CN109658939B - Method for identifying reason of call record non-connection - Google Patents
Method for identifying reason of call record non-connection Download PDFInfo
- Publication number
- CN109658939B CN109658939B CN201910076136.3A CN201910076136A CN109658939B CN 109658939 B CN109658939 B CN 109658939B CN 201910076136 A CN201910076136 A CN 201910076136A CN 109658939 B CN109658939 B CN 109658939B
- Authority
- CN
- China
- Prior art keywords
- reason
- telephone
- record
- database
- 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
- 238000000034 method Methods 0.000 title claims abstract description 62
- 230000011664 signaling Effects 0.000 claims abstract description 24
- 238000013145 classification model Methods 0.000 claims abstract description 13
- 238000010276 construction Methods 0.000 claims description 15
- 238000012549 training Methods 0.000 claims description 10
- 238000006073 displacement reaction Methods 0.000 claims description 9
- 238000004458 analytical method Methods 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 6
- 230000010354 integration Effects 0.000 claims description 5
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 description 14
- 230000008569 process Effects 0.000 description 5
- 238000002372 labelling Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000006855 networking Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000013398 bayesian method Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
Images
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
- G10L15/00—Speech recognition
- G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
- G10L15/063—Training
-
- 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
- G10L15/22—Procedures used during a speech recognition process, e.g. man-machine dialogue
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04M—TELEPHONIC COMMUNICATION
- H04M1/00—Substation equipment, e.g. for use by subscribers
- H04M1/64—Automatic arrangements for answering calls; Automatic arrangements for recording messages for absent subscribers; Arrangements for recording conversations
- H04M1/65—Recording arrangements for recording a message from the calling party
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04M—TELEPHONIC COMMUNICATION
- H04M3/00—Automatic or semi-automatic exchanges
- H04M3/42—Systems providing special services or facilities to subscribers
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04M—TELEPHONIC COMMUNICATION
- H04M3/00—Automatic or semi-automatic exchanges
- H04M3/42—Systems providing special services or facilities to subscribers
- H04M3/487—Arrangements for providing information services, e.g. recorded voice services or time announcements
Landscapes
- Engineering & Computer Science (AREA)
- Signal Processing (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Telephonic Communication Services (AREA)
- Telephone Function (AREA)
Abstract
The invention belongs to the field of voice recognition, and particularly relates to a method for recognizing a reason why a telephone is not connected. The method comprises the following steps: marking the reason of not switching on through signaling; if the reason cannot be obtained through signaling classification, extracting an audio fingerprint characteristic sequence from the telephone record to be identified, and utilizing the sequence to carry out retrieval in an audio fingerprint database; if the matched fingerprint is found, marking the reason of non-connection for the telephone to be identified according to the reason label of non-connection in the fingerprint key value; and if the matched fingerprint cannot be found, identifying the audio content as text content through automatic voice identification, classifying the text content in the unlinked reason document classification model by using a text classification method based on the content, and marking the to-be-identified telephone record by using the unlinked reason classification result obtained by classification. The method can be used for identifying the recording file in an off-line manner and identifying the streaming telephone voice, has higher universality and can be suitable for different application scenes of a call center.
Description
Technical Field
The invention belongs to the field of voice processing, relates to the field of audio fingerprint identification and telephone outgoing call service, and particularly relates to a method for identifying a reason why a telephone is not connected.
Background
The active telephone outbound is an important value-added service in the telecommunication industry, and bears a plurality of important service contents such as public information push, service promotion, customer communication and service follow-up and the like.
In the active telephone outbound service, the client code number is the most important service resource, the service benefit is significantly restricted by the code number resource utilization rate, and especially for the number which is not successfully called, the processing strategy of the number directly influences the service operation level. For example: if a simple abandoning strategy is adopted for the code numbers of the calls which are not connected, on one hand, precious code number resources are wasted, and on the other hand, for some critical outbound services (such as customer communication, complaint feedback and service follow-up), serious communication errors and even service accidents are more likely to be caused; in contrast, if a strategy of repeatedly calling an unaccessed number is taken violently, precious line resources are repeatedly occupied inefficiently, a negative user experience is brought, and even further rejection, rejection and complaint of the user are caused. Therefore, for the reason that the analysis of the failure of the unconnected call is failed, different processing strategies are adopted according to the reason, which is a basic requirement of the outbound industry, and the reason for acquiring the unconnected call is the basis for realizing the requirement.
The most direct mode for obtaining the non-connection is to analyze the telephone signaling on line, namely, tracking the signaling representing the calling state in the calling process, and judging the reason of the non-connection according to the state signaling returned by the access telecommunication gateway under the condition that the telephone is not connected. However, the services in the current telecommunication industry are very different, the networking modes are various, and various service requirements are always considered. In this case, the networking environment of the call center is often very complex, and the networking signaling protocol may be extended and changed to different extents, so that the calling party cannot stably and accurately obtain the signaling in the call in many cases, and can only obtain the unlink prompt tone provided in an audio manner in the media exchange. In this case, the determination of the non-on cause based on the non-on alert tone is the only option.
