WO2018098932A1 - 一种基于语音识别的主动型搜索方法和装置 - Google Patents

一种基于语音识别的主动型搜索方法和装置 Download PDF

Info

Publication number
WO2018098932A1
WO2018098932A1 PCT/CN2017/076968 CN2017076968W WO2018098932A1 WO 2018098932 A1 WO2018098932 A1 WO 2018098932A1 CN 2017076968 W CN2017076968 W CN 2017076968W WO 2018098932 A1 WO2018098932 A1 WO 2018098932A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
search
user
voice
domain
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.)
Ceased
Application number
PCT/CN2017/076968
Other languages
English (en)
French (fr)
Inventor
项连志
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.)
Baidu Online Network Technology Beijing Co Ltd
Original Assignee
Baidu Online Network Technology Beijing 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 Baidu Online Network Technology Beijing Co Ltd filed Critical Baidu Online Network Technology Beijing Co Ltd
Publication of WO2018098932A1 publication Critical patent/WO2018098932A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/3331—Query processing
    • G06F16/334—Query execution
    • G06F16/3344—Query execution using natural language analysis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/3331—Query processing
    • G06F16/334—Query execution
    • G06F16/3343—Query execution using phonetics
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/30—Semantic analysis
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/02—Feature extraction for speech recognition; Selection of recognition unit
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/26—Speech to text systems
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L17/00—Speaker identification or verification techniques
    • G10L17/26—Recognition of special voice characteristics, e.g. for use in lie detectors; Recognition of animal voices

Definitions

  • the present invention relates to the field of computer technologies, and in particular, to an active search technology based on voice recognition.
  • the existing information acquisition mainly depends on the search engine, and the related information is obtained through the search engine.
  • This method largely solves the problem that people obtain information from massive information, but the following problems are applied when communicating: 1) Search The resulting information relevance depends on the keywords entered by the user. Most of today's search engines use keyword-based index to find information. When users can't build suitable keywords, the relevance of information will drop sharply. 2) The lag of information acquisition, after users build keywords, Entering keywords into the search engine to get relevant information, this series of processes has a lag relative to communication.
  • an active search method based on speech recognition comprises the following steps:
  • the step c comprises:
  • the voice information is translated into a natural language text based on the recognition of the voice information
  • the subject domain comprises at least one of the following:
  • the subject domain comprises an information domain, an information intent, and an information intent object
  • the step d includes:
  • the populated theme template mapping is expanded to a corresponding search structure.
  • the step d comprises:
  • the step e includes:
  • an active matching search is performed, and corresponding topic information is acquired and presented to the user.
  • the step e includes:
  • an active search device based on speech recognition wherein the search device comprises:
  • a identifying device configured to identify the received voice information, and obtain a voice feature vector of the voice information
  • a determining device configured to determine whether the voice feature vector matches a reference voice feature vector preset by the user, and obtain a corresponding determination result
  • a generating device configured to generate a subject domain corresponding to the voice information based on the identification of the voice information if the determination result is a mismatch
  • mapping device configured to expand the topic domain mapping into a corresponding search structure
  • a rendering device configured to perform an active matching search according to the search structure, acquire corresponding topic information, and present the information to the user.
  • said generating means is for:
  • the voice information is translated into a natural language text based on the recognition of the voice information
  • the subject domain comprises at least one of the following:
  • the subject domain comprises an information domain, an information intent and an information intent object, wherein the mapping device is configured to:
  • the populated theme template mapping is expanded to a corresponding search structure.
  • the mapping device is configured to expand the subject domain mapping to the search structure according to an attribute of a search engine corresponding to the active matching search.
  • the presentation device is for:
  • an active matching search is performed, and corresponding topic information is acquired and presented to the user.
  • the presentation device comprises:
  • An obtaining unit configured to perform an active matching search according to the search structure, and obtain corresponding candidate information
  • An integration unit configured to generate the topic information based on information extraction and integration of the candidate information
  • a presentation unit for presenting the topic information to the user.
  • a computer readable storage medium storing computer code, the method of any of the above being executed when the computer code is executed .
  • a computer apparatus comprising a memory and a processor, the memory storing computer code, the processor being configured to execute by executing the computer code The method of any of the preceding claims.
  • the present invention has the following advantages:
  • the invention continuously obtains the voice information actively, and determines whether the voice feature vector of the voice information matches the preset reference voice feature vector, that is, whether the sender of the voice information is a predetermined user, and the result of the judgment is Whether to start the condition of the search function, and then obtain the subject information of the information corresponding to the voice information through the search operation, and present the information to the user, help the user to acquire the unknown information in a timely manner, and solve the problem in the communication or other language environment. Symmetrical problem.
  • the present invention obtains the keywords in the voice information more accurately by adopting the natural semantic analysis technology, and expands or deletes the keywords according to the semantics of the voice information, thereby improving the accuracy and hit of the search results. Rate to enhance the user experience.
  • the present invention generates a subject domain corresponding to the voice information based on the identification of the voice information, and the information domain, the information intention, and the information intention object in the subject domain structure of the subject domain supplement and correct keywords in the subject domain, and improve The accuracy of the search reduces the noise during the search and improves the user experience.
  • FIG. 1 is a block diagram showing the structure of an active search device based on speech recognition according to an aspect of the present invention
  • FIG. 2 is a schematic diagram showing a search structure when a Baidu search engine is used according to an embodiment of the present invention
  • FIG. 3 is a flow chart showing a voice recognition based active search method according to another aspect of the present invention.
  • Computer device also referred to as “computer” in the context, is meant an intelligent electronic device that can perform predetermined processing, such as numerical calculations and/or logical calculations, by running a predetermined program or instruction, which can include a processor and The memory is executed by the processor to execute a predetermined process pre-stored in the memory to execute a predetermined process, or is executed by hardware such as an ASIC, an FPGA, a DSP, or the like, or a combination of the two.
  • Computer devices include, but are not limited to, servers, personal computers, notebook computers, tablets, smart phones, and the like.
  • the user equipment, the network equipment, the network, and the like are merely examples, and other existing or future possible computer equipment or networks, such as those applicable to the present invention, are also included in the scope of the present invention. It is included here by reference.
  • first first
  • second second
  • first first
  • second second
  • the search device 1 is a block diagram showing the structure of an active search device based on speech recognition according to an aspect of the present invention.
  • the search device 1 includes an identification device 101, a determination device 102, a generation device 103, a mapping device 104, and a presentation device 105.
  • the network device interacts with the user equipment, receives the voice information collected by the user equipment, identifies the received voice information, obtains the voice feature vector of the voice information, and determines the location. Whether the voice feature vector matches the reference voice feature vector preset by the user, and if the judgment result is a mismatch, based on the recognition of the voice information, generating a topic domain corresponding to the voice information, The topic domain mapping is extended to a corresponding search structure; an active matching search is performed according to the search structure, and corresponding topic information is obtained and returned to the user equipment, thereby being presented to the user.
  • the user equipment includes, but is not limited to, a personal computer, a portable computer, a tablet computer, a smart phone, a PDA, a virtual reality (VR) glasses, a virtual reality helmet, a smart earphone, and the like.
  • the user equipment collects voice information through the voice collection module thereon, and sends the collected voice information to the search device 1 through the network, so that the search device 1 receives the collected voice information and performs voice recognition, and the voice recognition result is obtained.
  • the corresponding topic information is obtained by searching and presented to the user.
  • the identification device 101 identifies the received voice information and obtains a voice feature vector of the voice information. Specifically, the user equipment continuously collects voice information through a specific module, such as a voice collection module, and then sends the voice information to the search device 1 through the network, or the search device 1 periodically passes an agreed communication manner, such as a communication protocol such as http and https. Obtaining the voice information collected by the user equipment directly from the user equipment, after the search apparatus 1 acquires the voice information, first filtering the noise information in the voice information by using a filtering device, and the filtering device receives the filtered information.
  • a specific module such as a voice collection module
  • the search device 1 periodically passes an agreed communication manner, such as a communication protocol such as http and https.
  • the speech information is pre-processed, for example, by sampling, quantization, pre-emphasis, windowing, and then, for example, by using Mel Frequency Cepstral Coefficient (MFCC) extraction method, linear predictive coding (linear predictive coding) , LPC) extraction method, high frequency domain emphasis extraction method or window function extraction method to extract speech feature vectors.
  • MFCC Mel Frequency Cepstral Coefficient
  • LPC linear predictive coding
  • high frequency domain emphasis extraction method or window function extraction method to extract speech feature vectors.
  • the speech feature vector is a speech feature parameter for identifying a speaker (voiceprint recognition), and by determining whether the two speech feature vectors match, the voice information corresponding to the two speech feature vectors can be substantially determined. Whether it is from the same user.
  • the Mel frequency is proposed based on the auditory characteristics of the human ear, which has a nonlinear relationship with the Hz frequency.
  • B can quickly obtain music related information corresponding to the voice information sent by A through the search device 1, specifically, in communication.
  • the live search function when the A sends out the voice information, the user equipment collects the voice information through the voice receiving module thereon, and then sends the voice information to the search device 1 through the network. After the search device 1 acquires the voice information, the identification device 101 receives the voice information.
  • the voice information is preprocessed, for example, performing Fast Fourier Transformation (FFT) on each frame, obtaining a spectrum, and then obtaining an amplitude spectrum, adding a Mel filter group to the amplitude spectrum, for all
  • FFT Fast Fourier Transformation
  • the filter output is logarithm (Logarithm), and further discrete Cosine Transform (DCT) can be used to obtain the speech feature parameter MMFC, and then the corresponding speech feature vector is obtained.
  • DCT discrete Cosine Transform
  • the determining device 102 determines whether the voice feature vector matches the reference voice feature vector preset by the user, and obtains a corresponding determination result.
  • the user sets a reference voice feature vector in the search device 1 in advance, for example, by receiving voice information of a predetermined user in advance, and identifying the voice information by the recognition device 101, obtaining a voice feature vector of the voice information, and then
  • the voice feature vector is stored as a reference voice feature vector for determining whether the voice feature vector of the voice information recognized by the search device 1 matches the reference voice feature vector, that is, whether the sender of the voice information recognized by the search device 1 is a predetermined user.
  • a corresponding judgment result is obtained, and the judgment result is further used as a condition for starting a search.
  • the predetermined user may be determined according to a specific use situation.
  • the user C sends voice information to the search device 1, the recognition device 101 recognizes the voice information, and obtains a voice feature vector of the voice information, and uses the voice feature vector as a reference voice feature vector through the reference voice feature vector. It can be judged whether the voice information subsequently received by the search device 1 is from the user C, and then responds according to the judgment result.
  • the search device 1 receives the voice information in the occasion, and recognizes the received voice information, obtains the voice feature vector x of the voice information, and determines the device 102 the voice
  • the feature vector x is compared with the reference speech feature vector to determine whether the two match, and the corresponding judgment result is obtained.
  • the voice information is the voice information sent by the user C itself, and no further processing is performed; If the result of the determination is a mismatch, the voice information is voice information sent by the user C, and the search device 1 performs a subsequent search operation, and obtains a corresponding search result and provides the user C, and the user C can obtain the search result according to the search result.
  • User C uses the search device 1 in a relatively noisy environment for user-based
  • the search device 1 receives the voice information in the occasion, and identifies the received voice information, obtains the voice feature vector y of the voice information, and the determining device 102 uses the voice feature vector y.
  • the voice information is the voice information sent by the user C itself, and the search device 1 performs a subsequent search operation, and obtains the corresponding search result and provides the user C to the user C; if the judgment result is not matched, the further deal with.
  • the generating means 103 If the result of the determination is a mismatch, the generating means 103 generates a subject domain corresponding to the voice information based on the identification of the voice information. Specifically, if the user wants to search for the corresponding information of the voice information sent by the user other than a predetermined user (including himself), the voice feature vector of the voice information sent by the predetermined user is used as the reference voice feature vector. If the result of the determination is a mismatch, the sender of the received voice information is not a predetermined user, that is, the sender of the received voice information is a user other than the predetermined user, and the voice feature vector conforms to the user for searching.
  • the condition generating means 103 identifies a specific acoustic model by the feature extraction technique and the pattern matching technique based on the voice information recognized by the identification device 101, and forms a certain language model by the acoustic model training, in the acoustic model and the language model.
  • the composed space is quickly optimized, the voice information is converted into text information, and then the text information is semantically analyzed.
  • the text information is segmented by a full-segment word segmentation technique to obtain a corresponding keyword. Then perform eigenvalue calculation according to the preset feature model, and match the most Subject field, and a keyword relating to the filling a domain; wherein the subject domain includes but is not limited to: an information domain; an information intent; an information intent object.
  • the subject domain (which can be regarded as a feature structure) can be obtained by the following methods: first, setting feature parameters by human annotation; second, grasping a large amount of basic data, performing machine learning, and generating feature parameters; second method sometimes combining artificial Regulate the way to correct it to achieve better results.
  • first setting feature parameters by human annotation
  • second grasping a large amount of basic data, performing machine learning, and generating feature parameters
  • second method sometimes combining artificial Regulate the way to correct it to achieve better results.
  • the same type of data or resources, and the services provided around the data or resources are called an information domain.
  • the information domain data is generally structured tabular data, with a primary key (primary attribute), the information domain is generally named after nouns; the information intent represents the user's operation on the information domain data, such as querying, querying the value of an attribute, Booking, dialing, etc., are generally named after verbs; the information intent object (object) describes the intent data, which is also a parameter required to implement the intent, and is embodied as a main keyword about the text information obtained by the word segmentation technique.
  • the above information domain, information intent, and information intent object can also be regarded as the subject domain structure of the subject domain.
  • the user D using the search device 1 uses the voice feature vector of the sent voice information as the reference voice feature vector, and the user D and the friend E discuss the matter of returning to the home, D said: I have already decided to return the train ticket to Qingdao, friend E Said: I have to return to Qingdao recently, I don't know if I can still book a train ticket from Beijing to Qingdao.
  • the identification device 101 respectively identifies the received voice information, obtains a voice feature vector corresponding to the voice information, and the determining device 102 determines the voice information recognized by the recognition device 101, and determines the voice feature vector of the voice information sent by the user D.
  • the judging device 102 judges the voice information sent by the user E, "I have to return to Qingdao recently, I don't know if I can still book the train ticket from Beijing to Qingdao".
  • the generating device 103 converts the speech information into text information, and performs segmentation by semantic analysis on the text information, after segmentation
  • the text is calculated according to the preset feature model, the optimal subject domain is matched, and the text content is filled into the subject domain to obtain a simple subject domain structure:
  • Intent predetermined
  • Mapping device 104 extends the subject domain mapping to a corresponding search structure.
  • the mapping device 104 maps the corresponding keywords in the filled subject domain according to a preset rule, for example, mapping through a certain template, and performing corresponding expansion, for example, according to information extension of the template.
  • a preset rule for example, mapping through a certain template
  • performing corresponding expansion for example, according to information extension of the template.
  • a part of the keywords supplement the keyword information in the subject domain, and get the corresponding complete search structure.
  • the presentation device 105 performs an active matching search according to the search structure, acquires corresponding topic information, and presents it to the user. Specifically, the presentation device 105 performs an active matching search according to the search result determined by the mapping device 104, and obtains a large number of related search results, and the rendering device 105 determines the search result, and if the search result is the thematic information, directly presents To the user; if the search result is non-themeization information, the presentation device 105 performs the thematic information extraction, and presents the main information after the refined aggregation to the user through the screen of the smart machine or the wearable headset that receives the voice message. To help users quickly obtain the specific content involved in the voice information, and obtain unknown information in a timely and accurate manner.
  • the method for the presentation device 105 to acquire the corresponding topic information is, for example, 1) for a single document, using an automatic summarization technique, for example, an automatic summarization based on understanding or an automatic summarization based on structure, to obtain the obtained search result. Summarizing and extracting, obtaining concise and coherent topic information; 2) For multiple documents, multiple documents can be grouped and classified by clustering method, and then the abstract information is extracted and then extracted by abstract technology. The clustering method needs to calculate the feature values of the documents corresponding to each search result, and then aggregates them according to the feature values to a certain category. Among them, calculate the characteristics of each document The characteristic parameters and seed document feature values required for the levy are trained in advance through the training model.
  • an automatic summarization technique for example, an automatic summarization based on understanding or an automatic summarization based on structure
  • the search device 1 provides a reference voice feature vector preset function for the user, by which the user presets the reference voice feature vector, and then determines whether the voice feature vector of the received voice information matches the reference voice feature vector, and the judgment result is Whether to start the condition of the search function, and organize the information obtained by the search, obtain the thematic information and present it to the user, and continuously receive the voice information actively in the communication or the complex voice environment, and independently judge the voice Information, understanding and searching for information related to the voice information, and feeding back relevant information or abstracts to the user in real time, so that the user can acquire unknown information in time and quickly, and solve the problem of information asymmetry in communication.
  • the reference speech feature vector is set as the reference speech feature vector of the user's own speech information, and the search device 1 can continuously receive the speech information actively, if the sender of the received speech information is the The user does not perform further processing.
  • the search device 1 will provide the user with relevant explanations and principles of the contents of the doctor in real time, so that The user has the basic ability to continue to ask questions, obtain the desired information, and relieve the information asymmetry between doctors and patients to a certain extent. 3) When negotiating the business, the user can be more professional through the search device 1 for the questions or concepts raised by the customer.
  • search Set real-time information provided by the user in a conversation with other people can be involved in multi-topic in many fields, the two sides to improve communication effectiveness and scalability. Further, the search device 1 will promote the information inclusiveness to a new level, and in use will provide a new way of obtaining for those who do not know how to obtain information.
  • the generating means 103 is configured to: 1) if the result of the determination is a mismatch, translate the voice information into a natural language text based on the recognition of the voice information; 2) the natural language text A natural semantic analysis is performed to generate a subject domain corresponding to the voice information. Specifically, if the voice feature vector of the received voice information and the user-preset base language The tone feature vector does not match, that is, the sender of the received voice information is not a predetermined user, and the generating device 103 performs analog-to-digital conversion on the voice information based on the recognition of the voice information to obtain audio of the voice information.
  • the generating device 103 obtains the keyword in the voice information more accurately by adopting the natural semantic analysis technology, and expands or deletes the keyword according to the semantics of the voice information, thereby improving the accuracy of the search result and Hit rate to enhance the user experience.
  • the subject domain includes an information domain, an information intent, and an information intent object
  • the mapping device 104 determines a corresponding topic template according to the information domain and the information intent of the subject domain;
  • the information intent object generates an extended extended keyword list, and the subject domain and the extended extended keyword list are filled into the theme template; and the filled theme template mapping is expanded to a corresponding search structure.
  • the subject domain structure of the subject domain includes: an information domain, an information intent, and an information intent object
  • the information domain of the subject domain After the text content converted into the voice information is filled into the subject domain, the information domain of the subject domain, the information intent, and the information intent object Each has its own corresponding keyword.
  • the mapping device 104 acquires a corresponding theme template in the theme template library by pattern matching according to the information domain and the information intent in the subject domain, and the information intent object (main keyword) generates the extended extension keyword by combining the preset rules in the theme template.
  • a list, the content of the subject domain and the extended extended keyword list are respectively filled into corresponding positions of the theme template, wherein the extended extended keyword may be a keyword plus a keyword, and the extension may be extended.
  • the words are populated into the corresponding positions of the theme template to make the content of the theme template more complete and improve the hit rate of the search.
  • the information intent object (primary keyword) and the extended extension keyword together constitute a theme template and various search keywords of the corresponding search structure.
  • the search mode is pre-set in the theme template, and the search mode is, for example, a search instruction, and a specific search structure preset by each search instruction (where the specific search structure is a search structure of the theme template, corresponding to the search engine) Search structure is different), etc., content mapping of the theme template Expands to certain properties of the search structure corresponding to the default search engine.
  • the theme template library is a database for storing a theme template in advance. For example, a large number of voice information samples can be collected and analyzed, and a large number of theme templates are obtained and stored in the theme template library.
  • the extended keyword makes the search keyword richer, and it is easier to obtain the result desired by the user, and at the same time, the noise information in the search keyword can be reduced, the unnecessary search is reduced, and the keyword is more accurate.
  • the voice information recognized by the identification device 101 is: "I have recently returned to Qingdao, I don't know if I can still book a train ticket from Beijing to Qingdao," and the mapping device 104 generates the subject domain generated by the generating device 103.
  • the corresponding theme template is obtained in the theme template library through pattern matching, and the content of the subject domain is correspondingly filled into the theme template, and the information intention object (main keyword: Beijing) , Qingdao) combined with the preset rules in the theme template, such as the date limit of the preset tickets in the theme template is within one week, and according to the information field for the train information, the theme template built-in excludes the transportation modes such as airplanes, buses, boats, etc. Therefore, the list of extended extension keywords is generated: date: within one week; transportation mode: - aircraft, - bus, - ship, among them, "within a week” is an increased keyword, "aircraft, bus, ship” is a reduced keyword.
  • the search device 1 searches by using a Baidu search engine
  • the content of the theme template is mapped to a search structure corresponding to the Baidu search engine, for example, the content of the theme template is mapped to the search structure shown in FIG. 2,
  • the active matching search is performed, and the corresponding body information is obtained and presented to the user.
  • 2 illustrates a framework of a search structure when using a Baidu search engine, in accordance with one embodiment of the present invention.
  • the mapping device 104 extends the subject domain mapping to the search structure according to an attribute of the search engine corresponding to the active matching search. Specifically, the mapping device 104 constructs a search instruction according to the information domain of the subject domain, the information intent, the information intent object, or any combination of the three, and then searches for the corresponding search engine based on the active matching search. The difference in attributes generates a corresponding search structure. For example, when the user searches using the Baidu search engine, according to the recognition of the received user voice information, the generated search structure framework corresponding to the subject domain is as shown in FIG. 2, according to the received voice information, in the input bar of FIG. A keyword in the search structure corresponding to the voice information is automatically embedded to perform a search operation.
  • the presenting device 105 performs an active matching search according to the search structure corresponding to the subject domain, and combines the cognitive computing, and acquires corresponding topic information and presents it to the user.
  • the presentation device 104 according to the search structure corresponding to the topic domain, combines the cognitive calculation, matches the information that satisfies the search structure, acquires the topic information of the information, and presents the information to the user, for example, learning and analyzing the combination.
  • the keyword information in the search structure is used to perceive the user's needs according to the keyword information, so that the user obtains information with high relevance, and organizes the search result into a theme, obtains corresponding topic information, and presents the information to the user.
  • the user is described to improve the intelligence of the search device 1 for performing automatic matching search.
  • the cognitive calculation realizes the simulation of the human brain thinking mode and cognition by calculating the simple unit calculation by hierarchically accumulating the unstructured information.
  • the presentation device 105 comprises: an acquisition unit 1051 (not shown), an integration unit 1052 (not shown) and a presentation unit 1053 (not shown).
  • the obtaining unit 1051 performs an active matching search according to the search structure corresponding to the subject domain. So, the corresponding candidate information is obtained. Specifically, the obtaining unit 1051 performs an automatic matching search according to the search structure, and obtains a large number of related search results.
  • the display interface of the presentation device 105 is limited, and the time for the user to obtain information is also limited, so the related search needs to be performed. As a result, the candidate information is refined and processed and presented to the user.
  • the integration unit 1052 generates the topic information based on the information extraction integration of the candidate information. Specifically, a resource and rule library is preset, and the integration unit 1052 determines a topic according to the search structure, and then extracts, from the information of the search result, the information related to the search result by using the resource and the rule stored in the rule library. A description vector of the topic, based on the description vector, generates global information related to the topic as topic information, so that the user quickly acquires relevant knowledge through the global information. For example, when the user queries "his rash", the search device 1 has a large number of search results after the search, and is limited by the presentation device 105, and cannot present all the results to the user at one time, and the user can quickly acquire the main knowledge.
  • the integration unit 1052 summarizes the urticaria as the following subject information: wheal; rubella; skin disease; sudden onset of red lumps on local or systemic skin; rapid onset, rapid disappearance; itching; allergies. After seeing the above general information, the user can get a general understanding of the urticaria disease.
  • the presentation unit 1053 presents the topic information to the user.
  • the presentation manner in which the topic information is presented to the user includes, but is not limited to: being presented by the audio device; being presented by the video device; being presented by the wearable device.
  • the audio device is, for example, a smart headset, a smart headset, etc.
  • the video device is, for example, virtual reality (VR) glasses, smart glasses capable of freely switching between ordinary glasses functions and projection functions; wearable devices include smart watches, smart devices Helmets, smart clothes, etc.
  • VR virtual reality
  • FIG. 3 is a flow chart showing a voice recognition based active search method according to another aspect of the present invention.
  • the search device 1 identifies the received voice information, and obtains a voice feature vector of the voice information. Specifically, the user equipment continuously collects voice information through a specific module, such as a voice collection module, and then sends the message to the network through the network.
  • the device 1 or the search device 1 periodically obtains the voice information collected by the user equipment directly from the user equipment by using a communication protocol such as http, https, etc., after the search device 1 obtains the voice information, first
  • the filtering device filters out the noise information in the voice information, and in step S301, the searching device 1 performs pre-processing on the received filtered voice information, for example, through a sampling, quantization, pre-emphasis, windowing preprocessing process.
  • the speech feature vector is a speech feature parameter for identifying a speaker (voiceprint recognition), and by determining whether the two speech feature vectors match, the voice information corresponding to the two speech feature vectors can be substantially determined. Whether it is from the same user.
  • the Mel frequency is proposed based on the auditory characteristics of the human ear, which has a nonlinear relationship with the Hz frequency.
  • B can quickly obtain music related information corresponding to the voice information sent by A through the search device 1, specifically, in communication.
  • B activates the search function by clicking the corresponding function area on the application interface corresponding to the search device 1 on the user equipment, or by other means.
  • the user equipment collects the voice information through the voice receiving module thereon. And then transmitting it to the search device 1 through the network.
  • the search device 1 acquires the voice information, in step S301, the search device 1 performs pre-processing on the received voice information, for example, performing fast fast on each frame.
  • FFT Fast Fourier Transformation
  • MMFC Discrete Cosine Transform
  • the search device 1 determines whether the speech feature vector matches the reference speech feature vector preset by the user, and obtains a corresponding determination result.
  • the user sets the reference voice feature vector in the search device 1 in advance, for example, by receiving the voice information of the predetermined user in advance, and obtaining the voice feature vector of the voice information by the search device 1 identifying the voice information in step S301. And storing the speech feature vector as a reference speech feature vector, and determining whether the speech feature vector of the speech information recognized by the search device 1 matches the reference speech feature vector, that is, whether the issuer of the speech information recognized by the search device 1 is For the predetermined user, the corresponding judgment result is obtained, and the judgment result is used as a condition for starting the search.
  • the predetermined user may be determined according to a specific use situation.
  • the search device 1 recognizes the voice information, and obtains a voice feature vector of the voice information, and uses the voice feature vector as a reference voice feature vector.
  • the reference speech feature vector can determine whether the speech information subsequently received by the search device 1 is from the user C, and then respond accordingly according to the judgment result. For example, 1) expert consultation occasion, the expert tells a lot of ideas and opinions, the search device 1 receives the voice information in the occasion, and recognizes the received voice information, and obtains the voice feature vector x of the voice information, in steps In S302, the search device 1 compares the phonetic feature vector x with the reference phone feature vector to determine whether the two match, and obtains a corresponding judgment result.
  • the voice information is sent by the user C himself.
  • the voice information is not further processed; if the result of the determination is not matched, the voice information is voice information sent by the user C, and the search device 1 performs a subsequent search operation, and obtains the corresponding search result and provides the user C with the corresponding search result.
  • User C can understand the ideas and opinions of the experts according to the search results, and effectively communicate and consult with the experts to improve the efficiency and quality of user C and expert communication; 2) User C uses in a relatively noisy environment.
  • the search device 1 performs a voice search based on voice information sent by the user C, and the search device 1 receives the field Converging voice information, and identifying the received voice information to obtain the voice information
  • the speech feature vector y in step S302, the search device 1 compares the speech feature vector y with the reference speech feature vector to determine whether the two match, and obtains a corresponding determination result. If the determination result is a match, the The voice information is the voice information sent by the user C itself, and the search device 1 performs a subsequent search operation, and obtains the corresponding search result and provides the user C to the user C; if the judgment result is not matched, no further processing is performed.
  • step S303 the search device 1 generates a subject domain corresponding to the voice information based on the recognition of the voice information. Specifically, if the user wants to search for the corresponding information of the voice information sent by the user other than a predetermined user (including himself), the voice feature vector of the voice information sent by the predetermined user is used as the reference voice feature vector. If the result of the determination is a mismatch, the sender of the received voice information is not a predetermined user, that is, the sender of the received voice information is a user other than the predetermined user, and the voice feature vector conforms to the user for searching.
  • the search device 1 identifies a specific acoustic model by the feature extraction technique and the pattern matching technique based on the recognized speech information, and forms a certain language model by the acoustic model training, in the acoustic model and The space composed of the language model is quickly optimized, the voice information is transformed into text information, and then the text information is semantically analyzed, for example, the word segmentation technique is used to segment the text information, and the corresponding key is obtained. Word, then perform eigenvalue calculation according to the preset feature model, matching the most The subject domain is filled and the keyword is populated into the subject domain; wherein the subject domain includes but is not limited to: an information domain; an information intent; and an information intent object.
  • the subject domain (which can be regarded as a feature structure) can be obtained by the following methods: first, setting feature parameters by human annotation; second, grasping a large amount of basic data, performing machine learning, and generating feature parameters; second method sometimes combining artificial Regulate the way to correct it to achieve better results.
  • first setting feature parameters by human annotation
  • second grasping a large amount of basic data, performing machine learning, and generating feature parameters
  • second method sometimes combining artificial Regulate the way to correct it to achieve better results.
  • the same type of data or resources, and the services provided around the data or resources are called an information domain.
  • the information domain data is generally structured tabular data, with a primary key (primary attribute), the information domain is generally named after nouns; the information intent represents the user's operation on the information domain data, such as querying, querying the value of an attribute, Booking, dialing, etc., generally named after verbs; information intention object (object) describes the intended data, is also realized
  • the parameters required for the graph are embodied as the main keywords related to the text information obtained by the word segmentation technique.
  • the above information domain, information intent, and information intent object can also be regarded as the subject domain structure of the subject domain.
  • the user D using the search device 1 uses the voice feature vector of the sent voice information as the reference voice feature vector, and the user D and the friend E discuss the matter of returning to the home, D said: I have already decided to return the train ticket to Qingdao, friend E Said: I have to return to Qingdao recently, I don't know if I can still book a train ticket from Beijing to Qingdao.
  • the search device 1 respectively identifies the received voice information, and obtains a voice feature vector corresponding to the voice information respectively.
  • the search device 1 determines the recognized voice information, and determines the user D. If the voice feature vector of the voice information matches the reference voice feature vector, no further processing is performed.
  • step S302 the search device 1 determines the voice message sent by the user E. "I have to return to Qingdao recently. I don't know if I can still When the voice feature vector of the train ticket to Beijing to Qingdao does not match the reference voice feature vector, that is, when the issuer of the voice message is not the user, the search device 1 converts the voice information into text in step S303. Information, and the word information is segmented by semantic analysis, and the text after the word segmentation is calculated according to the preset feature model, the optimal subject domain is matched, and the text content is filled into the subject domain to obtain a simple subject domain. structure:
  • Intent predetermined
  • step S304 the search device 1 expands the subject domain mapping to a corresponding search. structure.
  • step S304 the search device 1 maps the keywords corresponding to the filled subject domain according to a preset rule, for example, by using a template, and performs corresponding expansion, for example, according to the
  • the information extension of the template generates a part of keywords, supplements the keyword information in the subject domain, and obtains a corresponding complete search structure.
  • step S305 the search device 1 performs an active matching search according to the search structure, acquires corresponding topic information, and presents it to the user. Specifically, in step S305, the search device 1 performs an active matching search according to the search result determined in step S304, and obtains a large number of related search results. In step S305, the search device 1 determines the search result.
  • the search device 1 When the search result is the thematic information, it is directly presented to the user; if the search result is the non-thematic information, in step S305, the search device 1 performs the thematic information extraction, through the screen of the smart machine or has the receiving
  • the wearable headset of the voice message presents the user with the main information after the refining and aggregation, helps the user to quickly obtain the specific content involved in the voice information, and acquires the unknown information in time and accurately.
  • the method for the search device 1 to acquire the corresponding topic information is, for example, 1) for a single document, using an automatic summarization technique, for example, based on an understanding of automatic summarization techniques or a structure-based automatic summarization technique.
  • the search results are summarized and extracted to obtain concise and coherent topic information; 2) for multiple documents, multiple documents can be grouped, classified, and then subjected to abstraction techniques by clustering methods.
  • the topic information is extracted, wherein the clustering method needs to calculate feature values of documents corresponding to each search result, and then aggregates to a certain category according to the feature values.
  • the feature parameters and the seed document feature values required to calculate the feature values of each document are trained in advance through the training model.
  • the search device 1 provides a reference voice feature vector preset function for the user, by which the user presets the reference voice feature vector, and then determines whether the voice feature vector of the received voice information matches the reference voice feature vector, and the judgment result is Whether to start the conditions of the search function, and organize the information obtained by the search, obtain the thematic information and present it to the user, and realize the continuous and active voice receiving in the communication or the complex voice environment.
  • the reference speech feature vector is set as the reference speech feature vector of the user's own speech information, and the search device 1 can continuously receive the speech information actively, if the sender of the received speech information is the The user does not perform further processing.
  • the search device 1 will provide the user with relevant explanations and principles of the contents of the doctor in real time, so that The user has the basic ability to continue to ask questions, obtain the desired information, and relieve the information asymmetry between doctors and patients to a certain extent. 3) When negotiating the business, the user can be more professional through the search device 1 for the questions or concepts raised by the customer.
  • search Set real-time information provided by the user in a conversation with other people can be involved in multi-topic in many fields, the two sides to improve communication effectiveness and scalability. Further, the search device 1 will promote the information inclusiveness to a new level, and in use will provide a new way of obtaining for those who do not know how to obtain information.
  • the search apparatus 1 is configured to: 1) if the determination result is a mismatch, translate the voice information into a natural language text based on the recognition of the voice information; 2) The natural language text performs natural semantic parsing to generate a subject domain corresponding to the voice information.
  • step S303 the search device 1 is based on the Identifying the voice information, performing analog-to-digital conversion on the voice information, obtaining an audio interval of the voice information, extracting feature quantities of the audio data of the audio interval, and identifying the voice information as corresponding based on the feature vector Natural language text; then in step S303, the search device 1 performs natural semantic analysis on the natural language text to obtain a corresponding parsing result, according to the parsing result A subject domain corresponding to the voice information is generated.
  • step S303 the search device 1 obtains the keyword in the voice information more accurately by adopting the natural semantic analysis technology, and expands or deletes the keyword according to the semantics of the voice information to improve the search.
  • the accuracy of the results and the hit rate improve the user experience.
  • the subject domain includes an information domain, an information intent, and an information intent object
  • the search apparatus 1 determines a corresponding topic template according to the information domain of the subject domain, the information intent; according to the theme template And generating, by the information intent object, an extended extended keyword list, filling the subject domain and the extended extended keyword list into the theme template; and expanding the filled theme template mapping into a corresponding search structure.
  • the search device 1 acquires a corresponding theme template in the theme template library by pattern matching according to the information domain and the information intention in the subject domain, and the information intention object (main keyword) Generating an extended extended keyword list according to a preset rule in the theme template, and filling the content of the topic domain and the extended extended keyword list into corresponding positions of the theme template, wherein the extended extended keyword may be a keyword.
  • the keyword may also be reduced, and the extended extended keyword is filled into the corresponding position of the theme template, so that the content of the theme template is more complete, and the hit rate of the search is improved.
  • the information intent object (primary keyword) and the extended extension keyword together constitute a theme template and various search keywords of the corresponding search structure.
  • the search mode is pre-set in the theme template, and the search mode is, for example, a search instruction, and a specific search structure preset by each search instruction (where the specific search structure is a search structure of the theme template, corresponding to the search engine)
  • the search structure is different, etc., and the content mapping of the theme template is extended to certain attributes of the search structure corresponding to the preset search engine.
  • the theme template library is a database for storing a theme template in advance. For example, a large number of voice information samples can be collected for analysis and processing, and a large number of theme modules are obtained.
  • the board is saved in the theme template library.
  • the extended keyword makes the search keyword richer, and it is easier to obtain the result desired by the user, and at the same time, the noise information in the search keyword can be reduced, the unnecessary search is reduced, and the keyword is more accurate.
  • the voice information recognized by the search device 1 is: "I have recently returned to Qingdao, I don't know if I can still book a train ticket from Beijing to Qingdao", in step S304, search The device 1 obtains a corresponding theme template in the theme template library by pattern matching according to the information field (train), information intention (predetermined) in the generated subject domain, and fills the content of the topic domain into the theme template correspondingly, and the information intention Objects (main keywords: Beijing, Qingdao) combined with preset rules in the theme template, such as the date limit of the preset tickets in the theme template is within one week, and according to the information field for the train information, the theme template is built-in to exclude aircraft, Bus, boat and other modes of transportation, so generate extended extension keyword list: date: within one week; transportation mode: - aircraft, - bus, - ship, where "within a week" is the keyword of increase, "aircraft, bus, ship For the reduced keywords,
  • the search device 1 searches by using a Baidu search engine
  • the content of the theme template is mapped to a search structure corresponding to the Baidu search engine, for example, the theme template is
  • the content is mapped to the search structure shown in FIG. 2, and an active matching search is performed, and the corresponding body information is obtained and presented to the user.
  • 2 illustrates a framework of a search structure when using a Baidu search engine, in accordance with one embodiment of the present invention.
  • the search device 1 expands the subject domain mapping to the search structure according to an attribute of the search engine corresponding to the active matching search.
  • the search apparatus 1 constructs a search instruction according to the information domain of the subject domain, the information intent, the information intent object, or any combination of the three, and then searches based on the active matching.
  • Corresponding search engine attributes are different, and a corresponding search structure is generated. For example, when the user searches using the Baidu search engine, according to the recognition of the received user voice information, the generated search structure framework corresponding to the subject domain is as shown in FIG. 2, according to the received voice information, in the input bar of FIG. A keyword in the search structure corresponding to the voice information is automatically embedded to perform a search operation.
  • the search device 1 performs an active matching search according to the search structure corresponding to the subject domain, and combines the cognitive calculation to acquire corresponding topic information and present it to the user.
  • the search device 1 matches the cognitive structure, matches the information that satisfies the search structure, and obtains the topic information of the information, and then presents the information to the user according to the search structure corresponding to the topic domain, for example, Learning and analyzing the keyword information in the search structure, and perceiving the user's needs according to the keyword information, so that the user obtains information with high relevance, and the subjective organization of the search result is performed to obtain corresponding topic information.
  • the cognitive calculation realizes the simulation of the human brain thinking mode and cognition by calculating the simple unit calculation by hierarchically accumulating the unstructured information.
  • step S305 comprises: sub-step 3051 (not shown), sub-step 3052 (not shown) and sub-step 3053 (not shown).
  • sub-step S3051 the search device 1 performs an active matching search according to the search structure corresponding to the subject domain, and acquires corresponding candidate information. Specifically, in sub-step S3051, the search apparatus 1 performs an automatic matching search according to the search structure, and obtains a large number of related search results. In step S305, the display interface of the search apparatus 1 is limited, and the user is obtained. The time for taking the information is also limited. Therefore, the plurality of related search results need to be used as candidate information, and the candidate information is refined and presented to the user.
  • the search device 1 In sub-step S3052, the search device 1 generates the subject information based on the information extraction integration of the candidate information. Specifically, a resource and rule library is preset, and in sub-step S3052, the search device 1 determines a topic according to the search structure, and then uses the resources and rules stored in the resource and rule library to obtain information from the search result. Extracting a description vector related to the topic, and generating global information related to the topic as topic information based on the description vector, so that the user quickly acquires relevant knowledge through the global information. For example, when the user queries "his rash", the search device 1 has a large number of search results after the search, and is limited by the search device 1, and cannot present all the results to the user at one time, and the user can quickly acquire the main knowledge.
  • a resource and rule library is preset
  • the search device 1 determines a topic according to the search structure, and then uses the resources and rules stored in the resource and rule library to obtain information from the search result. Extracting a description vector related to the topic,
  • the search device 1 summarizes the urticaria as the following subject information: wheal; rubella; skin disease; sudden onset of red lumps on local or systemic skin; rapid onset, rapid disappearance; allergy. After seeing the above general information, the user can get a general understanding of the urticaria disease.
  • the search device 1 presents the topic information to the user.
  • the presentation manner in which the topic information is presented to the user includes, but is not limited to: being presented by the audio device; being presented by the video device; being presented by the wearable device.
  • the audio device is, for example, a smart headset, a smart headset, etc.
  • the video device is, for example, virtual reality (VR) glasses, smart glasses capable of freely switching between ordinary glasses functions and projection functions; wearable devices include smart watches, smart devices Helmets, smart clothes, etc.
  • VR virtual reality
  • the present invention can be implemented in software and/or a combination of software and hardware.
  • the various devices of the present invention can be implemented using an application specific integrated circuit (ASIC) or any other similar hardware device.
  • the software program of the present invention may be executed by a processor to implement the steps or functions described above.
  • the software program (including related data structures) of the present invention can be stored in a computer readable recording medium such as a RAM memory, a magnetic or optical drive or a floppy disk and the like.
  • some of the steps or functions of the present invention may be implemented in hardware, for example, as a circuit that cooperates with a processor to perform various steps or functions.
  • An active search method based on speech recognition comprising the following steps:
  • step c comprises:
  • the voice information is translated into a natural language text based on the recognition of the voice information
  • the populated theme template mapping is expanded to a corresponding search structure.
  • step d comprises:
  • step e comprises:
  • an active matching search is performed, and corresponding topic information is acquired and presented to the user.
  • step e comprises:
  • search device based on speech recognition, wherein the search device comprises:
  • a identifying device configured to identify the received voice information, and obtain a voice feature vector of the voice information
  • a determining device configured to determine whether the voice feature vector matches a reference voice feature vector preset by the user, and obtain a corresponding determination result
  • a generating device configured to generate a subject domain corresponding to the voice information based on the identification of the voice information if the determination result is a mismatch
  • mapping device configured to expand the topic domain mapping into a corresponding search structure
  • a rendering device configured to perform an active matching search according to the search structure, acquire corresponding topic information, and present the information to the user.
  • the voice information is translated into a natural language text based on the recognition of the voice information
  • the subject domain comprises an information domain, an information intent, and an information intent object
  • the mapping device is configured to:
  • the populated theme template mapping is expanded to a corresponding search structure.
  • mapping device is for:
  • an active matching search is performed, and corresponding topic information is acquired and presented to the user.
  • An obtaining unit configured to perform an active matching search according to the search structure, and obtain corresponding candidate information
  • An integration unit configured to generate the topic information based on information extraction and integration of the candidate information
  • a presentation unit for presenting the topic information to the user.
  • a computer readable storage medium storing computer code, the method of any one of clauses 1 to 7 being executed when the computer code is executed.
  • a computer program product when the computer program product is executed by a computer device, the method of any one of clauses 1 to 7 being performed.
  • a computer device comprising a memory and a processor, the memory storing computer code, the processor being configured to execute the computer code to perform any of clauses 1 to 7 Said method.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Artificial Intelligence (AREA)
  • Human Computer Interaction (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种基于语音识别的主动型搜索方法和装置，该方法包括：识别接收的语音信息，获得语音信息的语音特征向量（S301）；判断语音特征向量与用户预设的基准语音特征向量是否匹配，获得对应的判断结果（S302）；若判断结果为不匹配，基于对语音信息的识别，生成与语音信息对应的主题域（S303）；将主题域映射扩展为对应的搜索结构（S304）；根据搜索结构进行主动匹配搜索，获取对应的主题信息并呈现给用户（S305）。将接收的语音信息的语音特征向量与基准语音特征向量是否匹配，作为是否启动搜索功能的条件，实时向用户反馈相关信息或摘要的功能，使用户能够及时、快速获取未知信息，解决沟通中或者其他语言环境中信息不对称的问题。

Description

一种基于语音识别的主动型搜索方法和装置
相关申请的交叉引用
本申请享有2016年12月01日提交的专利申请号为201611091688.4、名称为“一种基于语音识别的主动型搜索方法和装置”的中国专利申请的优先权，该在先申请的内容以引用方式合并于此。”
技术领域
本发明涉及计算机技术领域，尤其涉及一种基于语音识别的主动型搜索技术。
背景技术
在人类生活中，沟通是最基本的社会活动，但经常出现沟通不畅的场景，例如，在社交场合，一方聊起了音乐，另一方由于对该领域不甚了解而略显尴尬接不下去；在业务洽谈时，业务方提出了新概念和方法，沟通受阻；在专家咨询场合，专家讲述了众多的理念和意见，听众发现跟不上了，进一步的沟通无法展开。综上可以发现，沟通受阻将会影响沟通效率，也可能会导致商业、人脉建立等机会的流失。究其根本原因在于人与人之间由于信息广度和深度的不对等(即信息不对称)，以及信息获取滞后造成的。
现有的信息获取主要依赖于搜索引擎，通过搜索引擎获得相关的信息，这种方式很大程度上解决了人们从海量信息中获取信息的难题，但应用于沟通时存在以下问题：1)搜索得到的信息相关度依赖于用户输入的关键词。当今的搜索引擎大多是以关键词为索引的方式去查找信息，当用户无法构建合适的关键词时，信息的相关度将急剧下降；2)信息获取的滞后性，用户在构建关键词之后，将关键词输入搜索引擎得到相关信息，这一系列过程相对于沟通存在滞后性。
因此，如何提供一种基于语音识别的主动型搜索技术，帮助用户及时获取未知信息，成为本领域技术人员亟需解决的技术问题之一。
发明内容
本发明的目的是提供一种基于语音识别的主动型搜索方法和装置。
根据本发明的一个方面，提供一种基于语音识别的主动型搜索方法，其中，该方法包括以下步骤：
a.识别接收的语音信息，获得所述语音信息的语音特征向量；
b.判断所述语音特征向量与用户预设的基准语音特征向量是否匹配，获得对应的判断结果；
c.若所述判断结果为不匹配，基于对所述语音信息的识别，生成与所述语音信息对应的主题域
d.将所述主题域映射扩展为对应的搜索结构；
e.根据所述搜索结构进行主动匹配搜索，获取对应的主题信息并呈现给所述用户。
优选地，所述步骤c包括：
若所述判断结果为不匹配，基于对所述语音信息的识别，将所述语音信息翻译成自然语言文本；
对所述自然语言文本进行自然语义解析，生成与所述语音信息对应的主题域。
优选地，所述主题域包括以下至少任一项：
信息领域；
信息意图；
信息意图对象。
更优选地，其中，所述主题域包括信息领域、信息意图和信息意图对象，其中，所述步骤d包括：
根据所述主题域的信息领域、信息意图确定对应的主题模板；
根据所述主题模板，结合所述信息意图对象生成外延扩展关键词列表，将所述主题域和所述外延扩展关键词列表填充至所述主题模板；
将填充后的所述主题模板映射扩展为对应的搜索结构。
优选地，所述步骤d包括：
根据所述主动匹配搜索所对应的搜索引擎的属性，将所述主题域映射扩展为所述搜索结构。
优选地，所述步骤e包括：
根据所述搜索结构，并结合认知计算，进行主动匹配搜索，获取对应的主题信息并呈现给所述用户。
优选地，所述步骤e包括：
根据所述搜索结构进行主动匹配搜索，获取对应的候选信息；
基于对所述候选信息的信息提取整合，生成所述主题信息；
将所述主题信息呈现给所述用户。
根据本发明的另一个方面，还提供了一种基于语音识别的主动型搜索装置，其中，该搜索装置包括：
识别装置，用于识别接收的语音信息，获得所述语音信息的语音特征向量；
判断装置，用于判断所述语音特征向量与用户预设的基准语音特征向量是否匹配，获得对应的判断结果；
生成装置，用于若所述判断结果为不匹配，基于对所述语音信息的识别，生成与所述语音信息对应的主题域；
映射装置，用于将所述主题域映射扩展为对应的搜索结构；
呈现装置，用于根据所述搜索结构进行主动匹配搜索，获取对应的主题信息并呈现给所述用户。
优选地，所述生成装置用于：
若所述判断结果为不匹配，基于对所述语音信息的识别，将所述语音信息翻译成自然语言文本；
对所述自然语言文本进行自然语义解析，生成与所述语音信息对应 的主题域。
优选地，所述主题域包括以下至少任一项：
信息领域；
信息意图；
信息意图对象。
优选地，所述主题域包括信息领域、信息意图和信息意图对象，其中，所述映射装置用于：
根据所述主题域的信息领域、信息意图确定对应的主题模板；
根据所述主题模板，结合所述信息意图对象生成外延扩展关键词列表，将所述主题域和所述外延扩展关键词列表填充至所述主题模板；
将填充后的所述主题模板映射扩展为对应的搜索结构。
优选地，所述映射装置用于根据所述主动匹配搜索所对应的搜索引擎的属性，将所述主题域映射扩展为所述搜索结构。
优选地，所述呈现装置用于：
根据所述搜索结构，并结合认知计算，进行主动匹配搜索，获取对应的主题信息并呈现给所述用户。
优选地，所述呈现装置包括：
获取单元，用于根据所述搜索结构进行主动匹配搜索，获取对应的候选信息；
整合单元，用于基于对所述候选信息的信息提取整合，生成所述主题信息；
呈现单元，用于将所述主题信息呈现给所述用户。
根据本发明的又一个方面，还提供了一种计算机可读存储介质，所述计算机可读存储介质存储有计算机代码，当所述计算机代码被执行时，如上任一项所述的方法被执行。
根据本发明的再一个方面，还提供了一种计算机程序产品，当所述计算机程序产品被计算机设备执行时，如上任一项所述的方法被执行。
根据本发明的再一个方面，还提供了一种计算机设备，所述计算机设备包括存储器和处理器，所述存储器中存储有计算机代码，所述处理器被配置来通过执行所述计算机代码以执行如上任一项所述的方法。
与现有技术相比，本发明具有以下优点：
本发明通过持续主动获取语音信息，并且判断所述语音信息的语音特征向量与预先设置的基准语音特征向量是否匹配，即，判断所述语音信息的发出者是否为预定用户，将判断的结果作为是否启动搜索功能的条件，然后通过搜索操作，获得与所述语音信息相对应的信息的主题信息，并呈现给用户，帮助用户及时、快速获取未知信息，解决沟通中或者其他语言环境中信息不对称的问题。
进一步地，本发明通过采用自然语义分析技术，更加准确地获得语音信息中的关键词，并且根据所述语音信息的语义对所述关键词进行扩展或者删减，提高搜索结果的准确性和命中率，提升用户的使用体验。
进一步地，本发明基于对语音信息的识别，生成与该语音信息对应的主题域，主题域的主题域结构中的信息领域、信息意图和信息意图对象补充并修正主题域中的关键词，提高搜索的准确率，减少搜索过程中的噪音，提升了用户的使用体验。
附图说明
通过阅读参照以下附图所作的对非限制性实施例所作的详细描述，本发明的其它特征、目的和优点将会变得更明显：
图1示出根据本发明一个方面的一种基于语音识别的主动型搜索装置的结构示意图；
图2示出根据本发明一个实施例的使用百度搜索引擎时的搜索结构示意图；
图3示出根据本发明另一个方面的一种基于语音识别的主动型搜索方法的流程示意图。
附图中相同或相似的附图标记代表相同或相似的部件。
具体实施方式
在更加详细地讨论示例性实施例之前应当提到的是，一些示例性实施例被描述成作为流程图描绘的处理或方法。虽然流程图将各项操作描述成顺序的处理，但是其中的许多操作可以被并行地、并发地或者同时实施。此外，各项操作的顺序可以被重新安排。当其操作完成时所述处理可以被终止，但是还可以具有未包括在附图中的附加步骤。所述处理可以对应于方法、函数、规程、子例程、子程序等等。
在上下文中所称“计算机设备”，也称为“电脑”，是指可以通过运行预定程序或指令来执行数值计算和/或逻辑计算等预定处理过程的智能电子设备，其可以包括处理器与存储器，由处理器执行在存储器中预存的存续指令来执行预定处理过程，或是由ASIC、FPGA、DSP等硬件执行预定处理过程，或是由上述二者组合来实现。计算机设备包括但不限于服务器、个人电脑、笔记本电脑、平板电脑、智能手机等。
需要说明的是，所述用户设备、网络设备和网络等仅为举例，其他现有的或今后可能出现的计算机设备或网络如可适用于本发明，也应包含在本发明保护范围以内，并以引用方式包含于此。
后面所讨论的方法(其中一些通过流程图示出)可以通过硬件、软件、固件、中间件、微代码、硬件描述语言或者其任意组合来实施。当用软件、固件、中间件或微代码来实施时，用以实施必要任务的程序代码或代码段可以被存储在机器或计算机可读介质(比如存储介质)中。(一个或多个)处理器可以实施必要的任务。
这里所公开的具体结构和功能细节仅仅是代表性的，并且是用于描述本发明的示例性实施例的目的。但是本发明可以通过许多替换形式来具体实现，并且不应当被解释成仅仅受限于这里所阐述的实施例。
应当理解的是，虽然在这里可能使用了术语“第一”、“第二”等等来描述各个单元，但是这些单元不应当受这些术语限制。使用这些术语仅仅是为了将一个单元与另一个单元进行区分。举例来说，在不背离示例 性实施例的范围的情况下，第一单元可以被称为第二单元，并且类似地第二单元可以被称为第一单元。这里所使用的术语“和/或”包括其中一个或更多所列出的相关联项目的任意和所有组合。
应当理解的是，当一个单元被称为“连接”或“耦合”到另一单元时，其可以直接连接或耦合到所述另一单元，或者可以存在中间单元。与此相对，当一个单元被称为“直接连接”或“直接耦合”到另一单元时，则不存在中间单元。应当按照类似的方式来解释被用于描述单元之间的关系的其他词语(例如“处于...之间”相比于“直接处于...之间”，“与...邻近”相比于“与...直接邻近”等等)。
这里所使用的术语仅仅是为了描述具体实施例而不意图限制示例性实施例。除非上下文明确地另有所指，否则这里所使用的单数形式“一个”、“一项”还意图包括复数。还应当理解的是，这里所使用的术语“包括”和/或“包含”规定所陈述的特征、整数、步骤、操作、单元和/或组件的存在，而不排除存在或添加一个或更多其他特征、整数、步骤、操作、单元、组件和/或其组合。
还应当提到的是，在一些替换实现方式中，所提到的功能/动作可以按照不同于附图中标示的顺序发生。举例来说，取决于所涉及的功能/动作，相继示出的两幅图实际上可以基本上同时执行或者有时可以按照相反的顺序来执行。
下面结合附图对本发明作进一步详细描述。
图1示出根据本发明一个方面的一种基于语音识别的主动型搜索装置的结构示意图。搜索装置1包括：识别装置101、判断装置102、生成装置103、映射装置104和呈现装置105。
以下以搜索装置1位于网络设备中为例进行详细描述，该网络设备与用户设备进行交互，接收用户设备收集的语音信息，识别接收的语音信息，获得所述语音信息的语音特征向量，判断所述语音特征向量与用户预设的基准语音特征向量是否匹配，若所述判断结果为不匹配，基于对所述语音信息的识别，生成与所述语音信息对应的主题域，将所述 主题域映射扩展为对应的搜索结构；根据所述搜索结构进行主动匹配搜索，获取对应的主题信息并返回给该用户设备，从而呈现给所述用户。
在此，所述用户设备包括但不限于个人电脑、便携式电脑、平板电脑、智能手机、PDA、虚拟现实(Virtual Reality，VR)眼镜、虚拟现实头盔、智能耳机等。用户设备通过其上的语音收集模块收集语音信息，并将所述收集的语音信息通过网络发送给搜索装置1，以便搜索装置1接收所述收集的语音信息后进行语音识别，将语音识别的结果作为是否启动搜索的条件，通过搜索获取对应的主题信息并呈现给用户。
其中，识别装置101识别接收的语音信息，获得所述语音信息的语音特征向量。具体地，用户设备通过其上的特定模块，例如语音收集模块，持续收集语音信息，然后通过网络发送给搜索装置1，或者搜索装置1定期地通过约定的通信方式，如http、https等通信协议，直接从用户设备获取用户设备收集的所述语音信息，搜索装置1获取所述语音信息后，首先通过滤波装置过滤掉所述语音信息中的噪音信息，识别装置101对接收的过滤后的所述语音信息进行预处理，例如通过采样、量化、预加重、加窗的预处理过程，然后例如通过使用Mel频率倒谱参数(Mel Frequency Cepstral Coefficient，MFCC)提取法、线性预测编码(linear predictive coding，LPC)提取法、高频域强调提取法(high frequency domain emphasis extraction method)或窗口函数提取法来提取语音特征向量。其中，所述语音特征向量是一种用于识别说话人(声纹识别)的语音特征参数，通过判断两个语音特征向量是否匹配，便可大体确定所述两个语音特征向量对应的语音信息是否出自同一用户。其中，Mel频率是基于人耳听觉特性提出来的，它与Hz频率成非线性对应关系。
例如，在社交场合，如果用户A聊起了音乐，用户B对该话题不了解，则B可以通过搜索装置1快速获取与A所发出的语音信息相对应的音乐相关信息，具体地，在交流前，B通过点击用户设备上与所述搜索装置1对应的应用界面上相应的功能区域，或者其他方式激 活搜索功能，在A发出语音信息时，用户设备通过其上的语音接收模块收集语音信息，然后通过网络将其发送至搜索装置1，搜索装置1获取所述语音信息后，识别装置101对接收的所述语音信息进行预处理，例如，对每一帧进行快速傅里叶变换(Fast Fourier Transformation，FFT)，求频谱，进而求得幅度谱，对幅度谱加Mel滤波器组，对所有的滤波器输出做对数运算(Logarithm)，再进一步做离散余弦变换(Discrete Cosine Transform，DCT)可得语音特征参数MMFC，进而得出对应的语音特征向量。
本领域技术人员应能理解，上述获得所述语音特征向量的方法仅为举例，现有的或者今后可能出现的获得语音特征向量的方法如可适用于本发明，都应包含在本发明的保护范围内，并在此以引用的方式包含于此。
判断装置102判断所述语音特征向量与用户预设的基准语音特征向量是否匹配，获得对应的判断结果。具体地，用户预先在搜索装置1中设置基准语音特征向量，例如通过预先接收预定用户的语音信息，并通过识别装置101识别所述语音信息，获得所述语音信息的语音特征向量，然后将该语音特征向量存储为基准语音特征向量，用于判断搜索装置1识别的语音信息的语音特征向量是否与所述基准语音特征向量匹配，即搜索装置1识别的语音信息的发出者是否为预定用户，获得对应的判断结果，进而将所述判断结果作为是否启动搜索的条件。其中，所述预定用户可以根据具体的使用情况确定。
例如，用户C对所述搜索装置1发出语音信息，识别装置101识别该语音信息，并获得该语音信息的语音特征向量，将该语音特征向量作为基准语音特征向量，通过该基准语音特征向量，可以判断搜索装置1后续接收的语音信息是否出自用户C，然后根据判断结果，做出相应的响应。例如，1)专家咨询场合，专家讲述了众多的理念和意见，搜索装置1接收该场合中的语音信息，并识别接收的所述语音信息，获得所述语音信息的语音特征向量x，判断装置102将该语音 特征向量x与所述基准语音特征向量进行比较，判断两者是否匹配，获得对应的判断结果，如果判断结果为匹配，则所述语音信息为用户C自己发出的语音信息，则不作进一步处理；如果判断结果为不匹配，则所述语音信息为非用户C发出的语音信息，则搜索装置1进行后续搜索操作，并获得对应的搜索结果提供给用户C，用户C便可以根据所述搜索结果理解专家讲述的理念和意见，并与该专家进行有效的沟通和咨询，提高用户C与专家沟通的效率和质量；2)用户C在相对嘈杂的环境中使用所述搜索装置1，进行基于用户C发出的语音信息的语音搜索时，搜索装置1接收该场合中的语音信息，并识别接收的所述语音信息，获得所述语音信息的语音特征向量y，判断装置102将该语音特征向量y与所述基准语音特征向量进行比较，判断两者是否匹配，获得对应的判断结果，若判断结果为匹配，则所述语音信息为用户C自己发出的语音信息，搜索装置1，进行后续搜索操作，并获得对应的搜索结果提供给用户C；若判断结果为不匹配，则不作进一步处理。
若所述判断结果为不匹配，生成装置103基于对所述语音信息的识别，生成与所述语音信息对应的主题域。具体地，若用户想搜索除某一预定用户(包括自己)以外的其他用户所发出的语音信息的对应信息，将该预定用户所发出的语音信息的语音特征向量作为基准语音特征向量，若所述判断结果为不匹配，则所述接收的语音信息的发出者不是预定用户，即，所述接收的语音信息的发出者为除预定用户以外的其他用户，所述语音特征向量符合用户进行搜索的条件，生成装置103基于识别装置101识别的所述语音信息，通过特征提取技术和模式匹配技术识别出特定的声学模型，通过所述声学模型训练形成一定的语言模型，在声学模型和语言模型组成的空间进行快速寻优，将所述语音信息转变成文本信息，然后对所述文本信息进行语义分析，例如采用全切分的分词技术对所述文本信息进行分词，得到对应的关键词，然后按照预设的特征模型进行特征值计算，匹配最优主题域，并将所述关键词填充至主题 域；其中，所述主题域包括但不限于：信息领域(domain)；信息意图(intent)；信息意图对象(object)。其中，主题域(可看成特征结构)可以通过以下方式获得：一是人为标注设置特征参数；二是抓取大量的基础数据，进行机器学习，生成特征参数；第二种方法有时也结合人工监管纠正的方式，以达到更好地效果。其中，同一类型的数据或资源，以及围绕数据或资源提供的服务称为一个信息领域(domain)。信息领域数据一般是结构化的表格数据，有一个主键(主属性)，信息领域一般以名词命名；信息意图(intent)代表用户对信息领域数据的操作，如查询、查询某一个属性的值、预订、拨打等，一般以动词命名；信息意图对象(object)描述意图的数据，也是实现意图所需要的参数，具体表现为通过分词技术得出的关于所述文本信息的主关键词。上述信息领域、信息意图和信息意图对象也可看作该主题域的主题域结构。
例如，使用搜索装置1的用户D将所发出的语音信息的语音特征向量作为基准语音特征向量，用户D和朋友E一起讨论回老家的事宜，D说：我已经定好回青岛的火车票，朋友E说：我最近也要回青岛，不知道现在还能不能订到北京到青岛的火车票。识别装置101分别识别接收的上述语音信息，获得所述语音信息分别对应的语音特征向量，判断装置102分别对识别装置101识别的语音信息进行判断，判断用户D发出的语音信息的语音特征向量与所述基准语音特征向量匹配，则不作进一步处理，判断装置102判断出用户E发出的语音信息“我最近也要回青岛，不知道现在还能不能订到北京到青岛的火车票”的语音特征向量与基准语音特征向量不匹配时，即所述语音信息的发出者不是用户时，生成装置103将所述语音信息转变成文字信息，并对所述文字信息通过语义分析进行分词，对分词后的文本按照预设的特征模型进行特征值计算，匹配最优主题域，并将文本内容填充至主题域，得到简单的主题域结构：
领域：火车；
意图：预定；
意图对象：
{
始发站：北京；
到达站：青岛；
}
本领域技术人员应能理解，上述主题域仅为举例，现有的或者今后可能出现的主题域如可适用于本发明都应包含在本发明的保护范围内，并在此以引用的方式包含于此。
映射装置104将所述主题域映射扩展为对应的搜索结构。
具体地，映射装置104将被填充后的主题域中对应的关键词根据预设的规则进行映射，例如，通过某一模板进行映射，并进行相应扩展，例如，根据所述模板的信息扩展生成一部分关键词，补充主题域中的关键词信息，得到对应的完整的搜索结构。
呈现装置105根据所述搜索结构进行主动匹配搜索，获取对应的主题信息并呈现给所述用户。具体地，呈现装置105根据映射装置104所确定的搜索结果进行主动匹配搜索，获得大量相关搜索结果，呈现装置105对所述搜索结果进行判断，若所述搜索结果为主题化信息时，直接呈现给用户；若所述搜索结果为非主题化信息时，呈现装置105进行主题化的信息提取，通过智能机的屏幕或具有接收语音消息的穿戴式耳机，向用户呈现经过提炼聚集之后的主要信息，帮助用户快速获取所述语音信息涉及的具体内容，及时准确的获取未知信息。
在此，呈现装置105获取对应的主题信息的方法例如，1)对于单个文档，采用自动文摘技术，例如，基于理解的自动文摘技术或者基于结构的自动文摘技术对获得到的所述搜索结果进行概括、提取，获得简洁、连贯的主题信息；2)对于多个文档，可以将多个文档通过聚类方法，将内容相似的搜索结果进行分组、分类，然后再采用文摘技术提取主题信息，其中，所述聚类方法需要计算每个搜索结果所对应的文档的特征值，然后根据特征值聚合到某个类别下。其中，计算每个文档的特 征值所需的特征参数和种子文档特征值是预先通过训练模型训练好的。
在此，搜索装置1为用户提供基准语音特征向量预设功能，用户通过该功能预先设置基准语音特征向量，然后判断接收的语音信息的语音特征向量与基准语音特征向量是否匹配，将判断结果作为是否启动搜索功能的条件，并将搜索获得的信息进行主题化组织，获得主题化信息并呈现给用户，实现了在沟通中或者复杂的语音环境下持续主动接收语音信息，并自主判断所述语音信息，理解并搜索获取与所述语音信息相关的信息，实时向用户反馈相关信息或摘要，使用户能够及时、快速获取未知信息，解决交流中信息不对称的问题。例如，1)在接受咨询服务时，将基准语音特征向量设定为用户自身语音信息的基准语音特征向量，搜索装置1可以持续主动接收语音信息，若所述接收的语音信息的发出者为该用户，则不作进一步处理，若发出者为其他人，则生成所述语音信息对应的主题域，并根据所述主题域进行搜索，为用户获取与接收的所述语音信息对应的背景主题信息、具体知识要点或者技术框架等信息，及时为用户提供准确、全面专业的解答；2)假设用户正在接受医疗服务，搜索装置1将向该用户实时提供医生所述内容的相关解释和原理，让该用户有基础能够持续发问，获得想要的信息，一定程度上缓解医患之间的信息不对称；3)在洽谈业务时，对于客户所提的问题或概念，用户通过搜索装置1可以更专业和全面的回答问题，有利于下一步合作的展开；4)日常生活中，利用搜索装置1实时提供的信息，用户在与其他人交谈是可以涉及多领域多主题，提高双方沟通的有效性和扩展性。进一步地，搜索装置1将把信息普惠性提升到一个新的层级，在使用上将会给那些不懂如何获取信息的人员提供一个全新的获取方式。
优选地，所述生成装置103用于：1)若所述判断结果为不匹配，基于对所述语音信息的识别，将所述语音信息翻译成自然语言文本；2)对所述自然语言文本进行自然语义解析，生成与所述语音信息对应的主题域。具体地，若接收的语音信息的语音特征向量与用户预设的基准语 音特征向量不匹配，即所述接收的语音信息的发出者不是预定用户，生成装置103基于对所述语音信息的识别，将所述语音信息进行模数转换，得出所述语音信息的音频区间，提取出该音频区间的音频数据的特征量，基于该特征向量将所述语音信息识别为对应的自然语言文本；然后生成装置103对所述自然语言文本进行自然语义解析，获得对应的解析结果，根据所述解析结果生成与所述语音信息对应的主题域。
在此，生成装置103通过采用自然语义分析技术，更加准确的获得语音信息中的关键词，并且根据所述语音信息的语义对所述关键词进行扩展或者删减，提高搜索结果的准确性和命中率，提升用户的使用体验。
优选地，所述主题域包括信息领域、信息意图和信息意图对象，其中，所述映射装置104根据所述主题域的信息领域、信息意图确定对应的主题模板；根据所述主题模板，结合所述信息意图对象生成外延扩展关键词列表，将所述主题域和所述外延扩展关键词列表填充至所述主题模板；将填充后的所述主题模板映射扩展为对应的搜索结构。
具体地，当主题域的主题域结构包括：信息领域、信息意图和信息意图对象时，上述将语音信息转化成的文本内容填充至主题域之后，主题域的信息领域、信息意图和信息意图对象都各自有对应的关键词。映射装置104根据主题域中的信息领域、信息意图，通过模式匹配在主题模板库中获取对应的主题模板，并且信息意图对象(主关键词)结合主题模板中预设的规则生成外延扩展关键词列表，将主题域的内容和所述外延扩展关键词列表分别填充至主题模板对应位置，其中，所述外延扩展关键词可以是加关键词，也可以是减关键词，将所述外延扩展关键词填充至主题模板的对应位置中，以使主题模板各项内容更加完整，提高搜索的命中率。其中，所述信息意图对象(主关键词)和外延扩展关键词共同组成主题模板以及对应的搜索结构的各项搜索关键词。其中，所述主题模板中预设有搜索模式，所述搜索模式例如为搜索指令，每一个搜索指令预设的特定搜索结构(其中，特定搜索结构为主题模板的搜索结构，与搜索引擎对应的搜索结构不同)等，所述主题模板的内容映射 扩展到预设的搜索引擎对应的搜索结构的某些属性中。其中，所述主题模板库为预先设置用于存储主题模板的数据库，例如可以通过采集大量的语音信息样本，进行分析处理，得出大量主题模板，存入所述主题模板库中。
在此，所述扩展关键词使搜索关键词更加丰富，更容易得到用户想要的结果，并且同时能减少搜索关键词中的噪音信息，减少不必要的搜索，使得关键词更加精确。根据主题域获取对应的主题模板，将主题域内容和对应于该主题模板生成的外延扩展关键词填充至主题模板中，将主题模板直接映射成搜索引擎对应的搜索结构，提高了搜索关键词的准确度，减小了搜索引擎的压力，提高了搜索的命中率，为用户呈现更加精确的匹配结果，增加了用户实时快速获取信息的能力。
例如，接前述例子，识别装置101识别的语音信息为：“我最近也要回青岛，不知道现在还能不能订到北京到青岛的火车票”，映射装置104根据生成装置103生成的主题域中的信息领域(火车)、信息意图(预定)，通过模式匹配在主题模板库中获取对应的主题模板，将主题域的内容对应填充至主题模板中，并且信息意图对象(主关键词：北京、青岛)结合主题模板中预设的规则例如主题模板中预设各种票的日期限制是一周以内，并且根据信息领域为火车这一信息，主题模板内置排除飞机、大巴、船等交通方式，因此生成外延扩展关键词列表：日期：一周以内；交通方式：-飞机、-大巴、-船，其中，“一周以内”为增加的关键词，“飞机、大巴、船”为缩减的关键词，将主题域内容和外延扩展关键词对应填充至获取主题模板：
主题模板名：列车&查询
搜索指令：
出发地：{北京}；
目的地：{青岛}；
日期：{可选，若未设置，可选最近一周}
优先类别：{可选}
排除的交通方式：{飞机、大巴、船}
若搜索装置1采用百度搜索引擎进行搜索，则将上述主题模板的内容映射至与百度搜索引擎对应的搜索结构中，例如，将上述主题模板的内容映射至与图2示出的搜索结构中，进行主动匹配搜索，获取对应的主体信息呈现给用户。图2示出根据本发明一个实施例的使用百度搜索引擎时的搜索结构的框架。
优选地，所述映射装置104根据所述主动匹配搜索所对应的搜索引擎的属性，将所述主题域映射扩展为所述搜索结构。具体地，映射装置104根据所述主题域的所述信息领域、所述信息意图、所述信息意图对象或三者中的任意组合构建搜索指令，然后基于所述主动匹配搜索所对应的搜索引擎属性的不同，生成对应的搜索结构。例如，当用户使用百度搜索引擎进行搜索时，根据对接收的用户语音信息的识别，生成的对应于主题域的搜索结构框架如图2所示，根据接收的语音信息，图2的输入条中会自动嵌入与所述语音信息对应的搜索结构中的关键词，进行搜索操作。
优选地，所述呈现装置105根据所述主题域所对应的搜索结构，并结合认知计算，进行主动匹配搜索，获取对应的主题信息并呈现给所述用户。具体地，呈现装置104根据所述主题域所对应的搜索结构，结合认知计算，匹配满足所述搜索结构的信息，获取所述信息的主题信息后呈现给用户，例如，学习并分析组合所述搜索结构中的关键词信息，根据所述关键词信息感知用户的需求，以为用户获得相关度高的信息，并将所述搜索结果进行主题化组织，获取对应的主题信息，并呈现给所述用户，提高搜索装置1进行自动匹配搜索的智能性。其中，所述认知计算通过将简单的单元计算通过层级累加，将无结构的信息进行计算，实现对人脑思考方式和认知的模拟。
优选地，所述呈现装置105包括：获取单元1051(未示出)、整合单元1052(未示出)和呈现单元1053(未示出).
获取单元1051根据所述主题域所对应的搜索结构进行主动匹配搜 索，获取对应的候选信息。具体地，获取单元1051根据所述搜索结构进行自动匹配搜索，得到大量相关的搜索结果，呈现装置105的显示界面有限，而且用户获取信息的时间也是有限的，因此需要将所述大量相关的搜索结果作为候选信息，对所述候选信息进行提炼处理之后呈现给用户。
整合单元1052基于对所述候选信息的信息提取整合，生成所述主题信息。具体地，预先设置一资源和规则库，整合单元1052根据所述搜索结构确定主题，然后通过所述资源和规则库中存储的资源和规则，从所述搜索结果的信息中提取出涉及所述主题的描述向量，基于所述描述向量，生成涉及该主题的全局性信息作为主题信息，以便于用户通过所述全局性信息快速获取相关知识。例如，用户查询“荨麻疹”的情况，搜索装置1进行搜索之后会有大量的搜索结果，受呈现装置105的限制，不能一次性将所有结果呈现给用户，而且为了用户能够快速的获取主要知识，整合单元1052将荨麻疹总结为以下主题信息：风团；风疹团；皮肤病；局部或全身性皮肤上突然成片出现红色肿块；发病快，消失快；剧痒；过敏。用户再看到上述全局性信息之后便可以大体了解荨麻疹这一疾病。
呈现单元1053将所述主题信息呈现给所述用户。具体地，将主题信息呈现给所述用户的呈现方式包括但不限于：通过音频设备呈现；通过视频设备呈现；通过穿戴式设备呈现。所述音频设备例如为智能耳麦，智能耳机等；所述视频设备例如为虚拟现实(VR)眼镜，能够在普通眼镜功能和放映功能中自由切换的智能眼镜等；穿戴式设备包括智能手表、智能头盔、智能服饰等。
图3示出根据本发明另一个方面的一种基于语音识别的主动型搜索方法的流程示意图。
其中，在步骤S301中，搜索装置1识别接收的语音信息，获得所述语音信息的语音特征向量。具体地，用户设备通过其上的特定模块，例如语音收集模块，持续收集语音信息，然后通过网络发送给搜 索装置1，或者搜索装置1定期地通过约定的通信方式，如http、https等通信协议，直接从用户设备获取用户设备收集的所述语音信息，搜索装置1获取所述语音信息后，首先通过滤波装置过滤掉所述语音信息中的噪音信息，在步骤S301中，搜索装置1对接收的过滤后的所述语音信息进行预处理，例如通过采样、量化、预加重、加窗的预处理过程，然后例如通过使用Mel频率倒谱参数(Mel Frequency Cepstral Coefficient，MFCC)提取法、线性预测编码(linear predictive coding，LPC)提取法、高频域强调提取法(high frequency domain emphasis extraction method)或窗口函数提取法来提取语音特征向量。其中，所述语音特征向量是一种用于识别说话人(声纹识别)的语音特征参数，通过判断两个语音特征向量是否匹配，便可大体确定所述两个语音特征向量对应的语音信息是否出自同一用户。其中，Mel频率是基于人耳听觉特性提出来的，它与Hz频率成非线性对应关系。
例如，在社交场合，如果用户A聊起了音乐，用户B对该话题不了解，则B可以通过搜索装置1快速获取与A所发出的语音信息相对应的音乐相关信息，具体地，在交流前，B通过点击用户设备上与所述搜索装置1对应的应用界面上相应的功能区域，或者其他方式激活搜索功能，在A发出语音信息时，用户设备通过其上的语音接收模块收集语音信息，然后通过网络将其发送至搜索装置1，搜索装置1获取所述语音信息后，在步骤S301中，搜索装置1对接收的所述语音信息进行预处理，例如，对每一帧进行快速傅里叶变换(Fast Fourier Transformation，FFT)，求频谱，进而求得幅度谱，对幅度谱加Mel滤波器组，对所有的滤波器输出做对数运算(Logarithm)，再进一步做离散余弦变换(Discrete Cosine Transform，DCT)可得语音特征参数MMFC，进而得出对应的语音特征向量。
本领域技术人员应能理解，上述获得所述语音特征向量的方法仅为举例，现有的或者今后可能出现的获得语音特征向量的方法如可适用于本发明，都应包含在本发明的保护范围内，并在此以引用的方式 包含于此。
在步骤S302中，搜索装置1判断所述语音特征向量与用户预设的基准语音特征向量是否匹配，获得对应的判断结果。具体地，用户预先在搜索装置1中设置基准语音特征向量，例如通过预先接收预定用户的语音信息，并通过搜索装置1在步骤S301中识别所述语音信息，获得所述语音信息的语音特征向量，然后将该语音特征向量存储为基准语音特征向量，用于判断搜索装置1识别的语音信息的语音特征向量是否与所述基准语音特征向量匹配，即搜索装置1识别的语音信息的发出者是否为预定用户，获得对应的判断结果，进而将所述判断结果作为是否启动搜索的条件。其中，所述预定用户可以根据具体的使用情况确定。
例如，用户C对所述搜索装置1发出语音信息，在步骤S301中，搜索装置1识别该语音信息，并获得该语音信息的语音特征向量，将该语音特征向量作为基准语音特征向量，通过该基准语音特征向量，可以判断搜索装置1后续接收的语音信息是否出自用户C，然后根据判断结果，做出相应的响应。例如，1)专家咨询场合，专家讲述了众多的理念和意见，搜索装置1接收该场合中的语音信息，并识别接收的所述语音信息，获得所述语音信息的语音特征向量x，在步骤S302中，搜索装置1将该语音特征向量x与所述基准语音特征向量进行比较，判断两者是否匹配，获得对应的判断结果，如果判断结果为匹配，则所述语音信息为用户C自己发出的语音信息，则不作进一步处理；如果判断结果为不匹配，则所述语音信息为非用户C发出的语音信息，则搜索装置1进行后续搜索操作，并获得对应的搜索结果提供给用户C，用户C便可以根据所述搜索结果理解专家讲述的理念和意见，并与该专家进行有效的沟通和咨询，提高用户C与专家沟通的效率和质量；2)用户C在相对嘈杂的环境中使用所述搜索装置1，进行基于用户C发出的语音信息的语音搜索时，搜索装置1接收该场合中的语音信息，并识别接收的所述语音信息，获得所述语音信息的 语音特征向量y，在步骤S302中，搜索装置1将该语音特征向量y与所述基准语音特征向量进行比较，判断两者是否匹配，获得对应的判断结果，若判断结果为匹配，则所述语音信息为用户C自己发出的语音信息，搜索装置1进行后续搜索操作，并获得对应的搜索结果提供给用户C；若判断结果为不匹配，则不作进一步处理。
若所述判断结果为不匹配，在步骤S303中，搜索装置1基于对所述语音信息的识别，生成与所述语音信息对应的主题域。具体地，若用户想搜索除某一预定用户(包括自己)以外的其他用户所发出的语音信息的对应信息，将该预定用户所发出的语音信息的语音特征向量作为基准语音特征向量，若所述判断结果为不匹配，则所述接收的语音信息的发出者不是预定用户，即，所述接收的语音信息的发出者为除预定用户以外的其他用户，所述语音特征向量符合用户进行搜索的条件，在步骤S301中，搜索装置1基于识别的所述语音信息，通过特征提取技术和模式匹配技术识别出特定的声学模型，通过所述声学模型训练形成一定的语言模型，在声学模型和语言模型组成的空间进行快速寻优，将所述语音信息转变成文本信息，然后对所述文本信息进行语义分析，例如采用全切分的分词技术对所述文本信息进行分词，得到对应的关键词，然后按照预设的特征模型进行特征值计算，匹配最优主题域，并将所述关键词填充至主题域；其中，所述主题域包括但不限于：信息领域(domain)；信息意图(intent)；信息意图对象(object)。其中，主题域(可看成特征结构)可以通过以下方式获得：一是人为标注设置特征参数；二是抓取大量的基础数据，进行机器学习，生成特征参数；第二种方法有时也结合人工监管纠正的方式，以达到更好地效果。其中，同一类型的数据或资源，以及围绕数据或资源提供的服务称为一个信息领域(domain)。信息领域数据一般是结构化的表格数据，有一个主键(主属性)，信息领域一般以名词命名；信息意图(intent)代表用户对信息领域数据的操作，如查询、查询某一个属性的值、预订、拨打等，一般以动词命名；信息意图对象(object)描述意图的数据，也是实现意 图所需要的参数，具体表现为通过分词技术得出的关于所述文本信息的主关键词。上述信息领域、信息意图和信息意图对象也可看作该主题域的主题域结构。
例如，使用搜索装置1的用户D将所发出的语音信息的语音特征向量作为基准语音特征向量，用户D和朋友E一起讨论回老家的事宜，D说：我已经定好回青岛的火车票，朋友E说：我最近也要回青岛，不知道现在还能不能订到北京到青岛的火车票。在步骤S301中，搜索装置1分别识别接收的上述语音信息，获得所述语音信息分别对应的语音特征向量，在步骤S302中，搜索装置1分别对识别的语音信息进行判断，判断用户D发出的语音信息的语音特征向量与所述基准语音特征向量匹配，则不作进一步处理，在步骤S302中，搜索装置1判断出用户E发出的语音信息“我最近也要回青岛，不知道现在还能不能订到北京到青岛的火车票”的语音特征向量与基准语音特征向量不匹配时，即所述语音信息的发出者不是用户时，在步骤S303中，搜索装置1将所述语音信息转变成文字信息，并对所述文字信息通过语义分析进行分词，对分词后的文本按照预设的特征模型进行特征值计算，匹配最优主题域，并将文本内容填充至主题域，得到简单的主题域结构：
领域：火车；
意图：预定；
意图对象：
{
始发站：北京；
到达站：青岛；
}
本领域技术人员应能理解，上述主题域仅为举例，现有的或者今后可能出现的主题域如可适用于本发明都应包含在本发明的保护范围内，并在此以引用的方式包含于此。
在步骤S304中，搜索装置1将所述主题域映射扩展为对应的搜索 结构。
具体地，在步骤S304中，搜索装置1将被填充后的主题域中对应的关键词根据预设的规则进行映射，例如，通过某一模板进行映射，并进行相应扩展，例如，根据所述模板的信息扩展生成一部分关键词，补充主题域中的关键词信息，得到对应的完整的搜索结构。
在步骤S305中，搜索装置1根据所述搜索结构进行主动匹配搜索，获取对应的主题信息并呈现给所述用户。具体地，在步骤S305中，搜索装置1根据其在步骤S304中所确定的搜索结果进行主动匹配搜索，获得大量相关搜索结果，在步骤S305中，搜索装置1对所述搜索结果进行判断，若所述搜索结果为主题化信息时，直接呈现给用户；若所述搜索结果为非主题化信息时，在步骤S305中，搜索装置1进行主题化的信息提取，通过智能机的屏幕或具有接收语音消息的穿戴式耳机，向用户呈现经过提炼聚集之后的主要信息，帮助用户快速获取所述语音信息涉及的具体内容，及时准确的获取未知信息。
在此，在步骤S305中，搜索装置1获取对应的主题信息的方法例如，1)对于单个文档，采用自动文摘技术，例如，基于理解的自动文摘技术或者基于结构的自动文摘技术对获得到的所述搜索结果进行概括、提取，获得简洁、连贯的主题信息；2)对于多个文档，可以将多个文档通过聚类方法，将内容相似的搜索结果进行分组、分类，然后再采用文摘技术提取主题信息，其中，所述聚类方法需要计算每个搜索结果所对应的文档的特征值，然后根据特征值聚合到某个类别下。其中，计算每个文档的特征值所需的特征参数和种子文档特征值是预先通过训练模型训练好的。
在此，搜索装置1为用户提供基准语音特征向量预设功能，用户通过该功能预先设置基准语音特征向量，然后判断接收的语音信息的语音特征向量与基准语音特征向量是否匹配，将判断结果作为是否启动搜索功能的条件，并将搜索获得的信息进行主题化组织，获得主题化信息并呈现给用户，实现了在沟通中或者复杂的语音环境下持续主动接收语音 信息，并自主判断所述语音信息，理解并搜索获取与所述语音信息相关的信息，实时向用户反馈相关信息或摘要，使用户能够及时、快速获取未知信息，解决交流中信息不对称的问题。例如，1)在接受咨询服务时，将基准语音特征向量设定为用户自身语音信息的基准语音特征向量，搜索装置1可以持续主动接收语音信息，若所述接收的语音信息的发出者为该用户，则不作进一步处理，若发出者为其他人，则生成所述语音信息对应的主题域，并根据所述主题域进行搜索，为用户获取与接收的所述语音信息对应的背景主题信息、具体知识要点或者技术框架等信息，及时为用户提供准确、全面专业的解答；2)假设用户正在接受医疗服务，搜索装置1将向该用户实时提供医生所述内容的相关解释和原理，让该用户有基础能够持续发问，获得想要的信息，一定程度上缓解医患之间的信息不对称；3)在洽谈业务时，对于客户所提的问题或概念，用户通过搜索装置1可以更专业和全面的回答问题，有利于下一步合作的展开；4)日常生活中，利用搜索装置1实时提供的信息，用户在与其他人交谈是可以涉及多领域多主题，提高双方沟通的有效性和扩展性。进一步地，搜索装置1将把信息普惠性提升到一个新的层级，在使用上将会给那些不懂如何获取信息的人员提供一个全新的获取方式。
优选地，在步骤S303中，搜索装置1用于：1)若所述判断结果为不匹配，基于对所述语音信息的识别，将所述语音信息翻译成自然语言文本；2)对所述自然语言文本进行自然语义解析，生成与所述语音信息对应的主题域。具体地，若接收的语音信息的语音特征向量与用户预设的基准语音特征向量不匹配，即所述接收的语音信息的发出者不是预定用户，在步骤S303中，搜索装置1基于对所述语音信息的识别，将所述语音信息进行模数转换，得出所述语音信息的音频区间，提取出该音频区间的音频数据的特征量，基于该特征向量将所述语音信息识别为对应的自然语言文本；然后在步骤S303中，搜索装置1对所述自然语言文本进行自然语义解析，获得对应的解析结果，根据所述解析结果 生成与所述语音信息对应的主题域。
在此，在步骤S303中，搜索装置1通过采用自然语义分析技术，更加准确的获得语音信息中的关键词，并且根据所述语音信息的语义对所述关键词进行扩展或者删减，提高搜索结果的准确性和命中率，提升用户的使用体验。
优选地，所述主题域包括信息领域、信息意图和信息意图对象，其中，在步骤S304中，搜索装置1根据所述主题域的信息领域、信息意图确定对应的主题模板；根据所述主题模板，结合所述信息意图对象生成外延扩展关键词列表，将所述主题域和所述外延扩展关键词列表填充至所述主题模板；将填充后的所述主题模板映射扩展为对应的搜索结构。
具体地，当主题域的主题域结构包括：信息领域、信息意图和信息意图对象时，上述将语音信息转化成的文本内容填充至主题域之后，主题域的信息领域、信息意图和信息意图对象都各自有对应的关键词，在步骤S304中，搜索装置1根据主题域中的信息领域、信息意图，通过模式匹配在主题模板库中获取对应的主题模板，并且信息意图对象(主关键词)结合主题模板中预设的规则生成外延扩展关键词列表，将主题域的内容和所述外延扩展关键词列表分别填充至主题模板对应位置，其中，所述外延扩展关键词可以是加关键词，也可以是减关键词，将所述外延扩展关键词填充至主题模板的对应位置中，以使主题模板各项内容更加完整，提高搜索的命中率。其中，所述信息意图对象(主关键词)和外延扩展关键词共同组成主题模板以及对应的搜索结构的各项搜索关键词。其中，所述主题模板中预设有搜索模式，所述搜索模式例如为搜索指令，每一个搜索指令预设的特定搜索结构(其中，特定搜索结构为主题模板的搜索结构，与搜索引擎对应的搜索结构不同)等，所述主题模板的内容映射扩展到预设的搜索引擎对应的搜索结构的某些属性中。其中，所述主题模板库为预先设置用于存储主题模板的数据库，例如可以通过采集大量的语音信息样本，进行分析处理，得出大量主题模 板，存入所述主题模板库中。
在此，所述扩展关键词使搜索关键词更加丰富，更容易得到用户想要的结果，并且同时能减少搜索关键词中的噪音信息，减少不必要的搜索，使得关键词更加精确。根据主题域获取对应的主题模板，将主题域内容和对应于该主题模板生成的外延扩展关键词填充至主题模板中，将主题模板直接映射成搜索引擎对应的搜索结构，提高了搜索关键词的准确度，减小了搜索引擎的压力，提高了搜索的命中率，为用户呈现更加精确的匹配结果，增加了用户实时快速获取信息的能力。
例如，接前述例子，在步骤S301中，搜索装置1识别的语音信息为：“我最近也要回青岛，不知道现在还能不能订到北京到青岛的火车票”，在步骤S304中，搜索装置1根据生成的主题域中的信息领域(火车)、信息意图(预定)，通过模式匹配在主题模板库中获取对应的主题模板，将主题域的内容对应填充至主题模板中，并且信息意图对象(主关键词：北京、青岛)结合主题模板中预设的规则例如主题模板中预设各种票的日期限制是一周以内，并且根据信息领域为火车这一信息，主题模板内置排除飞机、大巴、船等交通方式，因此生成外延扩展关键词列表：日期：一周以内；交通方式：-飞机、-大巴、-船，其中，“一周以内”为增加的关键词，“飞机、大巴、船”为缩减的关键词，将主题域内容和外延扩展关键词对应填充至获取主题模板：
主题模板名：列车&查询
搜索指令：
出发地：{北京}；
目的地：{青岛}；
日期：{可选，若未设置，可选最近一周}
优先类别：{可选}
排除的交通方式：{飞机、大巴、船}
若搜索装置1采用百度搜索引擎进行搜索，则将上述主题模板的内容映射至与百度搜索引擎对应的搜索结构中，例如，将上述主题模板的 内容映射至与图2示出的搜索结构中，进行主动匹配搜索，获取对应的主体信息呈现给用户。图2示出根据本发明一个实施例的使用百度搜索引擎时的搜索结构的框架。
优选地，在步骤S304中，搜索装置1根据所述主动匹配搜索所对应的搜索引擎的属性，将所述主题域映射扩展为所述搜索结构。具体地，在步骤S304中，搜索装置1根据所述主题域的所述信息领域、所述信息意图、所述信息意图对象或三者中的任意组合构建搜索指令，然后基于所述主动匹配搜索所对应的搜索引擎属性的不同，生成对应的搜索结构。例如，当用户使用百度搜索引擎进行搜索时，根据对接收的用户语音信息的识别，生成的对应于主题域的搜索结构框架如图2所示，根据接收的语音信息，图2的输入条中会自动嵌入与所述语音信息对应的搜索结构中的关键词，进行搜索操作。
优选地，在步骤S305中，搜索装置1根据所述主题域所对应的搜索结构，并结合认知计算，进行主动匹配搜索，获取对应的主题信息并呈现给所述用户。具体地，在步骤S304中，搜索装置1根据所述主题域所对应的搜索结构，结合认知计算，匹配满足所述搜索结构的信息，获取所述信息的主题信息后呈现给用户，例如，学习并分析组合所述搜索结构中的关键词信息，根据所述关键词信息感知用户的需求，以为用户获得相关度高的信息，并将所述搜索结果进行主题化组织，获取对应的主题信息，并呈现给所述用户，提高搜索装置1进行自动匹配搜索的智能性。其中，所述认知计算通过将简单的单元计算通过层级累加，将无结构的信息进行计算，实现对人脑思考方式和认知的模拟。
优选地，步骤S305包括：子步骤3051(未示出)、子步骤3052(未示出)和子步骤3053(未示出).
在子步骤S3051中，搜索装置1根据所述主题域所对应的搜索结构进行主动匹配搜索，获取对应的候选信息。具体地，在子步骤S3051中，搜索装置1根据所述搜索结构进行自动匹配搜索，得到大量相关的搜索结果，在步骤S305中，搜索装置1的显示界面有限，而且用户获 取信息的时间也是有限的，因此需要将所述大量相关的搜索结果作为候选信息，对所述候选信息进行提炼处理之后呈现给用户。
在子步骤S3052中，搜索装置1基于对所述候选信息的信息提取整合，生成所述主题信息。具体地，预先设置一资源和规则库，在子步骤S3052中，搜索装置1根据所述搜索结构确定主题，然后通过所述资源和规则库中存储的资源和规则，从所述搜索结果的信息中提取出涉及所述主题的描述向量，基于所述描述向量，生成涉及该主题的全局性信息作为主题信息，以便于用户通过所述全局性信息快速获取相关知识。例如，用户查询“荨麻疹”的情况，搜索装置1进行搜索之后会有大量的搜索结果，受搜索装置1的限制，不能一次性将所有结果呈现给用户，而且为了用户能够快速的获取主要知识，在子步骤S3052中，搜索装置1将荨麻疹总结为以下主题信息：风团；风疹团；皮肤病；局部或全身性皮肤上突然成片出现红色肿块；发病快，消失快；剧痒；过敏。用户再看到上述全局性信息之后便可以大体了解荨麻疹这一疾病。
在子步骤S3053中，搜索装置1将所述主题信息呈现给所述用户。具体地，将主题信息呈现给所述用户的呈现方式包括但不限于：通过音频设备呈现；通过视频设备呈现；通过穿戴式设备呈现。所述音频设备例如为智能耳麦，智能耳机等；所述视频设备例如为虚拟现实(VR)眼镜，能够在普通眼镜功能和放映功能中自由切换的智能眼镜等；穿戴式设备包括智能手表、智能头盔、智能服饰等。
需要注意的是，本发明可在软件和/或软件与硬件的组合体中被实施，例如，本发明的各个装置可采用专用集成电路(ASIC)或任何其他类似硬件设备来实现。在一个实施例中，本发明的软件程序可以通过处理器执行以实现上文所述步骤或功能。同样地，本发明的软件程序(包括相关的数据结构)可以被存储到计算机可读记录介质中，例如，RAM存储器，磁或光驱动器或软磁盘及类似设备。另外，本发明的一些步骤或功能可采用硬件来实现，例如，作为与处理器配合从而执行各个步骤或功能的电路。
对于本领域技术人员而言，显然本发明不限于上述示范性实施例的细节，而且在不背离本发明的精神或基本特征的情况下，能够以其他的具体形式实现本发明。因此，无论从哪一点来看，均应将实施例看作是示范性的，而且是非限制性的，本发明的范围由所附权利要求而不是上述说明限定，因此旨在将落在权利要求的等同要件的含义和范围内的所有变化涵括在本发明内。不应将权利要求中的任何附图标记视为限制所涉及的权利要求。此外，显然“包括”一词不排除其他单元或步骤，单数不排除复数。系统权利要求中陈述的多个单元或装置也可以由一个单元或装置通过软件或者硬件来实现。第一，第二等词语用来表示名称，而并不表示任何特定的顺序。
在权利要求书中规定了各个实施例的各个方面。在下列编号条款中规定了各个实施例的这些和其他方面：
1.一种基于语音识别的主动型搜索方法，其中，该方法包括以下步骤：
a.识别接收的语音信息，获得所述语音信息的语音特征向量；
b.判断所述语音特征向量与用户预设的基准语音特征向量是否匹配，获得对应的判断结果；
c.若所述判断结果为不匹配，基于对所述语音信息的识别，生成与所述语音信息对应的主题域；
d.将所述主题域映射扩展为对应的搜索结构；
e.根据所述搜索结构进行主动匹配搜索，获取对应的主题信息并呈现给所述用户。
2.根据条款1所述的方法，其中，所述步骤c包括：
若所述判断结果为不匹配，基于对所述语音信息的识别，将所述语音信息翻译成自然语言文本；
对所述自然语言文本进行自然语义解析，生成与所述语音信息对应的主题域。
3.根据条款1或2所述的方法，其中，所述主题域包括以下至少 任一项：
信息领域；
信息意图；
信息意图对象。
4.根据条款3所述的方法，其中，所述主题域包括信息领域、信息意图和信息意图对象，其中，所述步骤d包括：
根据所述主题域的信息领域、信息意图确定对应的主题模板；
根据所述主题模板，结合所述信息意图对象生成外延扩展关键词列表，将所述主题域和所述外延扩展关键词列表填充至所述主题模板；
将填充后的所述主题模板映射扩展为对应的搜索结构。
5.根据条款1至4中任一项所述的方法，其中，所述步骤d包括：
根据所述主动匹配搜索所对应的搜索引擎的属性，将所述主题域映射扩展为所述搜索结构。
6.根据条款1至5中任一项所述的方法，其中，所述步骤e包括：
根据所述搜索结构，并结合认知计算，进行主动匹配搜索，获取对应的主题信息并呈现给所述用户。
7.根据条款1至6中任一项所述的方法，其中，所述步骤e包括：
根据所述搜索结构进行主动匹配搜索，获取对应的候选信息；
基于对所述候选信息的信息提取整合，生成所述主题信息；
将所述主题信息呈现给所述用户。
8.一种基于语音识别的主动型搜索装置，其中，该搜索装置包括：
识别装置，用于识别接收的语音信息，获得所述语音信息的语音特征向量；
判断装置，用于判断所述语音特征向量与用户预设的基准语音特征向量是否匹配，获得对应的判断结果；
生成装置，用于若所述判断结果为不匹配，基于对所述语音信息的识别，生成与所述语音信息对应的主题域；
映射装置，用于将所述主题域映射扩展为对应的搜索结构；
呈现装置，用于根据所述搜索结构进行主动匹配搜索，获取对应的主题信息并呈现给所述用户。
9.根据条款8所述的搜索装置，其中，所述生成装置用于：
若所述判断结果为不匹配，基于对所述语音信息的识别，将所述语音信息翻译成自然语言文本；
对所述自然语言文本进行自然语义解析，生成与所述语音信息对应的主题域。
10.根据条款8或9所述的搜索装置，其中，所述主题域包括以下至少任一项：
信息领域；
信息意图；
信息意图对象。
11.根据条款10所述的搜索装置，其中，所述主题域包括信息领域、信息意图和信息意图对象，其中，所述映射装置用于：
根据所述主题域的信息领域、信息意图确定对应的主题模板；
根据所述主题模板，结合所述信息意图对象生成外延扩展关键词列表，将所述主题域和所述外延扩展关键词列表填充至所述主题模板；
将填充后的所述主题模板映射扩展为对应的搜索结构。
12.根据条款8至11中任一项所述的搜索装置，其中，所述映射装置用于：
根据所述主动匹配搜索所对应的搜索引擎的属性，将所述主题域映射扩展为所述搜索结构。
13.根据条款8至12中任一项所述的搜索装置，其中，所述呈现装置用于：
根据所述搜索结构，并结合认知计算，进行主动匹配搜索，获取对应的主题信息并呈现给所述用户。
14.根据条款8至13中任一项所述的搜索装置，其中，所述呈现装置包括：
获取单元，用于根据所述搜索结构进行主动匹配搜索，获取对应的候选信息；
整合单元，用于基于对所述候选信息的信息提取整合，生成所述主题信息；
呈现单元，用于将所述主题信息呈现给所述用户。
15.一种计算机可读存储介质，所述计算机可读存储介质存储有计算机代码，当所述计算机代码被执行时，如条款1至7中任一项所述的方法被执行。
16.一种计算机程序产品，当所述计算机程序产品被计算机设备执行时，如条款1至7中任一项所述的方法被执行。
17.一种计算机设备，所述计算机设备包括存储器和处理器，所述存储器中存储有计算机代码，所述处理器被配置来通过执行所述计算机代码以执行如条款1至7中任一项所述的方法。

Claims (17)

  1. 一种基于语音识别的主动型搜索方法，其中，该方法包括以下步骤：
    a.识别接收的语音信息，获得所述语音信息的语音特征向量；
    b.判断所述语音特征向量与用户预设的基准语音特征向量是否匹配，获得对应的判断结果；
    c.若所述判断结果为不匹配，基于对所述语音信息的识别，生成与所述语音信息对应的主题域；
    d.将所述主题域映射扩展为对应的搜索结构；
    e.根据所述搜索结构进行主动匹配搜索，获取对应的主题信息并呈现给所述用户。
  2. 根据权利要求1所述的方法，其中，所述步骤c包括：
    若所述判断结果为不匹配，基于对所述语音信息的识别，将所述语音信息翻译成自然语言文本；
    对所述自然语言文本进行自然语义解析，生成与所述语音信息对应的主题域。
  3. 根据权利要求1或2所述的方法，其中，所述主题域包括以下至少任一项：
    信息领域；
    信息意图；
    信息意图对象。
  4. 根据权利要求3所述的方法，其中，所述主题域包括信息领域、信息意图和信息意图对象，其中，所述步骤d包括：
    根据所述主题域的信息领域、信息意图确定对应的主题模板；
    根据所述主题模板，结合所述信息意图对象生成外延扩展关键词列表，将所述主题域和所述外延扩展关键词列表填充至所述主题模板；
    将填充后的所述主题模板映射扩展为对应的搜索结构。
  5. 根据权利要求1至4中任一项所述的方法，其中，所述步骤d包括：
    根据所述主动匹配搜索所对应的搜索引擎的属性，将所述主题域映射扩展为所述搜索结构。
  6. 根据权利要求1至5中任一项所述的方法，其中，所述步骤e包括：
    根据所述搜索结构，并结合认知计算，进行主动匹配搜索，获取对应的主题信息并呈现给所述用户。
  7. 根据权利要求1至6中任一项所述的方法，其中，所述步骤e包括：
    根据所述搜索结构进行主动匹配搜索，获取对应的候选信息；
    基于对所述候选信息的信息提取整合，生成所述主题信息；
    将所述主题信息呈现给所述用户。
  8. 一种基于语音识别的主动型搜索装置，其中，该搜索装置包括：
    识别装置，用于识别接收的语音信息，获得所述语音信息的语音特征向量；
    判断装置，用于判断所述语音特征向量与用户预设的基准语音特征向量是否匹配，获得对应的判断结果；
    生成装置，用于若所述判断结果为不匹配，基于对所述语音信息的识别，生成与所述语音信息对应的主题域；
    映射装置，用于将所述主题域映射扩展为对应的搜索结构；
    呈现装置，用于根据所述搜索结构进行主动匹配搜索，获取对应的主题信息并呈现给所述用户。
  9. 根据权利要求8所述的搜索装置，其中，所述生成装置用于：
    若所述判断结果为不匹配，基于对所述语音信息的识别，将所述语音信息翻译成自然语言文本；
    对所述自然语言文本进行自然语义解析，生成与所述语音信息对应的主题域。
  10. 根据权利要求8或9所述的搜索装置，其中，所述主题域包括以下至少任一项：
    信息领域；
    信息意图；
    信息意图对象。
  11. 根据权利要求10所述的搜索装置，其中，所述主题域包括信息领域、信息意图和信息意图对象，其中，所述映射装置用于：
    根据所述主题域的信息领域、信息意图确定对应的主题模板；
    根据所述主题模板，结合所述信息意图对象生成外延扩展关键词列表，将所述主题域和所述外延扩展关键词列表填充至所述主题模板；
    将填充后的所述主题模板映射扩展为对应的搜索结构。
  12. 根据权利要求8至11中任一项所述的搜索装置，其中，所述映射装置用于：
    根据所述主动匹配搜索所对应的搜索引擎的属性，将所述主题域映射扩展为所述搜索结构。
  13. 根据权利要求8至12中任一项所述的搜索装置，其中，所述呈现装置用于：
    根据所述搜索结构，并结合认知计算，进行主动匹配搜索，获取对应的主题信息并呈现给所述用户。
  14. 根据权利要求8至13中任一项所述的搜索装置，其中，所述呈现装置包括：
    获取单元，用于根据所述搜索结构进行主动匹配搜索，获取对应的候选信息；
    整合单元，用于基于对所述候选信息的信息提取整合，生成所述主题信息；
    呈现单元，用于将所述主题信息呈现给所述用户。
  15. 一种计算机可读存储介质，所述计算机可读存储介质存储有计算机代码，当所述计算机代码被执行时，如权利要求1至7中任一项所 述的方法被执行。
  16. 一种计算机程序产品，当所述计算机程序产品被计算机设备执行时，如权利要求1至7中任一项所述的方法被执行。
  17. 一种计算机设备，所述计算机设备包括存储器和处理器，所述存储器中存储有计算机代码，所述处理器被配置来通过执行所述计算机代码以执行如权利要求1至7中任一项所述的方法。
PCT/CN2017/076968 2016-12-01 2017-03-16 一种基于语音识别的主动型搜索方法和装置 Ceased WO2018098932A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201611091688.4A CN108132952B (zh) 2016-12-01 2016-12-01 一种基于语音识别的主动型搜索方法和装置
CN201611091688.4 2016-12-01

Publications (1)

Publication Number Publication Date
WO2018098932A1 true WO2018098932A1 (zh) 2018-06-07

Family

ID=62241092

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/076968 Ceased WO2018098932A1 (zh) 2016-12-01 2017-03-16 一种基于语音识别的主动型搜索方法和装置

Country Status (2)

Country Link
CN (1) CN108132952B (zh)
WO (1) WO2018098932A1 (zh)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110060681A (zh) * 2019-04-26 2019-07-26 广东昇辉电子控股有限公司 具有智能语音识别功能的智能网关的控制方法
CN110110146A (zh) * 2019-04-12 2019-08-09 深圳壹账通智能科技有限公司 基于人工智能的视频短片搜索方法、装置、介质和设备
CN110782511A (zh) * 2019-09-09 2020-02-11 天脉聚源(杭州)传媒科技有限公司 动态改变虚拟主播形象的方法、系统、装置和存储介质
CN111343022A (zh) * 2020-02-28 2020-06-26 上海万得维进出口有限公司 通过与用户直接交互的方式实现智能设备进行网络配置处理的方法及其系统
CN112562652A (zh) * 2020-12-02 2021-03-26 湖南翰坤实业有限公司 一种基于Untiy引擎的语音处理方法及系统
CN112800782A (zh) * 2021-01-29 2021-05-14 中国科学院自动化研究所 融合文本语义特征的语音翻译方法、系统、设备

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109343806B (zh) * 2018-08-09 2023-07-21 维沃移动通信有限公司 一种信息显示方法及终端
CN109065055B (zh) * 2018-09-13 2020-12-11 三星电子（中国）研发中心 基于声音生成ar内容的方法、存储介质和装置
CN111178081B (zh) * 2018-11-09 2023-07-21 中移（杭州）信息技术有限公司 语义识别的方法、服务器、电子设备及计算机存储介质
CN111291168A (zh) * 2018-12-07 2020-06-16 北大方正集团有限公司 图书检索方法、装置及可读存储介质
CN110853615B (zh) * 2019-11-13 2022-05-27 北京欧珀通信有限公司 一种数据处理方法、装置及存储介质
CN111105796A (zh) * 2019-12-18 2020-05-05 杭州智芯科微电子科技有限公司 无线耳机控制装置及控制方法、语音控制设置方法和系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101075435A (zh) * 2007-04-19 2007-11-21 深圳先进技术研究院 一种智能聊天系统及其实现方法
US20140081643A1 (en) * 2012-09-14 2014-03-20 Avaya Inc. System and method for determining expertise through speech analytics
CN105068661A (zh) * 2015-09-07 2015-11-18 百度在线网络技术（北京）有限公司 基于人工智能的人机交互方法和系统
CN105159568A (zh) * 2015-08-31 2015-12-16 百度在线网络技术（北京）有限公司 一种在输入界面中进行音乐搜索的方法和装置
CN105723449A (zh) * 2013-11-06 2016-06-29 系统翻译国际有限公司 基于对录取的语音数据的核心语提取的言语内容分析系统和利用该系统的索引方法及言语内容分析方法

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102467541B (zh) * 2010-11-11 2016-06-15 深圳市世纪光速信息技术有限公司 一种情境搜索方法及系统
CN102497391B (zh) * 2011-11-21 2016-03-09 宇龙计算机通信科技(深圳)有限公司 服务器和提示方法
KR101878488B1 (ko) * 2011-12-20 2018-08-20 한국전자통신연구원 대화 연관 컨텐츠 제공 방법 및 장치
CN102880645B (zh) * 2012-08-24 2015-12-16 上海云叟网络科技有限公司 语义化的智能搜索方法
CN104836720B (zh) * 2014-02-12 2022-02-25 北京三星通信技术研究有限公司 交互式通信中进行信息推荐的方法及装置
CN105095406A (zh) * 2015-07-09 2015-11-25 百度在线网络技术（北京）有限公司 一种基于用户特征的语音搜索方法及装置
CN105279227B (zh) * 2015-09-11 2020-07-28 百度在线网络技术（北京）有限公司 同音词的语音搜索处理方法及装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101075435A (zh) * 2007-04-19 2007-11-21 深圳先进技术研究院 一种智能聊天系统及其实现方法
US20140081643A1 (en) * 2012-09-14 2014-03-20 Avaya Inc. System and method for determining expertise through speech analytics
CN105723449A (zh) * 2013-11-06 2016-06-29 系统翻译国际有限公司 基于对录取的语音数据的核心语提取的言语内容分析系统和利用该系统的索引方法及言语内容分析方法
CN105159568A (zh) * 2015-08-31 2015-12-16 百度在线网络技术（北京）有限公司 一种在输入界面中进行音乐搜索的方法和装置
CN105068661A (zh) * 2015-09-07 2015-11-18 百度在线网络技术（北京）有限公司 基于人工智能的人机交互方法和系统

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110110146A (zh) * 2019-04-12 2019-08-09 深圳壹账通智能科技有限公司 基于人工智能的视频短片搜索方法、装置、介质和设备
CN110060681A (zh) * 2019-04-26 2019-07-26 广东昇辉电子控股有限公司 具有智能语音识别功能的智能网关的控制方法
CN110782511A (zh) * 2019-09-09 2020-02-11 天脉聚源(杭州)传媒科技有限公司 动态改变虚拟主播形象的方法、系统、装置和存储介质
CN111343022A (zh) * 2020-02-28 2020-06-26 上海万得维进出口有限公司 通过与用户直接交互的方式实现智能设备进行网络配置处理的方法及其系统
CN112562652A (zh) * 2020-12-02 2021-03-26 湖南翰坤实业有限公司 一种基于Untiy引擎的语音处理方法及系统
CN112562652B (zh) * 2020-12-02 2024-01-19 湖南翰坤实业有限公司 一种基于Untiy引擎的语音处理方法及系统
CN112800782A (zh) * 2021-01-29 2021-05-14 中国科学院自动化研究所 融合文本语义特征的语音翻译方法、系统、设备
CN112800782B (zh) * 2021-01-29 2023-10-03 中国科学院自动化研究所 融合文本语义特征的语音翻译方法、系统、设备

Also Published As

Publication number Publication date
CN108132952B (zh) 2022-03-15
CN108132952A (zh) 2018-06-08

Similar Documents

Publication Publication Date Title
CN108132952B (zh) 一种基于语音识别的主动型搜索方法和装置
US12217740B2 (en) Detection and/or enrollment of hot commands to trigger responsive action by automated assistant
CN113468302B (zh) 组合共享询问线的多个搜索查询的参数
CN111694940B (zh) 一种用户报告的生成方法及终端设备
CN111459290B (zh) 交互意图确定方法及装置、计算机设备及存储介质
EP4036910B1 (en) Dynamic and/or context-specific hot words to invoke automated assistant
CN110634472B (zh) 一种语音识别方法、服务器及计算机可读存储介质
KR20190108523A (ko) 음성신호 기반 인공지능을 이용한 감정분석을 제공하는 감성일기 서비스 제공 서버 및 방법
CN111033494A (zh) 用于多个搜索机器人和行为机器人的计算体系架构以及相关的设备和方法
WO2017112813A1 (en) Multi-lingual virtual personal assistant
US11640426B1 (en) Background audio identification for query disambiguation
CN120353899B (zh) 心理疗愈导向的大模型对话智能体
CN112236765A (zh) 基于生成的子查询的集合确定复合查询的响应内容
CN114064943A (zh) 会议管理方法、装置、存储介质及电子设备
CN113903335B (zh) 一种用户意图识别方法、用户意图识别装置和存储介质
CN112883350B (zh) 一种数据处理方法、装置、电子设备以及存储介质
US12596878B1 (en) System for inferring saliency in a multivariate time series derived from periodic conversation with fine-tuned large language model
JP7169031B1 (ja) プログラム、情報処理装置、情報処理システム、情報処理方法、情報処理端末
KR102925156B1 (ko) 온디바이스 기반 능동형 대화 메시지 생성 방법 및 이를 활용한 사용자 기기
CN116955601B (zh) 多模态的情感识别方法和装置
JP7169030B1 (ja) プログラム、情報処理装置、情報処理システム、情報処理方法、情報処理端末
TWI866860B (zh) 一種摘要主從式對話資訊生成方法及其裝置
JP2026028078A (ja) システム
JP2026062176A (ja) システム
JP2026050665A (ja) システム

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17875997

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 04/10/2019)

122 Ep: pct application non-entry in european phase

Ref document number: 17875997

Country of ref document: EP

Kind code of ref document: A1