresponding with condition code storehouse feature compensation code storehouse is built
It is vertical so that translation system can reduce hardware cost in practical application.The foundation in spectrum signature compensation codes storehouse uses following steps：
Massage voice reading is carried out with same section of representational word, declaimer is various standard men, female voice under certain mother tongue system
Voice, massage voice reading person should be the mother tongue system standard base voice.
To the condition code of standard base voice extraction standard base voice, wherein, the extraction of the condition code of standard base voice is with building
The extracting method of condition code is identical during vertical dialect phonetic condition code storehouse.
Feature compensation code and the condition code of dialect phonetic are combined, and the condition code after being combined again with standard base voice passes through n times
Verification is corrected, and is combined after being corrected to the n times verification of dialect phonetic feature compensation code and is shown that standard base phonetic feature code is being permitted
Perhaps in error range, this feature compensation codes are the feature compensation code of the dialect phonetic, and searching number is assigned by this feature compensation codes
It is incorporated into feature compensation code storehouse.Available for later stage real-time dialect or the language translation system of personalized linguistic context.
Establish behind condition code storehouse and feature compensation code storehouse, produced with the corresponding source voice of feature compensation code is compound
Standard base voice under correspondence mother tongue system；The acquisition of feature compensation code has two ways：
Feature compensation code obtains step：Similarity retrieval is carried out by the feature compensation code and feature compensation code storehouse of source voice
Obtain corresponding feature compensation code.I.e. in the case of unknown languages, to source voice carry out condition code extraction, then with standard base language
The feature compensation code of sound generation source voice, the feature compensation code of source voice carries out maximum comparability with feature compensation code storehouse and retrieved
To corresponding feature compensation code, obtained feature compensation code carries out the compound standard produced under corresponding mother tongue system with source voice again
Base voice.
Feature compensation code, which is obtained, to be set manually by user.In the case of being applicable known languages, source voice directly with
The compound standard base voice produced under correspondence mother tongue system of given feature compensation code.
Standard base voice is converted into the voice or text of object language, and exports voice or text.
The present invention can translate the dialect for specifying languages according to user's request, can also translate the dialect of unknown languages.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
Any modifications, equivalent substitutions and improvements made within refreshing and principle etc., should be included in the scope of the protection.
Claims (9)
1. a kind of multi-language translation method of dialect linguistic context, it is characterised in that comprise the following steps：
Step 1, set up various dialect phonetics under mother tongue system condition code composition condition code storehouse；
Step 2, the feature compensation code for producing according to the difference of condition code and standard base phonetic feature code dialect, composition characteristic compensation
Code storehouse；
Step 3, with the compound standard base voice produced under correspondence mother tongue system of the corresponding source voice of feature compensation code；
Step 4, standard base voice are converted into the voice or text of object language.
2. a kind of multi-language translation method of dialect linguistic context according to claim 1, it is characterised in that condition code storehouse is built
Cube method comprises the following steps：
Various dialect phonetic samples under step 1.1, acquisition mother tongue system；
Step 1.2, to dialect phonetic sample carry out pretreatment remove redundancy section, pass through bandpass filter carry out pre-filtering processing
Preemphasis is carried out by a high-pass filter again afterwards, it is laggard to being multiplied by Hamming window after the dialect phonetic sample noise abatement after preemphasis
Row end-point detection；
Step 1.3, the pretreated dialect phonetic sample of Hamming window will be multiplied by carry out spectrum analysis, and then carry out feature extraction,
The formant of dialect phonetic sample after feature extraction spectrum analysis, pitch period feature, MFCC and LPCC parameter attributes code；
Step 1.4 carries out the formant of acquisition, pitch period feature, MFCC and LPCC parameter attributes code after de-redundancy to multiple
Dialect phonetic file carries out the probability distribution statistical of condition code, and finding out its denominator will as the condition code of the dialect phonetic
Element；
Step 1.5 by the code with condition code key element use compress mode to recompile for 64 bytes dialect phonetic feature
Code, assigns searching number by this feature code and is incorporated into condition code storehouse.
3. the multi-language translation method of a kind of dialect linguistic context according to claim 2, it is characterised in that right in step 1.3
The extraction step of the formant of dialect phonetic after spectrum analysis includes：
Smooth spectrum is obtained after homomorphic filtering to the dialect phonetic after spectrum analysis discrete Fourier transform is asked to the spectrum again,
Then the formant parameter of voice signal is extracted with DFT spectrums.
4. the multi-language translation method of a kind of dialect linguistic context according to claim 2, it is characterised in that right in step 1.3
The extraction step of the pitch period feature of dialect phonetic after spectrum analysis includes：
Pitch period feature is extracted using average magnitude difference function method to the dialect phonetic after spectrum analysis.
5. the multi-language translation method of a kind of dialect linguistic context according to claim 2, it is characterised in that right in step 1.3
The extraction step of the MFCC parameters of dialect phonetic after spectrum analysis includes：
Dialect phonetic after spectrum analysis is subjected to Short Time Fourier Transform and obtains its frequency spectrum, then ask spectrum amplitude square energy
Amount spectrum, bandpass filtering is carried out with triangle filter equalizer, the number of wave filter is close with critical band number, if wave filter number is M, filter
Obtain being output as after ripple：X (k), k=l, 2 ..., M, the output to wave filter group are taken the logarithm, and are then made 2M points and are become against Fourier
Change and can obtain MFCC parameters.
6. the multi-language translation method of a kind of dialect linguistic context according to claim 2, it is characterised in that right in step 1.3
The extraction step of the LPCC parameters of dialect phonetic after spectrum analysis includes：
Dialect phonetic after spectrum analysis is carried out to the inverse Z-transform of logarithm modular function after transform, become by the Fourier of signal
Change, the logarithm of modulus, then Fourier transformation of negating obtain LPCC parameters.
7. a kind of multi-language translation method of dialect linguistic context according to claim 1, it is characterised in that feature in step 2
The foundation in compensation codes storehouse uses following steps：
Step 2.1, the standard base speech samples obtained under mother tongue system, the condition code of extraction standard base voice；
Step 2.2, com-parison and analysis is carried out to dialect phonetic condition code under mother tongue system and standard base phonetic feature code, drawn general meaning
Respective probability distribution variances frequency spectrum；
Step 2.3, by the difference frequency spectrum carry out radix-minus-one complement superposition draw dialect phonetic feature compensation code；
Step 2.3, feature compensation code and the condition code of dialect phonetic are combined, and the condition code after being combined again with standard base voice is passed through
N times verification is corrected, and is combined after being corrected to the n times verification of dialect phonetic feature compensation code and is shown that standard base phonetic feature code exists
In the range of allowable error, this feature compensation codes are the feature compensation code of the dialect phonetic, and this feature compensation codes are assigned and retrieved
Number it is incorporated into feature compensation code storehouse.
8. according to a kind of multi-language translation method of any described dialect linguistic context of claim 1-7, it is characterised in that：Step 3
In feature compensation code obtaining step：Similarity retrieval is carried out by the feature compensation code of source voice with feature compensation code storehouse to obtain
Corresponding feature compensation code.
9. according to a kind of multi-language translation method of any described dialect linguistic context of claim 1-7, it is characterised in that：Step 3
In feature compensation code obtain be to be set manually by user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710368832.2A CN107248409A (en) | 2017-05-23 | 2017-05-23 | A kind of multi-language translation method of dialect linguistic context |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710368832.2A CN107248409A (en) | 2017-05-23 | 2017-05-23 | A kind of multi-language translation method of dialect linguistic context |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107248409A true CN107248409A (en) | 2017-10-13 |
Family
ID=60016797
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710368832.2A Pending CN107248409A (en) | 2017-05-23 | 2017-05-23 | A kind of multi-language translation method of dialect linguistic context |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107248409A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109446535A (en) * | 2018-10-22 | 2019-03-08 | 内蒙古工业大学 | A kind of illiteracy Chinese nerve machine translation method based on triangle framework |
| CN113836945A (en) * | 2021-09-23 | 2021-12-24 | 平安科技（深圳）有限公司 | Intention recognition method and device, electronic equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1225567A2 (en) * | 2001-01-23 | 2002-07-24 | Matsushita Electric Industrial Co., Ltd. | Method and apparatus for speech recognition |
| CN201403118Y (en) * | 2008-12-12 | 2010-02-10 | 康佳集团股份有限公司 | Device with dialect translating function and mobile terminal |
| CN102074234A (en) * | 2009-11-19 | 2011-05-25 | 财团法人资讯工业策进会 | Voice variation model building device and method as well as voice recognition system and method |
| CN102117614A (en) * | 2010-01-05 | 2011-07-06 | 索尼爱立信移动通讯有限公司 | Personalized text-to-speech synthesis and personalized speech feature extraction |
| CN102543073A (en) * | 2010-12-10 | 2012-07-04 | 上海上大海润信息系统有限公司 | Shanghai dialect phonetic recognition information processing method |
-
2017
- 2017-05-23 CN CN201710368832.2A patent/CN107248409A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1225567A2 (en) * | 2001-01-23 | 2002-07-24 | Matsushita Electric Industrial Co., Ltd. | Method and apparatus for speech recognition |
| CN201403118Y (en) * | 2008-12-12 | 2010-02-10 | 康佳集团股份有限公司 | Device with dialect translating function and mobile terminal |
| CN102074234A (en) * | 2009-11-19 | 2011-05-25 | 财团法人资讯工业策进会 | Voice variation model building device and method as well as voice recognition system and method |
| CN102117614A (en) * | 2010-01-05 | 2011-07-06 | 索尼爱立信移动通讯有限公司 | Personalized text-to-speech synthesis and personalized speech feature extraction |
| CN102543073A (en) * | 2010-12-10 | 2012-07-04 | 上海上大海润信息系统有限公司 | Shanghai dialect phonetic recognition information processing method |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109446535A (en) * | 2018-10-22 | 2019-03-08 | 内蒙古工业大学 | A kind of illiteracy Chinese nerve machine translation method based on triangle framework |
| CN113836945A (en) * | 2021-09-23 | 2021-12-24 | 平安科技（深圳）有限公司 | Intention recognition method and device, electronic equipment and storage medium |
| WO2023045186A1 (en) * | 2021-09-23 | 2023-03-30 | 平安科技（深圳）有限公司 | Intention recognition method and apparatus, and electronic device and storage medium |
| CN113836945B (en) * | 2021-09-23 | 2024-04-16 | 平安科技（深圳）有限公司 | Intention recognition method, device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107945805B (en) | A kind of across language voice identification method for transformation of intelligence | |
| Gawali et al. | Marathi isolated word recognition system using MFCC and DTW features | |
| Schlippe et al. | Grapheme-to-phoneme model generation for Indo-European languages | |
| Ali et al. | Advances in dialectal arabic speech recognition: A study using twitter to improve egyptian asr | |
| CN110853629A (en) | Speech recognition digital method based on deep learning | |
| Bharali et al. | A comparative study of different features for isolated spoken word recognition using HMM with reference to Assamese language | |
| CN113744722A (en) | Off-line speech recognition matching device and method for limited sentence library | |
| Tantawi et al. | A deep learning approach for automatic speech recognition of The Holy Qur’ān recitations | |
| Cardenas et al. | Siminchik: A speech corpus for preservation of southern quechua | |
| CN107248409A (en) | A kind of multi-language translation method of dialect linguistic context | |
| Mukherjee et al. | REARC-a Bangla Phoneme recognizer | |
| Abbas et al. | Pashto Spoken Digits database for the automatic speech recognition research | |
| Gedam et al. | Development of automatic speech recognition of Marathi numerals-a review | |
| Ahmed et al. | Technique for automatic sentence level alignment of long speech and transcripts. | |
| Mahmood | Arabic speaker recognition system based on phoneme fusion | |
| Sharan et al. | Speaker-independent recognition system for continuous Hindi speech using probabilistic model | |
| Chit et al. | Myanmar continuous speech recognition system using fuzzy logic classification in speech segmentation | |
| Liu et al. | Supra-Segmental Feature Based Speaker Trait Detection. | |
| Jin et al. | On continuous speech recognition of Indian English | |
| Coats | A pipeline for the large-scale acoustic analysis of streamed content | |
| Zubi et al. | Arabic Dialects System using Hidden Markov Models (HMMs) | |
| Laleye et al. | Automatic text-independent syllable segmentation using singularity exponents and rényi entropy | |
| Sreeraj et al. | Automatic dialect recognition using feature fusion | |
| Shinde et al. | Isolated Word Recognition System based on LPC and DTW Technique | |
| Gadekar et al. | Analysis of speech recognition techniques |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171013 |
|
| RJ01 | Rejection of invention patent application after publication |