CN106844343A - Instruction results screening plant - Google Patents
Instruction results screening plant Download PDFInfo
- Publication number
- CN106844343A CN106844343A CN201710048185.7A CN201710048185A CN106844343A CN 106844343 A CN106844343 A CN 106844343A CN 201710048185 A CN201710048185 A CN 201710048185A CN 106844343 A CN106844343 A CN 106844343A
- Authority
- CN
- China
- Prior art keywords
- result
- good
- clouds
- score value
- field
- 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.)
- Granted
Links
- 238000012216 screening Methods 0.000 title claims abstract description 60
- 238000004458 analytical method Methods 0.000 claims abstract description 123
- 238000004364 calculation method Methods 0.000 claims abstract description 25
- 230000002452 interceptive effect Effects 0.000 claims abstract description 19
- 238000001514 detection method Methods 0.000 claims abstract description 18
- 238000004891 communication Methods 0.000 claims description 33
- 230000000875 corresponding effect Effects 0.000 description 28
- 238000000034 method Methods 0.000 description 18
- 230000008569 process Effects 0.000 description 17
- 238000010586 diagram Methods 0.000 description 6
- 235000013399 edible fruits Nutrition 0.000 description 6
- 238000011156 evaluation Methods 0.000 description 4
- 238000000605 extraction Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000018199 S phase Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/28—Constructional details of speech recognition systems
- G10L15/32—Multiple recognisers used in sequence or in parallel; Score combination systems therefor, e.g. voting systems
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/50—Network service management, e.g. ensuring proper service fulfilment according to agreements
- H04L41/5041—Network service management, e.g. ensuring proper service fulfilment according to agreements characterised by the time relationship between creation and deployment of a service
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Machine Translation (AREA)
Abstract
The invention provides a kind of instruction results screening plant, it is arranged in the interactive voice terminal that user is held, the local candidate result of local semantic analysis device and the high in the clouds candidate result of high in the clouds semantic analysis device are received respectively and screening is carried out to two kinds of candidate results draw semantic analysis result, including：Unanimously judge acquisition unit, judge local candidate result and whether consistent candidate result is deposited in the candidate result of high in the clouds and is obtained；It is good at field detection unit, decision instruction type belongs to local and is good at field or field is good in high in the clouds；Score value Adjustable calculation portion, calculating is adjusted according to field score value regulation rule is good at successively；Consistent score value calculating part, is calculated consistent results score value；And the first result configuration part, the order by consistent candidate result according to consistent results score value from high to low arranged, and by predetermined ranking before consistent candidate result be set as semantic analysis result.
Description
Technical field
The present invention relates to a kind of instruction results screening plant, and in particular to a kind of instruction being arranged in interactive voice terminal
As a result screening plant.
Background technology
Using widely in current intelligent electronic device, its basis is exactly speech recognition and language to man-machine interactive system
Justice analyzes both technologies.Speech recognition refers to that people's one's voice in speech is converted into corresponding text, and semantic analysis is then by text
Originally be analyzed, obtain computer it will be appreciated that or perform instruction.Semantic analysis include participle, syntactic analysis, sentence pattern matching,
Keyword such as obtains at the step, and wherein each step is required for applying to the database for prestoring or setting, such as word storehouse,
Syntactic base, Sentence Template storehouse etc..
In the prior art, semantic analysis is more completes in the cloud server of supplier, due to can in cloud server
The database volume of storage is larger, therefore with being related to the extensive advantage in field.
In the mode of above-mentioned high in the clouds analysis, intelligent electronic device needs to keep networking with the cloud server moment, once net
Network is interrupted cannot then complete analysis.Some intelligent electronic devices simultaneously do not have network environment steady in a long-term (for example, being arranged on vapour
Intelligent electronic device in the transport facilitys such as car may enter without network environment in the process of moving), this is allowed for so
Intelligent electronic device need to set local semantic analysis device in therein.Storage volume due to intelligent electronic device etc.
Hardware condition is limited, and the local field to be analyzed of semantic analysis device is also limited, in the phonetic order that user sends
Often cannot get correct result when being related to wider field.
In addition, in the semantic analysis of prior art, each step can draw multiple intermediate results.For example, will use
When phonetic order corresponding text instruction in family carries out instruction type or instruction template matching, it will usually draw with Different matching value
Multiple instruction type or template；When carrying out keyword extraction to text instruction, multiple keywords can be generally also drawn.But
It is that generally only one optimal intermediate result of selection carries out subsequent step for the semantic analysis of prior art, and this allows for each step
It is rapid to be likely to lose some information.When the phonetic order that user is sent has certain ambiguity, such information loses pole
There is a possibility that final analysis result differs greatly with the original idea of user.
The content of the invention
To solve the above problems, there is provided one kind can realize that local and high in the clouds semantic analysis is combined, and can be as far as possible
Retain the result of each step to reduce the semantic analysis equipment of information loss, present invention employs following technical scheme：
The invention provides a kind of instruction results screening plant, it is arranged in the interactive voice terminal that user is held, leads to
The first communication network and the local semantic analysis device communication connection being arranged in the interactive voice terminal are crossed, by the second communication
Network and the high in the clouds semantic analysis device communication connection being arranged in the server of semantic analysis service supplier, receive this respectively
The ground local candidate result of semantic analysis device and the high in the clouds candidate result of high in the clouds semantic analysis device and to two kinds of candidates knot
Fruit carries out screening and draws semantic analysis result, and each candidate result is comprising instruction type, keyword and by corresponding instruction
The score value that type matching value and keyword score value are constituted, it is characterised in that including：It is consistent to judge acquisition unit, judge local
Candidate result with the presence or absence of instruction type and the consistent candidate result of keyword and obtains all of with the candidate result of high in the clouds
Consistent candidate result；Field detection unit is good at, is judged that the instruction type of consistent candidate result belongs to successively and is locally good at field also
It is that field is good in high in the clouds；Score value Adjustable calculation portion, to be good at field score value regulation rule consistent to each successively according to predetermined
The score value of local candidate result and high in the clouds candidate result in candidate result is adjusted calculating respectively；Consistent score value is calculated
Portion, meter is summed up by score value after the adjustment of local candidate result and high in the clouds candidate result in each consistent candidate result
Calculate, obtain consistent results score value；And the first result configuration part, by consistent candidate result according to consistent results score value from height
Arranged to low order, and by predetermined ranking before consistent candidate result be set as semantic analysis result.
The instruction results screening plant that the present invention is provided, can also have following technical characteristic：Wherein, predetermined ranking is the
Four.
The instruction results screening plant that the present invention is provided, can also have following technical characteristic：Wherein, when the field of being good at is sentenced
When determining the result of determination in portion and being not belonging to high in the clouds and be good at field to be not only not belonging to locally to be good at field, consistent score value calculating part general
The score value of local candidate result and high in the clouds candidate result in the consistent candidate result directly sums up calculating, obtains consistent
As a result score value.
The instruction results screening plant that the present invention is provided, can also include：Same type packet portion, when unanimously judging acquisition unit
Judged result the candidate result with identical instruction type in local candidate result and high in the clouds candidate result is remembered when being no
It is a same type group；Same type component value calculating part, the score value of the candidate result in same same type group is added
To the corresponding same type component value of the same type group；Optimal same type group configuration part, by same type component value highest same type
Group is set as optimal same type group；Field judging part is most good at, judges whether the instruction type of optimal same type group belongs to local
Most it is good at field or field is most good in high in the clouds；Second result configuration part, when the instruction type of optimal same type group belongs to locally most
When being good at field, the local candidate result of score value highest in the optimal same type group is set as semantic analysis result, when optimal
The instruction type of same type group belongs to high in the clouds when being most good at field, and score value highest high in the clouds candidate in the optimal same type group is tied
Fruit is set as semantic analysis result.
The instruction results screening plant that the present invention is provided, can also have following technical characteristic：Wherein, when being most good at field
The judged result of judging part is that the instruction type of optimal same type group had not only been not belonging to locally most be good at field but also be not belonging to high in the clouds most
When being good at field, the field detection unit of being good at judges that the instruction type of candidate result in optimal same type group is locally to be good at neck successively
Field is good in domain or high in the clouds, and score value Adjustable calculation portion is according to being good at field score value regulation rule successively to optimal same type group
The score value of middle candidate result is adjusted calculating, and the second result configuration part is by score value in optimal same type group after Adjustable calculation
Highest candidate result is set as semantic analysis result.
The instruction results screening plant that the present invention is provided, can also have following technical characteristic：Wherein, it is good at field score value
Regulation rule is：When the result of determination for being good at field detection unit is locally good at field to belong to, by the pass of local candidate result
Keyword score value is multiplied by a local regulation coefficient more than 1, when the result of determination for being good at field detection unit is arrogated to oneself to belong to high in the clouds
During field long, the keyword score value of high in the clouds candidate result is multiplied by a high in the clouds regulation coefficient more than 1.
Present invention also offers a kind of semantic analysis equipment, with the language being arranged in the interactive voice terminal that user is held
Sound conversion equipment is communicated to connect, and the semantic text that be converted into for user speech instruction by reception speech apparatus is simultaneously literary to the semanteme
Originally it is analyzed, it is characterised in that including：High in the clouds semantic analysis device, is arranged on the server of semantic analysis service supplier
It is interior, semantic text is received by the first communication network and high in the clouds semantic analysis is carried out to semantic text, obtain multiple high in the clouds candidates
As a result；Local semantic analysis device, is arranged in interactive voice terminal, receives semantic text and the semantic text is carried out locally
Semantic analysis, obtains multiple local candidate results；Instruction results screening plant, is arranged in interactive voice terminal, by first
Communication network and be arranged in the interactive voice terminal local semantic analysis device communication connection, by the second communication network with
The high in the clouds semantic analysis device communication connection in the server of semantic analysis service supplier is arranged on, receives local semantic respectively
The local candidate result of analytical equipment and the high in the clouds candidate result of high in the clouds semantic analysis device are simultaneously carried out to two kinds of candidate results
Screening draws semantic analysis result, wherein, instruction results screening plant is instruction results screening plant as described above.
Invention effect and effect
Instruction results screening plant of the invention, consistent candidate's knot can be successively judged due to being good at field detection unit
The instruction type of fruit belongs to local and is good at field or field is good in high in the clouds, and score value Adjustable calculation portion can be according to arrogating to oneself accordingly
Field long carries out score value adjustment respectively, therefore is carried out in the multiple candidate results drawn to high in the clouds and local semantic analysis device
More accurate instruction results can be drawn according to different field screenings of being good at during screening.
Brief description of the drawings
Fig. 1 is that semantic analysis equipment of the present invention in embodiment constitutes schematic diagram；
Fig. 2 is the block diagram of instruction results screening plant of the present invention in embodiment；
Fig. 3 is the workflow diagram of semantic analysis equipment；
The semantic analysis process flow chart that Fig. 4 is carried out by local semantic analysis device or high in the clouds semantic analysis device；
Fig. 5 is the instruction screening process figure of instruction results screening plant.
Specific embodiment
Specific embodiment of the invention is illustrated below in conjunction with drawings and Examples.
<Embodiment>
Fig. 1 is that semantic analysis equipment of the present invention in embodiment constitutes schematic diagram.
As shown in figure 1, semantic analysis equipment 1 includes local semantic analysis device 2, instruction results screening plant 3 and cloud
End semantic analysis device 4.
Local semantic analysis device 2 and instruction results screening plant 3 are arranged at an interactive voice held by user
In terminal 5, the two is communicated to connect by first communication network 6.A voice conversion is additionally provided with interactive voice terminal 5
Equipment 7, the voice for user to be spoken accordingly is converted to text, and the speech apparatus 7 are also by the first communication network 6
Communicated to connect with local semantic analysis device 2.
High in the clouds semantic analysis device 4 is arranged in the server 8 of semantic analysis service supplier, by the second communication network
9 communicate to connect with speech apparatus 7 and instruction results screening plant 3 respectively, receive the text that speech apparatus 7 are changed out
This, instruction results screening plant 3 is transferred to after carrying out semantic analysis by the result of semantic analysis by the second communication network 9.Together
When, the File Transfer that speech apparatus 7 will be also changed out by the first communication network 6 to local semantic analysis device 2, locally
Semantic analysis device 2 is transferred to instruction results screening plant 3 after drawing semantic analysis result by the first communication network 6.
Local semantic analysis device 2 and high in the clouds semantic analysis device 4 are made for the text received from speech apparatus 7
Analysis process all same, its purpose is primarily to show user wishes what does, that is, obtain when user sends voice
Go out the instruction results corresponding with user speech.In analysis, mainly drawn and text instruction's phase by various sorting algorithms
The instruction type (i.e. user wishes the action carried out) matched somebody with somebody, keyword (the i.e. user that command adapted thereto is drawn by keyword extraction
Wish the object of implementation action), while also obtaining reflecting the instruction type matching value of instruction type and text instruction's matching degree
With the keyword score value of reflection keyword confidence level.Therefore, an instruction type and corresponding keyword constitute a finger
The candidate result of order, corresponding instruction type matching value and keyword score value are the score value for constituting the candidate result.
In the present embodiment, interactive voice terminal 5 is arranged in automobile, and the first communication network 6 is the low coverage in the automobile
From wireless-transmission network or wire transmission network, the second communication network 9 is wide area network (such as Internet).
Fig. 2 is the block diagram of instruction results screening plant of the present invention in embodiment.
As shown in Fig. 2 instruction results screening plant 3 includes unanimously judging acquisition unit 11, is good at field judging part 12, scoring
Value Adjustable calculation portion 13, consistent score value calculating part 14, the first result configuration part 15, same type packet portion 16, optimal same type
Group configuration part 17, be most good at field judging part 18, the second result configuration part 19, same type score value calculating part, communication unit 21 and
Control unit 22.
It is consistent judge acquisition unit 11 for judge pass through local candidate result and the high in the clouds candidate knot that communication unit 21 is received
Whether there is instruction type and the consistent candidate result of keyword in fruit, and obtained when there is consistent candidate result all
Consistent candidate result.
Field detection unit 12 is good at after unanimously judging that acquisition unit 11 gets all consistent candidate results, this is judged successively
The instruction type of some consistent candidate results belongs to local and is good at field or field is good in high in the clouds.Wherein, be locally good at field and
It is pre-set all in accordance with actual conditions that field is good in high in the clouds, for example, in the present embodiment, due to the storage of interactive voice tag memory
When having the music store of user, therefore instruction type for " broadcasting music ", locally it is easier to draw accurate result, the instruction type
To be locally good at field.
Score value Adjustable calculation portion 13 is used to being good at field judging part 12 and judging the instruction type of certain candidate result
Belong to local to be good at field or carry out score value Adjustable calculation to the candidate result after field is good in high in the clouds.
The Adjustable calculation that score value Adjustable calculation portion 13 is carried out is used and is good at field score value regulation rule, i.e.,：When being good at
Field detection unit 12 judges that the instruction type of a local candidate result belongs to locally when being good at field, by the local candidate result
Keyword score value be multiplied by the corresponding local regulation coefficient of instruction type；Judge one when field detection unit 12 is good at
The instruction type of high in the clouds candidate result belongs to high in the clouds when being good at field, and the keyword score value of the high in the clouds candidate result is multiplied by into one
The corresponding high in the clouds regulation coefficient of the individual instruction type.Wherein, local regulation coefficient and high in the clouds regulation coefficient are according to instruction type
It is different and preset, and it is all higher than 1.In addition, when the instruction type of candidate result had not only been not belonging to locally to be good at field but also not
When belonging to high in the clouds and being good at field, score value Adjustable calculation portion 13 does not make any adjustment to it.
