CN110942783B - Group call type crank call classification method based on audio multistage clustering - Google Patents
Group call type crank call classification method based on audio multistage clustering Download PDFInfo
- Publication number
- CN110942783B CN110942783B CN201910978660.XA CN201910978660A CN110942783B CN 110942783 B CN110942783 B CN 110942783B CN 201910978660 A CN201910978660 A CN 201910978660A CN 110942783 B CN110942783 B CN 110942783B
- Authority
- CN
- China
- Prior art keywords
- audio
- clustering
- comparison
- group
- group call
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
- G10L25/51—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/60—Information retrieval; Database structures therefor; File system structures therefor of audio data
- G06F16/65—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/60—Information retrieval; Database structures therefor; File system structures therefor of audio data
- G06F16/68—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/686—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using information manually generated, e.g. tags, keywords, comments, title or artist information, time, location or usage information, user ratings
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/26—Speech to text systems
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04M—TELEPHONIC COMMUNICATION
- H04M3/00—Automatic or semi-automatic exchanges
- H04M3/22—Arrangements for supervision, monitoring or testing
Abstract
The invention relates to a group call type crank call classification method based on audio multistage clustering, which comprises the following steps: s100, dividing an audio pool comprising a plurality of audio data into a plurality of equal groups, sequentially performing feature extraction and feature comparison on each group, and further performing cluster analysis to obtain audio clusters; s200, voice transcription is carried out on the audio, and keyword library retrieval comparison is carried out on the text subjected to voice transcription to obtain a keyword comparison result; s300, performing audio library retrieval comparison on the audio clustering to obtain an audio clustering result; and S400, merging and analyzing the keyword comparison result and the audio clustering result to obtain the automatically classified group call type crank call. The invention has the beneficial effects that: the group call type harassing call can be effectively detected and found; the method combines the means of key words, text transcription and the like, realizes automatic classification of the crank calls, saves labor cost and improves efficiency.
Description
Technical Field
The invention relates to the field of audio identification, in particular to a group call type harassing call classification method based on audio multistage clustering.
Background
At present, various technical means are provided in China to realize the detection of harassing calls, wherein aiming at group call type calls, the detection and the discovery are carried out through audio characteristic comparison and audio clustering technology. However, along with the continuous deepening of the disturbance audio data treatment work, lawless persons continuously update the techniques to confront the disturbance audio data simultaneously, and great challenges are brought to the current treatment work.
In addition, the conventional audio feature comparison and audio clustering technology is to perform clustering analysis once in a group after audio data are coarsely grouped, so that the clustering analysis of the audio data is completed. The quality of the clustering analysis result completely depends on the quality of the same audio data grouping, and as the audio grouping is random grouping, part of the same harassing audios cannot be grouped into one group and cannot be aggregated together, so that the difficulty in harassing audio treatment is objectively increased.
With the increasingly severe form of countermeasure, the content of the group call type telephone is changeable, the treatment workload is increased, and at present, no content-based classification algorithm is used for the detection of the group call type telephone process. And the prior art has the following defects:
(1) according to the traditional clustering algorithm, certain defects of group call type crank calls are found, and all crank calls cannot be detected and found;
(2) the traditional crank call discovery does not combine score judgment such as key words and text transcription, and crank calls cannot be effectively classified.
Disclosure of Invention
The invention aims to solve at least one of technical problems in the prior art, and provides a group call type harassing call classification method based on audio multistage clustering, which realizes automatic classification, saves labor cost and improves efficiency.
The technical scheme of the invention comprises a method for classifying group call type crank calls based on audio multistage clustering, which is characterized by comprising the following steps: s100, dividing an audio pool comprising a plurality of audio data into a plurality of equivalent groups, sequentially performing feature extraction and feature comparison on each group, and further performing clustering analysis to obtain audio clusters; s200, voice transcription is carried out on the audio, and keyword library retrieval comparison is carried out on the text subjected to voice transcription to obtain a keyword comparison result; s300, performing audio library retrieval comparison on the audio clustering to obtain an audio clustering result; and S400, combining and analyzing the keyword comparison result and the audio clustering result to obtain the automatically classified group call type harassing calls.
According to the method for classifying the group call type crank calls based on the audio multistage clustering, S100 specifically comprises the following steps: s110, inputting an audio pool comprising a plurality of audio data, and randomly dividing the audio pool into N groups, wherein the maximum number of each group is M audio; s120, after feature extraction and feature comparison, each group of audio is subjected to clustering analysis according to a threshold value; s130, aggregating the clustering results of each group again to form a final clustering result; and S140, circularly executing the steps S110 to S130 until the clustering processing of all the audio data is completed.
According to the method for classifying the group call type crank calls based on the audio multistage cluster, S120 further comprises the following steps: if the audio clustering is successful, the secondary analysis is not participated; and if the audio clustering is unsuccessful, putting the audio into an audio pool again, and executing the steps S110-140 again.
According to the method for classifying the group call type crank calls based on the audio multistage cluster, M, N and the cycle execution times can be set in a user-defined mode.
According to the method for classifying the group call type crank calls based on the audio multistage clustering, the S200 specifically comprises the following steps: s210, comparing the audio with a harassment audio library to obtain an audio similarity score; s220, audio clustering is carried out on the audio to obtain audio clustering result information
According to the method for classifying the group call type crank calls based on the audio multistage cluster, the S300 specifically comprises the following steps: and (4) transcribing the audio files in the audio clusters into texts according to a set proportion, and judging whether the texts are harassing calls or not by combining a keyword identification method to obtain a keyword comparison result.
According to the method for classifying the group call type crank calls based on the audio multistage clustering, S400 specifically comprises the following steps: s410, summarizing and associating the results of S200 and S300 to obtain the audio comparison similarity score and the audio cluster number of each audio file; and S420, aiming at each audio cluster, judging and marking the corresponding harassment type according to the success of the audio comparison result, the comparison of the audio comparison similarity score with the set threshold value and the keyword comparison result, and automatically classifying according to the harassment type judgment and marking.
The beneficial effects of the invention are as follows: the group call type harassing call can be effectively detected and found; the method combines the means of key words, text transcription and the like, realizes automatic classification of the crank calls, saves labor cost and improves efficiency.
Drawings
The invention is further described below with reference to the accompanying drawings and examples;
FIG. 1 illustrates an overall flow diagram according to an embodiment of the invention;
FIG. 2 is a flow diagram illustrating packet clustering according to an embodiment of the present invention;
fig. 3 is a general flow chart of the classification of a group call type crank call according to the embodiment of the invention.
Detailed Description
Reference will now be made in detail to the present preferred embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like reference numerals refer to like elements throughout.
In the description of the present invention, the meaning of several is one or more, and the meaning of a plurality is more than two, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated or implicitly indicating the precedence of the technical features indicated.
In the description of the present invention, unless otherwise explicitly defined, such as set forth, a person skilled in the art can reasonably determine the specific meaning of the above-mentioned words in the present invention in combination with the details of the technical solution.
The noun explains:
audio library: and intercepting the audio clips with fixed lengths, and converting the audio clips into a model library for audio matching after feature extraction.
Clustering: the same feature audio is aggregated into one category.
FIG. 1 shows a general flow diagram according to an embodiment of the invention. The process includes steps S100 to S400 as follows: s100, dividing an audio pool comprising a plurality of audio data into a plurality of equal groups, sequentially performing feature extraction and feature comparison on each group, and further performing cluster analysis to obtain audio clusters; s200, voice transcription is carried out on the audio, and keyword library retrieval comparison is carried out on the text subjected to voice transcription to obtain a keyword comparison result; s300, performing audio library retrieval comparison on the audio clustering to obtain an audio clustering result; and S400, merging and analyzing the keyword comparison result and the audio clustering result to obtain the automatically classified group call type crank call.
FIG. 2 is a flow chart illustrating packet clustering according to an embodiment of the present invention. The method specifically comprises the following steps:
the method comprises the steps of finding group call type crank calls in mass data by utilizing the characteristics of batch call of the group call type crank calls and comprehensively utilizing relevant technologies such as audio comparison, audio clustering and voice-to-text conversion, then comprehensively scoring by combining and combining a keyword analysis technology, a voice-to-text technology and the like, and finally outputting crank call classification information.
