CN109741734A - A kind of speech evaluating method, device and readable medium - Google Patents

A kind of speech evaluating method, device and readable medium Download PDF

Info

Publication number
CN109741734A
CN109741734A CN201910175867.3A CN201910175867A CN109741734A CN 109741734 A CN109741734 A CN 109741734A CN 201910175867 A CN201910175867 A CN 201910175867A CN 109741734 A CN109741734 A CN 109741734A
Authority
CN
China
Prior art keywords
speech evaluating
target object
model
speech
evaluation
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
Application number
CN201910175867.3A
Other languages
Chinese (zh)
Other versions
CN109741734B (en
Inventor
钟贵平
刘顺鹏
李宝祥
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Orion Star Technology Co Ltd
Original Assignee
Beijing Orion Star Technology Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beijing Orion Star Technology Co Ltd filed Critical Beijing Orion Star Technology Co Ltd
Priority to CN201910175867.3A priority Critical patent/CN109741734B/en
Publication of CN109741734A publication Critical patent/CN109741734A/en
Application granted granted Critical
Publication of CN109741734B publication Critical patent/CN109741734B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Landscapes

  • Electrically Operated Instructional Devices (AREA)

Abstract

The embodiment of the invention discloses a kind of speech evaluating method, device and readable medium, it is related to technical field of voice recognition, in method provided by the invention, gets the voice messaging to be evaluated of either objective object；Determine the corresponding speech evaluating model of the target object；According to determining speech evaluating model, the voice messaging to be evaluated of the target object is evaluated and tested.So, it can be the different speech evaluating model of different Object Selections, so that identified speech evaluating model is more suitable for the evaluation and test of target object pronunciation, and then reasonable speech evaluating is provided as a result, improving the enthusiasm that target object learns language in degree for target object.

Description

A kind of speech evaluating method, device and readable medium
Technical field
The present invention relates to technical field of voice recognition more particularly to a kind of speech evaluating methods, device and readable medium.
Background technique
With economic globalization, people understand the humanistic cultures of various countries more and more, and learn the feelings of other country's language Condition is more and more common, in order to meet the language learning demand of target object, some speech evaluating software applications and give birth to, e.g., English Fluently says using etc., and learner is in learning process, and whether which rank standard or pronunciation reach to learner's pronunciation is measurement The key element of one learner's level of learning.
And the speech evaluating method being commonly used is, speech evaluating model is trained using the voice corpus of other country, It is then based on trained speech evaluating model to evaluate and test the voice of learner, the speech evaluating model needs are tighter Lattice, when learner pronounces to improve but when the not up to calculated updating result of speech evaluating model, speech evaluating model still can Export cannot upgrading as a result, this has seriously affected the learning initiative of learner, the especially enthusiasm of influence beginner, even It will lead to beginner and generate sense of defeat, learner's experience is poor.In addition, the pronunciation of study other country's language is extremely difficult to other country native country Human hair sound, therefore, the voice that learner issues is evaluated and tested using above-mentioned speech evaluating method, and there are certain irrationalities.
Therefore, how method that is a kind of reasonable and being suitable for learner's study and speech evaluating foreign language is provided, is learned with improving Habit person's enthusiasm and learning experience are one of the problem of being worthy of consideration.
Summary of the invention
The embodiment of the present invention provides a kind of speech evaluating method, device and readable medium, to be reasonably that learner is true Speech evaluating model is determined, to improve learner's enthusiasm and learning experience.
In a first aspect, the embodiment of the present invention provides a kind of speech evaluating method, comprising:
Get the voice messaging to be evaluated of either objective object；
Determine the corresponding speech evaluating model of the target object；
According to determining speech evaluating model, the voice messaging to be evaluated of the target object is evaluated and tested.
Second aspect, the embodiment of the present invention provide a kind of speech evaluating device, comprising:
Acquiring unit, for getting the voice messaging to be evaluated of either objective object；
Determination unit, for determining the corresponding speech evaluating model of the target object；
Speech evaluating unit, for believing the voice to be evaluated of the target object according to determining speech evaluating model Breath is evaluated and tested.
The third aspect, the embodiment of the present invention provide a kind of electronic equipment, including memory, processor and are stored in described deposit On reservoir and the computer program that can run on the processor；The processor is realized when executing described program such as the present invention The described in any item speech evaluating methods provided.
Fourth aspect, the embodiment of the present invention provide a kind of computer readable storage medium, are stored thereon with computer program, It realizes when the program is executed by processor such as the step in described in any item speech evaluating methods provided by the invention.
The invention has the advantages that:
Speech evaluating method, device and readable medium provided in an embodiment of the present invention, get either objective object to Evaluate and test voice messaging；Determine the corresponding speech evaluating model of the target object；According to determining speech evaluating model, to described The voice messaging to be evaluated of target object is evaluated and tested.It so, can be the different voice of different target object recognitions Model is evaluated and tested, so that identified speech evaluating model is more suitable for the evaluation and test of target object pronunciation, and then is given for target object Reasonable speech evaluating in degree as a result, improve the enthusiasm of target object study language (such as foreign language) out.
Other features and advantages of the present invention will be illustrated in the following description, also, partly becomes from specification It obtains it is clear that understand through the implementation of the invention.The objectives and other advantages of the invention can be by written explanation Specifically noted structure is achieved and obtained in book, claims and attached drawing.
Detailed description of the invention
The drawings described herein are used to provide a further understanding of the present invention, constitutes a part of the invention, this hair Bright illustrative embodiments and their description are used to explain the present invention, and are not constituted improper limitations of the present invention.In the accompanying drawings:
Fig. 1 is the flow diagram of speech evaluating method provided in an embodiment of the present invention；
Fig. 2 is one of the flow diagram of the speech evaluating model of determining target object provided in an embodiment of the present invention；
Fig. 3 is the two of the flow diagram of the speech evaluating model of determining target object provided in an embodiment of the present invention；
Fig. 4 is the three of the flow diagram of the speech evaluating model of determining target object provided in an embodiment of the present invention；
Fig. 5 is the flow diagram of the speech evaluating result provided in an embodiment of the present invention for obtaining voice messaging to be evaluated；
Fig. 6 is the flow diagram that determination provided in an embodiment of the present invention meets grade promotion condition；
Fig. 7 is one of the flow diagram of speech evaluating model training process provided in an embodiment of the present invention；
Fig. 8 a is the two of the flow diagram of speech evaluating model training process provided in an embodiment of the present invention；
Fig. 8 b is the training logical schematic of the speech evaluating submodel of multiple evaluation and test grades provided in an embodiment of the present invention；
Fig. 9 is the structural schematic diagram of speech evaluating device provided in an embodiment of the present invention；
Figure 10 is the hardware configuration signal of the terminal device 100 provided in an embodiment of the present invention for implementing speech evaluating method Figure.
Specific embodiment
Speech evaluating method, device and readable medium provided in an embodiment of the present invention, to be reasonably that learner determines Speech evaluating model, to improve learner's enthusiasm and learning experience.
Below in conjunction with Figure of description, preferred embodiment of the present invention will be described, it should be understood that described herein Preferred embodiment only for the purpose of illustrating and explaining the present invention and is not intended to limit the present invention, and in the absence of conflict, this hair The feature in embodiment and embodiment in bright can be combined with each other.
Speech evaluating method provided in an embodiment of the present invention can be applied to that there is the speech evaluating of speech evaluating function to answer With in program, which can be installed on terminal device, which can be mobile phone, plate electricity Brain, all kinds of wearable devices, PDA (Personal Digital Assistant, palm PC) etc..
Implement the process of speech evaluating method substantially based on above-mentioned terminal device are as follows: when either objective object is based on upper predicate When sound evaluates and tests application program progress speech evaluating, the available voice messaging to be evaluated to the target object of terminal device, so After determine the corresponding speech evaluating model of the target object, further according to determining speech evaluating model, to the target object to Evaluation and test voice messaging is evaluated and tested.So, the speech evaluating mould for being more suitable for the target object can be determined for target object Type, and then reasonable speech evaluating is provided as a result, improving target object in degree learns the positive of language for target object Property.
It should be noted that due to terminal device processing capacity compared to server for, still there is certain limit System, therefore in order to improve the speed of determining speech evaluating model and improve the service of speech evaluating application program, it is provided by the invention Speech evaluating method is also applied in server, it may be assumed that the speech evaluating application program in terminal device can will collect The voice messaging to be evaluated of target object be sent to server, target object voice to be evaluated is then being got by server After information, determine the corresponding speech evaluating model of target object, and according to determining speech evaluating model, to target object to Evaluation and test voice messaging is evaluated and tested, and then speech evaluating result is sent to application program by server, and is triggered in terminal device Speech evaluating result is shown on the display page of offer.
After the application scenarios for having introduced speech evaluating method provided in an embodiment of the present invention, with reference next to following each Figure describes to speech evaluating method provided in an embodiment of the present invention in detail, as shown in Figure 1, being provided in an embodiment of the present invention The flow diagram of speech evaluating method, may comprise steps of:
S11, the voice messaging to be evaluated for getting either objective object.
S12, the corresponding speech evaluating model of the target object is determined.
In this step, the phonetic study level of different objects is different in the present invention, the present invention is based on the starting point, It is suitble to the speech evaluating model of the object for different object recommendations, so, speech evaluating model of the object based on recommendation Not only can preferably learning pronunciation, but also the speech evaluating model recommended can provide objectively evaluation knot for the pronunciation of object Fruit.
In the embodiment of the present invention, different language corresponds to different speech evaluating models, and the voice of the same languages is commented Survey model can with more than one, such as, it may be considered that the regional information of object or the pronunciation level of object are commented to configure voice Model is surveyed, in this way, could be that target object determines the speech evaluating model for being suitble to the target object.
S13, according to determining speech evaluating model, the voice messaging to be evaluated of the target object is evaluated and tested.
Optionally, after determining the speech evaluating model of target object based on the voice messaging to be evaluated got, mesh Mark object can be carried out based on the speech evaluating model determined with reading to learn, then with reading to get target in learning process The voice messaging of object, and target object is carried out with the voice messaging for reading to issue in learning process using the speech evaluating model Speech evaluating.
After the embodiment provided based on step S12 determines the speech evaluating model of target object, the speech evaluating Model is to be suitble to the model that learns at present of target object, then speech evaluating application program can be with triggering terminal equipment to target pair As showing the speech evaluating model determined, in this way, target object is carried out based on the speech evaluating model with reading the same of study When, which can carry out speech evaluating to the voice messaging to be evaluated that target object issues, to record in time Speech evaluating in the study language procedures of target object is as a result, target object can improve oneself based on speech evaluating result Pronunciation level, to effectively transfer the learning initiative of target object.
Based on any of the above-described embodiment, the corresponding speech evaluating model of the determination target object in step S12, including with Lower two kinds of possible implementations:
In first way, it can use and entry evaluation is carried out to the voice messaging to be evaluated of target object, based on preliminary Assessment result determines the speech evaluating model of target object, and step S12 can be specifically executed according to process shown in Fig. 2, including Following steps:
S21, entry evaluation test is carried out to the voice messaging to be evaluated of target object, obtains the first of voice messaging to be evaluated Walk point value of evaluation.
In this step, first the voice institute that target object issues can be identified by the voice messaging to be evaluated based on target object Then the languages of category carry out just the voice messaging to be evaluated of target object using the corresponding universal phonetic evaluation and test model of the languages Step assessment test, the entry evaluation score value of available voice messaging to be evaluated, it may be assumed that orient the specific pronunciation water of target object It is flat.For example, languages belonging to the voice messaging to be evaluated that target object issues are English, then utilize using English as the object of mother tongue The obtained corresponding speech evaluating model of English languages of English corresponding first kind voice training data training to target object Voice messaging to be evaluated tentatively evaluated and tested, obtain the entry evaluation score value of voice messaging to be evaluated.Optionally, above-mentioned English The training process of the corresponding speech evaluating model of languages can be implemented in advance, when getting the corresponding voice messaging to be evaluated of English When, the speech evaluating model of the English languages is directly searched, model implementation steps S21 is then utilized.
Optionally, it can also utilize using the languages as the training of the first kind voice training data of the languages of the object of mother tongue The speech evaluating model of the obtained languages carries out entry evaluation test to the voice messaging to be evaluated of target object, obtains to be evaluated Survey the entry evaluation score value of voice messaging.
S22, according to the corresponding relationship between point value of evaluation section and speech evaluating model, determine the entry evaluation score value Corresponding speech evaluating model.
S23, by the corresponding speech evaluating model of the entry evaluation score value, be determined as the speech evaluating of the target object Model.
In specific implementation, when target object, which is based on speech evaluating application program, learns foreign languages, in order to rationally visitor Seeing ground is that the target object determines the speech evaluating model for being suitble to the target object, and speech evaluating application program can be with triggering terminal Equipment shows " please say a Duan Yuyin " to target object, and then target object can be led to based on the text shown on terminal device The corresponding voice messaging of the microphone input text on terminal device is crossed, thus speech evaluating application program can be got For carrying out entry evaluation, target object voice messaging to be evaluated.It is of course also possible to start to learn based on target object When first segment carry out entry evaluation with reading voice, be suitble to the speech evaluating model of the target object to determine.The present invention is implemented Example is not defined particular content, form and the acquisition modes of the voice messaging for carrying out entry evaluation.
Herein on basis, the corresponding relationship being pre-configured between point value of evaluation section and speech evaluating model, for example, For each speech evaluating model, a point value of evaluation section can be configured for the speech evaluating model and so work as base When step S21 determines the voice messaging to be evaluated corresponding entry evaluation score value of target object, entry evaluation can be determined Point value of evaluation section belonging to score value, and then can determine that the corresponding voice of entry evaluation score value is commented based on above-mentioned corresponding relationship Model is surveyed, is the model for being suitble to the current pronunciation level of target object based on this speech evaluating model determined.
It should be noted that when speech evaluating model includes multiple speech evaluating submodels, and each speech evaluating submodule It is then for all speech evaluating model (including speech evaluating when configuring corresponding relationship when type corresponds to different evaluation and test grades Submodel) point value of evaluation section is respectively configured, in this way, can also accurately determine for target object than better suited speech evaluating Model.
In the second way, under same languages, the object pronunciation of different geographical is also different from, therefore the present invention is also It proposes under same languages, the regional information based on target object is the speech evaluating mould that target object determines suitable target object Type, specifically can be according to flow implementation step S12 shown in Fig. 3, comprising the following steps:
S31, the corresponding target area information of the target object is determined.
It is alternatively possible to determine the corresponding target area information of target object according to following two ways:
Mode one: speech recognition is carried out to the voice messaging to be evaluated, obtains the pronunciation character of the target object；With And according to the corresponding relationship between pronunciation character and regional information, determine the corresponding target area information of the voice messaging.
It specifically, can also be in order to determine which object is carrying out phonetic study and evaluation and test, in the embodiment of the present invention The voice messaging to be evaluated of target object, Jin Erti are got according to the voice messaging acquisition methods to be evaluated in first way The pronunciation character of the voice messaging to be evaluated is taken, and then according to the pronunciation character, determines corresponding target area information.
Herein on basis, the method that can use prior art offer carries out the voice messaging to be evaluated of target object Speech recognition extracts the pronunciation character of target object, then can analyze out which people by analyzing voice messaging The voice messaging with above-mentioned pronunciation character can be issued, the target area information of target object can be determined based on this.Tool When body is implemented, when the target object of different geographical learns same languages, what the voice issued still had differences, it is based on this thing It is real, same languages can be directed to, can the voice messagings of the object to each region issues in advance the languages carry out feature and mention It takes, it is possible thereby to extract the pronunciation character of the object in each area, then can establish pronunciation character and characterize the region Corresponding relationship between regional information, can after determining the pronunciation character of voice messaging to be evaluated of target object based on this To be based on above-mentioned corresponding relationship, the corresponding target area information of target object is determined.Specifically, voice to be evaluated is being extracted Each pronunciation character in the pronunciation character and above-mentioned corresponding relationship can be carried out similarity calculation by the pronunciation character of information, Target area information of the corresponding regional information of the maximum pronunciation character of similarity as target object can then be chosen.
Mode two: according to the attribute information of the target object, the corresponding target area information of the target object is determined.
Specifically, target object is when using speech evaluating application program, record for convenience oneself study schedule and The information such as speech evaluating result when study every time, target object can be registered based on speech evaluating application program.If target Object is registered, then can extract the attribute information of target object from the registration information of target object, be then based on The corresponding target area information of target object can be determined by stating the information such as native place, nationality, the location that attribute information includes.
Optionally, the present invention in target object attribute information can be, but not limited to include target object user name, Nationality, native place, the now information such as inhabitation address and age.
Optionally, the present invention in regional information can for by the Five continents (Asia, Europe, America, Africa and Oceania) into Row, which divides, obtains the corresponding regional information in each continent, i.e. the corresponding regional information in a continent, such as Chinese and Japan are equal Belong to Asia, then Chinese and Japanese regional information are the regional information for characterizing Asia；It can also be based on the same continent, It is divided according to the country for including in the continent, i.e., the corresponding regional information of one country, for example, the corresponding region of China Information, the corresponding regional information of Japan, and the corresponding regional information of country variant is different；The same state can also be directed to Family carries out dividing acquisition regional information according to the area feature of the country or nationality, by taking China as an example, the corresponding ground in Henan Province Domain information, corresponding regional information in Hebei province etc., or the nationality based on China are divided, and different nationalities correspond to different Regional information etc. is also based on province and is finely divided to obtain the regional information etc. for saving each city, also wants other certainly Region zones method is determined to characterize the regional information of each region hereby based on dividing condition.For various regions configuration of territory It when domain information, can be configured according to existing region representation method, with Chinese geography position division methods, the ground such as different provinces Area, area code is different, then area code can be used as the regional information of each region.
After dividing region and uniquely characterizing the regional information of the region for each ground configuration of territory, it can be based on each The pronunciation character of object in region meets the speech evaluating model of the object of the region for the ground configuration of territory one, then establishes Corresponding relationship between regional information and the speech evaluating model of configuration, to be in time that target object is determined for compliance with the target pair The speech evaluating model of elephant.
S32, according to the corresponding relationship between regional information and speech evaluating model, determine that the target area information is corresponding Speech evaluating model.
In this step, after determining the target area information of target object using any of the above-described kind of mode, it can be based on Corresponding relationship between above-mentioned regional information and speech evaluating model is determined for compliance with the speech evaluating model of the target object.
S33, the speech evaluating model that will be determined, are determined as the speech evaluating model of the target object.
In this step, in the corresponding relationship between regional information and speech evaluating model, when speech evaluating model only has At one, then the speech evaluating model determined in step S32 is directly determined as to the speech evaluating model of target object.
Based on any of the above-described embodiment, by taking speech evaluating result is speech evaluating score value as an example, when speech evaluating model packet Containing multiple speech evaluating submodels, and when each speech evaluating submodel corresponds to different evaluation and test grades, namely commented there are multiple When surveying the speech evaluating submodel of grade (value of L1, L2, L3 ... ..., Ln, n are based on depending on actual conditions), then step is based on The speech evaluating submodel of the corresponding multiple evaluation and test grades of the target area information for the target object that S32 can be identified, then exist It, can be first between minimum evaluation and test grade and highest evaluation and test grade when determining the speech evaluating submodel of suitable target object Evaluation and test grade in centre (for convenience, the intermediate evaluation and test grade between minimum evaluation and test grade and highest evaluation and test grade is denoted as First intermediate grade) speech evaluating submodel the voice messaging to be evaluated of target object is evaluated and tested.Specifically, work as voice When evaluating and testing model comprising even number speech evaluating submodel, then intermediate evaluation and test grade can be corresponding for numerical value high in intermediate value Evaluation and test grade as intermediate evaluation and test grade, such as there is L1~L4 totally 4 evaluation and test grades, L4 is that highest evaluates and tests grade, and L1 is most Grade is surveyed in lower assessment, then using L3 as the intermediate evaluation and test grade between minimum evaluation and test grade L1 and highest evaluation and test grade L4.
If 1, the speech evaluating score value obtained based on the first middle-bracket speech evaluating submodel is lower than first centre etc. Corresponding first point value of evaluation of grade then shows that the current pronunciation level of target object is also not suitable for the first middle-bracket voice and comments Submodel is surveyed, then on the basis of the first intermediate grade, determines the intermediate grade between the first intermediate grade and minimum evaluation and test grade (being denoted as the second intermediate grade)；Then the language to be evaluated of the second middle-bracket speech evaluating submodel evaluation and test target object is utilized Message breath；If the first intermediate grade and minimum evaluation and test grade are close to that is, there is no intermediate evaluation and test grades, then by minimum evaluation and test grade Speech evaluating model of the corresponding speech evaluating submodel as target object.If minimum evaluation and test grade and the first intermediate grade it Between include that even number evaluates and tests grade, then using the corresponding evaluation and test grade of numerical value high in intermediate value as intermediate evaluation and test grade.
A) if the speech evaluating score value obtained based on the second middle-bracket speech evaluating submodel is higher than second centre etc. The corresponding third point value of evaluation of grade, then show the current pronunciation level of target object in the second intermediate grade and the first intermediate grade Between, then it can continue to determine the intermediate grade (being denoted as third intermediate grade) between the first intermediate grade and the second intermediate grade Corresponding speech evaluating submodel is then based on the middle-bracket voice messaging submodel of third to the language to be evaluated of target object Message breath is evaluated and tested, until determining the speech evaluating submodel for being suitble to target object pronunciation level；If the first intermediate grade With the second intermediate grade close to then commenting the voice that the corresponding speech evaluating submodel of third intermediate grade is determined as target object Survey model；It, will be high in intermediate value if including that even number evaluates and tests grade between the first intermediate grade and the second intermediate grade The corresponding evaluation and test grade of numerical value is as third intermediate grade.
B) if the speech evaluating score value obtained based on the second middle-bracket speech evaluating submodel is lower than second centre etc. The corresponding third point value of evaluation of grade, but be higher than the 4th point value of evaluation, then show that the current pronunciation level of target object is suitble to second The corresponding speech evaluating submodel of intermediate grade and the pronunciation level that oneself can also be improved based on the speech evaluating submodel, then Determine that the second middle-bracket speech evaluating submodel is the speech evaluating model of target object；
C) if the speech evaluating score value obtained based on the second middle-bracket speech evaluating submodel is lower than second centre etc. Corresponding 4th point value of evaluation of grade, then show that the current pronunciation level of target object is not suitable for the corresponding voice of the second intermediate grade Submodel is evaluated and tested, and shows that the pronunciation level of target object, then can be with the between the second intermediate grade and the lowest class On the basis of two intermediate grades, the intermediate evaluation and test grade that can be continued between minimum evaluation and test grade and the second intermediate grade is corresponding Speech evaluating submodel carry out speech evaluating, until determine be suitble to target object pronunciation level speech evaluating submodel.
If 2, the speech evaluating score value obtained based on the first middle-bracket speech evaluating submodel is higher than first centre etc. Corresponding second point value of evaluation of grade, the second point value of evaluation are greater than the first point value of evaluation, then can on the basis of the first intermediate grade, The corresponding speech evaluating submodel of intermediate evaluation and test grade between grade and the first intermediate grade is evaluated and tested to target pair using highest The voice messaging to be evaluated of elephant is evaluated and tested, and then determines suitable target pair according to above-mentioned redirect procedure according to speech evaluating result The speech evaluating submodel of elephant；If highest evaluates and tests grade and the first intermediate grade close to directly by highest evaluation and test grade correspondence Speech evaluating submodel be determined as the speech evaluating model of target object；If highest is evaluated and tested between grade and the first intermediate grade Grade is evaluated and tested including even number, then using the corresponding evaluation and test grade of numerical value high in intermediate value as highest evaluation and test grade and first Intermediate evaluation and test grade between intermediate grade；
If the speech evaluating score value 3, obtained based on the first middle-bracket speech evaluating submodel is among first etc. Between corresponding first point value of evaluation of grade and the second point value of evaluation, the second point value of evaluation is greater than the first point value of evaluation), then show mesh The current pronunciation level of object is marked to be suitble to the corresponding speech evaluating submodel of the first intermediate grade and be based on the speech evaluating submodule Type can also improve the pronunciation level of oneself, it is determined that the first middle-bracket speech evaluating submodel is the voice of target object Evaluate and test model.
Specifically, the first point value of evaluation in the embodiment of the present invention, the second point value of evaluation, third point value of evaluation and the 4th are commented Estimate score value value can according to the actual situation depending on.
It include 3 evaluation and test grades (L1, L2 and L3) with the corresponding speech evaluating model of the target area information of target object Speech evaluating submodel for be illustrated, then can use L2 evaluation and test grade speech evaluating submodel to target object Voice messaging to be evaluated is evaluated and tested, however, it is determined that going out speech evaluating result is good (speech evaluating score value is at 80 points or so), then Show that the current pronunciation level of target object is relatively suitble to the speech evaluating submodel of L2 evaluation and test grade and also room for promotion, The speech evaluating submodel of L2 evaluation and test grade is then determined as the speech evaluating model of target object, be relatively suitble to target object into Row learns with reading；If it is determined that speech evaluating result is outstanding (speech evaluating score value is at 95 points or so), then show target object Current pronunciation level is preferable, and the speech evaluating submodel that L2 evaluates and tests grade helps not the raising of the pronunciation level of target object Greatly, then the speech evaluating submodel of L3 evaluation and test grade is determined as to the speech evaluating model of target object；When determining that voice comments When survey result is poor (speech evaluating score value is at 40 points or so), then show that the unsuitable L2 of the current pronunciation level of target object is commented The speech evaluating submodel of grade is surveyed, the basis of target object is not good enough, then the speech evaluating submodel of L1 evaluation and test grade is true It is set to the speech evaluating model of target object.
Based on this, it can determine that comparison is suitble to the speech evaluating model of target object study and learning outcome evaluation and test, make Target object when being carried out based on the speech evaluating model determined with reading study, can improve the pronunciation level of target object and And it can also improve the learning initiative of target object.
Based on any of the above-described embodiment, in practical applications, target object based on current speech evaluation and test model carry out with During reading study, theoretically, the pronunciation level of target object should be continuously improved, therefore there are current voices to comment The raising that model is not enough to the pronunciation level of support target object is surveyed, therefore the embodiment of the present invention passes through multiple evaluation and test grades of setting Speech evaluating submodel, and different brackets corresponds to different pronunciation levels, in this way when the speech evaluating submodule of grade is surveyed in lower assessment When type is not suitable for target object, then target object can be triggered using the speech evaluating submodel of high evaluation and test grade and carries out phonetics It practises and speech evaluating can specifically wrap in this way, the pronunciation level of target object can be improved according to flow implementation shown in Fig. 4 Include following steps:
S41, based on the corresponding speech evaluating submodel of current evaluation and test grade to the voice messaging to be evaluated of target object into Row evaluation and test, obtains the speech evaluating result of voice messaging to be evaluated.
This step, during target object carries out phonetic study based on the speech evaluating model of current evaluation and test grade, currently The speech evaluating submodel for evaluating and testing grade also can carry out voice to the voice messaging to be evaluated issued in target object learning process Evaluation and test, so as to obtain the speech evaluating of voice messaging to be evaluated as a result, can be based on the determination of speech evaluating result in turn It is no to upgrade.
S42, whether judge the speech evaluating result determined based on the corresponding speech evaluating submodel of current evaluation and test grade Meet grade promotion condition, if so, thening follow the steps S43；Otherwise, step S44 is executed.
S43, the evaluation and test grade after upgrading is determined.
Specifically, when determining the evaluation and test grade after upgrading, a grade can be risen under normal circumstances, if current evaluation and test etc. Grade is L1, then can promote evaluation and test grade to L2, then subsequent to target object using the corresponding speech evaluating submodel of L2 Voice messaging to be evaluated evaluated and tested；Certainly can also determine continuous setting number speech evaluating score value and upgrade threshold it Between difference can suggest rising 2 or 3 grades if difference is larger, in order to more accurately be that target object determines suitable comment Grade is surveyed, the corresponding speech evaluating submodel of evaluation and test grade after evaluating and testing grade after determining upgrading, after can use upgrading The voice messaging to be evaluated of target object is evaluated and tested, if speech evaluating score value is lower, can suggest reducing an evaluation and test Grade, i.e. triggering terminal equipment show that " it is recommended that using speech evaluating submodel of low level-one " is carried out with reading to learn to target object It practises, while can also show the option of a "Yes" and "No", target object agrees to that above-mentioned suggestion can then click "Yes", different Meaning then clicks "No", can be with the personalized individual opinion for consulting target object by the way that this function is arranged.
For example, when the speech evaluating submodel that target object is currently based on L2 evaluation and test grade is carried out with reading to learn, based on step Rapid S42 can be counted when carrying out speech evaluating in preset duration every time, and obtained each secondary speech evaluating score value is above L2 etc. The corresponding upgrade threshold of grade, and when determining the grade after upgrading is that L3 evaluates and tests grade, then triggering terminal equipment is to target pair As show can " congratulations upgrade to L3 evaluation and test grade ", the speech evaluating submodel of L3 evaluation and test grade is then passed through into terminal device Target object is showed, so, target object can be carried out based on the L3 speech evaluating submodel for evaluating and testing grade with reading Study and speech evaluating.
S44, according to the corresponding speech evaluating submodel of evaluation and test grade after upgrading, to the language to be evaluated of the target object Message breath carries out speech evaluating.
When being carried out based on the speech evaluating submodel of L3 evaluation and test grade with reading study due to target object, the hair of target object Sound level can be continuously improved, it is thus possible to which depositing the speech evaluating submodel that L3 evaluates and tests grade over time cannot help Target object improves the case where pronunciation level, therefore is carried out based on the speech evaluating submodel of L3 evaluation and test grade with reading in target object When study and speech evaluating, it also will continue to execution and judge to determine based on the corresponding speech evaluating submodel of evaluation and test grade after upgrading Whether speech evaluating result out meets grade promotion condition, to guarantee speech evaluating suitable for target object matching in time Model.
S45, triggering terminal equipment show upgrading failure as a result, and in the voice messaging to be evaluated for getting target object Afterwards, step S41 is continued to execute.
Optionally, step S45 is optional step, can not show that upgrading is lost to target object when determining upgrading failure It is losing as a result, continue with the speech evaluating submodel voice messaging to be evaluated subsequent to target object of current evaluation and test grade into Row evaluation and test.
This step, can be in target object actively selection upgrading, if the speech evaluating submodule based on current evaluation and test grade Type determines that target object is unsatisfactory for promotion condition, then can be with the displaying upgrading failure of triggering terminal equipment as a result, in this way, target pair As that can also know oneself pronunciation level in time.
In this step, if judging in step S42, only less time in preset duration speech evaluating score value is higher than upgrading threshold Value, then show target object there may be supernormal performance situation, and pronunciation level is unstable, therefore also needs to continue based on current evaluation and test etc. The speech evaluating submodel of grade continues with reading study and speech evaluating.In order to determine that target object is suitble to reading to learn in time Speech evaluating submodel can continue to execute step S41 to improve the pronunciation level of target object, that is, continue based on current The speech evaluating submodel of evaluation and test grade evaluates and tests the voice messaging of target object.Target pair can be constantly monitored in this way As if appropriate for current speech evaluating submodel, and then recommend for target object the speech evaluating submodel being more suitable for.
It can be walked according to flow implementation shown in fig. 5 based on any of the above-described embodiment as a kind of possible implementation Rapid S41, comprising the following steps:
S51, feature extraction is carried out to the voice messaging to be evaluated of target object, extracts audio frequency characteristics.
Specifically, it can use existing audio feature extraction model and spy carried out to the voice messaging to be evaluated of target object Sign is extracted, and the audio frequency characteristics extracted can be, but not limited to include MFCC (Mel-Frequency Cepstral Coefficients, mel-frequency cepstral analysis) feature and/or FBANK (Filter BANK) feature etc..
S52, the speech evaluating submodel that the audio frequency characteristics are input to current evaluation and test grade, obtain for measuring target The probability matrix of object pronunciation level.
In this step, when the speech evaluating submodel based on current evaluation and test grade is trained DNN and HMM model, It, can be with output probability matrix when the audio frequency characteristics of extraction being then input to trained DNN model.
S53, processing is decoded to the probability matrix, determines routing information, the corresponding path of the routing information is deposited Contain the corresponding received pronunciation information of voice that the target object issues.
In this step, the probability matrix is input in decoder and is decoded, decoder can be used trained HMM model carries out state transfer processing to probability matrix, i.e., is transported using the state transition probability of HMM model to probability matrix It calculates, so as to obtain a routing information, the corresponding received pronunciation of voice which has target object to issue Information.
Plays voice messaging of the embodiment of the present invention can be, but not limited to as the standard phoneme of word each in received pronunciation Sequence.
S54, the received pronunciation information found according to the routing information believe the voice to be evaluated of the target object Breath is given a mark, and the speech evaluating result of the voice messaging to be evaluated is obtained.
Specifically, the routing information determined based on step S53, the available voice issued to target object are corresponding The corresponding standard aligned phoneme sequence of the word that received pronunciation information includes, on the other hand, can be parsed out target object sending to Evaluate and test the aligned phoneme sequence of each word in voice messaging, then by the aligned phoneme sequence parsed and the standard aligned phoneme sequence of storage into Row comparison, so as to obtain the speech evaluating of voice messaging to be evaluated as a result, the speech evaluating result can be speech evaluating Score value etc..
Optionally, evaluation result is obtained in addition to the phoneme using word carries out speech evaluating, it can also be according to following processes Carry out speech evaluating obtain speech evaluating as a result, specifically: by voice messaging to be evaluated dissect at phoneme one by one, then determine The evaluation result of each phoneme；It is then based on the evaluation result of each phoneme, determines the evaluation result for the phoneme that word includes, And the evaluation result for the factor for based on word including evaluates and tests word to obtain the evaluation result of word, then the evaluation and test by word As a result the sentence is evaluated and tested, obtains the evaluation result of the sentence；The finally evaluation and test of evaluation result and single based on phoneme As a result it carries out being averaging with the evaluation result of sentence or weighted average handles to obtain the speech evaluating result of voice messaging to be evaluated.
It is that speech evaluating divides in speech evaluating result as a kind of possible implementation based on any of the above-described embodiment When value, grade promotion condition can be met according to process shown in fig. 6 determination, comprising the following steps:
Based on the corresponding speech evaluating submodel of current evaluation and test grade to the target object in S61, statistics preset duration The speech evaluating score value that is obtained when being evaluated and tested respectively of voice messaging to be evaluated.
If the speech evaluating score value of S62, continuous setting number are all larger than the corresponding upgrade threshold of current evaluation and test grade, determine Meet grade promotion condition.
In step S61 and S62, the corresponding upgrade threshold of difference evaluation and test grade can be the same or different, and upgrading The value of threshold value can according to the actual situation depending on.Specifically, speech evaluating submodel based on current evaluation and test grade can be The voice messaging to be evaluated issued when study all previous to target object in continuous a period of time (preset duration) carries out speech evaluating, It can be obtained by preset duration in this way, the speech evaluating score value of each voice messaging to be evaluated, if continuously setting number Speech evaluating score value be above the corresponding upgrade threshold of current evaluation and test grade, then can determine the pronunciation level of target object It increases, the current speech evaluating submodel for evaluating and testing grade has been no longer desirable for target object, therefore can suggest target object Learnt based on high-grade speech evaluating submodel, i.e. execution step S43.
Specifically, the setting number in the present invention, preset duration can set value according to the actual situation.
Optionally, determining that above-mentioned promotion condition is in addition to having setting number when meeting promotion condition in preset duration Speech evaluating score value is greater than outside the corresponding upgrade threshold of current evaluation and test grade, can also count target object and be based on current evaluation and test etc. The quantity of study (quantity of the word of such as statistical learning, sentence and/or article) of the speech evaluating submodel of grade, then determines and learns Whether habit amount is higher than the corresponding quantity of study upgrade threshold of current evaluation and test grade, currently evaluates and tests the corresponding quantity of study of grade when both meeting Upgrade threshold, and meet the speech evaluating score value that there is setting number in preset duration and be greater than the corresponding upgrading of current evaluation and test grade When threshold value, just determination meets promotion condition.
Optionally, due to be accomplished that the corresponding speech evaluating model of different language is different and same in the present invention The corresponding speech evaluating model of languages different geographical information is also different, therefore believes in the present invention for any languages and any region Breath, can train the corresponding speech evaluating model of the regional information or speech evaluating under the languages according to process shown in Fig. 7 Submodel is illustrated by taking speech evaluating model as an example, comprising the following steps:
S71, it is directed to any languages and any regional information, obtained using the languages as the first of the languages of the object of mother tongue Second class voice training data of class voice training data, the languages of object in the corresponding region of the regional information.
Specifically, in training speech evaluating model, key is training data.Using the languages as the English regional information For characterization China, then obtaining using English is the English Phonetics data of the object of mother tongue as first kind voice training data, with And in regional object English Phonetics data as the second class voice training data, when choosing training data, the first kind Voice training data and the second class voice training data should be that pronunciation level is good, the English Phonetics data with learning value.
S72, using the first kind voice training data and the second class voice training data to speech evaluating model It is trained, obtains the corresponding speech evaluating model of the regional information under the languages.
Specifically, the corresponding speech evaluating model of the regional information under the languages is obtained by using step S72 training, by It considers during model training using the languages as the standard pronunciation of the object of mother tongue, while considering target object location The pronunciation situation of object in area, so that the corresponding speech evaluating model of the regional information under the languages that training obtains It is more suitable for the phonetic study and speech evaluating of object in the corresponding region of the regional information, while is based on the speech evaluating model pair It is more pertinent that the voice messaging to be evaluated of target object in the region carries out the speech evaluating result that speech evaluating obtains.For example, When target object --- Chinese are based on English study application program when studying English, then the speech evaluating determined for Chinese Model is the corresponding speech evaluating model of regional information that the region characterized under English languages includes China.
It can be according to flow implementation shown in Fig. 8 a as a kind of possible implementation based on any of the above-described embodiment Step S72, comprising the following steps:
S81, using the first kind voice training data, training obtains initial speech evaluation and test model.
S82, using the second class voice training data, initial speech evaluation and test model is adjusted, is somebody's turn to do The corresponding speech evaluating model of the regional information under languages.
It specifically,, can be first with using the languages as mother tongue for same languages when implementing process shown in Fig. 8 a The first kind voice training data training of object obtains initial speech evaluation and test model and carries out the parameter information of storage model, then In the corresponding speech evaluating model of regional information each under training the languages, mould can be evaluated and tested based on stored initial speech The parameter information of type first adjusts the speech evaluating model, then for each regional information is based on, is corresponded to using the regional information Region in the second class voice training data of object further adjust the speech evaluating model, thus can the languages The lower corresponding speech evaluating model of the regional information.So, it does not need to obtain the corresponding voice of other regional informations every time Re -training initial speech evaluates and tests model when evaluating and testing model, greatly reduces workload.
By using process shown in Fig. 8 a, so that it is the language that the speech evaluating model that training obtains, which may learn mother tongue, The pronunciation of the object of kind, while can also learn to the pronunciation of the object in the corresponding region of the regional information so to make The speech evaluating model that must be obtained is more suitable the study of the target object in the corresponding region of regional information, and is based on the mould The speech evaluating result of voice messaging to be evaluated in type learning process can more embody the pronunciation level of target object.
Optionally, when speech evaluating model includes speech evaluating submodel, each speech evaluating submodel corresponds to different Grade is evaluated and tested, in the speech evaluating submodel of each evaluation and test grade of training, the language of different evaluation and test grades can be obtained in advance Then sound training data is directed to each grade, the first kind voice training of the evaluation and test grade is utilized according to method shown in Fig. 8 a Second class voice training data of data and the evaluation and test grade are trained, and obtain the corresponding speech evaluating submodule of the evaluation and test grade Type.
Optionally, speech evaluating model be trained deep neural network (Deep Neutral Network, DNN) with Hidden Markov (Hidden Markov Model, HMM) model, then the embodiment of the present invention is utilizing first kind voice training number According to the substantially process with the second class voice training data training speech evaluating model are as follows: using first kind voice training data to mixed It closes Gauss model GMM and HMM model is trained, obtain trained HMM, then utilize the second class voice training data and instruction The HMM perfected is trained DNN, and the corresponding speech evaluating model of the regional information under the languages can be obtained.
It specifically, can be first with first kind voice training data to mixed Gauss model (Gauss of mixture Models, GMM) and HMM model be trained, obtain a relatively good alignment model, in order to further increase alignment effect, The second class voice training data, DNN model and HMM model can be used and carry out alignment training, that is, keep trained HMM model Constant, also referred to as state transition probability is constant in state transition model, only utilizes the second class voice training data training DNN model. During DNN model training, the state transition probability in trained HMM model can be used to calculate every wheel DNN model Trained frame intersects entropy loss, carries out backpropagation then to update the weight of neuron in DNN model, it is hereby achieved that instruction DNN model and HMM model are packaged into a unified file, training can be obtained by the DNN model perfected after DNN is trained The corresponding speech evaluating model of the regional information under the good languages.
It should be noted that when speech evaluating model includes speech evaluating submodel, with reference to shown in Fig. 8 b, then in order to instruct Practise different grades of speech evaluating submodel, can to by voice training data carry out graduation divide obtain it is different grades of First kind voice training data and the second class voice training data are then different using the training of different grades of voice training data DNN and HMM model, so as to obtain different grades of speech evaluating model.
For example, why voice training data give in the speech evaluating submodel of three evaluation and test grades of languages 1 in Fig. 8 b One out, be because regional information is consistent with the regional information of languages 1, then voice training data are first kind voice training number According to, however the pronunciation level height of the study languages 1 of object is different in the corresponding region of the regional information of languages 1, therefore languages 1 Corresponding speech evaluating model also includes the speech evaluating submodel of 3 evaluation and test grades, in training speech evaluating submodel, first 3 kinds of different grades of first kind voice training data are chosen according to evaluation and test grade, then i.e.: L1 grade is mother tongue with languages 1 The first kind voice training data of object, L2 grade with languages 1 for the first kind voice training data of the object of mother tongue and L3 grade is the first kind voice training data of the object of mother tongue with languages 1, then directly according to shown in step S81 in Fig. 8 a The step of the corresponding speech evaluating submodel of each evaluation and test grade can be obtained.
The corresponding regional information of languages 2 is not identical as regional information 1 in Fig. 8 b, that is, belongs to different areas, with 2 English of languages Language, for the corresponding region of regional information 1 is China, when the pronunciation to be studied English due to Chinese, pronunciation level is irregular, The speech evaluating model of the corresponding English of old place domain information 1 may include the speech evaluating submodel of multiple grades, and Fig. 8 b is with three It is illustrated for a grade (L1, L2 and L3), then chooses 3 kinds of different grades of English languages in advance also according to evaluation and test grade First kind voice training data, i.e. in Fig. 8 b L1 grade using English as the first kind voice training data of the object of mother tongue, L2 grade is the first kind voice training data of the object of mother tongue and L3 grade using English as the object of mother tongue using English Then first kind voice training data choose the second class language of the English languages of object in 3 kinds of different grades of regionals again Sound training data, i.e. the second of the English languages of object in the corresponding regional of regional information 1 of L1 grade in Fig. 8 b Class voice training data, L2 grade the corresponding regional of regional information 1 in object English languages the second class voice Second class voice training number of the English languages of training data and the object in the corresponding regional of the regional information of L3 grade 1 According to then L1 grade is right as the regional information 1 of the first kind voice training data of the object of mother tongue and utilization L1 grade using English Second class voice training data of the English languages for the object in regional answered obtain L1 evaluation and test etc. according to the process of Fig. 8 a The speech evaluating submodel of the corresponding English of regional information 1 of grade, similarly available L2 and L3 evaluates and tests the regional information 1 of grade The speech evaluating submodel of corresponding English.
By obtaining speech evaluating model according to the training of process shown in Fig. 7 and Fig. 8 a, so that the voice that training obtains is commented Model is surveyed, has not only learnt the pronunciation situation using the languages as the object of mother tongue, but also taken into account the corresponding region of regional information The pronunciation situation of middle object, so that the speech evaluating model for being more suitable for target object study and speech evaluating is filtered out, such one Come, not only effectively increases the pronunciation level of target object but also improve the enthusiasm of target object phonetic study.
Based on the same inventive concept, a kind of speech evaluating device is additionally provided in the embodiment of the present invention, due to above-mentioned apparatus The principle solved the problems, such as is similar to speech evaluating method, therefore the implementation of above-mentioned apparatus may refer to the implementation of method, repetition Place repeats no more.
As shown in figure 9, being the structural schematic diagram of speech evaluating device provided in an embodiment of the present invention, comprising:
Acquiring unit 91, for getting the voice messaging to be evaluated of either objective object；
Determination unit 92, for determining the corresponding speech evaluating model of the target object；
Speech evaluating unit 93, for according to determining speech evaluating model, to the voice to be evaluated of the target object Information is evaluated and tested.
Optionally, the determination unit 92 is specifically used for carrying out entry evaluation test to the voice messaging to be evaluated, obtain Obtain the entry evaluation score value of the voice messaging to be evaluated；According to the corresponding pass between point value of evaluation section and speech evaluating model System, determines the corresponding speech evaluating model of the entry evaluation score value；By the corresponding speech evaluating mould of the entry evaluation score value Type is determined as the speech evaluating model of the target object.
Optionally, the determination unit 92 is specifically used for determining the corresponding target area information of the target object；According to Corresponding relationship between regional information and speech evaluating model determines the corresponding speech evaluating model of the target area information； The speech evaluating model that will be determined is determined as the speech evaluating model of the target object.
Optionally, the determination unit 92 is specifically used for carrying out speech recognition to the voice messaging to be evaluated, obtains institute State the pronunciation character of target object；And according to the corresponding relationship between pronunciation character and regional information, determine the voice letter Cease corresponding target area information；Or the attribute information according to the target object, determine the corresponding mesh of the target object Mark regional information.
Optionally, the speech evaluating model includes multiple speech evaluating submodels, and each speech evaluating submodel is corresponding Different evaluation and test grades；Then
The speech evaluating unit 93, if specifically for being obtained based on the corresponding speech evaluating submodel of current evaluation and test grade Evaluation and test grade of the speech evaluating as a result, meet grade promotion condition, after determining upgrading；It is corresponding according to the evaluation and test grade after upgrading Speech evaluating submodel, speech evaluating is carried out to the voice messaging to be evaluated of the target object.
Optionally, the speech evaluating result includes speech evaluating score value；Then
The speech evaluating unit 93, specifically for being commented in statistics preset duration based on the corresponding voice of current evaluation and test grade Survey the speech evaluating score value obtained respectively when submodel evaluates and tests the voice messaging to be evaluated of the target object；If continuous The speech evaluating score value of setting number is all larger than the corresponding upgrade threshold of current evaluation and test grade, and determination meets grade promotion condition.
Optionally, training obtains the speech evaluating model by the following method: believing for any languages and any region Breath is obtained using the languages as the first kind voice training data of the languages of the object of mother tongue, the corresponding region of the regional information In object the languages the second class voice training data；Utilize the first kind voice training data and the second class language Sound training data is trained speech evaluating model, obtains the corresponding speech evaluating model of the regional information under the languages.
Optionally, using the first kind voice training data and the second class voice training data to speech evaluating mould Type is trained, and obtains the corresponding speech evaluating model of the regional information under the languages, comprising:
Using the first kind voice training data, training obtains initial speech evaluation and test model；Utilize the second class language Sound training data is adjusted initial speech evaluation and test model, obtains the corresponding voice of the regional information under the languages and comment Survey model.
For convenience of description, above each section is divided by function describes respectively for each module (or unit).Certainly, exist Implement to realize the function of each module (or unit) in same or multiple softwares or hardware when the present invention.
Based on the same technical idea, the embodiment of the invention also provides a kind of electronic equipment, with reference to electricity shown in Fig. 10 The hardware structural diagram of sub- equipment, the electronic equipment include: one or more processors 1010 and memory 1020, Figure 10 In by taking a processor 1010 as an example.
Wherein, the electronic equipment for executing speech evaluating method can also include: input unit 1030 and output device 1040.
Processor 1010, memory 1020, input unit 1030 and output device 1040 can by bus or other Mode connects, in Figure 10 for being connected by bus.
Memory 1020 is used as a kind of non-volatile computer readable storage medium storing program for executing, can be used for storing non-volatile software journey Sequence, non-volatile computer executable program and module, such as the corresponding program of speech evaluating method in the embodiment of the present invention Instruction/module/unit (for example, attached acquiring unit shown in Fig. 9 91, determination unit 92 and speech evaluating unit 93).Processor 1010 non-volatile software program, instruction and the module/units being stored in memory 1020 by operation, thereby executing clothes The speech evaluating that the various function application and data processing of business device or intelligent terminal, i.e. realization above method embodiment provide Method.
Memory 1020 may include storing program area and storage data area, wherein storing program area can store operation system Application program required for system, at least one function；Storage data area can be stored is created according to using for robot controller The data etc. built.In addition, memory 1020 may include high-speed random access memory, it can also include non-volatile memories Device, for example, at least a disk memory, flush memory device or other non-volatile solid state memory parts.In some embodiments In, optional memory 1020 includes the memory remotely located relative to processor 1010, these remote memories can pass through It is connected to the network to robot controller.The example of above-mentioned network include but is not limited to internet, intranet, local area network, Mobile radio communication and combinations thereof.
Input unit 1030 can receive the number or character information of input, and generate the user with robot controller Setting and the related key signals input of function control.Output device 1040 may include that display screen etc. shows equipment.
One or more of modules are stored in the memory 1020, when by one or more of processors When 1010 execution, the speech evaluating method in above-mentioned any means embodiment is executed.
Based on the same technical idea, the embodiment of the invention also provides a kind of computer storage mediums.The computer Readable storage medium storing program for executing is stored with computer executable instructions, before the computer executable instructions are for executing the computer State either step described in either method.
It should be understood by those skilled in the art that, the embodiment of the present invention can provide as method, system or computer program Product.Therefore, complete hardware embodiment, complete software embodiment or reality combining software and hardware aspects can be used in the present invention Apply the form of example.Moreover, it wherein includes the computer of computer usable program code that the present invention, which can be used in one or more, The computer program implemented in usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.) produces The form of product.
The present invention be referring to according to the method for the embodiment of the present invention, the process of equipment (system) and computer program product Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce A raw machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for real The device for the function of being specified in present one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates, Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one The step of function of being specified in a box or multiple boxes.
Although preferred embodiments of the present invention have been described, it is created once a person skilled in the art knows basic Property concept, then additional changes and modifications can be made to these embodiments.So it includes excellent that the following claims are intended to be interpreted as It selects embodiment and falls into all change and modification of the scope of the invention.
Obviously, various changes and modifications can be made to the invention without departing from essence of the invention by those skilled in the art Mind and range.In this way, if these modifications and changes of the present invention belongs to the range of the claims in the present invention and its equivalent technologies Within, then the present invention is also intended to include these modifications and variations.

Claims (10)

1. a kind of speech evaluating method characterized by comprising
Get the voice messaging to be evaluated of either objective object；
Determine the corresponding speech evaluating model of the target object；
According to determining speech evaluating model, the voice messaging to be evaluated of the target object is evaluated and tested.
2. the method as described in claim 1, which is characterized in that determine the corresponding speech evaluating model of the target object, wrap It includes:
Entry evaluation test is carried out to the voice messaging to be evaluated, obtains the entry evaluation point of the voice messaging to be evaluated Value；
According to the corresponding relationship between point value of evaluation section and speech evaluating model, the corresponding language of the entry evaluation score value is determined Sound evaluates and tests model；
By the corresponding speech evaluating model of the entry evaluation score value, it is determined as the speech evaluating model of the target object.
3. the method as described in claim 1, which is characterized in that determine the corresponding speech evaluating model of the target object, wrap It includes:
Determine the corresponding target area information of the target object；
According to the corresponding relationship between regional information and speech evaluating model, determine that the corresponding voice of the target area information is commented Survey model；
The speech evaluating model that will be determined is determined as the speech evaluating model of the target object.
4. method as claimed in claim 3, which is characterized in that determine the corresponding target area information of the target object, wrap It includes:
Speech recognition is carried out to the voice messaging to be evaluated, obtains the pronunciation character of the target object；And according to pronunciation Corresponding relationship between feature and regional information determines the corresponding target area information of the voice messaging；Or
According to the attribute information of the target object, the corresponding target area information of the target object is determined.
5. method according to any of claims 1-4, which is characterized in that the speech evaluating model includes that multiple voices are commented Submodel is surveyed, each speech evaluating submodel corresponds to different evaluation and test grades；Then according to determining speech evaluating model, to described The voice messaging to be evaluated of target object is evaluated and tested, further includes:
If the speech evaluating obtained based on the corresponding speech evaluating submodel of current evaluation and test grade is as a result, meet grade upgrading item Part, the evaluation and test grade after determining upgrading；
According to the corresponding speech evaluating submodel of evaluation and test grade after upgrading, to the voice messaging to be evaluated of the target object into Row speech evaluating.
6. the method as described in claim 1, which is characterized in that training obtains the speech evaluating model by the following method:
For any languages and any regional information, obtains and instructed by the first kind voice of the languages of the object of mother tongue of the languages Practice the second class voice training data of data, the languages of object in the corresponding region of the regional information；
Speech evaluating model is trained using the first kind voice training data and the second class voice training data, Obtain the corresponding speech evaluating model of the regional information under the languages.
7. method as claimed in claim 6, which is characterized in that utilize the first kind voice training data and second class Voice training data are trained speech evaluating model, obtain the corresponding speech evaluating model of the regional information under the languages, Include:
Using the first kind voice training data, training obtains initial speech evaluation and test model；
Using the second class voice training data, initial speech evaluation and test model is adjusted, obtaining should under the languages The corresponding speech evaluating model of regional information.
8. a kind of speech evaluating device characterized by comprising
Acquiring unit, for getting the voice messaging to be evaluated of either objective object；
Determination unit, for determining the corresponding speech evaluating model of the target object；
Speech evaluating unit, for according to determining speech evaluating model, to the voice messaging to be evaluated of the target object into Row evaluation and test.
9. a kind of electronic equipment, including memory, processor and it is stored on the memory and can transports on the processor Capable computer program；It is characterized in that, the processor is realized when executing described program such as any one of claim 1~7 institute The speech evaluating method stated.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the program is by processor The step in speech evaluating method as described in any one of claims 1 to 7 is realized when execution.
CN201910175867.3A 2019-03-08 2019-03-08 Voice evaluation method and device and readable medium Active CN109741734B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910175867.3A CN109741734B (en) 2019-03-08 2019-03-08 Voice evaluation method and device and readable medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910175867.3A CN109741734B (en) 2019-03-08 2019-03-08 Voice evaluation method and device and readable medium

Publications (2)

Publication Number Publication Date
CN109741734A true CN109741734A (en) 2019-05-10
CN109741734B CN109741734B (en) 2021-07-20

Family

ID=66369915

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910175867.3A Active CN109741734B (en) 2019-03-08 2019-03-08 Voice evaluation method and device and readable medium

Country Status (1)

Country Link
CN (1) CN109741734B (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110489454A (en) * 2019-07-29 2019-11-22 北京大米科技有限公司 A kind of adaptive assessment method, device, storage medium and electronic equipment
CN110491369A (en) * 2019-07-24 2019-11-22 北京大米科技有限公司 Appraisal procedure, device, storage medium and the electronic equipment of spoken grade
CN110797044A (en) * 2019-08-22 2020-02-14 腾讯科技（深圳）有限公司 Audio data processing method and device, computer equipment and storage medium
CN111816210A (en) * 2020-06-23 2020-10-23 华为技术有限公司 Voice scoring method and device
CN111915407A (en) * 2020-08-10 2020-11-10 政采云有限公司 Object compliance detection method and system
CN112017694A (en) * 2020-08-25 2020-12-01 天津洪恩完美未来教育科技有限公司 Voice data evaluation method and device, storage medium and electronic device
CN112908359A (en) * 2021-01-31 2021-06-04 云知声智能科技股份有限公司 Voice evaluation method and device, electronic equipment and computer readable medium

Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110144990A1 (en) * 2009-12-16 2011-06-16 International Business Machines Corporation Rating speech naturalness of speech utterances based on a plurality of human testers
CN103761975A (en) * 2014-01-07 2014-04-30 苏州思必驰信息科技有限公司 Method and device for oral evaluation
US8870575B2 (en) * 2010-08-03 2014-10-28 Industrial Technology Research Institute Language learning system, language learning method, and computer program product thereof
CN104464423A (en) * 2014-12-19 2015-03-25 科大讯飞股份有限公司 Calibration optimization method and system for speaking test evaluation
CN105681920A (en) * 2015-12-30 2016-06-15 深圳市鹰硕音频科技有限公司 Network teaching method and system with voice recognition function
CN105957516A (en) * 2016-06-16 2016-09-21 百度在线网络技术（北京）有限公司 Switching method and device for multiple voice identification models
CN106782603A (en) * 2016-12-22 2017-05-31 上海语知义信息技术有限公司 Intelligent sound evaluating method and system
CN106959943A (en) * 2016-01-11 2017-07-18 阿里巴巴集团控股有限公司 Languages recognize update method and device
KR20180048136A (en) * 2016-11-02 2018-05-10 한국전자통신연구원 Method for pronunciation assessment and system using the method
US9984682B1 (en) * 2016-03-30 2018-05-29 Educational Testing Service Computer-implemented systems and methods for automatically generating an assessment of oral recitations of assessment items
CN109065020A (en) * 2018-07-28 2018-12-21 重庆柚瓣家科技有限公司 The identification storehouse matching method and system of multilingual classification
KR20190012419A (en) * 2017-07-27 2019-02-11 한국전자통신연구원 System and method for evaluating speech fluency automatically

Patent Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110144990A1 (en) * 2009-12-16 2011-06-16 International Business Machines Corporation Rating speech naturalness of speech utterances based on a plurality of human testers
US8870575B2 (en) * 2010-08-03 2014-10-28 Industrial Technology Research Institute Language learning system, language learning method, and computer program product thereof
CN103761975A (en) * 2014-01-07 2014-04-30 苏州思必驰信息科技有限公司 Method and device for oral evaluation
CN104464423A (en) * 2014-12-19 2015-03-25 科大讯飞股份有限公司 Calibration optimization method and system for speaking test evaluation
CN105681920A (en) * 2015-12-30 2016-06-15 深圳市鹰硕音频科技有限公司 Network teaching method and system with voice recognition function
CN106959943A (en) * 2016-01-11 2017-07-18 阿里巴巴集团控股有限公司 Languages recognize update method and device
US9984682B1 (en) * 2016-03-30 2018-05-29 Educational Testing Service Computer-implemented systems and methods for automatically generating an assessment of oral recitations of assessment items
CN105957516A (en) * 2016-06-16 2016-09-21 百度在线网络技术（北京）有限公司 Switching method and device for multiple voice identification models
KR20180048136A (en) * 2016-11-02 2018-05-10 한국전자통신연구원 Method for pronunciation assessment and system using the method
CN106782603A (en) * 2016-12-22 2017-05-31 上海语知义信息技术有限公司 Intelligent sound evaluating method and system
KR20190012419A (en) * 2017-07-27 2019-02-11 한국전자통신연구원 System and method for evaluating speech fluency automatically
CN109065020A (en) * 2018-07-28 2018-12-21 重庆柚瓣家科技有限公司 The identification storehouse matching method and system of multilingual classification

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
陈嘉华: "《基于深度学习的英语语音识别与发音质量评价》", 《中国优秀硕士学位论文全文数据库 哲学与人文科学辑》 *

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110491369A (en) * 2019-07-24 2019-11-22 北京大米科技有限公司 Appraisal procedure, device, storage medium and the electronic equipment of spoken grade
CN110489454A (en) * 2019-07-29 2019-11-22 北京大米科技有限公司 A kind of adaptive assessment method, device, storage medium and electronic equipment
WO2021018232A1 (en) * 2019-07-29 2021-02-04 北京大米科技有限公司 Adaptive evaluation method and apparatus, storage medium, and electronic device
CN110797044A (en) * 2019-08-22 2020-02-14 腾讯科技（深圳）有限公司 Audio data processing method and device, computer equipment and storage medium
CN110797044B (en) * 2019-08-22 2024-02-06 腾讯科技（深圳）有限公司 Audio data processing method, device, computer equipment and storage medium
CN111816210A (en) * 2020-06-23 2020-10-23 华为技术有限公司 Voice scoring method and device
CN111816210B (en) * 2020-06-23 2022-08-19 华为技术有限公司 Voice scoring method and device
CN111915407A (en) * 2020-08-10 2020-11-10 政采云有限公司 Object compliance detection method and system
CN112017694A (en) * 2020-08-25 2020-12-01 天津洪恩完美未来教育科技有限公司 Voice data evaluation method and device, storage medium and electronic device
CN112017694B (en) * 2020-08-25 2021-08-20 天津洪恩完美未来教育科技有限公司 Voice data evaluation method and device, storage medium and electronic device
CN112908359A (en) * 2021-01-31 2021-06-04 云知声智能科技股份有限公司 Voice evaluation method and device, electronic equipment and computer readable medium

Also Published As

Publication number Publication date
CN109741734B (en) 2021-07-20

Similar Documents

Publication Publication Date Title
CN109741734A (en) A kind of speech evaluating method, device and readable medium
CN107578775B (en) Multi-classification voice method based on deep neural network
CN105448292B (en) A kind of time Speech Recognition System and method based on scene
CN108255934B (en) Voice control method and device
CN102194454B (en) Equipment and method for detecting key word in continuous speech
CN108428446A (en) Audio recognition method and device
CN110246487A (en) Optimization method and system for single pass speech recognition modeling
CN108711420A (en) Multilingual hybrid model foundation, data capture method and device, electronic equipment
CN107316638A (en) A kind of poem recites evaluating method and system, a kind of terminal and storage medium
CN110265040A (en) Training method, device, storage medium and the electronic equipment of sound-groove model
CN111833853A (en) Voice processing method and device, electronic equipment and computer readable storage medium
CN107871496B (en) Speech recognition method and device
CN111862942B (en) Method and system for training mixed speech recognition model of Mandarin and Sichuan
CN107133303A (en) Method and apparatus for output information
JP2017016566A (en) Information processing device, information processing method and program
CN102918591A (en) Geotagged environmental audio for enhanced speech recognition accuracy
CN107710192A (en) Measurement for the automatic Evaluation of conversational response
CN110322900A (en) A kind of method of phonic signal character fusion
CN102142253A (en) Voice emotion identification equipment and method
CN103594087A (en) Method and system for improving oral evaluation performance
CN106898176A (en) Analysis of the students method, analysis of the students server and point-of-reading system based on talking pen
CN108986798B (en) Processing method, device and the equipment of voice data
CN102184654B (en) Reading supervision method and device
CN110459242A (en) Change of voice detection method, terminal and computer readable storage medium
CN103730032A (en) Method and system for controlling multimedia data

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