In a traditional call center with a small scale, the operation of calling out is completed by a call center seat staff, and the seat staff can listen to ring back tone, busy tone and voice prompt in the telephone while operating, so that the reason of not connecting is directly marked. However, for a larger-scale call center, in order to implement management more effectively, a call strategy of splitting calls and services is usually adopted, that is, unified dialing is performed by an outbound platform, and then a call is forwarded to an idle seat person after the call is connected. More importantly, with the continuous progress of the artificial intelligence level, a computer replaces an artificial intelligent outbound system, and the missed calls are automatically classified by the computer system according to the reason, so that the labor consumption can be greatly saved, the operation cost of the call center is reduced, and the work efficiency of the call center is improved.
In continental areas of China, when a telephone is not connected, telephone prompt tones returned by an exchange are generally divided into the following two cases: one is the single frequency telephone automatic switching network tone defined in the national standard GB 3380-82, and the other is a voice broadcast recording specified by the operator, such as "you call busy", "you call out of service", etc. Also in more cases a mix of two alert tones, such as a ring-back voice alert. For the signal sound of the single-frequency switching network conforming to the national standard, the signal sound can be identified by a simple spectrum analysis method, and for the latter, the most common means is to label the reason of non-connection by manual marking, namely manually marking according to busy sound, ringing sound and voice prompt in the recording. This approach is extremely inefficient because it requires manual listening to the entire recording of an unanswered call and one call can only be handled at one time per person. This manual approach can severely reduce the efficiency and reliability of service operation as call center services expand and traffic grows to scale. In this case, it becomes a necessary choice to efficiently process the non-on recorded audio by the computer system and detect the cause of the non-on. With this as a trigger, detection of the reason why the outgoing call is not connected based on the artificial intelligence technology becomes a technology with important application value.
Disclosure of Invention
In order to provide accurate reasons for the missed call, the invention provides a database construction method for analyzing the reasons of the missed call, which comprises the following steps:
acquiring a certain number of unaccessed call recordings;
marking the reason of non-connection of the record of the non-connected phone to obtain a database of the record of the non-connected phone-the reason of non-connection marking;
and extracting an audio fingerprint characteristic sequence from the record of the un-switched call, and taking the corresponding reason of the un-switched call as a key value to obtain an un-switched call record-audio fingerprint characteristic database.
The database construction method for analyzing the reason of the missed call as described above, wherein: and performing voice recognition on the record of the un-switched telephone to obtain text information, and performing un-switched reason text classification modeling on the text information to obtain an un-switched prompt tone text classification model and storing the un-switched prompt tone text classification model in a database.
The database construction method for analyzing the reason of the missed call as described above, wherein: and marking the reason of not connecting according to the online signaling or the voice content.
The database construction method for analyzing the reason of the missed call as described above, wherein: and the audio fingerprint characteristic sequence is encoded by a time-frequency domain differential symbol with a fixed length to serve as an audio fingerprint of a single-frame voice signal.
The database construction method for analyzing the reason of the missed call as described above, wherein: and for a single sound recording file, combining each frame of audio fingerprint with each frame of time information to form a fingerprint characteristic sequence of the whole sound recording file.
The database construction method for analyzing the reason of the missed call as described above, wherein: the database is based on a hash table index structure, an index key value of the hash table index structure is an audio fingerprint characteristic, the content of an indexed unit of the hash table index structure is audio frame information corresponding to the key value, and the audio frame information comprises text information of an unconnected recording and the position of a corresponding audio frame in the recording.
The invention also comprises a method for identifying the reason of the call disconnection, which comprises the following steps:
marking the reason of not switching on through signaling;
if the reason cannot be obtained through signaling classification, extracting an audio fingerprint characteristic sequence from the telephone record to be identified, and utilizing the sequence to carry out retrieval in an audio fingerprint database; if the matched fingerprint is found, marking the reason of non-connection for the telephone to be identified according to the reason label of non-connection in the fingerprint key value;
and if the matched fingerprint cannot be found, identifying the audio content as text content through automatic voice identification, classifying the text content in the unlinked reason document classification model by using a text classification method based on the text content, and labeling the to-be-identified telephone record by using the unlinked reason classification result obtained by classification.
The invention also comprises another method for identifying the reason of the call disconnection, which comprises the following steps:
extracting audio fingerprints frame by frame for the call records which are not connected to obtain a call record audio fingerprint sequence to be identified;
performing single-frame fuzzy retrieval on the telephone recording audio fingerprint sequence to be identified in an audio fingerprint database to obtain a characteristic frame of the candidate original recording audio fingerprint characteristic;
integrating the characteristic frames of the candidate original recording audio fingerprint characteristics, and performing sequence matching with the to-be-identified telephone recording audio fingerprint sequence to obtain a candidate original recording matching sequence set;
searching the ordered candidate original recording matching sequence set in a call record-reason not-connection database, and selecting a plurality of reason not-connection candidates;
and screening the plurality of unconnected reason candidates to obtain the most possible unconnected reason of the telephone.
The method for identifying the reason why the telephone is not connected comprises the following steps: when the single-frame fuzzy retrieval is carried out, a plurality of fingerprint digits which are possibly interfered and have mismatch are determined according to the frequency band energy, a series of expansion candidate fingerprints are generated by taking the bit combination negation, and the retrieval is carried out together with the original fingerprints.
The method for identifying the reason why the telephone is not connected comprises the following steps: when the audio fingerprint feature integration and the sequence matching are carried out, the relative displacement of the feature frames of the candidate original recording audio fingerprint features and the feature frames of the to-be-matched telephone recording in the respective sequences is firstly calculated, when the proportion of the matching number under the same relative displacement in the original telephone recording meets the threshold requirement, the feature frames with the same relative displacement are integrated into the segments in the original telephone recording in the database, namely the candidate original recording matching sequences, and the proportion of the matching number in the original telephone recording is used as the matching hit rate score of the candidate original recording matching sequences.
The method for identifying the reason why the telephone is not connected comprises the following steps: when the plurality of unlinked reason candidates are screened, obtaining matching scores by using the matching hit rate scores of the candidate original sound recording matching sequences and the accumulated scores of the candidate unlinked reasons, sequencing the candidate unlinked reasons by using the matching scores, and selecting a final unlinked reason result if the highest score is higher than a preset decision threshold; the accumulated score refers to the similarity evaluation of the matched part when the ordered candidate original sound recording matching sequence set is searched in a database.
The method for identifying the reason why the telephone is not connected comprises the following steps: the setting method of the decision threshold comprises the following steps: and calculating the matching score distribution of the actual non-connection reasons and other non-connection reasons of the test data in the audio fingerprint database through a group of test data independent of the training data, and calculating the threshold value according to the optimal distinguishability principle.
The invention also comprises a method for identifying the reason of the missed call, which comprises a database construction step aiming at the reason analysis of the missed call; and identifying unlinked call records based on the database.
The method for identifying the reason why the telephone is not connected comprises the following steps: the method also comprises a database updating method, and specifically comprises the following steps:
updating the missed call record-missed call reason marking database by using the newly added missed call record and the missed call reason;
updating the record-audio fingerprint characteristic database of the un-switched telephone by using the newly added record of the un-switched telephone and the audio fingerprint thereof;
updating a text classification model of the reason of non-connection by using the newly added non-connection telephone record and the text information thereof;
updating the signaling list of the reason of non-connection by using the newly added sound record of the non-connection call and the signaling information thereof;
the newly added missed call record refers to a missed call record capable of identifying the reason of missed call.
The method for identifying the reason why the telephone is not connected comprises the following steps:
and if the newly-added un-connected call records have unmatched segments, taking out the segments as training feature sequences corresponding to the un-connected reasons to update the un-connected call record-audio fingerprint feature database.
Compared with the prior art, the method has the following advantages:
the invention establishes an unconnected reason audio database based on an audio fingerprint technology, comprehensively utilizes a voice recognition technology and the audio fingerprint technology, adopts a small amount of manual labels at the initial stage to generate an initial discrimination model and an audio feature library, and iteratively updates the strategy of the model at the later stage; on one hand, a judgment result is directly given to the unmatched voice records successfully matched in the audio feature library by using an efficient and accurate audio fingerprint technology, on the other hand, unmatched prompt tones which are not matched are classified by using a voice recognition keyword extraction and combining an artificial correction method, and the classified audio is added into the audio feature library. By the method, the automatic judgment of the reason why the outgoing call is not connected can be efficiently and accurately realized under the condition of little manual intervention.
By utilizing the single-frame fuzzy retrieval method, suspected or similar audios can be screened from the audio database of the unconnected reason, and subsequent integration and long sequence matching are carried out on the basis, so that the missing detection is reduced to the maximum extent, and the recall rate of the system is improved. The method divides the audio to be detected into frame level characteristics for fuzzy matching, has no sequential constraint on time sequence, has concurrent operation property, improves the system retrieval speed, and provides a technical basis for real-time audio retrieval.
By using the method of integration and sequence matching, multi-level clues can be integrated, and the method is extended from single frame matching to longer frame sequence matching. On one hand, the fault tolerance of the system can be improved, and the system has better adaptability to audio frequency with noise interference or channel influence; on the other hand, the smaller labeling unit can reduce the data labeling cost and increase the application range of the system.
Save the manual work, reduce the operation cost: the system only needs to manually carry out a very small amount of auditory identification and marking in the operation process, and can carry out uninterrupted processing within 24 hours.
Fast and efficient: the method adopts an automatic distinguishing technology combining voice recognition and audio fingerprints, can simultaneously process multiple calls which are not connected on a single computer which is commonly configured, and the distinguishing speed of each call is far higher than the speed of manual listening and distinguishing.
The method can be used for identifying the recording file in an off-line manner and identifying the streaming telephone voice, has higher universality and can be suitable for different application scenes of a call center.
Drawings
FIG. 1 is a flow chart of an embodiment of a database construction method for missed call reason analysis according to the present invention;
FIG. 2 is a flow chart of another embodiment of a database construction method for analyzing reasons of a missed call according to the present invention;
FIG. 3 is a flow chart of an embodiment of a method for identifying a reason why a call is not connected according to the present invention;
fig. 4 is a flowchart of another embodiment of a method for identifying a reason why a phone call is not connected according to the present invention.
Detailed Description
The present invention will be described in detail with reference to examples.
First, the terms of art related to the present invention are explained as follows:
1. audio fingerprint technology "
Audio fingerprinting technology refers to the extraction of unique numerical features in a piece of Audio in the form of an identifier by a specific algorithm for identifying a large number of sound samples or tracking the location of a location sample in a database. The audio fingerprint is used as a core algorithm of a content automatic identification technology, and is widely applied to the fields of music identification, copyright content monitoring and broadcasting, content library duplicate removal, television second screen interaction and the like.
The audio fingerprint technology is completed by extracting data characteristics in sound and comparing the content to be identified with an established audio fingerprint database. The identification process is not influenced by the storage format, the coding mode, the code rate and the compression technology of the audio. The matching of audio fingerprints is a highly accurate match, independent of file meta information, watermarking and file hash values. Specific implementations of audio fingerprinting features include a variety of methods, such as: hash methods, complex cepstrum, wavelet analysis, etc.
2. "automatic speech recognition":
in recent years, particularly in 2009, with the development of deep learning research in the field of machine learning and the accumulation of large data corpora, speech recognition technology has been developed dramatically. On one hand, as the deep learning research in the field of machine learning is introduced into the training of the voice recognition acoustic model, the accuracy of the acoustic model is greatly improved, and the method becomes the fastest progress in the voice recognition technology in recent 20 years.
On the other hand, most of the mainstream speech recognition decoders at present adopt a decoding network based on a finite state machine (WFST), and the decoding network can integrate a language model, a dictionary and an acoustic shared tone word set into a large decoding network, so that the decoding speed is greatly improved, and a foundation is provided for the real-time application of speech recognition.
In addition, with the rapid development of the internet and the popularization and application of mobile terminals such as mobile phones, a large amount of linguistic data in the aspect of texts or voices can be obtained from multiple channels at present, so that abundant resources are provided for training of language models and acoustic models in voice recognition, and the construction of general large-scale language models and acoustic models becomes possible.
3. Text classification techniques
The text classification problem is defined by selecting a corresponding category from predefined category labels based on the content of a document. The basic steps of Chinese text classification are Chinese word segmentation, feature extraction, model training, class prediction and the like, and text classification based on statistics generally needs better labeled corpora as a training set, trains out a model, and classifies unclassified texts by using the model. The commonly used statistical characteristics mainly include chi-square statistics, information gain, mutual information, probability ratio, cross entropy and other methods.
Fig. 1 is a database construction method for analyzing reasons of a missed call according to the present invention, which includes the following steps:
step S11: acquiring a certain number of unaccessed call recordings;
step S12: and marking the reason of non-connection of the record of the non-connected phone to obtain a database of the record of the non-connected phone-the reason of non-connection. During the marking, the reason of the missed call can be judged by the telephone access platform through signaling, and the reason of the missed call can also be judged according to the missed call prompt tone (for example, the contents of ' power off ', line busy ', and the like).
And step S13, extracting the audio fingerprint characteristic sequence from the record of the un-switched call, and obtaining the record of the un-switched call-audio fingerprint characteristic database by taking the corresponding reason of the un-switched call as a key value.
The record of the un-switched call can comprise information such as an un-switched prompt tone, a ringing tone, a polyphonic ringtone and the like, and in order to obtain a better effect, the information such as the polyphonic ringtone and the like can be removed, and only the un-switched prompt tone is reserved as the record of the un-switched call.
The audio fingerprint feature sequence can be encoded by a time-frequency domain differential symbol with a fixed length as an audio fingerprint of a single-frame voice signal.
For a single sound recording file, the fingerprint characteristic sequence of the whole sound recording file can be formed by combining each frame of audio fingerprint with each frame of time information.
Preferably, as shown in fig. 2, step S24 may be further performed: and carrying out voice recognition on the un-switched telephone or the un-switched prompt tone to obtain corresponding text information, and storing the text information into a database. And text classification can be carried out on the text information by using the non-answering reason corresponding to the telephone recording. And when the texts are classified, extracting text classification features for each class of unconnected reasons based on the recognized texts to form an unconnected prompt tone text classification model.
Preferably, the database is built based on a hash table index structure. The index key value of the hash table index structure is an audio fingerprint characteristic, and the content of the indexed unit is audio frame information corresponding to the key value; the audio frame information comprises text information of the sound recording which is not connected and the position of the corresponding audio frame in the sound recording.
As shown in fig. 3, a method for identifying a reason why a phone call is not connected specifically includes the following steps:
step S31: marking the reason of the call connection failure through signaling for the call connection failure record to be identified;
step S32: for the unassociated call records which cannot be classified through the signaling, extracting an audio fingerprint characteristic sequence from the unassociated call records to be identified, and retrieving the audio fingerprint characteristic sequence in an audio fingerprint database; if the matched fingerprint is found, marking the reason of non-connection for the record of the non-connected phone to be identified according to the reason label of non-connection in the fingerprint key value;
and step S33, if no matched fingerprint is found, the audio content is recognized as text content through automatic voice recognition, classification is carried out in the unlinked reason document classification model by utilizing a text classification method based on the text content, and the unlinked call record to be recognized is marked by the classified unlinked reason classification result.
The text classification method may be a bayesian method, a decision tree method, a neural network method, or the like. If the text space of the telephone recording is shorter, a better effect can be obtained by adopting a naive Bayes classification method.
Fig. 3 also shows a database updating method for analyzing the reason of the missed call, which specifically includes the following steps:
step S34: updating the missed call record-missed call reason marking database by using the newly added missed call record and the missed call reason;
step S35: updating the record-audio fingerprint characteristic database of the un-switched telephone by using the newly added record of the un-switched telephone and the audio fingerprint thereof;
step S36: updating a text classification model of the reason of non-connection by using the newly added non-connection telephone record and the text information thereof;
and step S37, updating the signaling list of the reason of non-connection by the newly added recording of the non-connection call and the signaling information thereof.
The newly added non-call recording refers to non-call recording capable of identifying the reason of non-call.
If the newly added un-connected call records have unmatched segments, the segments are taken out to serve as training feature sequences corresponding to the un-connected reasons to update the un-connected call record-audio fingerprint feature database.
Fig. 4 is a method for identifying a reason why a phone call is not connected, which specifically includes the following steps:
and step S41, extracting audio fingerprints frame by frame for the unlinked call records to obtain the audio fingerprint sequence of the call records to be identified.
And step S42, retrieving the generated voice frequency fingerprint sequence of the telephone recording to be identified frame by frame in the voice frequency fingerprint database to obtain the characteristic frame of the candidate original voice frequency fingerprint characteristic.
The search may be a fuzzy search, for example, determining a number of fingerprint bits that may be interfered and have a mismatch according to the band energy, generating a series of extended candidate fingerprints by negating the bit combination, and performing the search in combination with the original fingerprint to reduce the mismatch caused by the local interference.
Step S43: and integrating the obtained characteristic frames of the candidate original recording audio fingerprint characteristics, and performing sequence matching with the to-be-identified telephone recording audio fingerprint sequence to obtain a candidate original recording matching sequence set.
Preferably, when the audio fingerprint feature integration and the sequence matching are performed, relative displacement of the feature frame of the candidate original recording audio fingerprint feature and the feature frame of the to-be-matched telephone recording in respective sequences is calculated, when the proportion of the matching number under the same relative displacement in the original telephone recording meets a threshold requirement, the feature frames with the same relative displacement are integrated into a segment in the original telephone recording in a database, namely a candidate original recording matching sequence, and the proportion of the matching number in the original telephone recording is used as the matching hit rate score of the candidate original recording matching sequence.
Step S44: and searching the obtained ordered candidate original sound recording matching sequence set in the un-connected call sound recording-un-connected reason database, and selecting a plurality of un-connected reason candidates.
Step S45: and screening the obtained multiple unlink reason candidates to obtain the most possible reason for the unlink call.
When a plurality of unlinked reason candidates are screened, matching scores are obtained by using the matching hit rate scores of the candidate original sound recording matching sequences and the accumulated scores of the candidate unlinked reasons, the candidate unlinked reasons are sorted (the matching scores for sorting can be obtained by adding, multiplying, weighting and summing the matching hit rate scores and the accumulated scores), and if the highest score is higher than a preset decision threshold, the final unlinked reason result is selected.
The accumulated score refers to the similarity evaluation of the part capable of matching when the ordered candidate original sound recording matching sequence set is searched in a database.
The preset decision threshold may be an empirical value, or the following setting method may be adopted: and calculating the matching score distribution of the actual non-connection reasons and other non-connection reasons of the test data in the audio fingerprint database through a group of test data independent of the training data, and calculating the threshold value according to the optimal distinguishability principle.
A method for identifying the reason why a phone is not connected can comprise a database construction step aiming at the reason analysis of the phone which is not connected; and a step of identifying unlinked call records based on the database.
The database construction step comprises the following steps: acquiring a certain number of unaccessed call recordings; marking the reason of non-connection of the record of the non-connected phone to obtain a database of the record of the non-connected phone-the reason of non-connection marking; extracting an audio fingerprint feature sequence from the record of the missed call, and taking the corresponding reason of the missed call as a key value to obtain a record-audio fingerprint feature database of the missed call;
wherein the step of identifying comprises: marking the reason of not switching on through signaling; if the reason cannot be obtained through signaling classification, extracting an audio fingerprint characteristic sequence from the telephone recording, and searching in an audio fingerprint database; if the matched fingerprint is found, marking the reason of non-connection for the telephone according to the reason label of non-connection in the fingerprint key value; and if the matched fingerprint cannot be found, obtaining the recognition content text of the word segmentation through automatic voice recognition, searching in the unlinked reason document classification model by using a text classification method, and classifying and labeling the telephone recording by using the unlinked reason obtained through searching.
In addition, the method also comprises an updating step of the database.
Although the present invention has been described with reference to the preferred embodiments, it is not intended to be limited thereto, and variations and modifications may be made by those skilled in the art without departing from the spirit and scope of the present invention.
Claims (6)
1. A database construction method for analyzing reasons of a missed call comprises the following steps:
acquiring a certain number of unaccessed call recordings;
marking the reason of non-connection of the record of the non-connected phone to obtain a database of the record of the non-connected phone-the reason of non-connection marking;
extracting an audio fingerprint feature sequence from the record of the missed call, and taking the corresponding reason of the missed call as a key value to obtain a record-audio fingerprint feature database of the missed call; for a single sound recording file, combining each frame of audio fingerprint with each frame of time information to form a fingerprint characteristic sequence of the whole sound recording file;
performing voice recognition on the record of the un-switched telephone to obtain text information, and performing un-switched reason text classification modeling on the text information to obtain an un-switched prompt tone text classification model and storing the un-switched prompt tone text classification model in a database;
marking the reason of not being connected according to the online signaling or the voice content;
the database is based on a hash table index structure, an index key value of the hash table index structure is an audio fingerprint characteristic, the content of an indexed unit of the hash table index structure is audio frame information corresponding to the key value, and the audio frame information comprises text information of an unconnected recording and the position of a corresponding audio frame in the recording;
the method also comprises a database updating method, and specifically comprises the following steps:
updating the missed call record-missed call reason marking database by using the newly added missed call record and the missed call reason;
updating the record-audio fingerprint characteristic database of the un-switched telephone by using the newly added record of the un-switched telephone and the audio fingerprint thereof;
updating a text classification model of the reason of non-connection by using the newly added non-connection telephone record and the text information thereof;
updating the signaling list of the reason of non-connection by using the newly added sound record of the non-connection call and the signaling information thereof;
the newly added missed call record refers to the missed call record capable of identifying the reason of missed call;
and if the newly-added un-connected call records have unmatched segments, taking out the segments as training feature sequences corresponding to the un-connected reasons to update the un-connected call record-audio fingerprint feature database.
2. The database construction method for the reason analysis of the missed call according to claim 1, wherein: and the audio fingerprint characteristic sequence is encoded by a time-frequency domain differential symbol with a fixed length to serve as an audio fingerprint of a single-frame voice signal.
3. A method for identifying a reason why a phone call is not connected comprises the following steps:
marking the reason of not switching on through signaling;
if the reason cannot be obtained through signaling classification, extracting an audio fingerprint characteristic sequence from the telephone record to be identified, and utilizing the sequence to carry out retrieval in an audio fingerprint database; if the matched fingerprint is found, marking the reason of non-connection for the telephone to be identified according to the reason label of non-connection in the fingerprint key value;
if the matched fingerprint cannot be found, identifying the audio content as text content through automatic voice identification, classifying the text content in a non-connected reason document classification model by using a text classification method based on the text content, and marking the to-be-identified telephone record by using a classification result of the non-connected reason obtained through classification;
extracting audio fingerprints frame by frame for the call records which are not connected to obtain a call record audio fingerprint sequence to be identified;
performing single-frame fuzzy retrieval on the telephone recording audio fingerprint sequence to be identified in an audio fingerprint database to obtain a characteristic frame of the candidate original recording audio fingerprint characteristic;
integrating the characteristic frames of the candidate original recording audio fingerprint characteristics, and performing sequence matching with the to-be-identified telephone recording audio fingerprint sequence to obtain a candidate original recording matching sequence set;
searching the ordered candidate original recording matching sequence set in a call record-reason not-connection database, and selecting a plurality of reason not-connection candidates;
screening the plurality of unconnected reason candidates to obtain the most possible unconnected reason of the telephone;
when the single-frame fuzzy retrieval is carried out, a plurality of fingerprint digits which are possibly interfered and have mismatch are determined according to the frequency band energy, a series of expansion candidate fingerprints are generated by taking the bit combination negation, and the retrieval is carried out together with the original fingerprints.
4. A method for identifying a reason why a telephone call has not been made as claimed in claim 3, wherein: when the audio fingerprint feature integration and the sequence matching are carried out, the relative displacement of the feature frames of the candidate original recording audio fingerprint features and the feature frames of the to-be-matched telephone recording in the respective sequences is firstly calculated, when the proportion of the matching number under the same relative displacement in the original telephone recording meets the threshold requirement, the feature frames with the same relative displacement are integrated into the segments in the original telephone recording in the database, namely the candidate original recording matching sequences, and the proportion of the matching number in the original telephone recording is used as the matching hit rate score of the candidate original recording matching sequences.
5. The method for identifying the reason why the phone is not connected as claimed in claim 4, wherein: when the plurality of unlinked reason candidates are screened, obtaining matching scores by using the matching hit rate scores of the candidate original sound recording matching sequences and the accumulated scores of the candidate unlinked reasons, sequencing the candidate unlinked reasons by using the matching scores, and selecting a final unlinked reason result if the highest score is higher than a preset decision threshold; the accumulated score refers to the similarity evaluation of the matched part when the ordered candidate original sound recording matching sequence set is searched in a database.
6. The method for identifying the reason why the phone is not connected as claimed in claim 5, wherein: the setting method of the decision threshold comprises the following steps: and calculating the matching score distribution of the actual non-connection reasons and other non-connection reasons of the test data in the audio fingerprint database through a group of test data independent of the training data, and calculating the threshold value according to the optimal distinguishability principle.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910076136.3A CN109658939B (en) | 2019-01-26 | 2019-01-26 | Method for identifying reason of call record non-connection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910076136.3A CN109658939B (en) | 2019-01-26 | 2019-01-26 | Method for identifying reason of call record non-connection |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109658939A CN109658939A (en) | 2019-04-19 |
| CN109658939B true CN109658939B (en) | 2020-12-01 |
Family
ID=66121641
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910076136.3A Active CN109658939B (en) | 2019-01-26 | 2019-01-26 | Method for identifying reason of call record non-connection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109658939B (en) |
Families Citing this family (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110290280B (en) * | 2019-05-28 | 2021-08-13 | 同盾控股有限公司 | Terminal state identification method and device and storage medium |
| CN110379444A (en) * | 2019-08-30 | 2019-10-25 | 北京太极华保科技股份有限公司 | Detection method and device, the electronic equipment of telephone state are judged by preposition media |
| CN111193940B (en) * | 2019-12-09 | 2021-07-06 | 腾讯科技（深圳）有限公司 | Audio playing method and device, computer equipment and computer readable storage medium |
| CN110995938B (en) * | 2019-12-13 | 2022-04-26 | 度小满科技(北京)有限公司 | Data processing method and device |
| CN111368527B (en) * | 2020-02-28 | 2023-06-20 | 上海汇航捷讯网络科技有限公司 | Key value matching method |
| CN111510566B (en) * | 2020-03-16 | 2021-05-28 | 深圳追一科技有限公司 | Method and device for determining call label, computer equipment and storage medium |
| CN111601002B (en) * | 2020-04-10 | 2021-07-30 | 北京捷通华声科技股份有限公司 | Client number state matching method and device, electronic equipment and storage medium |
| CN111629108A (en) * | 2020-04-27 | 2020-09-04 | 北京青牛技术股份有限公司 | Real-time identification method of call result |
| CN112235467B (en) * | 2020-10-09 | 2022-09-02 | 深圳壹账通智能科技有限公司 | Method and device for distinguishing types of disconnected telephones, electronic equipment and storage medium |
| CN112562654B (en) * | 2020-12-02 | 2022-11-11 | 天津车之家数据信息技术有限公司 | Audio classification method and computing device |
| CN112954118B (en) * | 2021-02-22 | 2022-09-30 | 上海井星信息科技有限公司 | Null number detection method and system based on frequency spectrum high point |
| CN113079259A (en) * | 2021-03-31 | 2021-07-06 | 北京智齿博创科技有限公司 | Outbound failure result detection method based on freeswitch and ASR technology |
| CN113312509A (en) * | 2021-06-22 | 2021-08-27 | 中国平安财产保险股份有限公司 | Telephone recall method, telephone recall device, computer equipment and storage medium |
| CN113269935B (en) * | 2021-07-16 | 2021-11-30 | 融讯伟业（北京）科技有限公司 | Visual weighing method and weighing system based on screen-free weighing device |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1768422A1 (en) * | 2004-06-22 | 2007-03-28 | ZTE Corporation | Method for protecting incoming calls in personalized ring back tone service |
| US9247057B2 (en) * | 2014-01-15 | 2016-01-26 | Verizon Patent And Licensing Inc. | System and method for providing proactive service assurance in emergency networks |
| CN106254696A (en) * | 2016-08-02 | 2016-12-21 | 北京京东尚科信息技术有限公司 | Outgoing call result determines method, Apparatus and system |
| CN107293307A (en) * | 2016-03-31 | 2017-10-24 | 阿里巴巴集团控股有限公司 | Audio-frequency detection and device |
| CN107580149A (en) * | 2017-08-28 | 2018-01-12 | 携程旅游网络技术（上海）有限公司 | The recognition methods of outgoing call failure cause, device, electronic equipment, storage medium |
| CN107622773A (en) * | 2017-09-08 | 2018-01-23 | 科大讯飞股份有限公司 | Audio feature extraction method and device and electronic equipment |
| CN108399913A (en) * | 2018-02-12 | 2018-08-14 | 北京容联易通信息技术有限公司 | High robust audio fingerprinting method and system |
| CN108924370A (en) * | 2018-07-23 | 2018-11-30 | 携程旅游信息技术（上海）有限公司 | Call center's outgoing call speech waveform analysis method, system, equipment and storage medium |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8509736B2 (en) * | 2002-08-08 | 2013-08-13 | Global Tel*Link Corp. | Telecommunication call management and monitoring system with voiceprint verification |
| US7783021B2 (en) * | 2005-01-28 | 2010-08-24 | Value-Added Communications, Inc. | Digital telecommunications call management and monitoring system |
| US8150020B1 (en) * | 2007-04-04 | 2012-04-03 | At&T Intellectual Property Ii, L.P. | System and method for prompt modification based on caller hang ups in IVRs |
| US8331919B1 (en) * | 2009-04-24 | 2012-12-11 | Nuance Communications, Inc. | System, method, and software program product for tracking call failures on a wireless phone |
| CN107911557A (en) * | 2017-11-30 | 2018-04-13 | 维沃移动通信有限公司 | The processing method and mobile terminal of a kind of missed call |
| CN108124065A (en) * | 2017-12-05 | 2018-06-05 | 浙江鹏信信息科技股份有限公司 | A kind of method junk call content being identified with disposal |
| CN109151220A (en) * | 2018-09-11 | 2019-01-04 | 中国—东盟信息港股份有限公司 | A kind of communication session call failure scene analysis system |
-
2019
- 2019-01-26 CN CN201910076136.3A patent/CN109658939B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1768422A1 (en) * | 2004-06-22 | 2007-03-28 | ZTE Corporation | Method for protecting incoming calls in personalized ring back tone service |
| US9247057B2 (en) * | 2014-01-15 | 2016-01-26 | Verizon Patent And Licensing Inc. | System and method for providing proactive service assurance in emergency networks |
| CN107293307A (en) * | 2016-03-31 | 2017-10-24 | 阿里巴巴集团控股有限公司 | Audio-frequency detection and device |
| CN106254696A (en) * | 2016-08-02 | 2016-12-21 | 北京京东尚科信息技术有限公司 | Outgoing call result determines method, Apparatus and system |
| CN107580149A (en) * | 2017-08-28 | 2018-01-12 | 携程旅游网络技术（上海）有限公司 | The recognition methods of outgoing call failure cause, device, electronic equipment, storage medium |
| CN107622773A (en) * | 2017-09-08 | 2018-01-23 | 科大讯飞股份有限公司 | Audio feature extraction method and device and electronic equipment |
| CN108399913A (en) * | 2018-02-12 | 2018-08-14 | 北京容联易通信息技术有限公司 | High robust audio fingerprinting method and system |
| CN108924370A (en) * | 2018-07-23 | 2018-11-30 | 携程旅游信息技术（上海）有限公司 | Call center's outgoing call speech waveform analysis method, system, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109658939A (en) | 2019-04-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109658939B (en) | Method for identifying reason of call record non-connection | |
| CN107798032B (en) | Method and device for processing response message in self-service voice conversation | |
| CN109819127B (en) | Method and system for managing crank calls | |
| US8219404B2 (en) | Method and apparatus for recognizing a speaker in lawful interception systems | |
| CN109451182B (en) | Detection method and device for fraud telephone | |
| US8346539B2 (en) | System and method for indexing voice mail messages by speaker | |
| US7817993B2 (en) | System and method for copying ring back tone transfer sound | |
| CN109842712B (en) | Call record generation method and device, computer equipment and storage medium | |
| US7415409B2 (en) | Method to train the language model of a speech recognition system to convert and index voicemails on a search engine | |
| CN1852354A (en) | Method and device for collecting user behavior characteristics | |
| CN103731832A (en) | System and method for preventing phone and short message frauds | |
| CN111131593B (en) | Crank call identification method and device | |
| CN104702759A (en) | Address list setting method and address list setting device | |
| WO2022206719A1 (en) | Outbound call failure result detection method based on freeswitch and asr | |
| CN111010484A (en) | Automatic quality inspection method for call recording | |
| US20040010408A1 (en) | Method and system for identifying a caller based on voice recognition | |
| CN101000767A (en) | Speech recognition equipment and method | |
| CN107357859A (en) | A kind of intelligent terminal for realizing that knowledge base shows automatically by voice collecting | |
| CN101072252A (en) | Method and device for identifying mobile phone number territoriality for mobile communication terminal | |
| CN116501844A (en) | Voice keyword retrieval method and system | |
| CN112509587B (en) | Method, device and equipment for dynamically matching mobile number and voiceprint and constructing index | |
| EP3744079B1 (en) | Answer machine detection method & apparatus | |
| CN114006982B (en) | Harassment number identification method based on classification gradient lifting algorithm | |
| US20210358504A1 (en) | System and method for obtaining voiceprints for large populations | |
| CN112908364B (en) | Telephone number state judging method and system |
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 |