In the analysis process, an audio multistage clustering method is adopted in the audio clustering process. The central idea is as follows: averagely grouping the audio data, carrying out cluster analysis on samples in a group, and storing the analysis result; on the basis of the last clustering analysis result, recombining and randomly grouping the audio data which are not successfully clustered, clustering and analyzing the samples in the groups, and merging the analysis result into the analysis result of the last time; and cycling sequentially until no new clusters are generated.
The specific process is described as follows:
(1) audio data is input and randomly divided into N groups, each group having a maximum of M audio frequencies.
(2) After feature extraction and feature comparison, each group is subjected to clustering analysis according to a threshold value, and if clustering is successful, secondary analysis is not involved; otherwise, the audio pool is put again to wait for the next analysis.
(3) And aggregating the clustering results of each group again to form a final clustering result.
(4) And (4) placing the audio failed in clustering into an audio pool for scattering, and executing the step (1) again after recombining.
(5) And repeating iteration until no new clustering result is generated.
After the method, all the group call type crank calls can be found and classified. In fact, the desired effect may be achieved over an infinite number of iterations, but at the expense of a significant amount of time. The grouping, the size of the group and the number of iterations are usually set for limitation, so that the harassing group call can be found as much as possible, and the analysis time can be prolonged.
Fig. 3 is a general flow chart of the classification of a group call type crank call according to the embodiment of the invention. The concrete steps are summarized as follows:
(1) and comparing the audio file with a harassment audio library to obtain an audio similarity score.
(2) And carrying out audio clustering on the audio files to obtain audio clustering result information.
(3) And (4) transferring the audio files in the audio cluster into texts according to a certain proportion, and judging whether the texts are harassing calls or not by combining a keyword recognition technology.
(4) And comprehensively analyzing the results of the steps and automatically classifying the crank calls.
Based on the embodiment of fig. 3, the invention further discloses a classification counting scheme of group call type harassing calls, which comprises the following steps:
and the classification module collects and associates the results of the audio comparison module and the audio clustering module to obtain the audio comparison similarity score and the audio clustering number of each audio file.
For each audio cluster, three cases can be classified according to whether the audio comparison result is successful, namely: comparing all the audio files in the category with the harassment audio library successfully, comparing partial audio files in the category with the harassment audio library successfully, and comparing all the audio files in the category with the harassment audio library unsuccessfully, and aiming at the former two conditions, the audio files in the category can be summarized as containing harassment audio files.
And aiming at each audio cluster, comparing the similarity score with a set threshold value according to audio comparison. If the audio similarity is greater than or equal to the threshold value, marking the audio similarity as a disturbance type I; conversely, if the audio similarity is less than the threshold, the audio is marked as disturbance type II; in addition, other audio files which are not successfully compared with the harassment audio library in the class are marked as harassment type III.
For each audio cluster, if the cluster does not contain a harassment audio file, the judgment needs to be carried out through the text content of the harassment audio file, currently, a keyword analysis technology is adopted to judge whether the harassment audio file is a harassment audio file, and if the harassment audio file is a harassment type IV.
By integrating the flows, it can be seen that the disturbance type I, the disturbance type II and the disturbance type III have strong relevance, and the disturbance type IV may be a new means for lawbreakers.
The present invention is further described below with reference to the above figures and flow. The following embodiments are merely used to more clearly illustrate the flow scheme of data analysis, and should not be taken as limiting the scope of the present invention.
The embodiments of the present invention have been described in detail with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the gist of the present invention.
Claims (6)
1. A method for classifying group call type crank calls based on audio multistage clustering is characterized by comprising the following steps:
s100, dividing an audio pool comprising a plurality of audio data into a plurality of equal groups, sequentially performing feature extraction and feature comparison on each group, and further performing cluster analysis to obtain audio clusters; the S100 specifically includes: s110, inputting an audio pool comprising a plurality of audio data, and randomly dividing the audio pool into N groups, wherein the maximum number of each group is M audio; s120, after feature extraction and feature comparison, each group of audio is subjected to clustering analysis according to a threshold value; s130, aggregating the clustering results of each group again to form a final clustering result; s140, circularly executing the steps S110-S130 until all audio data are clustered;
s200, voice transcription is carried out on the audio, and keyword library retrieval comparison is carried out on the text subjected to voice transcription to obtain a keyword comparison result;
s300, performing audio library retrieval comparison on the audio clustering to obtain an audio clustering result;
and S400, merging and analyzing the keyword comparison result and the audio clustering result to obtain the automatically classified group call type crank call.
2. The method for classifying group call-type harassing calls based on audio multistage clustering as claimed in claim 1, wherein said S120 further comprises:
if the audio clustering is successful, the secondary analysis is not participated;
and if the audio clustering is unsuccessful, putting the audio into an audio pool again, and executing the steps S110-140 again.
3. The method for classifying group call type crank calls based on audio multistage clustering according to claim 1, wherein: wherein M, N and the number of loop executions can be set by user.
4. The method for classifying group call type crank calls based on audio multistage clustering according to claim 1, wherein the S200 specifically comprises: and (4) transcribing the audio files in the audio clusters into texts according to a set proportion, and judging whether the texts are harassing calls or not by combining a keyword identification method to obtain a keyword comparison result.
5. The method for classifying group-call-type harassing calls based on audio multi-level clustering according to claim 1, wherein the S300 specifically comprises:
s310, comparing the audio with a harassment audio library to obtain an audio similarity score;
and S320, carrying out audio clustering on the audio to obtain audio clustering result information.
6. The method for classifying group call type crank calls based on audio multistage clustering according to claim 1, wherein the S400 specifically comprises:
s410, summarizing and associating the results of S200 and S300 to obtain the audio comparison similarity score and the audio cluster number of each audio file;
and S420, aiming at each audio cluster, judging and marking the corresponding harassment type according to the success of the audio comparison result, the comparison of the audio comparison similarity score with the set threshold value and the keyword comparison result, and automatically classifying according to the harassment type judgment and marking.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910978660.XA CN110942783B (en) | 2019-10-15 | 2019-10-15 | Group call type crank call classification method based on audio multistage clustering |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910978660.XA CN110942783B (en) | 2019-10-15 | 2019-10-15 | Group call type crank call classification method based on audio multistage clustering |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110942783A CN110942783A (en) | 2020-03-31 |
CN110942783B true CN110942783B (en) | 2022-06-17 |
Family
ID=69905809
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910978660.XA Active CN110942783B (en) | 2019-10-15 | 2019-10-15 | Group call type crank call classification method based on audio multistage clustering |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110942783B (en) |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102243641A (en) * | 2011-04-29 | 2011-11-16 | 西安交通大学 | Method for efficiently clustering massive data |
CN102663141A (en) * | 2012-05-17 | 2012-09-12 | 西安交通大学 | Multi-channel quantification and hierarchical clustering method based on multi-core parallel computation |
CN104240712A (en) * | 2014-09-30 | 2014-12-24 | 武汉大学深圳研究院 | Three-dimensional audio multichannel grouping and clustering coding method and three-dimensional audio multichannel grouping and clustering coding system |
CN108040185A (en) * | 2017-12-06 | 2018-05-15 | 福建天晴数码有限公司 | A kind of method and apparatus for identifying harassing call |
US10003688B1 (en) * | 2018-02-08 | 2018-06-19 | Capital One Services, Llc | Systems and methods for cluster-based voice verification |
CN109033084A (en) * | 2018-07-26 | 2018-12-18 | 国信优易数据有限公司 | A kind of semantic hierarchies tree constructing method and device |
CN109451182A (en) * | 2018-10-19 | 2019-03-08 | 北京邮电大学 | A kind of detection method and device of fraudulent call |
CN109600752A (en) * | 2018-11-28 | 2019-04-09 | 国家计算机网络与信息安全管理中心 | A kind of method and apparatus of depth cluster swindle detection |
CN109949798A (en) * | 2019-01-03 | 2019-06-28 | 刘伯涵 | Commercial detection method and device based on audio |
CN110312047A (en) * | 2019-06-24 | 2019-10-08 | 深圳市趣创科技有限公司 | The method and device of automatic shield harassing call |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10373611B2 (en) * | 2014-01-03 | 2019-08-06 | Gracenote, Inc. | Modification of electronic system operation based on acoustic ambience classification |
-
2019
- 2019-10-15 CN CN201910978660.XA patent/CN110942783B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102243641A (en) * | 2011-04-29 | 2011-11-16 | 西安交通大学 | Method for efficiently clustering massive data |
CN102663141A (en) * | 2012-05-17 | 2012-09-12 | 西安交通大学 | Multi-channel quantification and hierarchical clustering method based on multi-core parallel computation |
CN104240712A (en) * | 2014-09-30 | 2014-12-24 | 武汉大学深圳研究院 | Three-dimensional audio multichannel grouping and clustering coding method and three-dimensional audio multichannel grouping and clustering coding system |
CN108040185A (en) * | 2017-12-06 | 2018-05-15 | 福建天晴数码有限公司 | A kind of method and apparatus for identifying harassing call |
US10003688B1 (en) * | 2018-02-08 | 2018-06-19 | Capital One Services, Llc | Systems and methods for cluster-based voice verification |
CN109033084A (en) * | 2018-07-26 | 2018-12-18 | 国信优易数据有限公司 | A kind of semantic hierarchies tree constructing method and device |
CN109451182A (en) * | 2018-10-19 | 2019-03-08 | 北京邮电大学 | A kind of detection method and device of fraudulent call |
CN109600752A (en) * | 2018-11-28 | 2019-04-09 | 国家计算机网络与信息安全管理中心 | A kind of method and apparatus of depth cluster swindle detection |
CN109949798A (en) * | 2019-01-03 | 2019-06-28 | 刘伯涵 | Commercial detection method and device based on audio |
CN110312047A (en) * | 2019-06-24 | 2019-10-08 | 深圳市趣创科技有限公司 | The method and device of automatic shield harassing call |
Non-Patent Citations (2)
Title |
---|
Keyword Extraction and Clustering for Document Recommendation in Conversations;Maryam Habibi;《IEEE/ACM Transactions on Audio, Speech, and Language Processing》;20150219;全文 * |
基于数据挖掘技术实现骚扰电话识别;刘剑;《中国优秀硕士学位论文全文数据库》;20110815(第8期);I138-214 * |
Also Published As
Publication number | Publication date |
---|---|
CN110942783A (en) | 2020-03-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
KR102315732B1 (en) | Speech recognition method, device, apparatus, and storage medium | |
CN102882838B (en) | Authentication method and system applying verification code mechanism | |
CN107733869B (en) | Equipment identification method and device | |
US10019492B2 (en) | Stop word identification method and apparatus | |
CN105893351B (en) | Audio recognition method and device | |
CN105653620B (en) | Log analysis method and device of intelligent question-answering system | |
CN110659175A (en) | Log trunk extraction method, log trunk classification method, log trunk extraction equipment and log trunk storage medium | |
CN110866249A (en) | Method and device for dynamically detecting malicious code and electronic equipment | |
CN110765266B (en) | Method and system for merging similar dispute focuses of referee documents | |
CN113742292B (en) | Multithread data retrieval and access method of retrieved data based on AI technology | |
CN110942783B (en) | Group call type crank call classification method based on audio multistage clustering | |
CN110851675A (en) | Data extraction method, device and medium | |
CN113723501A (en) | Maximum diversity clustering construction method of pathogenic microorganism reference knowledge base | |
CN107133321B (en) | Method and device for analyzing search characteristics of page | |
WO2012159320A1 (en) | Method and device for clustering large-scale image data | |
CN111026940A (en) | Network public opinion and risk information monitoring system and electronic equipment for power grid electromagnetic environment | |
CN108647201B (en) | Classification identification method and system based on mobile application | |
CN113704287A (en) | Big data based data comparison analysis screening system and method | |
CN112464648A (en) | Industry standard blank feature recognition system and method based on multi-source data analysis | |
CN109240988B (en) | Method and system for preventing big data storage system from entering access imbalance state | |
CN113159178A (en) | Problem expansion method, device, server and medium | |
Li et al. | Multi-label classification based on association rules with application to scene classification | |
CN112733966A (en) | Cluster acquisition and identification method, system and storage medium | |
CN111191102A (en) | Fast search model training method based on big data retrieval and semantic analysis | |
EP4137970A1 (en) | Apparatus, method and computer program code for processing an audio metadata stream |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |