WO2020207035A1 - 骚扰电话拦截方法、装置、设备及存储介质 - Google Patents

骚扰电话拦截方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2020207035A1
WO2020207035A1 PCT/CN2019/121724 CN2019121724W WO2020207035A1 WO 2020207035 A1 WO2020207035 A1 WO 2020207035A1 CN 2019121724 W CN2019121724 W CN 2019121724W WO 2020207035 A1 WO2020207035 A1 WO 2020207035A1
Authority
WO
WIPO (PCT)
Prior art keywords
caller
voice
call
intention
connection request
Prior art date
Application number
PCT/CN2019/121724
Other languages
English (en)
French (fr)
Inventor
齐燕
Original Assignee
深圳壹账通智能科技有限公司
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 深圳壹账通智能科技有限公司 filed Critical 深圳壹账通智能科技有限公司
Publication of WO2020207035A1 publication Critical patent/WO2020207035A1/zh

Links

Classifications

    • 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
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04M—TELEPHONIC COMMUNICATION
    • H04M1/00—Substation equipment, e.g. for use by subscribers
    • H04M1/66—Substation equipment, e.g. for use by subscribers with means for preventing unauthorised or fraudulent calling
    • H04M1/663—Preventing unauthorised calls to a telephone set
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04M—TELEPHONIC COMMUNICATION
    • H04M1/00—Substation equipment, e.g. for use by subscribers
    • H04M1/66—Substation equipment, e.g. for use by subscribers with means for preventing unauthorised or fraudulent calling
    • H04M1/663—Preventing unauthorised calls to a telephone set
    • H04M1/665—Preventing unauthorised calls to a telephone set by checking the validity of a code
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04M—TELEPHONIC COMMUNICATION
    • H04M1/00—Substation equipment, e.g. for use by subscribers
    • H04M1/72—Mobile telephones; Cordless telephones, i.e. devices for establishing wireless links to base stations without route selection
    • H04M1/724—User interfaces specially adapted for cordless or mobile telephones
    • H04M1/72448—User interfaces specially adapted for cordless or mobile telephones with means for adapting the functionality of the device according to specific conditions
    • H04M1/72454—User interfaces specially adapted for cordless or mobile telephones with means for adapting the functionality of the device according to specific conditions according to context-related or environment-related conditions
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04M—TELEPHONIC COMMUNICATION
    • H04M1/00—Substation equipment, e.g. for use by subscribers
    • H04M1/72—Mobile telephones; Cordless telephones, i.e. devices for establishing wireless links to base stations without route selection
    • H04M1/724—User interfaces specially adapted for cordless or mobile telephones
    • H04M1/72484—User interfaces specially adapted for cordless or mobile telephones wherein functions are triggered by incoming communication events
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04M—TELEPHONIC COMMUNICATION
    • H04M2250/00—Details of telephonic subscriber devices
    • H04M2250/74—Details of telephonic subscriber devices with voice recognition means

Definitions

  • This application relates to the technical field of communication security, and in particular to a method, device, equipment and storage medium for intercepting harassing calls.
  • Harassing phone calls is one of the undesirable consequences of information leakage.
  • Existing communication devices such as mobile phones/landlines, and anti-harassment telephone systems are often incomplete. After the user answers the harassing call, it can be judged whether the incoming call is a harassing call, and then the user needs to manually pull the harassing call into the blacklist.
  • the anti-harassment mechanism of the call equipment is too intelligent.
  • the main purpose of this application is to provide a method, device, equipment and storage medium for intercepting harassing calls, aiming at the technical problem that the anti-harassment mechanism of existing communication equipment is too low in intelligence.
  • this application provides a method for intercepting harassing calls, which includes the following steps:
  • the step of conducting a voice conversation with the caller corresponding to the call connection request according to a preset conversation rule includes:
  • the target reply utterance is converted into voice output.
  • the step includes:
  • the step of extracting the intention features of the caller from the voice conversation is performed.
  • the step of extracting the caller's intention feature from the voice conversation and inputting the intention feature into a preset intention discriminator includes:
  • training samples and test samples where the training samples are the intent-labeled utterances
  • the accuracy test of the intention discrimination model is performed based on the characteristics of the test sample, and the optimal model parameters of the intention discrimination model are adjusted based on the test results.
  • the step of extracting the intention feature of the caller from the voice conversation includes:
  • the second keyword is compared with a preset image gallery to obtain an intent keyword in the second keyword, and the intent keyword is used as an intent feature of the caller.
  • the step of establishing a pre-connection with the caller terminal corresponding to the call connection request includes:
  • incoming call number is a whitelist number, output an incoming call reminder
  • a pre-connection is established with the incoming call terminal corresponding to the call connection request.
  • the following includes:
  • the present application also provides a harassing call interception device, which includes:
  • the intelligent voice module is configured to, after detecting the call connection request, establish a pre-connection with the caller terminal corresponding to the call connection request, and conduct a voice conversation with the caller corresponding to the call connection request according to a preset dialogue rule;
  • Intention discrimination module configured to extract the intention features of the caller from the voice dialogue, input the intention features into a preset intention discriminator, and obtain the discrimination result output by the intention discriminator;
  • the telephone processing module is configured to hang up the incoming call when it is determined that the call connection request is a harassing call according to the discrimination result.
  • this application also provides a harassing call interception device, which includes a processor, a memory, and computer-readable instructions stored on the memory and executable by the processor , Wherein when the computer-readable instructions are executed by the processor, the steps of the method for intercepting harassing calls are implemented.
  • the present application also provides a storage medium storing computer-readable instructions, where the computer-readable instructions are executed by a processor to implement the above-mentioned harassing call interception method. step.
  • FIG. 1 is a schematic structural diagram of a harassing call intercepting device in a hardware operating environment involved in a solution of an embodiment of the present application;
  • FIG. 2 is a schematic flowchart of an embodiment of a method for intercepting harassing calls according to the application
  • FIG. 3 is a schematic diagram of functional modules of an embodiment of an apparatus for intercepting harassing calls according to this application.
  • Figure 1 is a schematic diagram of the hardware structure of the harassing call interception device provided by this application.
  • the harassing call interception device can be a PC, or a mobile phone, a tablet computer, a portable computer, a desktop computer and other devices with a call function.
  • a mobile phone for calls in the following embodiments of the harassing call interception method of this application, Take a mobile phone as an example of a harassing call interception device for explanation.
  • the harassing call interception device may include components such as a processor 101 and a memory 201.
  • the processor 101 is connected to the memory 201, and computer readable instructions are stored on the memory 201.
  • the processor 101 can call the computer readable instructions stored in the memory 201, and implement the following implementations of the harassing call interception method Example steps.
  • the memory 201 can be used to store software programs and various data.
  • the memory 201 may mainly include a storage program area and a storage data area, where the storage program area may store an operating system, at least one application program required by a function (such as computer-readable instructions for interception of harassing calls), etc.; the storage data area may Including databases, etc.
  • the processor 101 is the control center of the harassing call intercepting device, which uses various interfaces and lines to connect the various parts of the entire harassing call intercepting device, and by running or executing the software programs and/or modules stored in the memory 201, and calling the
  • the data in the memory 201 performs various functions and processing data of the harassing call interception device, so as to perform overall monitoring of the harassing call interception device.
  • the structure of the harassing call intercepting device shown in FIG. 1 does not constitute a limitation on the harassing call intercepting device, and may include more or less components than shown in the figure, or a combination of certain components, or different The layout of the components.
  • This application provides a method for intercepting harassing calls.
  • FIG. 2 is a schematic flowchart of a first embodiment of a method for intercepting harassing calls according to this application.
  • the method for intercepting harassing calls includes the following steps:
  • Step S10 after detecting the call connection request, establish a pre-connection with the caller terminal corresponding to the call connection request, and conduct a voice conversation with the caller corresponding to the call connection request according to a preset dialogue rule;
  • the call connection request is the incoming call request.
  • the mobile phone After receiving the incoming call request, the mobile phone responds to the incoming call request and establishes a pre-connection with the incoming call terminal corresponding to the incoming call request, so as to realize the subsequent dialogue between the mobile phone and the incoming call terminal and the identification of harassment intent.
  • the pre-connection here refers to the call connection that the mobile phone automatically establishes with the caller terminal, and after the pre-connection is established, the smart voice module in the mobile phone automatically conducts a voice dialogue with the caller.
  • the call connection here can directly output the call reminder according to the preset settings To the user, so that the user knows whether there is an incoming call, or according to the preset setting not to remind the caller and output the caller's voice to the user.
  • the incoming call that is obviously a non-harassing phone can be identified, and the incoming call of the non-harassing phone can be output to prompt the user to answer the call, so as to avoid
  • the subsequent meaningless harassment intention recognition operation reduces the waste of operating resources and storage resources, facilitates the user to answer the call as soon as possible, and reduces the call delay caused by the harassment intention recognition.
  • the step of establishing a pre-connection with the caller terminal corresponding to the call connection request includes:
  • Step S101 After detecting the call connection request, compare the caller number corresponding to the call connection request with the numbers on the whitelist list to determine whether the caller number corresponding to the call connection request is a whitelist number;
  • the numbers on the whitelist are all non-harassing calls, which can include numbers stored in the address book or numbers whose historical call duration is longer than the preset duration, or non-harassing numbers marked by the user or special public numbers (such as hospital calls, China Mobile, China Unicom, etc.) Operator customer number, etc.).
  • the caller number corresponding to the call connection request is first identified as a non-harassing call.
  • the caller number can be compared with the number on the whitelist to determine whether the caller number is on the whitelist.
  • Step S102 if the incoming call number is a whitelisted number, output an incoming call reminder
  • the caller number is a whitelisted number
  • subsequent harassing intention recognition operations such as the establishment of a pre-connection with the caller terminal corresponding to the call connection request are no longer performed, avoiding unnecessary recognition operations, saving resources and improving user experience.
  • Step S103 If the incoming call number is not a whitelist number, a pre-connection is established with the incoming call terminal corresponding to the call connection request.
  • the caller number is not a whitelisted number
  • a pre-connection is established with the caller terminal corresponding to the call connection request, and the smart voice module in the mobile phone answers the call.
  • the call connection request may be immediately corresponding to the call connection request.
  • the caller terminal establishes a pre-connection, and the smart voice module in the mobile phone immediately answers the call; it can also establish a pre-connection with the caller terminal corresponding to the call connection request after detecting the call connection request for a preset duration to filter the call connection Request for harassing calls with a short duration, reducing the waste of resources for establishing pre-connections with such harassing calls.
  • the smart voice module in the mobile phone After establishing a pre-connection with the caller terminal corresponding to the call connection request, the smart voice module in the mobile phone conducts a voice dialogue with the caller.
  • the smart voice module in the mobile phone is configured with a dialogue management function to allow multiple rounds of dialogue with the caller.
  • the caller corresponding to the call connection request conducts a voice dialogue according to preset dialogue rules.
  • the dialogue content and dialogue order can be preset, and the mobile phone communicates with the incoming call according to the preset dialogue content and dialogue order.
  • the preset dialogue content and dialogue sequence are: "Who are you? Who are you looking for? What do you want?".
  • step S10 the step of performing a voice dialogue with the caller corresponding to the call connection request according to the preset dialogue rule in step S10 includes:
  • Step S104 Obtain the caller's audio corresponding to the call connection request, extract the caller's voice from the caller's audio based on voice endpoint detection technology, and convert the caller's voice into voice text;
  • the caller's audio is obtained, the caller's audio is semantically recognized, and the voice response is performed according to preset rules and the recognized caller's semantics to realize a voice dialogue with the caller.
  • the smart voice module on the mobile phone can ask questions first, and then obtain the caller's audio, or after detecting the caller's voice and obtaining the caller's audio, the corresponding voice response can be made according to the semantic recognition of the caller's audio.
  • the voice endpoint detection technology extracts the caller's voice from the caller's audio, avoiding the waste of resources due to too much silence, reducing interference with semantic recognition, and improving the accuracy of semantic recognition.
  • Voice endpoint detection technology refers to the technology to detect effective voice segments from continuous audio streams, including detecting the starting point and ending point of the effective voice segment, and identifying the effective voice segment by determining the energy level and change trajectory of the audio stream. If a certain audio stream is muted, the energy value of several consecutive frames of the audio stream will continue to be maintained at a low level, and the end points (starting and ending points) of the effective speech segment are at the critical point of energy change.
  • noise suppression is performed on the caller's audio to improve the accuracy of voice endpoint detection.
  • the caller’s voice in this embodiment is the effective voice segment obtained through voice endpoint detection, and then the caller’s voice is converted into voice text.
  • the effective voice segment may be one sentence or a continuous multi-sentence, which is converted into a corresponding sentence Or multiple sentences of speech text.
  • Step S105 Extract a first keyword from the voice text, and query a preset dialogue database according to the first keyword to obtain a target reply utterance;
  • the preset dialogue database stores well-defined dialogue data, including possible question and answer dialogues. Similar dialogues are retrieved in the dialogue database according to the first keyword, and the reply corresponding to the similar dialogue is obtained as the target reply utterance.
  • the first keyword here refers to the keyword in the current words spoken by the caller.
  • Step S106 Convert the target reply utterance into voice output.
  • the target reply utterance is converted into voice, and the voice is output to the caller terminal to realize a dialogue with the caller.
  • a voice database is preset, and the intelligent voice module can generate an imitated natural person voice based on the collected user voice (phone master voice) or voice in the preset voice database.
  • the intelligent voice module can imitate the voice of the owner, such as the voice characteristics of the owner, intonation, speech speed, etc., to generate imitated sounds.
  • the intelligent voice module imitates the human voice, which can reduce the vigilance of the harasser, and prevent the fabricated language from deceiving the intelligent voice module.
  • Step S20 extracting the intention feature of the caller from the voice dialogue, inputting the intention feature to a preset intention discriminator, and obtaining a discrimination result output by the intention discriminator;
  • Extract the intent keywords from the voice dialogue express the intent keywords as the corresponding feature vector, that is, the intent feature in this embodiment, and input the intent feature into the intent discriminator, and the intent discriminator will perform the intention discrimination based on the input intention feature , And output the results of intention determination, such as pure harassment, sales calls, etc.
  • the step of extracting the caller's intention feature from the voice conversation includes:
  • the intention keywords of the caller are the intention keywords of the caller.
  • All the voice of the caller can be obtained from the voice conversation, and the second keyword can be extracted from the voice of the caller, where the second keyword refers to the keyword extracted from all the voice of the caller; then from the second keyword
  • intent keywords you can compare the second keyword with the intent vocabulary in the preset intent vocabulary, and obtain words that are consistent with the intent vocabulary in the intent vocabulary or similar words from the second keyword as the intent keywords.
  • the intention keywords are expressed as corresponding feature vectors, that is, the caller's intention features.
  • the intention discriminator is a pre-trained intention discrimination model with annotated training corpus, which is used to classify and recognize the caller's intentions according to the intention features.
  • step S20 includes:
  • Step S21 Obtain training samples and test samples, where the training samples are utterances that have been intentionally annotated;
  • call text data including harassing call text and normal call text
  • the preset corpus can include harassing call text and normal call text found directly from the Internet, or Including the text of intention labeling after the voice of the local phone is converted to text.
  • the training samples are used to optimize the parameters of the model, and the test samples are used to evaluate the performance of the established intention discrimination model.
  • Step S22 extract the characteristics of the training sample and the test sample respectively, and calculate the optimal model parameters of the intention discrimination model through an iterative algorithm according to the characteristics of the training samples, and train the intention discrimination model containing the optimal model parameters;
  • the intention discrimination model is a neural network classification model, which can be a naive Bayes classifier.
  • the model parameters of the intention discrimination model are randomly set and not optimized.
  • Each training sample or test sample can be correspondingly represented as a feature vector composed of a set of features.
  • each word with a number or number (for example, from 1 to 36000) to realize the digitization of training samples or test samples, and use the respective digitized forms of training samples or test samples as their corresponding feature vectors.
  • iterative algorithms include gradient descent, conjugate gradient method and quasi-Newton method.
  • the optimal model parameters of the intention discrimination model can be calculated through any of the above iterative algorithms, and the intention discrimination model containing the optimal model parameters can be trained.
  • other methods can also be used to extract the characteristics of the training samples and test samples, such as vector space model VSM, information gain method, expected cross entropy, and so on.
  • Step S23 Perform an accuracy test on the intention discrimination model based on the characteristics of the test sample, and adjust the optimal model parameters of the intention discrimination model based on the test result.
  • the test sample is used to evaluate the accuracy of the intention discrimination model, and adjust the optimal model parameters of the intention discrimination model based on the test results to improve the accuracy of the discrimination model.
  • Step S30 Hang up the incoming call when it is determined that the call connection request is a nuisance call according to the discrimination result.
  • step S20 it includes: when it is determined that the call connection request is a normal call according to the discrimination result, outputting an incoming call reminder, and outputting the content of the voice dialogue in text form.
  • the text outputs the content of the dialogue between the intelligent voice module and the caller.
  • the user can conduct follow-up dialogues based on the content of the dialogue. For example, for the questions that have been asked in the intelligent dialogue stage (such as: who are you and what do you have), the user will answer the call There is no need to ask again, to realize the seamless connection between the intelligent dialogue and the user's answering dialogue, and reduce the dissatisfaction caused by the time delay caused by the intelligent assistant verification.
  • This embodiment establishes a pre-connection with the caller terminal corresponding to the call connection request after detecting the call connection request, and conducts a voice conversation with the caller corresponding to the call connection request according to preset dialogue rules; Extract the caller’s intention feature from the conversation, input the intention feature into a preset intention discriminator, and obtain the discrimination result output by the intention discriminator; after determining the call connection request as a harassment according to the discrimination result Hanging up the call during the call, that is, by pre-conducting a voice conversation with the caller, and using the voice content as the basis for harassment judgment, it can effectively avoid the harassment and fraud calls disguised as a normal number, and at the same time, it can intelligently intercept harassing calls.
  • the method includes:
  • Step S11 Collect the voiceprint information of the caller, compare and match the voiceprint information of the caller with the voiceprint information in a preset voiceprint database, and determine whether there is a voiceprint information in the preset voiceprint database. State the target voiceprint information matched by the caller’s voiceprint information;
  • This embodiment combines the caller’s voiceprint information to identify the “acquaintance’s new number”, reducing subsequent unnecessary harassment identification steps.
  • the voiceprint information of non-harassing telephone callers is pre-stored in the preset voiceprint database, and the voiceprint information of historical callers (the suspicion of harassing calls has been excluded) can be stored in the preset voiceprint database.
  • the caller’s voiceprint information can be collected and compared with the pre-stored voiceprint information, and it is determined whether there is a voiceprint with the caller’s voice in the preset voiceprint database.
  • the voiceprint information matches the target voiceprint information, where the voiceprint information of the caller can be collected while the voice of the caller is detected, and step S11 is performed to perform voiceprint recognition.
  • Step S12 if there is target voiceprint information matching the voiceprint information of the caller in the preset voiceprint database, output an incoming call reminder;
  • the incoming call reminder is directly output to remind the user to answer, avoid unnecessary verification, save resources and improve user experience.
  • Step S13 if there is no target voiceprint information matching the voiceprint information of the caller in the preset voiceprint database, then execute the step of extracting the caller’s intention features from the voice conversation .
  • This embodiment collects the voiceprint information of the caller when or after the caller corresponding to the call connection request is engaged in a voice conversation, and compares and matches the voiceprint information of the caller with the voiceprint information in the preset voiceprint database , To indirectly determine whether the caller is an acquaintance or a non-harassing call. When there is target voiceprint information matching the caller’s voiceprint information in the preset voiceprint database, it is determined that the call is not a harassing call. Otherwise, further harassment is required Recognition and judgment can identify the situation of acquaintances calling users with new phone numbers and directly remind users to answer, avoid unnecessary harassment recognition, save resources and improve user experience.
  • this application also provides a harassing call intercepting device corresponding to the steps of the above harassing call intercepting method.
  • Fig. 3 is a schematic diagram of the functional modules of the first embodiment of the device for intercepting harassing calls according to the present application.
  • the harassing call interception device of this application includes:
  • the intelligent voice module 10 is configured to, after detecting the call connection request, establish a pre-connection with the caller terminal corresponding to the call connection request, and conduct a voice dialogue with the caller corresponding to the call connection request according to preset dialogue rules;
  • the intention discrimination module 20 is configured to extract the intention features of the caller from the voice dialogue, input the intention features into a preset intention discriminator, and obtain the discrimination result output by the intention discriminator;
  • the telephone processing module 30 is configured to hang up the incoming call when it is determined that the call connection request is a nuisance call according to the judgment result.
  • the smart voice module 10 is also used to obtain the caller's audio corresponding to the call connection request, extract the caller's voice from the caller's audio based on voice endpoint detection technology, and convert the caller's voice into voice Text; extract a first keyword from the voice text, query a preset dialog database according to the first keyword to obtain a target reply utterance; convert the target reply utterance into a voice output.
  • the harassing call interception device of this application also includes:
  • the voiceprint judgment module is used to collect the voiceprint information of the caller, compare and match the voiceprint information of the caller with the voiceprint information in a preset voiceprint database, and determine whether the caller is in the preset voiceprint database. Whether there is target voiceprint information matching the voiceprint information of the caller;
  • the telephone processing module 30 is further configured to output an incoming call reminder if there is target voiceprint information matching the voiceprint information of the caller in the preset voiceprint database; if the preset voiceprint database does not exist If the target voiceprint information matches the voiceprint information of the caller, the step of extracting the intention feature of the caller from the voice conversation is executed.
  • the harassing call interception device of this application also includes:
  • the model training module is used to obtain training samples and test samples, where the training samples are the intent-labeled utterances; the characteristics of the training samples and the test samples are extracted respectively, and according to the characteristics of the training samples, an iterative algorithm is used to calculate the intent discrimination model
  • the optimal model parameters are used to train an intention discrimination model containing the optimal model parameters; the accuracy of the intention discrimination model is tested based on the characteristics of the test sample, and the optimal model parameters of the intention discrimination model are adjusted based on the test results.
  • the intention determination module 20 is also used to obtain the caller's voice from the voice conversation, and extract a second keyword from the caller's voice; and compare the second keyword with a preset Compare the image gallery to obtain the intention keyword in the second keyword, and use the intention keyword as the caller's intention feature.
  • the smart voice module 10 is also used to compare the caller number corresponding to the call connection request with the numbers on the whitelist after the call connection request is detected to determine whether the caller number corresponding to the call connection request is Is a whitelisted number; if the incoming call number is a whitelisted number, an incoming call reminder is output; if the incoming call number is not a whitelisted number, a pre-connection is established with the incoming call terminal corresponding to the call connection request.
  • the telephone processing module 30 is further configured to output an incoming call reminder when it is determined that the call connection request is a normal call according to the discrimination result, and output the content of the voice dialogue in text form.
  • the application also proposes a storage medium on which computer-readable instructions are stored.
  • the storage medium may be a non-volatile readable storage medium, may be the memory 201 in the harassing call interception device in FIG. 1, or may be a ROM (Read-Only Memory, read-only memory)/RAM (Random Access Memory, at least one of random access memory), magnetic disks, and optical disks.
  • the storage medium includes several instructions to enable a device with a processor (the harassing call interception device in the embodiment of this application, such as a mobile phone, a computer, etc.) ) Perform the method of each embodiment of the present application.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Security & Cryptography (AREA)
  • Signal Processing (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Environmental & Geological Engineering (AREA)
  • Computational Linguistics (AREA)
  • Telephonic Communication Services (AREA)
  • Telephone Function (AREA)

Abstract

本申请提供一种基于分类模型的骚扰电话拦截方法、装置、设备及存储介质，该方法包括：在检测到通话连接请求后，与所述通话连接请求对应的来电终端建立预连接，根据预设的对话规则与所述通话连接请求对应的来电者进行语音对话；从所述语音对话中提取所述来电者的意图特征，将所述意图特征输入预置的意图判别器，并获得所述意图判别器输出的判别结果；在根据所述判别结果确定所述通话连接请求为骚扰电话时，挂断来电。本申请可实现骚扰电话的智能拦截。

Description

骚扰电话拦截方法、装置、设备及存储介质
本申请要求于2019年04月12日提交中国专利局、申请号为201910301956.8、发明名称为“骚扰电话拦截方法、装置、设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在申请中。
技术领域
本申请涉及通信安全技术领域，尤其涉及一种骚扰电话拦截方法、装置、设备及存储介质。
背景技术
骚扰电话是信息泄露的不良后果之一。现有的通话设备，如手机/座机，防骚扰电话系统往往并不完善，在用户接听骚扰电话后才能判断来电是否为骚扰电话，且之后需要用户手动将骚扰电话拉入黑名单。通话设备的防骚扰机制智能化程度过低。
发明内容
本申请的主要目的在于提供一种骚扰电话拦截方法、装置、设备及存储介质，旨在现有通话设备的防骚扰机制智能化程度过低的技术问题。
为实现上述目的，本申请提供一种骚扰电话拦截方法，所述骚扰电话拦截方法包括以下步骤：
在检测到通话连接请求后，与所述通话连接请求对应的来电终端建立预连接，根据预设的对话规则与所述通话连接请求对应的来电者进行语音对话；
从所述语音对话中提取所述来电者的意图特征，将所述意图特征输入预置的意图判别器，并获得所述意图判别器输出的判别结果；
在根据所述判别结果确定所述通话连接请求为骚扰电话时，挂断来电。
可选地，所述根据预设的对话规则与所述通话连接请求对应的来电者进行语音对话的步骤包括：
获取所述通话连接请求对应的来电者音频，基于语音端点检测技术从所述来电者音频中提取来电者语音，并将所述来电者语音转化成语音文本；
从所述语音文本中提取第一关键词，根据所述第一关键词查询预置的对话库获得目标回复话语；
将所述目标回复话语转换成语音输出。
可选地，所述根据预设的对话规则与所述通话连接请求对应的来电者进行语音对话的步骤之后包括：
采集所述来电者的声纹信息，将所述来电者的声纹信息与预置声纹数据库中的声纹信息进行对比匹配，判断所述预置声纹数据库中是否存在与所述来电者的声纹信息匹配的目标声纹信息；
若所述预置声纹数据库中存在与所述来电者的声纹信息匹配的目标声纹信息，则输出来电提醒；
若所述预置声纹数据库中不存在与所述来电者的声纹信息匹配的目标声纹信息，则执行所述从所述语音对话中提取所述来电者的意图特征的步骤。
可选地，所述从所述语音对话中提取所述来电者的意图特征，将所述意图特征输入预置的意图判别器的步骤之前包括：
获取训练样本和测试样本，其中，训练样本为经过意图标注的话语；
分别提取训练样本和测试样本的特征，并根据训练样本的特征，通过迭代算法计算出意图判别模型的最优模型参数，训练出含最优模型参数的意图判别模型；
基于所述测试样本的特征对意图判别模型进行准确性测试，基于测试结果调整意图判别模型的最优模型参数。
可选地，所述从所述语音对话中提取所述来电者的意图特征的步骤包括：
从所述语音对话中获得所述来电者语音，从所述来电者语音中提取第二关键词；
将所述第二关键词与预置的意图库进行对比，获得所述第二关键词中的意图关键词，将所述意图关键词作为所述来电者的意图特征。
可选地，所述在检测到通话连接请求后，与所述通话连接请求对应的来电终端建立预连接的步骤包括：
在检测到通话连接请求后，将所述通话连接请求对应的来电号码与白名单列表上号码进行比较，以判断所述通话连接请求对应的来电号码是否为白名单号码；
若所述来电号码为白名单号码，则输出来电提醒；
若所述来电号码不是白名单号码，则与所述通话连接请求对应的来电终端建立预连接。
可选地，所述获得所述意图判别器输出的判别结果的步骤之后包括：
在根据所述判别结果确定所述通话连接请求为正常电话时，输出来电提醒，并以文字形式输出所述语音对话的内容。
此外，为实现上述目的，本申请还提供一种骚扰电话拦截装置，所述骚扰电话拦截装置包括：
智能语音模块，用于在检测到通话连接请求后，与所述通话连接请求对应的来电终端建立预连接，根据预设的对话规则与所述通话连接请求对应的来电者进行语音对话；
意图判别模块，用于从所述语音对话中提取所述来电者的意图特征，将所述意图特征输入预置的意图判别器，并获得所述意图判别器输出的判别结果；
电话处理模块，用于在根据所述判别结果确定所述通话连接请求为骚扰电话时，挂断来电。
此外，为实现上述目的，本申请还提供一种骚扰电话拦截设备，所述骚扰电话拦截设备包括处理器、存储器、以及存储在所述存储器上并可被所述处理器执行的计算机可读指令，其中所述计算机可读指令被所述处理器执行时，实现如上述的骚扰电话拦截方法的步骤。
此外，为实现上述目的，本申请还提供一种存储介质，所述存储介质上存储有计算机可读指令，其中所述计算机可读指令被处理器执行时，实现如上述的骚扰电话拦截方法的步骤。
本申请的一个或多个实施例的细节在下面的附图和描述中提出。本申请的其他特征和优点将从说明书、附图以及权利要求书变得明显。
附图说明
图1是本申请实施例方案涉及的硬件运行环境的骚扰电话拦截设备结构示意图；
图2为本申请骚扰电话拦截方法一实施例的流程示意图；
图3为本申请骚扰电话拦截装置一实施例的功能模块示意图。
本申请目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
具体实施方式
应当理解，此处所描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
请参见图1，图1为本申请所提供的骚扰电话拦截设备的硬件结构示意图。
骚扰电话拦截设备可以是PC，也可以是手机、平板电脑、便携计算机、台式计算机等具有通话功能的设备，鉴于通常使用手机进行通话，在本申请骚扰电话拦截方法的下述各实施例中，以手机作为骚扰电话拦截设备的一种示例进行解释说明。
骚扰电话拦截设备可以包括：处理器101以及存储器201等部件。在骚扰电话拦截设备中，处理器101与存储器201连接，存储器201上存储有计算机可读指令，处理器101可以调用存储器201中存储的计算机可读指令，并实现如下述骚扰电话拦截方法各实施例的步骤。
存储器201，可用于存储软件程序以及各种数据。存储器201可主要包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需的应用程序（比如应用于骚扰电话拦截的计算机可读指令）等；存储数据区可包括数据库等。处理器101，是骚扰电话拦截设备的控制中心，利用各种接口和线路连接整个骚扰电话拦截设备的各个部分，通过运行或执行存储在存储器201内的软件程序和/或模块，以及调用存储在存储器201内的数据，执行骚扰电话拦截设备的各种功能和处理数据，从而对骚扰电话拦截设备进行整体监控。
本领域技术人员可以理解，图1中示出的骚扰电话拦截设备结构并不构成对骚扰电话拦截设备的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。
基于上述硬件结构，提出本申请方法各个实施例。
本申请提供一种骚扰电话拦截方法。
参照图2，图2为本申请骚扰电话拦截方法第一实施例的流程示意图。
本实施例中，骚扰电话拦截方法包括以下步骤：
步骤S10，在检测到通话连接请求后，与所述通话连接请求对应的来电终端建立预连接，根据预设的对话规则与所述通话连接请求对应的来电者进行语音对话；
通话连接请求即来电请求，手机在接收到来电请求后，响应来电请求，与该来电请求对应的来电终端建立预连接，以便实现后续的手机与来电终端的对话和骚扰意图识别。这里的预连接指手机自动与来电终端建立的通话连接，且在建立预连接后通过手机中的智能语音模块自动与来电者进行语音对话，这里的通话连接可以根据预先的设置把来电提醒直接输出给用户，以便用户知晓当前是否有来电，也可根据预先设置不将来电提醒以及来电者语音输出给用户。
可选地，在与通话连接请求对应的来电终端建立预连接进行骚扰意图识别之前，可将明显属于非骚扰电话的来电识别出来，并将非骚扰电话的来电输出以提示用户接听来电，免除了后续无意义的骚扰意图识别操作，减少了运行资源、存储资源的浪费，便于用户尽快接听来电，减少因骚扰意图识别带来的通话延时。可选地，步骤S10中在检测到通话连接请求后，与所述通话连接请求对应的来电终端建立预连接的步骤包括：
步骤S101，在检测到通话连接请求后，将所述通话连接请求对应的来电号码与白名单列表上号码进行比较，以判断所述通话连接请求对应的来电号码是否为白名单号码；
白名单列表上号码都为非骚扰电话，可包括通讯录存储的号码或历史通话时长大于预设时长的号码或由用户标记的非骚扰号码或特殊公共号码（如医院电话、移动、联通等通信运营商客户号码等）。
在检测到通话连接请求后，首先根据通话连接请求对应的来电号码进行非骚扰电话的识别，可将来电号码与白名单列表上号码进行对比，判断来电号码是否为白名单列表上号码。
步骤S102，若所述来电号码为白名单号码，则输出来电提醒；
若来电号码是白名单号码，则直接提醒用户接听来电，不再执行与通话连接请求对应的来电终端建立预连接等后续的骚扰意图识别操作，避免不必要的识别操作，节约资源的同时提高了用户体验。
步骤S103，若所述来电号码不是白名单号码，则与所述通话连接请求对应的来电终端建立预连接。
若来电号码不是白名单号码，则与通话连接请求对应的来电终端建立预连接，由手机中智能语音模块接听来电。
可选地，还可以在检测到通话连接请求后或者在来电号码并非前述通讯录存储的号码或历史通话时长大于预设时长的号码或标记的非骚扰号码之后，立即与该通话连接请求对应的来电终端建立预连接，由手机中的智能语音模块立刻接听该来电；也可以在检测到通话连接请求持续预设时长后，再与该通话连接请求对应的来电终端建立预连接，以过滤通话连接请求持续时长很短的骚扰电话，减少与这种骚扰电话建立预连接的资源浪费。
在与通话连接请求对应的来电终端建立预连接后，由手机中的智能语音模块与来电者进行语音对话，手机中的智能语音模块配置对话管理功能，可与来电者进行多轮对话。
在一实施方式中，根据预设的对话规则与通话连接请求对应的来电者进行语音对话，可选地，可以预先设置对话内容以及对话顺序，由手机根据预先设置的对话内容以及对话顺序与来电者进行语音对话，例如，预先设置的对话内容以及对话顺序为：“你是谁？你找谁？你有什么事情？”。
在另一实施方式中，步骤S10中根据预设的对话规则与所述通话连接请求对应的来电者进行语音对话的步骤包括：
步骤S104，获取所述通话连接请求对应的来电者音频，基于语音端点检测技术从所述来电者音频中提取来电者语音，并将所述来电者语音转化成语音文本；
在与来电终端建立预连接后，获得来电者音频，对来电者音频进行语义识别，并根据预设规则以及识别的来电者语义进行语音回复，以实现与来电者的语音对话。可以由手机上智能语音模块首先提问，再获得来电者音频，也可以在检测到来电者发出声音、获得来电者音频后，根据对来电者音频的语义识别进行对应语音回复。
通过语音端点检测技术从来电者音频中提取来电者语音，避免因为太多静音导致资源浪费，并减少对语义识别的干扰，提升语义识别的准确性。语音端点检测技术指从连续的音频流中检测出有效语音段的技术，包括检测出有效语音段的起始点以及结束点，通过对音频流的能量大小以及变化轨迹的确定识别有效语音段，其中，若某段音频流为静音，则该段音频流的连续若干帧能量值持续维持在低水平，而有效语音段的端点（起始点以及结束点）则处于能量变化的临界点。
可选地，在基于语音端点检测技术从所述来电者音频中提取来电者语音的步骤之前，先对来电者音频进行噪声抑制，以提升语音端点检测的准确度。
本实施例中的来电者语音，即经语音端点检测获得的有效语音段，再将来电者语音转化成语音文本，有效语音段可能为一句或连续的多句话语，将其转化为对应的一句或多句语音文本。
步骤S105，从所述语音文本中提取第一关键词，根据所述第一关键词查询预置的对话库获得目标回复话语；
对语音文本进行分词以及语法分析，获得语音文本中各词语的的词性，并将词性为名词、动词以及形容词的词语作为第一关键词。预置的对话库中存储有定义好的对话数据，包括可能出现的问答对话，根据第一关键词在对话库中检索相似对话，并获得相似对话对应的回复作为目标回复话语。
这里的第一关键词，指来电者当前所说话语中的关键词。
步骤S106，将所述目标回复话语转换成语音输出。
将目标回复话语转换成语音，将该语音输出到来电终端，以实现与来电者的对话。
可选地，预置语音数据库，智能语音模块可根据采集的用户语音（手机机主语音）或预置语音数据库中的语音，生成模仿的自然人语音。可由智能语音模块模仿机主的说话方式，比如机主的声音特征，语调，语速等来产生模仿声音。由智能语音模块模仿人声，可以降低骚扰者的警惕，以防其编造语言欺骗智能语音模块。
步骤S20，从所述语音对话中提取所述来电者的意图特征，将所述意图特征输入预置的意图判别器，并获得所述意图判别器输出的判别结果；
从语音对话中提取意图关键词，将意图关键词表示为对应的特征向量，即本实施例中的意图特征，将意图特征输入意图判别器，由该意图判别器根据输入的意图特征进行意图判别，并输出意图判别结果，比如纯骚扰、推销等电话。
可选地，从语音对话中提取来电者的意图特征的步骤包括：
从所述语音对话中获得所述来电者语音，从所述来电者语音中提取第二关键词；将所述第二关键词与预置的意图库进行对比，获得所述第二关键词中的意图关键词，将所述意图关键词作为所述来电者的意图特征。
可从语音对话中获得来电者的所有语音，并从来电者语音中提取第二关键词，这里的第二关键词指从来电者的所有语音中提取的关键词；再从第二关键词中提取意图关键词，可以通过将第二关键词与预置的意图库中的意图词汇进行对比，从第二关键词中获取与意图库中意图词汇一致或为近义词的词作为意图关键词，将意图关键词表示为对应的特征向量，即来电者的意图特征。
意图判别器为预先用标注好的训练语料训练好的意图判别模型，用于根据意图特征对来电者意图进行分类、识别。
可选地，对于意图判别模型的训练，步骤S20之前包括：
步骤S21，获取训练样本和测试样本，其中，训练样本为经过意图标注的话语；
从预置语料库中获取通话文本数据（包含骚扰电话文本和正常通话文本）作为训练样本和测试样本来源，预置语料库中可包括直接从互联网中查找到的骚扰电话文本和正常通话文本，也可包括本机电话语音转文本后进行意图标注的文本。其中，训练样本用于优化模型的参数，而测试样本用于对建立的意图判别模型进行性能评价。
步骤S22，分别提取训练样本和测试样本的特征，并根据训练样本的特征，通过迭代算法计算出意图判别模型的最优模型参数，训练出含最优模型参数的意图判别模型；
意图判别模型是神经网络分类模型，可以为朴素贝叶斯分类器。初始情况下，意图判别模型的模型参数随机设定并未优化，可将每个训练样本或测试样本对应地表示为一个由一组特征组成的特征向量，可选地，因文字总数量的有限性，可以将每个字对应一个数字或号码（如从1到36000），实现训练样本或测试样本的数字化，将训练样本或测试样本各自的数字化形式作为各自对应的特征向量。
将训练样本对应的特征向量输入意图判别模型，通过优化方法迭代计算出最优的模型参数，训练出意图判别模型，将训练样本的特征输入意图判别模型中，并通过优化方法迭代计算出最优的模型参数，训练出意图判别模型，其中意图判别模型用于判断训练样本属于骚扰样本的概率。
其中，迭代算法包括梯度下降，共轭梯度法和拟牛顿法等。在可选实施中，可以通过上述任一迭代算法计算出意图判别模型的最优模型参数，训练出含最优模型参数的意图判别模型。此外，还可以采用其他方法分别提取训练样本和测试样本的特征，例如向量空间模型VSM、信息增益方法、期望交叉熵等。
步骤S23，基于所述测试样本的特征对意图判别模型进行准确性测试，基于测试结果调整意图判别模型的最优模型参数。
测试样本用于对意图判别模型进行准确性评价，并基于测试结果调整意图判别模型的最优模型参数，以提升判别模型的准确性。
步骤S30，在根据所述判别结果确定所述通话连接请求为骚扰电话时，挂断来电。
在判定为骚扰电话时，触发挂断电话，在判定为非骚扰电话时，接听来电。可选地，步骤S20之后包括：在根据所述判别结果确定所述通话连接请求为正常电话时，输出来电提醒，并以文字形式输出所述语音对话的内容。
文字输出智能语音模块与来电者的对话内容，用户可基于该对话内容进行后续对话，例如，对于在智能对话阶段已问的问题（如：你是谁，有什么事情），用户在接听电话后可不必再问，实现智能对话与用户接听对话的无缝对接，减少因智能助手验证导致的时间延迟导致的不满。
本实施例通过在检测到通话连接请求后，与所述通话连接请求对应的来电终端建立预连接，根据预设的对话规则与所述通话连接请求对应的来电者进行语音对话；从所述语音对话中提取所述来电者的意图特征，将所述意图特征输入预置的意图判别器，并获得所述意图判别器输出的判别结果；在根据所述判别结果确定所述通话连接请求为骚扰电话时，挂断来电，即通过预先与来电者进行语音对话，以语音内容为骚扰判断依据，可以有效地避免伪装成正常号码的骚扰诈骗来电，同时可以智能实现骚扰电话拦截。
进一步地，在本申请骚扰电话拦截方法的第二实施例中，步骤S10之后包括：
步骤S11，采集所述来电者的声纹信息，将所述来电者的声纹信息与预置声纹数据库中的声纹信息进行对比匹配，判断所述预置声纹数据库中是否存在与所述来电者的声纹信息匹配的目标声纹信息；
在日常生活中，常常存在熟人用更换的新电话号码打电话给用户，本实施例结合来电者的声纹信息识别“熟人新号”的情况，减少后续不必要的骚扰识别步骤。
预置声纹数据库中预存有非骚扰电话来电者的声纹信息，可以将历史来电者（已排除骚扰电话嫌疑）的声纹信息存入预置声纹数据库。
在智能语音模块与通话连接请求对应的来电者进行语音对话的过程中，可采集来电者声纹信息与预存的声纹信息进行对比，判断预置声纹数据库中是否存在与所述来电者的声纹信息匹配的目标声纹信息，其中，可以在检测到来电者语音的同时，采集来电者的声纹信息，执行步骤S11，进行声纹识别。
步骤S12，若所述预置声纹数据库中存在与所述来电者的声纹信息匹配的目标声纹信息，则输出来电提醒；
若有匹配的声纹信息，则判定来电不是骚扰电话，则无需进行后续的骚扰识别步骤，直接输出来电提醒，以提醒用户接听，避免不必要的验证，节约资源的同时提高了用户体验。
步骤S13，若所述预置声纹数据库中不存在与所述来电者的声纹信息匹配的目标声纹信息，则执行所述从所述语音对话中提取所述来电者的意图特征的步骤。
若无匹配的声纹信息，则说明不是用新电话号码的熟人，则需进行后续的骚扰识别步骤，以进一步准确拦截骚扰电话。
本实施例通过在与通话连接请求对应的来电者进行语音对话之时或之后，采集来电者的声纹信息，将来电者的声纹信息与预置声纹数据库中的声纹信息进行对比匹配，以间接判断来电者是否为熟人或非骚扰电话，在预置声纹数据库中存在与来电者的声纹信息匹配的目标声纹信息时，确定来电不是骚扰电话，反之，则需进一步进行骚扰识别判断，可以识别熟人用更换的新电话号码打电话给用户的情况并直接提醒用户接听，避免不必要的骚扰识别，节约资源的同时提高了用户体验。
此外，本申请还提供一种与上述骚扰电话拦截方法各步骤对应的骚扰电话拦截装置。
参照图3，图3为本申请骚扰电话拦截装置第一实施例的功能模块示意图。
在本实施例中，本申请骚扰电话拦截装置包括：
智能语音模块10，用于在检测到通话连接请求后，与所述通话连接请求对应的来电终端建立预连接，根据预设的对话规则与所述通话连接请求对应的来电者进行语音对话；
意图判别模块20，用于从所述语音对话中提取所述来电者的意图特征，将所述意图特征输入预置的意图判别器，并获得所述意图判别器输出的判别结果；
电话处理模块30，用于在根据所述判别结果确定所述通话连接请求为骚扰电话时，挂断来电。
进一步地，智能语音模块10，还用于获取所述通话连接请求对应的来电者音频，基于语音端点检测技术从所述来电者音频中提取来电者语音，并将所述来电者语音转化成语音文本；从所述语音文本中提取第一关键词，根据所述第一关键词查询预置的对话库获得目标回复话语；将所述目标回复话语转换成语音输出。
进一步地，本申请骚扰电话拦截装置还包括：
声纹判断模块，用于采集所述来电者的声纹信息，将所述来电者的声纹信息与预置声纹数据库中的声纹信息进行对比匹配，判断所述预置声纹数据库中是否存在与所述来电者的声纹信息匹配的目标声纹信息；
电话处理模块30，还用于若所述预置声纹数据库中存在与所述来电者的声纹信息匹配的目标声纹信息，则输出来电提醒；若所述预置声纹数据库中不存在与所述来电者的声纹信息匹配的目标声纹信息，则执行所述从所述语音对话中提取所述来电者的意图特征的步骤。
进一步地，本申请骚扰电话拦截装置还包括：
模型训练模块，用于获取训练样本和测试样本，其中，训练样本为经过意图标注的话语；分别提取训练样本和测试样本的特征，并根据训练样本的特征，通过迭代算法计算出意图判别模型的最优模型参数，训练出含最优模型参数的意图判别模型；基于所述测试样本的特征对意图判别模型进行准确性测试，基于测试结果调整意图判别模型的最优模型参数。
进一步地，所述意图判别模块20，还用于从所述语音对话中获得所述来电者语音，从所述来电者语音中提取第二关键词；将所述第二关键词与预置的意图库进行对比，获得所述第二关键词中的意图关键词，将所述意图关键词作为所述来电者的意图特征。
进一步地，智能语音模块10，还用于在检测到通话连接请求后，将所述通话连接请求对应的来电号码与白名单列表上号码进行比较，以判断所述通话连接请求对应的来电号码是否为白名单号码；若所述来电号码为白名单号码，则输出来电提醒；若所述来电号码不是白名单号码，则与所述通话连接请求对应的来电终端建立预连接。
进一步地，电话处理模块30，还用于在根据所述判别结果确定所述通话连接请求为正常电话时，输出来电提醒，并以文字形式输出所述语音对话的内容。
本申请还提出一种存储介质，其上存储有计算机可读指令。存储介质可以为非易失性可读存储介质，可以是图1的骚扰电话拦截设备中的存储器201，也可以是如ROM（Read-Only Memory，只读存储器）/RAM（Random Access Memory，随机存取存储器）、磁碟、光盘中的至少一种，存储介质包括若干指令用以使得一台具有处理器的设备(本申请实施例中的骚扰电话拦截设备，如手机，计算机等)执行本申请各个实施例的方法。
需要说明的是，在本文中，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者服务端不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者服务端所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括该要素的过程、方法、物品或者服务端中还存在另外的相同要素。
上述本申请实施例序号仅仅为了描述，不代表实施例的优劣。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。
以上仅为本申请的可选实施例，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种骚扰电话拦截方法，其中，所述骚扰电话拦截方法包括以下步骤：
    在检测到通话连接请求后，与所述通话连接请求对应的来电终端建立预连接，根据预设的对话规则与所述通话连接请求对应的来电者进行语音对话；
    从所述语音对话中提取所述来电者的意图特征，将所述意图特征输入预置的意图判别器，并获得所述意图判别器输出的判别结果；
    在根据所述判别结果确定所述通话连接请求为骚扰电话时，挂断来电；
    所述根据预设的对话规则与所述通话连接请求对应的来电者进行语音对话的步骤包括：
    获取所述通话连接请求对应的来电者音频，基于语音端点检测技术从所述来电者音频中提取来电者语音，并将所述来电者语音转化成语音文本；
    从所述语音文本中提取第一关键词，根据所述第一关键词查询预置的对话库获得目标回复话语；
    将所述目标回复话语转换成语音输出；
    所述从所述语音对话中提取所述来电者的意图特征的步骤包括：
    从所述语音对话中获得所述来电者语音，从所述来电者语音中提取第二关键词；
    将所述第二关键词与预置的意图库进行对比，获得所述第二关键词中的意图关键词，将所述意图关键词作为所述来电者的意图特征。
  2. 如权利要求1所述的骚扰电话拦截方法，其中，所述根据预设的对话规则与所述通话连接请求对应的来电者进行语音对话的步骤之后包括：
    采集所述来电者的声纹信息，将所述来电者的声纹信息与预置声纹数据库中的声纹信息进行对比匹配，判断所述预置声纹数据库中是否存在与所述来电者的声纹信息匹配的目标声纹信息；
    若所述预置声纹数据库中存在与所述来电者的声纹信息匹配的目标声纹信息，则输出来电提醒；
    若所述预置声纹数据库中不存在与所述来电者的声纹信息匹配的目标声纹信息，则执行所述从所述语音对话中提取所述来电者的意图特征的步骤。
  3. 如权利要求1所述的骚扰电话拦截方法，其中，所述从所述语音对话中提取所述来电者的意图特征，将所述意图特征输入预置的意图判别器的步骤之前包括：
    获取训练样本和测试样本，其中，训练样本为经过意图标注的话语；
    分别提取训练样本和测试样本的特征，并根据训练样本的特征，通过迭代算法计算出意图判别模型的最优模型参数，训练出含最优模型参数的意图判别模型；
    基于所述测试样本的特征对意图判别模型进行准确性测试，基于测试结果调整意图判别模型的最优模型参数。
  4. 如权利要求1所述的骚扰电话拦截方法，其中，所述在检测到通话连接请求后，与所述通话连接请求对应的来电终端建立预连接的步骤包括：
    在检测到通话连接请求后，将所述通话连接请求对应的来电号码与白名单列表上号码进行比较，以判断所述通话连接请求对应的来电号码是否为白名单号码；
    若所述来电号码为白名单号码，则输出来电提醒；
    若所述来电号码不是白名单号码，则与所述通话连接请求对应的来电终端建立预连接。
  5. 如权利要求1所述的骚扰电话拦截方法，其中，所述获得所述意图判别器输出的判别结果的步骤之后包括：
    在根据所述判别结果确定所述通话连接请求为正常电话时，输出来电提醒，并以文字形式输出所述语音对话的内容。
  6. 一种骚扰电话拦截装置，其中，所述骚扰电话拦截装置包括：
    智能语音模块，用于在检测到通话连接请求后，与所述通话连接请求对应的来电终端建立预连接，根据预设的对话规则与所述通话连接请求对应的来电者进行语音对话；
    意图判别模块，用于从所述语音对话中提取所述来电者的意图特征，将所述意图特征输入预置的意图判别器，并获得所述意图判别器输出的判别结果；
    电话处理模块，用于在根据所述判别结果确定所述通话连接请求为骚扰电话时，挂断来电；
    智能语音模块，还用于获取所述通话连接请求对应的来电者音频，基于语音端点检测技术从所述来电者音频中提取来电者语音，并将所述来电者语音转化成语音文本；从所述语音文本中提取第一关键词，根据所述第一关键词查询预置的对话库获得目标回复话语；将所述目标回复话语转换成语音输出；
    所述意图判别模块，还用于从所述语音对话中获得所述来电者语音，从所述来电者语音中提取第二关键词；将所述第二关键词与预置的意图库进行对比，获得所述第二关键词中的意图关键词，将所述意图关键词作为所述来电者的意图特征。
  7. 如权利要求6所述的骚扰电话拦截装置，其中，所述骚扰电话拦截装置还包括：
    声纹判断模块，用于采集所述来电者的声纹信息，将所述来电者的声纹信息与预置声纹数据库中的声纹信息进行对比匹配，判断所述预置声纹数据库中是否存在与所述来电者的声纹信息匹配的目标声纹信息；
    电话处理模块，还用于若所述预置声纹数据库中存在与所述来电者的声纹信息匹配的目标声纹信息，则输出来电提醒；若所述预置声纹数据库中不存在与所述来电者的声纹信息匹配的目标声纹信息，则执行所述从所述语音对话中提取所述来电者的意图特征的步骤。
  8. 如权利要求6所述的骚扰电话拦截装置，其中，所述骚扰电话拦截装置还包括：模型训练模块，用于获取训练样本和测试样本，其中，训练样本为经过意图标注的话语；分别提取训练样本和测试样本的特征，并根据训练样本的特征，通过迭代算法计算出意图判别模型的最优模型参数，训练出含最优模型参数的意图判别模型；基于所述测试样本的特征对意图判别模型进行准确性测试，基于测试结果调整意图判别模型的最优模型参数。
  9. 如权利要求6所述的骚扰电话拦截装置，其中，所述骚扰电话拦截装置还包括：智能语音模块，还用于在检测到通话连接请求后，将所述通话连接请求对应的来电号码与白名单列表上号码进行比较，以判断所述通话连接请求对应的来电号码是否为白名单号码；若所述来电号码为白名单号码，则输出来电提醒；若所述来电号码不是白名单号码，则与所述通话连接请求对应的来电终端建立预连接。
  10. 如权利要求6所述的骚扰电话拦截装置，其中，所述骚扰电话拦截装置还包括：电话处理模块，还用于在根据所述判别结果确定所述通话连接请求为正常电话时，输出来电提醒，并以文字形式输出所述语音对话的内容。
  11. 一种骚扰电话拦截设备，其中，所述骚扰电话拦截设备包括处理器、存储器、以及存储在所述存储器上并可被所述处理器执行的计算机可读指令，其中所述计算机可读指令被所述处理器执行时，实现如下步骤：
    在检测到通话连接请求后，与所述通话连接请求对应的来电终端建立预连接，根据预设的对话规则与所述通话连接请求对应的来电者进行语音对话；
    从所述语音对话中提取所述来电者的意图特征，将所述意图特征输入预置的意图判别器，并获得所述意图判别器输出的判别结果；
    在根据所述判别结果确定所述通话连接请求为骚扰电话时，挂断来电；
    所述根据预设的对话规则与所述通话连接请求对应的来电者进行语音对话的步骤包括：
    获取所述通话连接请求对应的来电者音频，基于语音端点检测技术从所述来电者音频中提取来电者语音，并将所述来电者语音转化成语音文本；
    从所述语音文本中提取第一关键词，根据所述第一关键词查询预置的对话库获得目标回复话语；
    将所述目标回复话语转换成语音输出；
    所述从所述语音对话中提取所述来电者的意图特征的步骤包括：
    从所述语音对话中获得所述来电者语音，从所述来电者语音中提取第二关键词；
    将所述第二关键词与预置的意图库进行对比，获得所述第二关键词中的意图关键词，将所述意图关键词作为所述来电者的意图特征。
  12. 如权利要求11所述的骚扰电话拦截设备，其中，所述计算机可读指令被所述处理器执行时，还实现以下步骤：
    采集所述来电者的声纹信息，将所述来电者的声纹信息与预置声纹数据库中的声纹信息进行对比匹配，判断所述预置声纹数据库中是否存在与所述来电者的声纹信息匹配的目标声纹信息；
    若所述预置声纹数据库中存在与所述来电者的声纹信息匹配的目标声纹信息，则输出来电提醒；
    若所述预置声纹数据库中不存在与所述来电者的声纹信息匹配的目标声纹信息，则执行所述从所述语音对话中提取所述来电者的意图特征的步骤。
  13. 如权利要求11所述的骚扰电话拦截设备，其中，所述计算机可读指令被所述处理器执行时，还实现以下步骤：
    获取训练样本和测试样本，其中，训练样本为经过意图标注的话语；
    分别提取训练样本和测试样本的特征，并根据训练样本的特征，通过迭代算法计算出意图判别模型的最优模型参数，训练出含最优模型参数的意图判别模型；
    基于所述测试样本的特征对意图判别模型进行准确性测试，基于测试结果调整意图判别模型的最优模型参数。
  14. 如权利要求11所述的骚扰电话拦截设备，其中，所述计算机可读指令被所述处理器执行时，还实现以下步骤：
    在检测到通话连接请求后，将所述通话连接请求对应的来电号码与白名单列表上号码进行比较，以判断所述通话连接请求对应的来电号码是否为白名单号码；
    若所述来电号码为白名单号码，则输出来电提醒；
    若所述来电号码不是白名单号码，则与所述通话连接请求对应的来电终端建立预连接。
  15. 如权利要求11所述的骚扰电话拦截设备，其中，所述计算机可读指令被所述处理器执行时，还实现以下步骤：
    在根据所述判别结果确定所述通话连接请求为正常电话时，输出来电提醒，并以文字形式输出所述语音对话的内容。
  16. 一种存储介质，其中，所述存储介质上存储有计算机可读指令，其中所述计算机可读指令被处理器执行时，实现如下步骤：
    在检测到通话连接请求后，与所述通话连接请求对应的来电终端建立预连接，根据预设的对话规则与所述通话连接请求对应的来电者进行语音对话；
    从所述语音对话中提取所述来电者的意图特征，将所述意图特征输入预置的意图判别器，并获得所述意图判别器输出的判别结果；
    在根据所述判别结果确定所述通话连接请求为骚扰电话时，挂断来电；
    所述根据预设的对话规则与所述通话连接请求对应的来电者进行语音对话的步骤包括：
    获取所述通话连接请求对应的来电者音频，基于语音端点检测技术从所述来电者音频中提取来电者语音，并将所述来电者语音转化成语音文本；
    从所述语音文本中提取第一关键词，根据所述第一关键词查询预置的对话库获得目标回复话语；
    将所述目标回复话语转换成语音输出；
    所述从所述语音对话中提取所述来电者的意图特征的步骤包括：
    从所述语音对话中获得所述来电者语音，从所述来电者语音中提取第二关键词；
    将所述第二关键词与预置的意图库进行对比，获得所述第二关键词中的意图关键词，将所述意图关键词作为所述来电者的意图特征。
  17. 如权利要求16所述的存储介质，其中，所述计算机可读指令被处理器执行时，还实现如下步骤：
    采集所述来电者的声纹信息，将所述来电者的声纹信息与预置声纹数据库中的声纹信息进行对比匹配，判断所述预置声纹数据库中是否存在与所述来电者的声纹信息匹配的目标声纹信息；
    若所述预置声纹数据库中存在与所述来电者的声纹信息匹配的目标声纹信息，则输出来电提醒；
    若所述预置声纹数据库中不存在与所述来电者的声纹信息匹配的目标声纹信息，则执行所述从所述语音对话中提取所述来电者的意图特征的步骤。
  18. 如权利要求16所述的存储介质，其中，所述计算机可读指令被处理器执行时，还实现如下步骤：
    获取训练样本和测试样本，其中，训练样本为经过意图标注的话语；
    分别提取训练样本和测试样本的特征，并根据训练样本的特征，通过迭代算法计算出意图判别模型的最优模型参数，训练出含最优模型参数的意图判别模型；
    基于所述测试样本的特征对意图判别模型进行准确性测试，基于测试结果调整意图判别模型的最优模型参数。
  19. 如权利要求16所述的存储介质，其中，所述计算机可读指令被处理器执行时，还实现如下步骤：
    在检测到通话连接请求后，将所述通话连接请求对应的来电号码与白名单列表上号码进行比较，以判断所述通话连接请求对应的来电号码是否为白名单号码；
    若所述来电号码为白名单号码，则输出来电提醒；
    若所述来电号码不是白名单号码，则与所述通话连接请求对应的来电终端建立预连接。
  20. 如权利要求16所述的存储介质，其中，所述计算机可读指令被处理器执行时，还实现如下步骤：
    在根据所述判别结果确定所述通话连接请求为正常电话时，输出来电提醒，并以文字形式输出所述语音对话的内容。
PCT/CN2019/121724 2019-04-12 2019-11-28 骚扰电话拦截方法、装置、设备及存储介质 WO2020207035A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910301956.8 2019-04-12
CN201910301956.8A CN110149441A (zh) 2019-04-12 2019-04-12 骚扰电话拦截方法、装置、设备及存储介质

Publications (1)

Publication Number Publication Date
WO2020207035A1 true WO2020207035A1 (zh) 2020-10-15

Family

ID=67589821

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/121724 WO2020207035A1 (zh) 2019-04-12 2019-11-28 骚扰电话拦截方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN110149441A (zh)
WO (1) WO2020207035A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114125155A (zh) * 2021-11-15 2022-03-01 天津市国瑞数码安全系统股份有限公司 一种基于大数据分析的骚扰电话检测方法及系统
CN114765645A (zh) * 2020-12-30 2022-07-19 上海博泰悦臻网络技术服务有限公司 一种来电处理方法及装置
US20220277740A1 (en) * 2021-02-26 2022-09-01 Walmart Apollo, Llc Methods and apparatus for improving search retrieval using inter-utterance context

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110149441A (zh) * 2019-04-12 2019-08-20 深圳壹账通智能科技有限公司 骚扰电话拦截方法、装置、设备及存储介质
CN112866470A (zh) * 2019-11-11 2021-05-28 宇龙计算机通信科技(深圳)有限公司 来电处理方法、装置、电子设备及介质
CN111092995A (zh) * 2020-01-02 2020-05-01 苏州思必驰信息科技有限公司 辅助接听电话的方法、电子设备及存储介质
CN111246008A (zh) * 2020-01-10 2020-06-05 三星电子（中国）研发中心 一种电话助理的实现方法、系统及装置
CN111343347A (zh) * 2020-03-19 2020-06-26 上海尊源通讯技术有限公司 一种通信装置、通信系统以及通信控制方法
CN111465021B (zh) * 2020-04-01 2023-06-09 北京中亦安图科技股份有限公司 基于图的骚扰电话识别模型构建方法
CN111683174B (zh) * 2020-06-01 2021-05-04 信雅达科技股份有限公司 来电处理方法、装置及系统
CN111988461A (zh) * 2020-06-29 2020-11-24 北京捷通华声科技股份有限公司 来电拦截方法及装置
CN112671968A (zh) * 2020-12-16 2021-04-16 平安普惠企业管理有限公司 骚扰电话的拦截方法、装置、计算机设备及存储介质
CN113037914A (zh) * 2021-03-01 2021-06-25 北京百度网讯科技有限公司 用于处理呼入电话的方法、相关装置及计算机程序产品
CN113726941A (zh) * 2021-08-30 2021-11-30 平安普惠企业管理有限公司 基于人工智能的骚扰电话监控方法、装置、设备及介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050243975A1 (en) * 2004-04-28 2005-11-03 International Business Machines Corporation Method and system of determining unsolicited callers
CN105007361A (zh) * 2015-06-13 2015-10-28 安徽味唯网络科技有限公司 一种智能自动屏蔽骚扰电话的方法
CN109525700A (zh) * 2018-12-25 2019-03-26 出门问问信息科技有限公司 来电识别方法、装置、计算机设备及可读存储介质
CN110149441A (zh) * 2019-04-12 2019-08-20 深圳壹账通智能科技有限公司 骚扰电话拦截方法、装置、设备及存储介质

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105187667A (zh) * 2015-10-09 2015-12-23 小米科技有限责任公司 来电处理方法及装置
CN105744057B (zh) * 2016-01-21 2019-02-22 平安科技（深圳）有限公司 一种智能语音对话交互方法和装置
CN106850931A (zh) * 2017-01-10 2017-06-13 捷开通讯(深圳)有限公司 防骚扰电话的方法及移动智能终端

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050243975A1 (en) * 2004-04-28 2005-11-03 International Business Machines Corporation Method and system of determining unsolicited callers
CN105007361A (zh) * 2015-06-13 2015-10-28 安徽味唯网络科技有限公司 一种智能自动屏蔽骚扰电话的方法
CN109525700A (zh) * 2018-12-25 2019-03-26 出门问问信息科技有限公司 来电识别方法、装置、计算机设备及可读存储介质
CN110149441A (zh) * 2019-04-12 2019-08-20 深圳壹账通智能科技有限公司 骚扰电话拦截方法、装置、设备及存储介质

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114765645A (zh) * 2020-12-30 2022-07-19 上海博泰悦臻网络技术服务有限公司 一种来电处理方法及装置
US20220277740A1 (en) * 2021-02-26 2022-09-01 Walmart Apollo, Llc Methods and apparatus for improving search retrieval using inter-utterance context
US11715469B2 (en) * 2021-02-26 2023-08-01 Walmart Apollo, Llc Methods and apparatus for improving search retrieval using inter-utterance context
CN114125155A (zh) * 2021-11-15 2022-03-01 天津市国瑞数码安全系统股份有限公司 一种基于大数据分析的骚扰电话检测方法及系统

Also Published As

Publication number Publication date
CN110149441A (zh) 2019-08-20

Similar Documents

Publication Publication Date Title
WO2020207035A1 (zh) 骚扰电话拦截方法、装置、设备及存储介质
US20230351536A1 (en) Detection and prevention of inmate to inmate message relay
WO2020213996A1 (en) Method and apparatus for interrupt detection
WO2015068947A1 (ko) 녹취된 음성 데이터에 대한 핵심어 추출 기반 발화 내용 파악 시스템과, 이 시스템을 이용한 인덱싱 방법 및 발화 내용 파악 방법
WO2021112642A1 (en) Voice user interface
US10621974B2 (en) Updating a voice template
CN109873907B (zh) 呼叫处理方法、装置、计算机设备及存储介质
WO2020139058A1 (en) Cross-device voiceprint recognition
WO2018128238A1 (ko) 디스플레이 장치를 이용한 가상 상담 시스템 및 방법
CN110310663A (zh) 违规话术检测方法、装置、设备及计算机可读存储介质
WO2015178600A1 (en) Speech recognition method and apparatus using device information
WO2019208860A1 (ko) 음성 인식 기술을 이용한 다자간 대화 기록/출력 방법 및 이를 위한 장치
WO2021251539A1 (ko) 인공신경망을 이용한 대화형 메시지 구현 방법 및 그 장치
WO2019172734A2 (ko) 데이터 마이닝 장치, 이를 이용하는 음성인식 방법 및 시스템
WO2015163684A1 (ko) 적어도 하나의 의미론적 유닛의 집합을 개선하기 위한 방법, 장치 및 컴퓨터 판독 가능한 기록 매체
WO2020042463A1 (zh) 基于生物特征识别的门禁解锁方法、装置、设备和介质
WO2022080774A1 (ko) 말 장애 평가 장치, 방법 및 프로그램
CN109688276A (zh) 一种基于人工智能技术的来电过滤系统及其方法
WO2020082766A1 (zh) 输入法的联想方法、装置、设备及可读存储介质
WO2020130498A1 (en) A command based interactive system and a method thereof
WO2020235910A1 (en) Text reconstruction system and method thereof
WO2023063718A1 (en) Method and system for device feature analysis to improve user experience
KR102114365B1 (ko) 음성인식 방법 및 장치
CN108040185B (zh) 一种识别骚扰电话的方法及设备
WO2019156427A1 (ko) 발화된 단어에 기초하여 화자를 식별하기 위한 방법 및 그 장치, 문맥 기반 음성 모델 관리 장치 및 그 방법

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: 19923907

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 02/02/2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19923907

Country of ref document: EP

Kind code of ref document: A1