Consistent score value calculating part 14 is used for local candidate result in consistent candidate result and high in the clouds candidate result
Score value sums up calculating, obtains the consistent results score value of the consistent candidate result.
When a local candidate result is consistent with a high in the clouds candidate result, that is, instruction type and keyword it is complete
When identical, be added for the score value of the two by consistent score value calculating part 14, obtains the score value of the consistent candidate result.For warp
The local candidate result or high in the clouds candidate result of the Adjustable calculation of score value Adjustable calculation portion 13 are crossed, consistent score value calculating part 14 will
The score value obtained after adjustment is added；For local candidate result or cloud without the Adjustable calculation of score value Adjustable calculation portion 13
Directly be added for its score value by end candidate result, consistent score value calculating part 14.
First result configuration part 15 be used for by all of consistent candidate result according to consistent results score value from high to low
Order arranged, and by predetermined ranking before consistent candidate result be set as semantic analysis result, by communication unit 21 will
The semantic analysis result is transferred to the next execution equipment, allows the next execution equipment to perform corresponding action.In the present embodiment, should
Predetermined ranking sets the consistent candidate result that consistent results score value comes front three for the 4th, i.e. the first result configuration part 15
It is set to final semantic analysis result.
Same type packet portion 16 is used in unanimously judging that acquisition unit 11 judges high in the clouds candidate result and local candidate result
By instruction type when there is no consistent candidate result of instruction type and keyword, but there is instruction type identical candidate result
Identical candidate result is designated as a same type group.
Be added for the score value of each candidate result in each same type group successively by same type component value calculating part 20,
Obtain the corresponding same type component value of the same type group, optimal same type group configuration part 17 is by same type component value highest one
Same type group is set as optimal same type group, is most good at field judging part 18 for judging the instruction type of the optimal same type group
Whether belong to local and be most good at field or field is most good in high in the clouds.
In the present embodiment, locally most it is good at field and is good at that field is most good in field, high in the clouds and high in the clouds is most good at local
Domain variability is incomplete same.To be locally most good at field and locally be good at as a example by field, " broadcasting music " " broadcasting radio " etc.
The local field for being easier to make Correct Analysis belongs to locally is good at field, but the analysis that " calling " this kind of high in the clouds is made can
Reliability is locally most good at field far below local then belonging to.
Different confidence levels of the aforementioned four field all in accordance with different instruction type done by local or high in the clouds when set in advance
Fixed, the local field of being good at contains and is locally most good at field in the present embodiment, and similarly the high in the clouds field of being good at contains high in the clouds most
It is good at field.Certainly, if there is inclusion relation can also set according to actual conditions, locally the field of being good at can not include or
Field is most good in part comprising local, and the high in the clouds field of being good at can also not include or field is most good in part comprising high in the clouds.
Second result configuration part 19 is used for should when the instruction type of optimal same type group belongs to and is locally most good at field
The local candidate result of score value highest is set as semantic analysis result in optimal same type group, in the instruction class of optimal same type group
Type belongs to and score value highest high in the clouds candidate result in the optimal same type group is set as into semantic analysis when field is most good in high in the clouds
As a result.When the instruction type of optimal same type group had not only been not belonging to locally most be good at field but also be not belonging to high in the clouds and be most good at field,
Candidate result in the optimal same type group is commented successively according to field score value regulation rule is good in score value Adjustable calculation portion 13
Score value is adjusted calculating, and score value highest candidate in optimal same type group after Adjustable calculation is tied in the second result configuration part 19
Fruit is set as semantic analysis result.
Below in conjunction with the workflow for illustrating semantic analysis equipment of the invention.
Fig. 3 is the workflow diagram of semantic analysis equipment.
As shown in figure 3, when user sends a voice, the semantic analysis equipment 1 of the embodiment of the present invention just by this once
Voice content be analyzed as phonetic order, its analysis process comprises the following steps：
Content Transformation for voice is corresponding text by step S1, speech apparatus 7, is obtained and phonetic order
Corresponding text instruction；
The text is instructed and is transferred to local semantic analysis dress by step S2, speech apparatus 7 by the first communication network 6
2 are put, while the text is instructed by the second communication network 9 being transmitted to high in the clouds semantic analysis device 4；
Step S3, the text instruction that 2 pairs, local semantic analysis device is received carries out semantic analysis, draws multiple local times
Select result；Meanwhile, the text instruction that 4 pairs, high in the clouds semantic analysis device is received carries out semantic analysis, draws multiple high in the clouds candidates
As a result.
Step S4, instruction results screening plant 3 receives local candidate result and high in the clouds candidate result and is instructed respectively
Result is screened, and screening is transmitted to the next execution equipment after obtaining semantic analysis result, allows the next execution equipment to perform corresponding action.
The semantic analysis process flow chart that Fig. 4 is carried out by local semantic analysis device or high in the clouds semantic analysis device.
As shown in figure 4, in step S3, the semanteme that local semantic analysis device 2 and high in the clouds semantic analysis device 4 are carried out point
Analysis process is identical, comprises the following steps：
Step S3-1, the text instruction to receiving carries out participle and attribute labeling, obtains multiple participle paths.Its
In, each participle path is comprising multiple words arranged according to certain word order, and each word is labeled with corresponding category
Property.The dictionary that participle process is based on prestoring is carried out, and the dictionary includes word and the corresponding many attribute of each word.Example
Such as, in text instruction " I want to listen see sea ", there are storage, therefore the text in " I ", " thinking ", " listening ", " seeing ", " sea " in dictionary
One participle path of instruction is " I/think/listen/see/sea "；Meanwhile, " seeing sea " also has storage in dictionary, and its attribute is " song
Song name ", therefore another participle path of text instruction is " I/think/listen/see sea ".
Step S3-2, characteristics extraction is carried out from multiple participle paths that step S3-1 is obtained, will be with particular community
Word as characteristics extraction out.Under normal circumstances, can be some verbs as characteristic value, for example " think ", " listening ",
" seeing ".
Step S3-3, extracting the characteristic value for obtaining according to step S3-2 using different matching algorithms carries out instruction type
Match somebody with somebody, i.e., classified using sorting algorithm according to characteristic value, obtain multiple corresponding instruction types and each instruction type with text
The instruction type matching value of this instruction.
Matching algorithm employed in the present embodiment is for direct template matching method, regular expression algorithm and based on engineering
The condition random field algorithm of habit.Three kinds of algorithms can be derived that multiple corresponding instruction types, each instruction type and text
Instruction is respectively provided with certain matching degree, and the matching degree draws while matching algorithm draws instruction type.When different
When algorithm draws same instruction type, the matching angle value that algorithms of different is drawn is multiplied by phase after the corresponding weighted value of the algorithm
Plus, what is obtained is exactly the instruction type matching value of the instruction type, and the weighted value of wherein algorithms of different is drawn according to the algorithm
The confidence level of result is set in advance.
Step S3-4, extracts keyword.Using participle path evaluation method from a plurality of participle path that step S3-1 draws
At least one optimal participle path is selected, then using the word with particular community in the optimal participle path as key
Word.Wherein, particular community is generally noun, such as name, road name, song name etc..
In the present embodiment, participle path evaluation method is to each word in a participle path according to default participle value storehouse
Language gives corresponding participle value, is then added the participle value of all words in this participle path, obtains commenting for the paths
Value, optimal participle path is one or more optimal participle paths of evaluation of estimate.For example, the participle road of " I wants to listen to see sea "
Footpath " I/think/listen/see/sea " and " I/think/listen/see sea " in, only " see/sea " different with the participle value of " seeing sea ", therefore the two
The evaluation of estimate for drawing is approached, and finally using this two participle paths as optimal participle path, extracting keyword therein is
" sea " and " seeing sea ".
Step S3-5, the multiple instruction type that step S3-3 is drawn is extracted the keyword for obtaining with step S3-4 and is combined,
Candidate result is formed, and sends candidate result to instruction results screening plant 3.Its cohesive process is needed according to different instruction class
The corresponding keyword attribute of type is carried out, for example, the keyword attribute corresponding to instruction type " listening " is song name, singer's name etc., because
This instruction type keyword to be combined is " seeing sea ".In each candidate result comprising instruction type, keyword with
And the score value being made up of corresponding instruction type matching value and keyword score value.
In the present embodiment, local semantic analysis device 2 and high in the clouds semantic analysis device 4 be not by all of candidate result
Instruction results screening plant 3 is transmitted to, but only the candidate result by score value more than predetermined value is transmitted to instruction results screening dress
3 are put, to ensure that the too low result of confidence level will not be transmitted to instruction results screening plant 3.
Fig. 5 is the instruction screening process figure of instruction results screening plant.
As shown in figure 5, instruction results screening fills 3 instruction screening process, i.e. step S4 comprises the following steps：
Step S4-1, receives local candidate result and high in the clouds candidate result respectively；
Step S4-2, unanimously judges that whether acquisition unit 11 judges local candidate result and has consistent in the candidate result of high in the clouds
Candidate result, step S4-3 is entered when judged result is to be, step S4-7 is entered when judged result is no；
Step S4-3, unanimously judges that acquisition unit 11 obtains consistent candidate result；
Step S4-4, is good at the instruction of the consistent candidate result that field detection unit 12 is got in determination step S4-3 successively
Type belongs to local and is good at field or field is good in high in the clouds, and score value Adjustable calculation portion 13 advises according to field score value adjustment is good at
Then and it is good at the judged result of field judging part 12 and carries out score value Adjustable calculation；
Step S4-5, consistent score value calculating part 14 ties local candidate result in consistent candidate result and high in the clouds candidate
The score value of fruit sums up calculating, obtains the consistent results score value of the consistent candidate result；
Step S4-6, the first result configuration part 15 by all of consistent candidate result according to consistent results score value from height to
Low order is arranged, and will be come the consistent candidate result of first 3 and be set as semantic analysis result, this instruction results sieve
Process is selected to terminate.
Instruction type identical candidate result is designated as a same type group by step S4-7, same type packet portion 16；
Step S4-8, the commenting each candidate result in each same type group successively of same type component value calculating part 20
Score value is added, and obtains the corresponding same type component value of the same type group；
Be set as same type component value one same type group of highest most by step S4-9, optimal same type group configuration part 17
Excellent same type group；
Step S4-10, is most good at field judging part 18 and judges whether the instruction type of the optimal same type group belongs to local
Most be good at field or field be most good in high in the clouds, when judged result for belong to it is local be most good at field when enter step S4-11, when sentencing
Disconnected result enters step S4-12 to belong to when field is most good in high in the clouds, when judged result for be both not belonging to locally most to be good at field,
It is not belonging to enter step S4-13 when field is most good in high in the clouds again；
Be set as the local candidate result of score value highest in optimal same type group by step 4-11, the second result configuration part 19
Semantic analysis result, this time instruction results screening process terminates.
Be set as score value highest high in the clouds candidate result in optimal same type group by step 4-12, the second result configuration part 19
Semantic analysis result, this time instruction results screening process terminates.
Step 4-13, is good at field detection unit 12 and judges that the instruction type of optimal same type group belongs to and be locally good at field also
It is that field is good in high in the clouds, when result of determination is good at field or when field is good in high in the clouds to belong to local, score value Adjustable calculation portion
13 bases are good at score value of the field score value regulation rule successively to candidate result in optimal same type group and are adjusted calculating；
Step 4-14, the second result configuration part 19 is by score value highest in optimal same type group after step 4-13 Adjustable calculations
Candidate result be set as semantic analysis result, this time instruction results screening process terminates.
After the semantic analysis result that communication unit 21 will draw in said process is transmitted to the next execution equipment, the next execution equipment
Corresponding action can be performed according to the situation of the semantic analysis result for receiving.For example, when in semantic analysis result with two
Individual above candidate result, and score value it is higher when, illustrate that these credible result degree are all higher, the next execution equipment can be ask
Ask which user selects to perform；When only one of which candidate result score value is higher, the next execution equipment can direct basis
The candidate result performs corresponding actions；When without score value candidate result higher, illustrate without draw can
Reliability result higher, the next execution equipment can allow user to re-emit phonetic order.
Embodiment is acted on and effect
Instruction results screening plant according to the present embodiment, consistent candidate can be successively judged due to being good at field detection unit
The instruction type of result belongs to local and is good at field or field is good in high in the clouds, and score value Adjustable calculation portion can be according to corresponding
Be good at field carries out score value adjustment respectively, therefore enters in the multiple candidate results drawn to high in the clouds and local semantic analysis device
More accurate instruction results can be drawn according to different field screenings of being good at during row screening.
When it is local draw consistent candidate result with high in the clouds when, judge and after score value is adjusted by above-mentioned field of being good at,
The consistent results score value for calculating is summed up using the score value after adjustment can preferably reflect the consistent results
It is whether accurate credible；When locally consistent candidate result is not drawn with high in the clouds, certain is not belonging to most in optimal same type group
In the case of being good at field, this adjustment can also be adjusted correspondingly to belonging to the different candidate results for being good at field, allow
The final result for drawing is more credible.
Due to being most good at field detection unit, thus some instruction types be substantially high in the clouds or it is local in one kind point
During the type that analysis just can accurately draw, corresponding candidate result directly can be set as semantic analysis result, exclude other one
Plant the interference produced by the result of the analytical equipment that can not accurately draw.
In addition, the semantic analysis equipment of the present embodiment can be screened the candidate result in multiple local and high in the clouds, and
And before the screening only by local semantic analysis device and high in the clouds semantic analysis device eliminate score value substantially it is relatively low, obvious
Incredible candidate result, therefore, it is possible to allow different candidate results to enter screening process as much as possible, retains more letters
Breath so that the result precision that screening draws is higher.
Claims (7)
1. a kind of instruction results screening plant, is arranged in the interactive voice terminal that user is held, by the first communication network
Be arranged in the interactive voice terminal local semantic analysis device communication connection, by the second communication network be arranged on language
High in the clouds semantic analysis device communication connection in the server of adopted Analysis Service supplier, receives the local semantic analysis respectively
The high in the clouds candidate result of the local candidate result of device and the high in the clouds semantic analysis device is simultaneously carried out to two kinds of candidate results
Screening draws semantic analysis result, and each described candidate result is comprising instruction type, keyword and by corresponding instruction class
The score value that type matching value and keyword score value are constituted, it is characterised in that including：
Unanimously judge acquisition unit, judge to whether there is the instruction class in the local candidate result and the high in the clouds candidate result
Type and the consistent candidate result of the keyword simultaneously obtain all of consistent candidate result；
Field detection unit is good at, is judged that the instruction type of the consistent candidate result belongs to successively and is locally good at field still
It is good at field in high in the clouds；
Score value Adjustable calculation portion, field score value regulation rule is good at successively to consistent candidate result each described according to predetermined
In local candidate result and the score value of high in the clouds candidate result be adjusted calculating respectively；
Consistent score value calculating part, by local candidate result and the tune of high in the clouds candidate result in each described consistent candidate result
Whole rear score value sums up calculating, obtains consistent results score value；And
First result configuration part, the order by the consistent candidate result according to the consistent results score value from high to low is carried out
Arrangement, and by predetermined ranking before described consistent candidate result be set as the semantic analysis result.
2. instruction results screening plant according to claim 1, it is characterised in that：
Wherein, the predetermined ranking is fourth.
3. instruction results screening plant according to claim 1, it is characterised in that：
Wherein, when the result of determination for being good at field detection unit is to be not only not belonging to described local be good at field and described in being not belonging to
When field is good in high in the clouds, the consistent score value calculating part is by the local candidate result in the consistent candidate result and high in the clouds
The score value of candidate result directly sums up calculating, obtains the consistent results score value.
4. instruction results screening plant according to claim 1, it is characterised in that also include：
Same type packet portion, when the consistent judged result for judging acquisition unit is for no by the local candidate result and described
The candidate result with instruction type described in identical is designated as a same type group in the candidate result of high in the clouds；
Same type component value calculating part, the score value of the candidate result in the same same type group is added
To the corresponding same type component value of the same type group；
Optimal same type group configuration part, optimal same type is set as by same type group described in the same type component value highest
Group；
Field judging part is most good at, is judged whether the instruction type of the optimal same type group belongs to and is locally most good at field
Or field is most good in high in the clouds；
Second result configuration part, when the instruction type of the optimal same type group belongs to locally most is good at field, by this
Local candidate result is set as the semantic analysis result described in score value highest in optimal same type group, when described optimal similar
The instruction type of type group belongs to high in the clouds when being most good at field, and high in the clouds described in score value highest in the optimal same type group is waited
Result is selected to be set as the semantic analysis result.
5. instruction results screening plant according to claim 4, it is characterised in that：
Wherein, when the instruction type that the judged result for being most good at field judging part is the optimal same type group neither
Belong to local when being most good at field and being not belonging to high in the clouds again and be most good at field, it is described be good at field detection unit judge successively it is described most
The instruction type of candidate result described in excellent same type group is locally to be good at field or field is good in high in the clouds,
Field score value regulation rule is good at according to successively in the optimal same type group in the score value Adjustable calculation portion
The score value of the candidate result is adjusted calculating,
The second result configuration part is by candidate described in the score value highest in the optimal same type group after Adjustable calculation
Result is set as the semantic analysis result.
6. instruction results screening plant according to claim 1, it is characterised in that：
Wherein, the field score value regulation rule of being good at is：
When the result of determination for being good at field detection unit for belong to it is described it is local be good at field when, by the local candidate result
The keyword score value be multiplied by one more than 1 local regulation coefficient,
When the result of determination for being good at field detection unit is good at field to belong to the high in the clouds, by the high in the clouds candidate result
The keyword score value be multiplied by one more than 1 high in the clouds regulation coefficient.
7. a kind of semantic analysis equipment, with the speech apparatus communication link being arranged in the interactive voice terminal that user is held
Connect, the semantic text that be converted into for user speech instruction by the reception speech apparatus is simultaneously analyzed to the semantic text,
It is characterised in that it includes：
High in the clouds semantic analysis device, is arranged in the server of semantic analysis service supplier, is received by the first communication network
The semantic text simultaneously carries out high in the clouds semantic analysis to the semantic text, obtains multiple high in the clouds candidate results；
Local semantic analysis device, is arranged in the interactive voice terminal, receives the semantic text and to the semantic text
Local semantic analysis is carried out, multiple local candidate results are obtained；
Instruction results screening plant, is arranged in the interactive voice terminal, by the first communication network be arranged on the voice
In interactive terminal local semantic analysis device communication connection, by the second communication network be arranged on semantic analysis service provision
High in the clouds semantic analysis device communication connection in the server of business, receives the local candidate of the local semantic analysis device respectively
The high in the clouds candidate result of result and the high in the clouds semantic analysis device simultaneously carries out screening to two kinds of candidate results and draws semantic point
Analysis result,
Wherein, the instruction results screening plant is the instruction results screening plant as any one of claim 1~6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710048185.7A CN106844343B (en) | 2017-01-20 | 2017-01-20 | Instruction results screening plant |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710048185.7A CN106844343B (en) | 2017-01-20 | 2017-01-20 | Instruction results screening plant |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106844343A true CN106844343A (en) | 2017-06-13 |
| CN106844343B CN106844343B (en) | 2019-11-19 |
Family
ID=59121055
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710048185.7A Active CN106844343B (en) | 2017-01-20 | 2017-01-20 | Instruction results screening plant |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106844343B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108806682A (en) * | 2018-06-12 | 2018-11-13 | 奇瑞汽车股份有限公司 | The method and apparatus for obtaining Weather information |
| CN109767763A (en) * | 2018-12-25 | 2019-05-17 | 苏州思必驰信息科技有限公司 | It is customized wake up word determination method and for determine it is customized wake up word device |
| CN110223679A (en) * | 2019-06-14 | 2019-09-10 | 南京机电职业技术学院 | A kind of voice recognition input devices |
| CN110276073A (en) * | 2019-06-14 | 2019-09-24 | 南京机电职业技术学院 | A kind of interactive mode Oral English Practice bearing calibration |
| CN111477225A (en) * | 2020-03-26 | 2020-07-31 | 北京声智科技有限公司 | Voice control method and device, electronic equipment and storage medium |
| CN112614490A (en) * | 2020-12-09 | 2021-04-06 | 北京罗克维尔斯科技有限公司 | Method, device, medium, equipment, system and vehicle for generating voice instruction |
| CN113593524A (en) * | 2021-01-26 | 2021-11-02 | 腾讯科技（深圳）有限公司 | Method and device for training acoustic model for accent recognition, and storage medium |
| CN114974228A (en) * | 2022-05-24 | 2022-08-30 | 名日之梦(北京)科技有限公司 | Rapid voice recognition method based on hierarchical recognition |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103020164A (en) * | 2012-11-26 | 2013-04-03 | 华北电力大学 | Semantic search method based on multi-semantic analysis and personalized sequencing |
| CN104063486A (en) * | 2014-07-03 | 2014-09-24 | 四川中亚联邦科技有限公司 | Big data distributed storage method and system |
| CN104580437A (en) * | 2014-12-30 | 2015-04-29 | 创新科存储技术（深圳）有限公司 | Cloud storage client and high-efficiency data access method thereof |
| US20160019299A1 (en) * | 2014-07-17 | 2016-01-21 | International Business Machines Corporation | Deep semantic search of electronic medical records |
| US20160379086A1 (en) * | 2015-06-29 | 2016-12-29 | International Business Machines Corporation | Systems and methods for inferring gender by fusion of multimodal content |
-
2017
- 2017-01-20 CN CN201710048185.7A patent/CN106844343B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103020164A (en) * | 2012-11-26 | 2013-04-03 | 华北电力大学 | Semantic search method based on multi-semantic analysis and personalized sequencing |
| CN104063486A (en) * | 2014-07-03 | 2014-09-24 | 四川中亚联邦科技有限公司 | Big data distributed storage method and system |
| US20160019299A1 (en) * | 2014-07-17 | 2016-01-21 | International Business Machines Corporation | Deep semantic search of electronic medical records |
| CN104580437A (en) * | 2014-12-30 | 2015-04-29 | 创新科存储技术（深圳）有限公司 | Cloud storage client and high-efficiency data access method thereof |
| US20160379086A1 (en) * | 2015-06-29 | 2016-12-29 | International Business Machines Corporation | Systems and methods for inferring gender by fusion of multimodal content |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108806682A (en) * | 2018-06-12 | 2018-11-13 | 奇瑞汽车股份有限公司 | The method and apparatus for obtaining Weather information |
| CN109767763A (en) * | 2018-12-25 | 2019-05-17 | 苏州思必驰信息科技有限公司 | It is customized wake up word determination method and for determine it is customized wake up word device |
| CN110223679A (en) * | 2019-06-14 | 2019-09-10 | 南京机电职业技术学院 | A kind of voice recognition input devices |
| CN110276073A (en) * | 2019-06-14 | 2019-09-24 | 南京机电职业技术学院 | A kind of interactive mode Oral English Practice bearing calibration |
| CN111477225A (en) * | 2020-03-26 | 2020-07-31 | 北京声智科技有限公司 | Voice control method and device, electronic equipment and storage medium |
| CN111477225B (en) * | 2020-03-26 | 2021-04-30 | 北京声智科技有限公司 | Voice control method and device, electronic equipment and storage medium |
| CN112614490A (en) * | 2020-12-09 | 2021-04-06 | 北京罗克维尔斯科技有限公司 | Method, device, medium, equipment, system and vehicle for generating voice instruction |
| CN112614490B (en) * | 2020-12-09 | 2024-04-16 | 北京罗克维尔斯科技有限公司 | Method, device, medium, equipment, system and vehicle for generating voice instruction |
| CN113593524A (en) * | 2021-01-26 | 2021-11-02 | 腾讯科技（深圳）有限公司 | Method and device for training acoustic model for accent recognition, and storage medium |
| CN114974228A (en) * | 2022-05-24 | 2022-08-30 | 名日之梦(北京)科技有限公司 | Rapid voice recognition method based on hierarchical recognition |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106844343B (en) | 2019-11-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106844343B (en) | Instruction results screening plant | |
| US11322153B2 (en) | Conversation interaction method, apparatus and computer readable storage medium | |
| CN105630876B (en) | Information acquisition method and device across application | |
| CN107832468B (en) | Demand recognition methods and device | |
| CN105336342B (en) | Speech recognition result evaluation method and system | |
| CN103544188B (en) | The user preference method for pushing of mobile Internet content and device | |
| CN107797984A (en) | Intelligent interactive method, equipment and storage medium | |
| CN107633227A (en) | A kind of fine granularity gesture identification method and system based on CSI | |
| CN107704453A (en) | A kind of word semantic analysis, word semantic analysis terminal and storage medium | |
| WO2017206661A1 (en) | Voice recognition method and system | |
| CN108447471A (en) | Audio recognition method and speech recognition equipment | |
| CN107992585A (en) | Universal tag method for digging, device, server and medium | |
| CN106874441A (en) | Intelligent answer method and apparatus | |
| CN109190124B (en) | Method and apparatus for participle | |
| CN109213844A (en) | A kind of text handling method, device and relevant device | |
| CN109033075B (en) | Intention matching method and device, storage medium and terminal equipment | |
| CN107104994A (en) | Audio recognition method, electronic installation and speech recognition system | |
| CN109657063A (en) | A kind of processing method and storage medium of magnanimity environment-protection artificial reported event data | |
| CN106407377A (en) | Search method and device based on artificial intelligence | |
| CN113778871A (en) | Mock testing method, device, equipment and storage medium | |
| CN109492081A (en) | Text information search and information interacting method, device, equipment and storage medium | |
| CN107239450A (en) | Natural language method is handled based on Interaction context | |
| CN106844651B (en) | Instruction results compare screening plant | |
| CN114973086A (en) | Video processing method and device, electronic equipment and storage medium | |
| CN104391969A (en) | User query statement syntactic structure determining method and device |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |