Summary of the invention
The main purpose of the present invention is to provide a kind of answering method based on speech analysis, device, equipment and computersReadable storage medium storing program for executing, it is intended to the technical issues of solving existing attendant consultation service efficiency of service low technical problem.
To achieve the above object, the present invention provides a kind of answering method based on speech analysis, described to be based on speech analysisAnswering method the following steps are included:
When receiving user's consulting instruction, the phonetic problem information of target user is obtained, and pass through and preset speech recognitionModel identifies the phonetic problem information, generates the corresponding text information of the phonetic problem information and emotional information；
According to the text information, the problem of judging the phonetic problem information, type is Products Show type or problemSeek advice from type；
If the problem of phonetic problem information type is Products Show type, according to the text information and the feelingsThread information scans for default words art forest, and the product information of recommended products is added to candidate words art template, obtains voiceThe corresponding Products Show of problem information talks about art, and shows the Products Show words art.
Optionally, described when receiving user's consulting instruction, the phonetic problem information of target user is obtained, and by pre-If speech recognition modeling identifies the phonetic problem information, the corresponding text information of the phonetic problem information and mood letter are generatedBefore the step of breath, further includes:
The default dialect family of languages is acquired respectively and the mandarin family of languages reads aloud voice data, and it is corresponding to extract each voice dataSpeech characteristic parameter, each speech characteristic parameter is formed into phonetic feature set；
Each speech characteristic parameter of preset ratio in the phonetic feature set is extracted, and special by each voiceIt levies parameter and constructs initial speech identification model；
By the iterative algorithm training initial speech identification model, and obtain the language for the speech recognition modeling that training obtainsRecognition accuracy is higher than the speech recognition modeling of preset threshold as default speech recognition modeling by sound recognition accuracy.
Optionally, described according to the text information, the problem of judging the phonetic problem information, type is Products ShowAfter the step of type or problem consulting type, further includes:
If the problem of phonetic problem information type is that problem seeks advice from type, according to the text information in default topicCorresponding issue-resolution is searched in library, and is scanned for according to the emotional information in default words art forest；
Described problem solution is added to candidate words art template, obtains the corresponding solution answer of phonetic problem informationArt, and show described problem solution answer art.
Optionally, described when receiving user's consulting instruction, the phonetic problem information of target user is obtained, and by pre-If speech recognition modeling identifies the phonetic problem information, the corresponding text information of the phonetic problem information and mood letter are generatedBefore the step of breath, further includes:
When receiving user's consulting instruction, the phonetic problem information of target user is obtained, the voice messaging is input toDefault speech recognition modeling carries out speech recognition by the default speech recognition modeling, it is corresponding to obtain the voice messagingText information；
Extract the target voice parameter in voice messaging by the default speech recognition modeling, to target voice parameter intoRow analysis, obtains voice mood；
The text information is pressed into default participle method and carries out word segmentation processing, obtains the corresponding Feature Words of the text information,And each Feature Words are compared with the standard words in default dictionary；
Acquisition and each matched target criteria word of Feature Words and the corresponding text mood of the target criteria word, andUsing the voice mood and the text mood as the corresponding emotional information of the voice messaging.
Optionally, if type is Products Show type the problem of the phonetic problem information, according to the textInformation and the emotional information scan for default words art forest, and the product information of recommended products is added to candidate words art mouldPlate obtains the corresponding Products Show words art of phonetic problem information, and the step of showing the Products Show words art specifically includes:
If the problem of phonetic problem information type is Products Show type, by the text information to default words artForest scans for, and obtains the corresponding text leaf node of the text information, and it is corresponding to obtain the text leaf nodeRecommended products；
Default words art forest is scanned for by the emotional information, obtains the corresponding mood leaf section of the emotional informationPoint, and obtain art template if the mood leaf node corresponds to；
The product information is added in the words art template, obtains institute by the product information for obtaining the recommended productsThe corresponding Products Show words art of phonetic problem information is stated, and shows the Products Show words art, so that staff can be according to instituteIt states Products Show and talks about art to target user progress corresponding product recommendation.
Optionally, the product information for obtaining the recommended products, is added to the words art mould for the product informationIn plate, the corresponding Products Show words art of the phonetic problem information is obtained, and show the Products Show words art, so as to the people that worksMember can be talked about according to the Products Show after the step of art carries out corresponding product recommendation to the target user, further includes:
Feedback voice messaging of the target user based on Products Show words art feedback is received, and by the backchannelMessage breath is input to default speech recognition modeling, identifies to obtain the feedback voice messaging by the default speech recognition modelingCorresponding feedback text information and feedback emotional information；
Feedback text information and the feedback emotional information, it is corresponding to obtain the feedback voice messaging described in binding analysisFeed back trend information；
When the feedback trend information is passive, the text leaf node is returned by the feedback trend informationIt traces back search, obtains secondary text leaf node and the corresponding secondary recommended products of the secondary text leaf node；
Backtracking search is carried out to the mood leaf node by the feedback trend information, obtains secondary mood leaf nodeAnd the corresponding secondary words art template of the secondary mood leaf node；
The afterproduct information is added to the secondary words by the afterproduct information for obtaining the secondary recommended productsSecondary recommendation words art is obtained in art template, and shows that art is talked about in secondary recommendations, so that staff secondary can push away according to describedIt recommends words art and carries out afterproduct recommendation to the target user.
Optionally, if type is Products Show type the problem of the phonetic problem information, according to the textInformation and the emotional information scan for default words art forest, and the product information of recommended products is added to candidate words art mouldPlate obtains the corresponding Products Show words art of phonetic problem information, and after the step of showing the Products Show words art, also wrapsIt includes:
Generating the user whether the Products Show words art meets the target user based on Products Show words art needsThe reminder message asked, and show that the reminder message can be based on the reminder message to the Products Show so as to staffWords art is fed back；
In the feedback command for not meeting user demand that reception is triggered based on the reminder message, reduces the product and push awayRecommend the corresponding recommendation of words art.
In addition, to achieve the above object, it is described to be based on language the present invention also provides a kind of question and answer system based on speech analysisCent analysis question and answer system include:
Speech recognition module, for obtaining the phonetic problem information of target user when receiving user's consulting instruction, andIdentify the phonetic problem information by default speech recognition modeling, generate the corresponding text information of the phonetic problem information andEmotional information；
Problem judgment module, type is production the problem of for judging the phonetic problem information according to the text informationProduct type of recommendation or problem seek advice from type；
Words art instructs module, if the problem of being used for phonetic problem information type is Products Show type, according to instituteIt states text information and the emotional information to scan for default words art forest, the product information of recommended products is added to candidateArt template is talked about, obtains the corresponding Products Show words art of phonetic problem information, and show the Products Show words art.
In addition, to achieve the above object, the question and answer equipment based on speech analysis that the present invention also provides a kind of is described to be based on languageThe question and answer equipment of cent analysis includes processor, memory and is stored on the memory and can be executed by the processorThe question and answer routine based on speech analysis, wherein described when being executed based on the question and answer routine of speech analysis by the processor, realityNow such as the step of the above-mentioned answering method based on speech analysis.
In addition, to achieve the above object, it is described computer-readable the present invention also provides a kind of computer readable storage mediumThe question and answer routine based on speech analysis is stored on storage medium, wherein it is described based on the question and answer routine of speech analysis by processorWhen execution, realize such as the step of the above-mentioned answering method based on speech analysis.
The present invention provides a kind of answering method based on speech analysis, i.e., when receiving user's consulting instruction, obtains meshThe phonetic problem information of user is marked, and the phonetic problem information is identified by default speech recognition modeling, generates the voiceThe corresponding text information of problem information and emotional information；According to the text information, the problem of judging the phonetic problem informationType is that Products Show type or problem seek advice from type；If the problem of phonetic problem information type is Products Show classType then scans for default words art forest according to the text information and the emotional information, and the product of recommended products is believedBreath is added to candidate words art template, obtains the corresponding Products Show words art of phonetic problem information, and show the Products Show wordsArt.By the above-mentioned means, the present invention identifies the voice messaging of client, the corresponding text information of voice messaging and feelings are obtainedThread information, and determine that the intention of user is used to scan for default words art forest according to text information and emotional informationThe interested target product in family and words art template；And will acquire the product information of target product, product information is added to words artIn template, the corresponding Products Show words art of phonetic problem information is obtained, efficiency of service is promoted, promotes user experience, solve existingThere is the technical issues of the technical issues of artificial counseling services inefficiency.
Specific embodiment
It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not intended to limit the present invention.
The present embodiments relate to the answering method based on speech analysis be mainly used in the question and answer based on speech analysisEquipment, being somebody's turn to do the question and answer equipment based on speech analysis can be PC, portable computer, mobile terminal etc. with display and processing functionEquipment.
Referring to Fig.1, Fig. 1 is the hardware configuration of the question and answer equipment based on speech analysis involved in the embodiment of the present inventionSchematic diagram.In the embodiment of the present invention, the question and answer equipment based on speech analysis may include processor 1001 (such as CPU), communicationBus 1002, user interface 1003, network interface 1004, memory 1005.Wherein, communication bus 1002 is for realizing these groupsConnection communication between part；User interface 1003 may include display screen (Display), input unit such as keyboard(Keyboard)；Network interface 1004 optionally may include standard wireline interface and wireless interface (such as WI-FI interface)；It depositsReservoir 1005 can be high speed RAM memory, be also possible to stable memory (non-volatilememory), such as diskMemory, memory 1005 optionally can also be the storage device independently of aforementioned processor 1001.
It will be understood by those skilled in the art that hardware configuration shown in Fig. 1 is not constituted and is asked based on speech analysisThe restriction for answering equipment may include perhaps combining certain components or different component cloth than illustrating more or fewer componentsIt sets.
With continued reference to Fig. 1, the memory 1005 in Fig. 1 as a kind of computer readable storage medium may include operation systemSystem, network communication module and the question and answer routine based on speech analysis.
In Fig. 1, network communication module is mainly used for connecting server, carries out data communication with server；And processor1001 can call the question and answer routine based on speech analysis stored in memory 1005, and execute provided in an embodiment of the present inventionAnswering method based on speech analysis.
The embodiment of the invention provides a kind of answering methods based on speech analysis.
It is that the present invention is based on the flow diagrams of the answering method first embodiment of speech analysis referring to Fig. 2, Fig. 2.
In the present embodiment, the answering method based on speech analysis the following steps are included:
Step S10 obtains the phonetic problem information of target user, and by default when receiving user's consulting instructionSpeech recognition modeling identifies the phonetic problem information, generates the corresponding text information of the phonetic problem information and mood letterBreath；
In the present embodiment, server presets speech recognition modeling, and the voice messaging of user, obtains voice messaging for identificationCorresponding text information and emotional information, the text information and emotional information then obtained according to identification are (pre- to default words art forestIf words art forest refers to art forest if the data structure for constructing more multiway trees in advance, presetting includes default produce in words art forestProduct tree and default words art tree, the product information for including in the leaf node of pre-set product tree, for example, father node in pre-set product treeFor product type, level-one leaf node is product category, and second level leaf node is product group, and three-level leaf node is each productInformation corresponding with each product etc.；Art template if the leaf node of default words art tree is corresponding, for example, level-one leaf node isArt template if interested, second level leaf node are art template if neutrality, and three-level leaf node is art template if passiveness) intoRow search seeks advice from the corresponding Products Show words art of instruction to obtain user.Specifically, staff is carrying out industry to target userWhen business processing, it can be asked questions according to user and trigger user's consulting instruction in this question and answer terminal, question and answer terminal is by presetting languageThe phonetic problem information of sound acquisition device acquisition client, wherein default voice acquisition device refers to pre-set voice collectingDevice, for example, default voice acquisition device can be the recording device in terminal, terminal sends out collected phonetic problem informationSend to server, the phonetic problem information that server receiving terminal is sent, with according to phonetic problem information carry out product inquiry orThe search of person's answer scheme.After server receives phonetic problem information, server is by presetting speech recognition mouldType is handled using phonetic problem information of the signal processing technology to target user, to extract the feature in phonetic problem informationData.Wherein, characteristic refers to then the tone color of voice, tone, audio preset speech recognition modeling according to acoustics, languageModel and dictionary, searching can export the corresponding word string of this feature data with maximum probability, and believe word string as customer voiceCease corresponding text information.In addition, server obtains the target voice parameter in characteristic, wherein target voice parameter packetIt includes: the speech parameter of volume, tone, frequency, word speed, pause and undulation of voice etc.；Server joins target voiceNumber is analyzed to obtain the corresponding emotional information of voice messaging.
Further, the step S10 is specifically included:
When receiving user's consulting instruction, the phonetic problem information of target user is obtained, the voice messaging is input toDefault speech recognition modeling carries out speech recognition by the default speech recognition modeling, it is corresponding to obtain the voice messagingText information；
Extract the target voice parameter in voice messaging by the default speech recognition modeling, to target voice parameter intoRow analysis, obtains voice mood；
The text information is pressed into default participle method and carries out word segmentation processing, obtains the corresponding Feature Words of the text information,And each Feature Words are compared with the standard words in default dictionary；
Acquisition and each matched target criteria word of Feature Words and the corresponding text mood of the target criteria word, andUsing the voice mood and the text mood as the corresponding emotional information of the voice messaging.
In the present embodiment, implementation one handles voice messaging by default speech recognition modeling, obtains mood(default speech recognition modeling is second by the phonetic problem information input to default speech recognition modeling for information, i.e. serverThe model of training in embodiment, does not repeat in the present embodiment), speech recognition is carried out by the default speech recognition modeling,That is, default speech recognition modeling extracts the corresponding speech characteristic parameter of phonetic problem information, wherein speech characteristic parameter includes:Voice tone, acoustic amplitudes, sound frequency etc., default speech recognition modeling obtain the phonetic problem to speech characteristic parameterThe corresponding text information of information.Server extracts the target voice in phonetic problem information by the default speech recognition modelingParameter, wherein target voice parameter is the special parameter extracted in phonetic problem information, such as the volume of voice, tone, frequencyThe speech parameter of rate, word speed, pause and undulation etc.；Server is analyzed according to target voice parameter, obtains feelingsThread information.That is, by target voice parameter, (the preset state table of comparisons presets speech parameter to server with the preset state table of comparisonsWith the emotional information table of comparisons) in pre-stored voice parameter be compared, obtain the corresponding emotional information of target voice parameter；ExampleSuch as, server inquires the default table of comparisons, obtains sound frequency in target voice parameter and exists in 1200-2000f/Hz, wave volume60-80 decibels of corresponding emotional informations are excitement.Target voice parameter is analyzed in the present embodiment, obtains emotional information,So that customer anger analysis is more accurate.
The text information and the voice feature data are combined analysis, obtain feelings by implementation two, serverThread information, that is, the phonetic problem information input to default speech recognition modeling is passed through the default speech recognition by serverModel carries out speech recognition, obtains the corresponding text information of the phonetic problem information, the present embodiment identical with above-described embodimentIn do not repeat.
The target voice parameter in phonetic problem information is extracted by the default speech recognition modeling, target voice is joinedNumber is analyzed, and the voice mood of client is obtained, that is, is preset speech parameter and the emotional information table of comparisons in server, is takenTarget voice parameter is compared by business device with the pre-stored voice parameter in the state table of comparisons, and it is corresponding to obtain target voice parameterVoice mood.
Text information is pressed default participle method and carries out word segmentation processing by server, is obtained the corresponding field of text information, is servicedDevice denoises each field, retains the Feature Words of reflection customer anger；Then, server by each Feature Words and is presetStandard words in dictionary are compared.Wherein, default segmentation methods refer to the pre-set Chinese character by target text informationSequence is cut into the algorithm of independent word one by one.
Each Feature Words are compared server with the standard words in default dictionary, that is, are previously provided in default dictionaryIndicate the standard words of the moods such as pleasure, anger, sorrow, happiness, server passes through fuzzy matching determination and Feature Words unisonance or synonymous markQuasi- word, server obtain and each matched target criteria word of Feature Words.Server obtains the corresponding text of each target criteria wordThis mood, for example, the corresponding text mood of " thanks " of target criteria word is " gratitude ".Server is by voice mood and text moodAs the corresponding emotional information of phonetic problem information.In the method implementation two for determining customer anger, server will be realizedIt is analyzed in mode one by target voice parameter, obtains emotional information as voice mood, then further according to text informationAnalysis obtains text mood, and using voice mood and text mood as the corresponding emotional information of phonetic problem information, so that withFamily Emotion identification is more accurate.
Step S20, according to the text information, the problem of judging the phonetic problem information, type is Products Show typeOr problem seeks advice from type；
In the present embodiment, phonetic problem information includes the consultation information of financial product class, such as seeks advice from that pre-sales personnel are some to be hadThe product information of financial product is closed, to buy the financial product and problem consulting category information that are suitble to oneself, such as in clientThe problems such as some operational issues or product generated when upper purchase financial products redeem relevant operational flow.By the textThe corresponding message identification of information is compared with the message identification of the message identification of Products Show type and problem consulting type,To judge that the phonetic problem information that user proposes is Products Show type or problem consulting type.
Step S30, if type is Products Show type the problem of the phonetic problem information, according to the text informationDefault words art forest is scanned for the emotional information, the product information of recommended products is added to candidate words art template,The corresponding Products Show words art of phonetic problem information is obtained, and shows the Products Show words art.
In the present embodiment, if it is determined that type is Products Show type the problem of the phonetic problem information, then determiningAfter stating the corresponding text information of phonetic problem information and emotional information, server is according to text information and emotional information to defaultWords art forest scans for, specifically, comprising:
1, server scans for the pre-set product tree in default words art forest according to text information, to obtain textThe corresponding text leaf node of information, and then using the corresponding product of text leaf node as recommended products；
2, server scans for the default words art tree in default words art forest according to emotional information, to obtain moodThe corresponding mood leaf node of information, art template talks about art template as candidate if server corresponds to mood leaf node；
3, server obtains the product information of recommended products, and the product information of recommended products is added to candidate words art mouldPlate, obtains phonetic problem information corresponding Products Show words art and is sent to terminal being shown.
Further, after the step S30, further includes:
If the problem of phonetic problem information type is that problem seeks advice from type, according to the text information in default topicCorresponding issue-resolution is searched in library, and is scanned for according to the emotional information in default words art forest；
Described problem solution is added to candidate words art template, obtains the corresponding solution answer of phonetic problem informationArt, and show described problem solution answer art.
In the present embodiment, the problem of often being asked previously according to user information and corresponding issue-resolution establish default topicLibrary, determining the phonetic problem information be problem seek advice from type, then by the corresponding text information of the phonetic problem information intoRow participle, and extract the keyword in the text information.It is searched in default exam pool according to the keyword, described pre-If searching the corresponding relevant issues information of the phonetic problem information in exam pool, and obtain that the relevant issues information is corresponding to askInscribe solution.Then corresponding candidate is searched in default words art forest according to the corresponding emotional information of the phonetic problem informationArt template is talked about, and described problem solution is added to the corresponding position of the candidate words art template, the voice is obtained and asksThe corresponding solution answer art of information is inscribed, and described problem solution answer art is shown by terminal interface, so that staff can rootAnswer is carried out to the phonetic problem information that target user proposes according to described problem solution answer art.In specific embodiment, may be used alsoTo be shown phonetic problem information and the relevant issues information of matched and searched in default exam pool simultaneously, so as to staff's rootIt whether is that target user needs to solve solution to the problem according to relevant issues confirmation described problem solution.
The present embodiment provides a kind of answering methods based on speech analysis, i.e., when receiving user's consulting instruction, obtainThe phonetic problem information of target user, and the phonetic problem information is identified by default speech recognition modeling, generate institute's predicateThe corresponding text information of sound problem information and emotional information；According to the text information, asking for the phonetic problem information is judgedInscribing type is that Products Show type or problem seek advice from type；If the problem of phonetic problem information type is Products Show classType then scans for default words art forest according to the text information and the emotional information, and the product of recommended products is believedBreath is added to candidate words art template, obtains the corresponding Products Show words art of phonetic problem information, and show the Products Show wordsArt.By the above-mentioned means, the present invention identifies the voice messaging of client, the corresponding text information of voice messaging and feelings are obtainedThread information, and determine that the intention of user is used to scan for default words art forest according to text information and emotional informationThe interested target product in family and words art template；And will acquire the product information of target product, product information is added to words artIn template, the corresponding Products Show words art of phonetic problem information is obtained, efficiency of service is promoted, promotes user experience, solve existingThere is the technical issues of the technical issues of artificial counseling services inefficiency.
It is that the present invention is based on the flow diagrams of the answering method second embodiment of speech analysis referring to Fig. 3, Fig. 3.
Based on above-mentioned embodiment illustrated in fig. 2, in the present embodiment, before the step S10, further includes:
Step S01, acquires the default dialect family of languages respectively and the mandarin family of languages reads aloud voice data, and extracts each voiceEach speech characteristic parameter is formed phonetic feature set by the corresponding speech characteristic parameter of data；
In the present embodiment, needed before establishing speech recognition modeling first acquire mandarin, Beijing native language, northeast dialect,Wu Fangyan, Jiangxi dialect, Hunan dialect, Hakka dialect, Fujian dialect, Guangdong dialect and another name for Sichuan Province dialect voice messaging, server will collectVoice messaging handled to obtain characteristic by preprocessing rule, wherein pretreatment mainly include preemphasis, adding window framingProcessing, end-point detection and noise reduction process Four processes.Preemphasis processing is that have using the difference of characteristics of signals and noise characteristicEffect ground handles signal, aggravates to the high frequency section of voice, and the influence of removal mouth and nose radiation increases the high frequency of voiceResolution ratio.Adding window sub-frame processing includes adding window and framing, wherein general framing method is the method for overlapping segmentation, former frameBe known as frame with the overlapping part of a later frame to move, and framing be the method that is weighted using the window of moveable finite length comeIt realizes, i.e., with certain window function, to form adding window voice signal, wherein window function generally uses Hamming window and rectangleWindow.End-point detection is that the starting point and end point of voice are found out from one section of given voice signal, correctly, is effectively heldPoint detection can not only reduce calculation amount and shorten the processing time, but also can also exclude the noise jamming of unvoiced segments, improve voiceThe accuracy of identification.
Step S02 extracts each speech characteristic parameter of preset ratio in the phonetic feature set, and passes through each instituteState speech characteristic parameter building initial speech identification model；
It is initial to constructing using characteristic after server extracts characteristic after pretreatment in the present embodimentSpeech recognition modeling, that is, building initial speech identification model is based on HMM (Hidden Markov Model, hidden MarkovModel) realize, hidden Markov model is substantially exactly to model to the feature for characterizing voice messaging in characteristic, is passed throughA large amount of statistics has been carried out to the phonetic feature in characteristic and has obtained model parameter, and iterative algorithm can use Baum-Model can also can be improved using through the improved Baum-Welch algorithm of K mean algorithm in Welch (Bao Muweierqi) algorithmAccuracy.
Step S03 by the iterative algorithm training initial speech identification model, and obtains the speech recognition that training obtainsRecognition accuracy is higher than the speech recognition modeling of preset threshold as default speech recognition mould by the speech recognition accuracy of modelType.
In the present embodiment, server passes through iterative algorithm training initial speech identification model, wherein speech recognition modelingTraining process is as follows: 1, constructing speech recognition modeling based on HMM model, and the initial parameter value of speech recognition modeling is arranged, joinNumber initial value can be by waiting divisions state or rule of thumb estimation setting；2, maximum the number of iterations and convergence threshold are setValue；3, state is carried out using characteristic of the Viterbi algorithm (Viterbi Algorithm, viterbi algorithm) to inputStaged operation；4, the parameter of the speech recognition modeling is updated by iterative algorithm (Baum-welch algorithm), and to spySign data are iterated training, constantly loop iteration, until reaching previously positioned the number of iterations or having restrained, at this point,Recognition accuracy is higher than the speech recognition of preset threshold by the speech recognition accuracy for obtaining the speech recognition modeling that training obtainsModel is as default speech recognition modeling.Wherein, preset threshold refers to that pre-set accuracy rate critical value, preset threshold can rootsIt is arranged according to concrete scene, such as is set as 99%, customer voice information is realized by training speech recognition modeling in the present embodimentAccurately identify.
Further, after the step S30, further includes:
Generating the user whether the Products Show words art meets the target user based on Products Show words art needsThe reminder message asked, and show that the reminder message can be based on the reminder message to the Products Show so as to staffWords art is fed back；
In the feedback command for not meeting user demand that reception is triggered based on the reminder message, reduces the product and push awayRecommend the corresponding recommendation of words art.
In the present embodiment, each product can correspond to multiple Products Show words arts, in order to promote user experience, preferential recommendationThe high Products Show of recommendation talks about art, wherein can determine that the product pushes away according to satisfaction of the user to Products Show words artRecommend the recommendation of words art.Specifically, it is showing that the Products Show words are postoperative, whether is further generating the Products Show words artMeet the reminder message of the user demand of the target user, the product is pushed away so that staff is based on field feedbackWords art is recommended to be fed back.If staff's feedback is to meet user's needs, i.e., the described Products Show words art is compared with by userIt welcomes, the Products Show words art can be increased.Otherwise, reduce the recommendation of the Products Show words art.In specific embodiment, InIt is preferential to show recommendation most when determining that corresponding Products Show words art has multiple according to the phonetic problem information of the target userArt is talked about in high recommendation.
It is that the present invention is based on the flow diagrams of the answering method 3rd embodiment of speech analysis referring to Fig. 4, Fig. 4.
Based on above-mentioned embodiment illustrated in fig. 2, in the present embodiment, the step S30 is specifically included:
Step S31, if type is Products Show type the problem of the phonetic problem information, by the text information pairDefault words art forest scans for, and obtains the corresponding text leaf node of the text information, and obtain the text leaf sectionThe corresponding recommended products of point；
In the present embodiment, server scans for default words art forest by text information, wherein pre- in the present embodimentIf multiway tree is divided into two types in words art forest, the first kind is the multiway tree (pre-set product book) comprising product information, treeEach node is both provided with different product and product related information；Second class is the multiway tree comprising sound template, tree it is everyA node is both provided with different sound templates (sound template can be understood as voice feedback mode)；Server presses text informationTo scanning for for the default first kind multiway tree for talking about art forest, server search obtains the corresponding text leaf of the text informationChild node, and obtain the corresponding recommended products of the text leaf node.
Step S32 scans for default words art forest by the emotional information, obtains the corresponding feelings of the emotional informationThread leaf node, and obtain art template if the mood leaf node corresponds to；
In the present embodiment, server (is called default words to the second class multiway tree in default words art forest by emotional informationArt tree) it scans for, the corresponding mood leaf node of the emotional information is obtained, server simultaneously obtains mood leaf node correspondenceIf art template.
Step S33 obtains the product information of the recommended products, and the product information is added to the words art templateIn, the corresponding Products Show words art of the phonetic problem information is obtained, and show the Products Show words art, so as to staffArt can be talked about according to the Products Show carry out corresponding product recommendation to the target user.
In the present embodiment, after server determines recommended products, server obtains the product information of recommended products, and willProduct information is added in words art template, obtains phonetic problem information corresponding Products Show words art and be sent to terminal being shownShow, carries out corresponding product recommendation to the target user so that staff can talk about art according to the Products Show；The present embodimentMiddle server selects corresponding product according to text information, allows the product that staff is lead referral more accurate,Art template if server is corresponded to according to emotional information selection, that is, server can be selected according to the tone of user in the present embodimentCorresponding voice feedback mode is selected, so that Products Show is more in line with Communication with Customer habit.
Further, after the step S33, further includes:
Feedback voice messaging of the target user based on Products Show words art feedback is received, and by the backchannelMessage breath is input to default speech recognition modeling, identifies to obtain the feedback voice messaging by the default speech recognition modelingCorresponding feedback text information and feedback emotional information；
In the present embodiment, server receives feedback voice messaging of the client based on Products Show words art feedback, and will feedbackVoice messaging is input to default speech recognition modeling, identifies to obtain feedback voice messaging pair by the default speech recognition modelingText information and feedback emotional information should be fed back, wherein identify to obtain by default speech recognition modeling in the present embodiment and answerThe middle implementation phase of voice messaging corresponding feedback text information and the specific implementation first embodiment for feeding back emotional informationTogether, the present embodiment does not repeat.
Feedback text information and the feedback emotional information, it is corresponding to obtain the feedback voice messaging described in binding analysisFeed back trend information；
In the present embodiment, text information and feedback emotional information binding analysis will be fed back in embodiment, that is, server will be anti-It presents text information and presses default participle method progress word segmentation processing, obtain the corresponding participle set of feedback text information；Server removalNoise word in participle set obtains the Feature Words for reflecting customer anger in participle set, and server is by the feature of customer angerWord feedback trend information corresponding with the emotional information conduct feedback voice messaging that identification obtains.For example, feedback text information are as follows:Alright, health insurance is introduced in trouble help, feedback emotional information be it is genial, then it is corresponding to obtain feedback voice messaging for serverFeedback tendency be it is genial.
When the feedback trend information is passive, the text leaf node is returned by the feedback trend informationIt traces back search, obtains secondary text leaf node and the corresponding secondary recommended products of the secondary text leaf node；
Backtracking search is carried out to the mood leaf node by the feedback trend information, obtains secondary mood leaf nodeAnd the corresponding secondary words art template of the secondary mood leaf node；
The afterproduct information is added to the secondary words by the afterproduct information for obtaining the secondary recommended productsSecondary recommendation words art is obtained in art template, and shows that art is talked about in secondary recommendations, so that staff secondary can push away according to describedIt recommends words art and carries out afterproduct recommendation to the target user.
In the present embodiment, after server obtains feedback trend information, server needs to be sentenced according to feedback trend informationIt is whether interested in the corresponding product of Products Show words art to determine client, specifically, comprising:
The feedback text information fed back in trend information is compared server with default emotional information；Wherein, it presetsEmotional information refers to the word of pre-set passive type, for example, I'm sorry, do not have to etc., server determines feedback text envelopeBreath is matched with default emotional information, then decision-feedback trend information is passiveness, that is, client is to the corresponding product of Products Show words artLose interest in, conversely, server determines that feedback text information and the preset standard information mismatch, then decision-feedback is inclined to letterBreath is positive, that is, interested in the corresponding product of Products Show words art, server determines that feedback trend information is positive Shi ZaijinRow is searched for, and is not repeated in the present embodiment.
For example, Products Show talks about art are as follows: me is needed to introduce health insurance a for you client can to Products Show talk about art intoThe feedback voice messaging of row feedback are as follows: how long the term of validity of health insurance a insurance is server Recognition feedback voice messaging:Does is 1, how long the term of validity of health insurance a insurance server determines that the feedback trend information of client is interested (positive)；ClothesBusiness device Recognition feedback voice messaging: 2, letting down, does not I need health insurance the feedback trend information for obtaining client is to lose interest in(passiveness).
The default words art forest is carried out by the feedback trend information and Products Show words art in the present embodimentBacktracking search obtains feedback recommendation words art and is sent to the terminal and is shown, specifically, comprising:
Step a, when the feedback trend information is passive, by the feedback trend information to the text leaf nodeBacktracking search is carried out, new text leaf node and the corresponding target product of the text leaf node are obtained；
Step b carries out backtracking search to the mood leaf node by the feedback trend information, obtains new mood leafArt template if child node and the mood leaf node are corresponding；
Step c obtains the product information of the target product, and the generation information is added in the words art template and is obtainedIt talks about art to feedback recommendation and is sent to the terminal and shown.
That is, when it is passive for feeding back trend information, that is to say, that client loses interest in the product in Products Show words art,Server needs are recommended new product again, and server is according to feedback trend information and Products Show words art to defaultWords art forest carries out backtracking search；For example, server determines corresponding node in Products Show words art first, and reversely obtain the sectionThe superior node of point obtains the corresponding product of superior node, and then, server carries out backtracking search by mood leaf node, obtainsArt template if taking new mood leaf node and the mood leaf node corresponding obtains the product letter of new target productBreath, by the generation information of new product be added to it is new if in art template, obtain feedback recommendation words art and be sent to the terminal intoRow display.
Feedback voice messaging of the collection of server client based on Products Show words art feedback in the present embodiment, and identify anti-Feedback voice messaging obtains feedback trend information；Server is in conjunction with the feedback trend information and Products Show words art to described defaultWords art forest scans for, so that Products Show is more intelligent, improves Products Show accuracy rate.
In addition, the embodiment of the present invention also provides a kind of question and answer system based on speech analysis.
It is that the present invention is based on the functional block diagrams of the question and answer system first embodiment of speech analysis referring to Fig. 5, Fig. 5.
In the present embodiment, the question and answer system based on speech analysis includes:
Speech recognition module 10, for obtaining the phonetic problem information of target user when receiving user's consulting instruction,And the phonetic problem information is identified by default speech recognition modeling, generate the corresponding text information of the phonetic problem informationAnd emotional information；
Problem judgment module 20, type is the problem of for judging the phonetic problem information according to the text informationProducts Show type or problem seek advice from type；
Words art instructs module 30, if the problem of being used for phonetic problem information type is Products Show type, basisThe text information and the emotional information scan for default words art forest, and the product information of recommended products is added to timeChoosing words art template obtains the corresponding Products Show words art of phonetic problem information, and shows the Products Show words art.
Wherein, modules and the above-mentioned answering method based on speech analysis in the above-mentioned question and answer system based on speech analysisEach step is corresponding in embodiment, and function and realization process no longer repeat one by one here.
In addition, the embodiment of the present invention also provides a kind of computer readable storage medium.
The question and answer routine based on speech analysis is stored on computer readable storage medium of the present invention, wherein described be based on languageWhen the question and answer routine of cent analysis is executed by processor, realize such as the step of the above-mentioned answering method based on speech analysis.
Wherein, the question and answer routine based on speech analysis, which is performed realized method, can refer to that the present invention is based on voices pointEach embodiment of the answering method of analysis, details are not described herein again.
It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to non-rowHis property includes, so that the process, method, article or the system that include a series of elements not only include those elements, andAnd further include other elements that are not explicitly listed, or further include for this process, method, article or system institute it is intrinsicElement.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including being somebody's turn to doThere is also other identical elements in the process, method of element, article or system.
The serial number of the above embodiments of the invention is only for description, does not represent the advantages or disadvantages of the embodiments.
Through the above description of the embodiments, those skilled in the art can be understood that above-described embodiment sideMethod can be realized by means of software and necessary general hardware platform, naturally it is also possible to by hardware, but in many casesThe former is more preferably embodiment.Based on this understanding, technical solution of the present invention substantially in other words does the prior artThe part contributed out can be embodied in the form of software products, which is stored in one as described aboveIn storage medium (such as ROM/RAM, magnetic disk, CD), including some instructions are used so that terminal device (it can be mobile phone,Computer, server, air conditioner or network equipment etc.) execute method described in each embodiment of the present invention.
The above is only a preferred embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hairEquivalent structure or equivalent flow shift made by bright specification and accompanying drawing content is applied directly or indirectly in other relevant skillsArt field, is included within the scope of the present invention.