Summary of the invention
In view of this, this application provides a kind of words art matching process and device, storage medium, computer equipments, mainlyPurpose is to solve the problems, such as how to improve customer service quality.
According to the one aspect of the application, a kind of words art matching process is provided, this method comprises:
The voice signal for obtaining client carries out speech recognition to the voice signal based on neural network model, obtains instituteThe corresponding text of predicate sound signal；
The keyword in the text is extracted, the first mood of client is determined according to the keyword extracted；
Emotion identification is carried out to the voice signal based on neural network model, obtains the voice signal corresponding secondMood；
The corresponding mood of the voice signal is determined according to first mood, the second mood and default rule；
It is corresponding from the reply data library lookup pre-established corresponding with text answer words art and/or with the moodPacify words art, text, the corresponding relationship for answering words art and/or mood are preserved in the reply data library, pacifies words artCorresponding relationship.
Optionally, speech recognition is carried out to the voice signal, the voice signal pair is obtained based on neural network modelThe text answered, comprising:
Acoustic model is constructed, wherein the acoustic model includes phoneme training pattern and the mixing based on memory unit connectionNeural network model；
The acoustic feature is input to the acoustic model by the acoustic feature for extracting the voice signal；
Phoneme recognition is carried out to the acoustic feature by the phoneme training pattern of trained completion, obtains phoneme recognitionAs a result；
Text region is carried out by the hybrid production style based on memory unit connection of trained completion, is obtainedText corresponding with the voice signal.
Optionally, the acoustic feature for extracting the voice signal, comprising:
Fourier transformation is carried out to the voice signal, the voice signal of time domain is converted to the energy spectrum of frequency domain；
The energy spectrum is inputted into triangular filter group, obtains the logarithmic energy of the triangular filter group output；
The acoustic feature that discrete cosine transform obtains the voice signal is carried out to the logarithmic energy.
Optionally, the second mood acquiring unit is further used for:
Multiple trained audios are obtained, the first acoustic feature vector sum first sample entropy feature of the trained audio is extracted,First sample entropy feature described in the first acoustic feature vector sum of each trained audio is merged respectively, is obtainedFirst emotion sound spectrum vector of each trained audio；
Dimension-reduction treatment is carried out to the first emotion sound spectrum vector, obtains the second emotion sound spectrum vector；
The second emotion sound spectrum vector of the corresponding trained audio of various moods is inputted into nerve net respectivelyNetwork model is trained, and obtains the corresponding sound spectrum vector mood model of various moods and trained template library is added；
The the second acoustics feature vector and the second Sample Entropy feature for extracting the voice signal, by second acoustic featureSecond Sample Entropy feature described in vector sum is merged, and the third emotion sound spectrum vector of the voice signal is obtained, byThree emotion sound spectrum vectors are compared and calculate with each sound spectrum vector mood model in the trained template libraryMood model matching degree exports corresponding second mood of maximum mood model matching degree.
Optionally, described that dimension-reduction treatment is carried out to the first emotion sound spectrum vector, it is special to obtain the second emotion sound spectrumLevy vector, comprising:
Dimension-reduction treatment is carried out to the first emotion sound spectrum vector using principal component analysis PCA algorithm, obtains secondEmotion sound spectrum vector.
Optionally, the neural network model is backpropagation BP neural network model.
Optionally, the method also includes:
Emotion identification is carried out to the voice signal of client and customer service dialogue overall process, generates emotional curve；
Service satisfaction is determined according to the emotional curve.
According to the another aspect of the application, a kind of words art coalignment is provided, which includes:
Voice recognition unit, for obtaining the voice signal of client, based on neural network model to the voice signal intoRow speech recognition obtains the corresponding text of the voice signal；
First mood determination unit determines visitor according to the keyword extracted for extracting the keyword in the textFirst mood at family；
Second mood acquiring unit is obtained for carrying out Emotion identification to the voice signal based on neural network modelCorresponding second mood of the voice signal；
Mood determination unit, for determining the voice according to first mood, the second mood and default ruleThe corresponding mood of signal；
Art matching unit is talked about, for talking about art from the reply data library lookup answer corresponding with the text pre-establishedAnd/or the corresponding relationship pacified words art, text is preserved in the reply data library, answers words art corresponding with the moodAnd/or mood, the corresponding relationship for pacifying words art.
Optionally, the voice recognition unit is further used for:
Acoustic model is constructed, wherein the acoustic model includes phoneme training pattern and the mixing based on memory unit connectionNeural network model；
The acoustic feature is input to the acoustic model by the acoustic feature for extracting the voice signal；
Phoneme recognition is carried out to the acoustic feature by the phoneme training pattern of trained completion, obtains phoneme recognitionAs a result；
Text region is carried out by the hybrid production style based on memory unit connection of trained completion, is obtainedText corresponding with the voice signal.
Optionally, the acoustic feature for extracting the voice signal, comprising:
Fourier transformation is carried out to the voice signal, the voice signal of time domain is converted to the energy spectrum of frequency domain；
The energy spectrum is inputted into triangular filter group, obtains the logarithmic energy of the triangular filter group output；
The acoustic feature that discrete cosine transform obtains the voice signal is carried out to the logarithmic energy.
Optionally, the second mood acquiring unit is further used for:
Multiple trained audios are obtained, the first acoustic feature vector sum first sample entropy feature of the trained audio is extracted,First sample entropy feature described in the first acoustic feature vector sum of each trained audio is merged respectively, is obtainedFirst emotion sound spectrum vector of each trained audio；
Dimension-reduction treatment is carried out to the first emotion sound spectrum vector, obtains the second emotion sound spectrum vector；
The second emotion sound spectrum vector of the corresponding trained audio of various moods is inputted into nerve net respectivelyNetwork model is trained, and obtains the corresponding sound spectrum vector mood model of various moods and trained template library is added；
The the second acoustics feature vector and the second Sample Entropy feature for extracting the voice signal, by second acoustic featureSecond Sample Entropy feature described in vector sum is merged, and the third emotion sound spectrum vector of the voice signal is obtained, byThree emotion sound spectrum vectors are compared and calculate with each sound spectrum vector mood model in the trained template libraryMood model matching degree exports corresponding second mood of maximum mood model matching degree.
Optionally, described that dimension-reduction treatment is carried out to the first emotion sound spectrum vector, it is special to obtain the second emotion sound spectrumLevy vector, comprising:
Dimension-reduction treatment is carried out to the first emotion sound spectrum vector using principal component analysis PCA algorithm, obtains secondEmotion sound spectrum vector.
Optionally, the neural network model is backpropagation BP neural network model.
Optionally, described device further include:
Emotional curve generation unit carries out Emotion identification for the voice signal to client and customer service dialogue overall process, rawAt emotional curve；
Service satisfaction determination unit determines service satisfaction according to the emotional curve.
According to the application another aspect, a kind of storage medium is provided, computer program, described program are stored thereon withAbove-mentioned words art matching process is realized when being executed by processor.
According to the application another aspect, a kind of computer equipment is provided, including storage medium, processor and be stored inOn storage medium and the computer program that can run on a processor, the processor realize above-mentioned words art when executing described programMatching process.
By above-mentioned technical proposal, a kind of method and device provided by the present application, storage medium, computer equipment pass throughSpeech recognition and/or Emotion identification are carried out to the voice signal of client, provide answer words art to contact staff and/or pacify wordsArt avoids contact staff due to lacking the case where experience leads to confusing communication, improves service quality.Also, the application is also rightThe voice signal of client and customer service dialogue overall process carries out Emotion identification, generates emotional curve, is determined and serviced according to emotional curveSatisfaction can further promote customer service quality.
Above description is only the general introduction of technical scheme, in order to better understand the technological means of the application,And it can be implemented in accordance with the contents of the specification, and in order to allow above and other objects, features and advantages of the application canIt is clearer and more comprehensible, below the special specific embodiment for lifting the application.
Specific embodiment
The application is described in detail below with reference to attached drawing and in conjunction with the embodiments.It should be noted that not conflictingIn the case of, the features in the embodiments and the embodiments of the present application can be combined with each other.
Current contact staff probably because lack experience cause confusing communication, influence service quality aiming at the problem that.The embodiment of the present application provides a kind of words art matching process, as shown in Figure 1, this method comprises:
S11: obtaining the voice signal of client, carries out speech recognition to the voice signal based on neural network model, obtainsObtain the corresponding text of the voice signal；
It should be noted that speech recognition is the skill for allowing computer voice signal to be changed by identification process textArt, it is therefore an objective to identify content described in client.
S12: extracting the keyword in the text, and the first mood of client is determined according to the keyword extracted；
It will be appreciated that the mood of client can be presented by specific keyword, the embodiment of the present application is by extracting clientKeyword in voice signal can primarily determine the first mood of client.When include in the keyword extracted " too slow " " tooThe keywords such as difference ", it may be determined that the first mood of client is indignation.
S13: Emotion identification is carried out to the voice signal based on neural network model, it is corresponding to obtain the voice signalSecond mood.
It will be appreciated that sound is the important carrier of information, it is the important channel of person to person's exchange, the sound of people not only wrapsVoice content information is contained, has further included emotional information.Emotion identification is the processing for the emotional information in voice signal, people'sEmotion variation can be reacted by the acoustic feature extracted in voice.
S14: the corresponding feelings of the voice signal are determined according to first mood, the second mood and default ruleThread；
It will be appreciated that the emotional information that the embodiment of the present application includes according to the keyword and voice signal itself extractedIt determines the first mood and the second mood respectively, and then (for example the first mood and the second mood is assigned not according to default ruleSame weight) determine the corresponding mood of voice signal, improve the accuracy of determining mood.
For example, the embodiment of the present application carries out the scoring of 1-100 to mood, and different scoring ranges corresponds to different feelingsThread.Such as 1-10 shows gladness, 11-30 indicates glad, and 31-50 meaning with thumb down, 51-60 indicates very dissatisfied, 61-80 indicate indignation, and 81-100 indicates very angry.
When determining the first mood, different keywords corresponds to different scorings, and different scorings characterizes different moods.When including keywords such as " too slow " " too poor ", ' mood scores 70, it may be determined that the first mood of client in the keyword extractedFor indignation.It similarly, is indignation carrying out the second mood that Emotion identification determines to voice signal based on neural network model, it is rightThe ' mood scores answered are 65.Further, different weights is assigned to the first mood and the second mood by Principal Component Analysis,Such as the weight of the first mood is 0.6, the weight of the second mood is 0.4, then the scoring of the mood finally determined is 70 × 0.6+ 65 × 0.4=68, corresponding mood are indignation.
S15: from the reply data library lookup that pre-establishes answer words art corresponding with the text and/or with the moodIt is corresponding to pacify words art.
It should be noted that preserved in the reply data library text, answer words art corresponding relationship and/or mood,Pacify the corresponding relationship of words art.The text is subjected to word segmentation processing, keyword is obtained, from the reply data library lookup pre-establishedArt is talked about in answer corresponding with keyword.Client is determined by the mood saved in reply data library and the corresponding relationship for pacifying words artCurrent mood.
It will be appreciated that telephone service needs to handle the complaint of customer, and voice system can automatically detect this feelingsThread reminds contact staff to pay attention to attitude, provides and pacify accordingly in time.Art coalignment if the embodiment of the present application leads toIt crosses and speech recognition and/or Emotion identification is carried out to the voice signal of client, provide answer words art to contact staff and/or pacify wordsArt avoids contact staff due to lacking the case where experience leads to confusing communication, improves service quality.
In practical applications, feature is extracted to voice signal, passes through weighted finite state machine (WeightedFinite State Transducer, WFST) voice messaging converted corresponding text information by network, to complete speech recognitionProcess.But this audio recognition method recognition accuracy is lower.In order to improve the accuracy rate of speech recognition, implement in the applicationIt is similar with the method in Fig. 1, wherein speech recognition, base are carried out to the voice signal in another words art matching process of exampleThe corresponding text of the voice signal is obtained in neural network model, comprising:
Acoustic model is constructed, wherein the acoustic model includes phoneme training pattern and the mixing based on memory unit connectionNeural network model；
The acoustic feature is input to the acoustic model by the acoustic feature for extracting the voice signal；
Phoneme recognition is carried out to the acoustic feature by the phoneme training pattern of trained completion, obtains phoneme recognitionAs a result；
Text region is carried out by the hybrid production style based on memory unit connection of trained completion, is obtainedText corresponding with the voice signal.
It will be appreciated that acoustic feature different in the sound of people characterizes different moods, recognition of speech signals can be passed throughAcoustic feature, using trained hybrid production style to voice signal carry out Emotion identification.
Preferably, before the acoustic feature for extracting the voice signal, this method further includes carrying out to voice signal pre-Processing, wherein pretreatment specifically includes: sampling processing and/or preemphasis processing and/or pre-filtering processing and/or windowing processAnd/or endpoint detection processing.
Hybrid production style by the trained completion based on memory unit connection is according to receivingRecognition result exports the text information opposite with the voice signal, is spoken by extracting after first pre-processing to primary speech signalIt learns feature and speech recognition is carried out by acoustic model again, improve the accuracy of speech recognition.
Optionally, the acoustic feature for extracting the voice signal, comprising:
Fourier transformation is carried out to the voice signal, the voice signal of time domain is converted to the energy spectrum of frequency domain；
The energy spectrum is inputted into triangular filter group, obtains the logarithmic energy of the triangular filter group output；
The acoustic feature that discrete cosine transform obtains the voice signal is carried out to the logarithmic energy.
It will be appreciated that carrying out Fourier transformation to voice signal, the voice signal of time domain is converted to the energy of frequency domainAmount spectrum；The energy spectrum is passed through to the triangular filter group of one group of Meier scale, the formant feature of prominent voice.It calculates eachThe logarithmic energy of filter group output.Logarithmic energy calculating after, by the triangular filter group output energy frequency spectrum pass through fromScattered cosine transform just can be obtained MFCC coefficient (mel frequency cepstrum coefficient), that is, MFCC acoustics is specialSign.
Specifically, Emotion identification is carried out to the voice signal, the voice signal pair is obtained based on neural network modelThe mood answered, comprising:
Multiple trained audios are obtained, the first acoustic feature vector sum first sample entropy feature of the trained audio is extracted,First sample entropy feature described in the first acoustic feature vector sum of each trained audio is merged respectively, is obtainedFirst emotion sound spectrum vector of each trained audio；
Dimension-reduction treatment is carried out to the first emotion sound spectrum vector, obtains the second emotion sound spectrum vector；
The second emotion sound spectrum vector of the corresponding trained audio of various moods is inputted into nerve net respectivelyNetwork model is trained, and obtains the corresponding sound spectrum vector mood model of various moods and trained template library is added；
The the second acoustics feature vector and the second Sample Entropy feature for extracting the voice signal, by second acoustic featureSecond Sample Entropy feature described in vector sum is merged, and the third emotion sound spectrum vector of the voice signal is obtained, byThree emotion sound spectrum vectors are compared and calculate with each sound spectrum vector mood model in the trained template libraryMood model matching degree exports the corresponding mood of maximum mood model matching degree.
It should be noted that Sample Entropy is a kind of new time series Complexity Measurement method, data vector is defined as in mDimension continues to keep the conditional probability of its similitude when increasing to m+1 dimension；Respectively by the first acoustic feature of each trained audio toAmount and first sample entropy feature carry out fusion be in order to acoustic feature vector emotion mood carry out various dimensions extraction, Sample EntropyValue is bigger, and the probability for generating new information is bigger, and sequence is more complicated, can pass through the voice signal dynamic change degree of different emotionsTo distinguish emotional category.
Preferably, described that dimension-reduction treatment is carried out to the first emotion sound spectrum vector, it is special to obtain the second emotion sound spectrumLevy vector, comprising:
Dimension-reduction treatment is carried out to the first emotion sound spectrum vector using principal component analysis PCA algorithm, obtains secondEmotion sound spectrum vector.
Optionally, the neural network model is backpropagation BP neural network model.
Preferably, the second emotion sound spectrum vector of the corresponding trained audio of various moods is inputted respectivelyNeural network model is trained, and is specifically included:
Network parameter initialization is carried out to backpropagation BP neural network model, wherein the network parameter includes: connectionWeight, connection threshold value, maximum study number, error precision；
Respectively by the reversed biography of the second emotion sound spectrum vector input of the corresponding trained audio of various moodsBP neural network model is broadcast to be trained.
It should be noted that backpropagation BP neural network (Back Propagation) is that one kind is calculated by error Back-PropagationThe Multi-layered Feedforward Networks of method training have the function of realizing any complex nonlinear mapping, with self-learning function and with oneThe advantages of fixed popularization, abstract ability, it can be used for pattern-recognition.
It should be noted that since BP neural network neuron node is numerous, when carrying out the calculating of output neuron node,It if the dimension of input neuron is excessive, will lead to computationally intensive, so that the building of BP neural network be made to complicate, reduce and trainEfficiency, therefore, it is necessary to the first emotion sound spectrum vector carry out dimension-reduction treatment, and PCA algorithm (principal component analysis,Principal ComponentAnalysis) it is a kind of method for being used to analyze data in multi-variate statistical analysis, it is with one kindSmall number of feature is described sample to reach the method for reducing feature space dimension, and the embodiment of the present application uses PCAAlgorithm carries out dimension-reduction treatment to the first emotion sound spectrum vector.
The embodiment of the present application is by carrying out the first acoustic feature vector sum first sample entropy feature of each trained audioFusion obtains the emotion sound spectrum vector of each trained audio, and the value of Sample Entropy is bigger, and the probability for generating new information is bigger, sequenceColumn are more complicated, can distinguish emotional category by the voice signal dynamic change degree of different emotions, ensure that emotional semantic classificationPerformance, improve mood classification accuracy rate；Meanwhile BP neural network is a kind of multilayer by Back Propagation Algorithm trainingFeedforward network has the function of realizing any complex nonlinear mapping, with self-learning function and with certain popularization, summaryIt is the advantages of ability, special to emotion sound spectrum using PCA algorithm before using BP neural network to the modeling of emotion sound spectrum vectorIt levies vector and carries out dimension-reduction treatment, so that the dimension for inputting neuron is reduced, and is subtracted when carrying out the calculating of output neuron nodeLack the calculation amount of BP output neuron, to simplify the building of BP neural network, improves training effectiveness；And it is a variety ofAccurately identifying for emotional characteristics may be implemented in the sound spectrum vector comprehensive matching of classification, improve Emotion identification flexibility,Convenience, tightness and recognition efficiency, can better adapt to the demand in intelligent hardware future, it is sustainable to complexity increasinglyThe intelligent hardware of growth completely, rapidly configure, and solves current Emotion identification complex disposal process, realizes difficultyThe technical issues of height, accuracy rate is low, low efficiency.
Fig. 2 shows the flow diagrams of another words art matching process provided by the embodiments of the present application, as shown in Fig. 2,This method comprises:
S21: obtaining the voice signal of client, carries out speech recognition to the voice signal based on neural network model, obtainsObtain the corresponding text of the voice signal；
S22: extracting the keyword in the text, and the first mood of client is determined according to the keyword extracted；
S23: Emotion identification is carried out to the voice signal based on neural network model, it is corresponding to obtain the voice signalSecond mood；
S24: the corresponding feelings of the voice signal are determined according to first mood, the second mood and default ruleThread；
S25: from the reply data library lookup that pre-establishes answer words art corresponding with the text and/or with the moodIt is corresponding to pacify words art；
S26: Emotion identification is carried out to the voice signal of client and customer service dialogue overall process, generates emotional curve；
It will be appreciated that the mood of the embodiment of the present application tracking description client, generates emotional curve, emotional curve characterization shouldThe curve of client's emotional change within the period talked with client.
S27: service satisfaction is determined according to the emotional curve.
It should be noted that the embodiment of the present application calculates service satisfaction according to emotional curve according to default rule, it is rightThe service quality of customer service is evaluated.
Wherein, the detailed process of step S21-S25 is similar in Fig. 1, and details are not described herein.
Client is required to evaluate this service at the end of service in traditional customer service system, however many clientsIt is unwilling to be evaluated, evaluation information is caused to lack.Voice signal of the embodiment of the present application to client and customer service dialogue overall processEmotion identification is carried out, emotional curve is generated, service satisfaction is determined according to emotional curve.Service satisfaction is fed back to visitor againClothes, to promote customer service quality, play the role of supervising again.
Fig. 3 shows a kind of structural schematic diagram for talking about art coalignment provided by the embodiments of the present application.As shown in figure 3, thisApplication embodiment device include:
Voice recognition unit 31, for obtaining the voice signal of client, based on neural network model to the voice signalSpeech recognition is carried out, the corresponding text of the voice signal is obtained；
First mood determination unit 32 is determined for extracting the keyword in the text according to the keyword extractedThe first mood of client；
Second mood acquiring unit 33 is obtained for carrying out Emotion identification to the voice signal based on neural network modelObtain corresponding second mood of the voice signal；
Mood determination unit 34, for determining institute's predicate according to first mood, the second mood and default ruleThe corresponding mood of sound signal；
Art matching unit 35 is talked about, for talking about art from the reply data library lookup answer corresponding with the text pre-establishedAnd/or the corresponding relationship pacified words art, text is preserved in the reply data library, answers words art corresponding with the moodAnd/or mood, the corresponding relationship for pacifying words art.
Art coalignment if the embodiment of the present application carries out speech recognition and/or mood by the voice signal to clientIdentification provides to contact staff and answers words art and/or pacify words art, contact staff is avoided to lead to confusing communication due to lacking experienceThe case where, improve service quality.
Voice recognition unit 31 is further used for:
Acoustic model is constructed, wherein the acoustic model includes phoneme training pattern and the mixing based on memory unit connectionNeural network model；
The acoustic feature is input to the acoustic model by the acoustic feature for extracting the voice signal；
Phoneme recognition is carried out to the acoustic feature by the phoneme training pattern of trained completion, obtains phoneme recognitionAs a result；
Text region is carried out by the hybrid production style based on memory unit connection of trained completion, is obtainedText corresponding with the voice signal.
Optionally, the acoustic feature for extracting the voice signal, comprising:
Fourier transformation is carried out to the voice signal, the voice signal of time domain is converted to the energy spectrum of frequency domain；
The energy spectrum is inputted into triangular filter group, obtains the logarithmic energy of the triangular filter group output；
The acoustic feature that discrete cosine transform obtains the voice signal is carried out to the logarithmic energy.
Second mood acquiring unit 33 is further used for:
Multiple trained audios are obtained, the first acoustic feature vector sum first sample entropy feature of the trained audio is extracted,First sample entropy feature described in the first acoustic feature vector sum of each trained audio is merged respectively, is obtainedFirst emotion sound spectrum vector of each trained audio；
Dimension-reduction treatment is carried out to the first emotion sound spectrum vector, obtains the second emotion sound spectrum vector；
The second emotion sound spectrum vector of the corresponding trained audio of various moods is inputted into nerve net respectivelyNetwork model is trained, and obtains the corresponding sound spectrum vector mood model of various moods and trained template library is added；
The the second acoustics feature vector and the second Sample Entropy feature for extracting the voice signal, by second acoustic featureSecond Sample Entropy feature described in vector sum is merged, and the third emotion sound spectrum vector of the voice signal is obtained, byThree emotion sound spectrum vectors are compared and calculate with each sound spectrum vector mood model in the trained template libraryMood model matching degree exports the corresponding mood of maximum mood model matching degree.
Optionally, described that dimension-reduction treatment is carried out to the first emotion sound spectrum vector, it is special to obtain the second emotion sound spectrumLevy vector, comprising:
Dimension-reduction treatment is carried out to the first emotion sound spectrum vector using principal component analysis PCA algorithm, obtains secondEmotion sound spectrum vector.
Optionally, the neural network model is backpropagation BP neural network model.
Optionally, described device further include:
Emotional curve generation unit carries out Emotion identification for the voice signal to client and customer service dialogue overall process, rawAt emotional curve；
Service satisfaction determination unit determines service satisfaction according to the emotional curve.
It should be noted that other of each functional unit involved by a kind of words art coalignment provided by the embodiments of the present applicationCorresponding description, can be with reference to the corresponding description in Fig. 1 and Fig. 2, and details are not described herein.
Based on above-mentioned method as depicted in figs. 1 and 2, correspondingly, the embodiment of the present application also provides a kind of storage medium,On be stored with computer program, the program realized when being executed by processor it is above-mentioned as depicted in figs. 1 and 2 if art matching process.
Based on this understanding, the technical solution of the application can be embodied in the form of software products, which producesProduct can store in a non-volatile memory medium (can be CD-ROM, USB flash disk, mobile hard disk etc.), including some instructionsWith so that computer equipment (can be personal computer, server or the network equipment an etc.) execution the application is eachArt matching process described in implement scene.
It is above-mentioned in order to realize based on above-mentioned method as shown in Figure 1 and Figure 2 and virtual bench embodiment shown in Fig. 3Purpose, the embodiment of the present application also provides a kind of computer equipments, are specifically as follows personal computer, server, the network equipmentDeng the entity device includes storage medium and processor；Storage medium, for storing computer program；Processor, for executingComputer program is to realize above-mentioned art matching process as depicted in figs. 1 and 2.
Optionally, which can also include user interface, network interface, camera, radio frequency (RadioFrequency, RF) circuit, sensor, voicefrequency circuit, WI-FI module etc..User interface may include display screen(Display), input unit such as keyboard (Keyboard) etc., optional user interface can also connect including USB interface, card readerMouthful etc..Network interface optionally may include standard wireline interface and wireless interface (such as blue tooth interface, WI-FI interface).
It will be understood by those skilled in the art that a kind of computer equipment structure provided by the embodiments of the present application is not constituted pairThe restriction of the entity device may include more or fewer components, perhaps combine certain components or different component clothIt sets.
It can also include operating system, network communication module in storage medium.Operating system is that management computer equipment is hardThe program of part and software resource supports the operation of message handling program and other softwares and/or program.Network communication module is usedCommunication between each component in realization storage medium inside, and communicated between other hardware and softwares in the entity device.
Through the above description of the embodiments, those skilled in the art can be understood that the application can borrowIt helps software that the mode of necessary general hardware platform is added to realize, hardware realization can also be passed through.Pass through the skill of application the applicationArt scheme carries out speech recognition and/or Emotion identification to the voice signal of client, provided to contact staff answer words art and/orWords art is pacified, avoids contact staff due to lacking the case where experience leads to confusing communication, improves service quality.Also, this ShenEmotion identification also please is carried out to the voice signal of client and customer service dialogue overall process, generates emotional curve, it is true according to emotional curveDetermine service satisfaction, can further promote customer service quality.
It will be appreciated by those skilled in the art that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, module in attached drawing orProcess is not necessarily implemented necessary to the application.It will be appreciated by those skilled in the art that the mould in device in implement sceneBlock can according to implement scene describe be distributed in the device of implement scene, can also carry out corresponding change be located at be different fromIn one or more devices of this implement scene.The module of above-mentioned implement scene can be merged into a module, can also be into oneStep splits into multiple submodule.
Above-mentioned the application serial number is for illustration only, does not represent the superiority and inferiority of implement scene.Disclosed above is only the applicationSeveral specific implementation scenes, still, the application is not limited to this, and the changes that any person skilled in the art can think of is allThe protection scope of the application should be fallen into.