Call recording detection and analysis method based on AI

Abstract

The invention discloses a call recording detection and analysis method based on AI, and relates to the technical field of call recording detection, and the method comprises the following steps: a client is configured in advance to carry out the collection of call audio, and a recording file is obtained; carrying out audio encryption on the sound recording file and uploading the sound recording file to a server; the server side detects and analyzes the sound recording file, performs call transcription on the sound recording file, obtains the theme category of the current call transcription text, and obtains the emotion category according to the current call transcription text; and generating a comment text according to the acquired theme category and emotion category and a preset comment template. The call recording is analyzed from multiple dimensions of theme categories and voice emotions, call content and related information can be grasped more comprehensively and accurately, careless omissions and subjective judgment errors possibly occurring in manual quality inspection are avoided, valuable service insight is provided for enterprises, the enterprises are helped to adjust strategies and optimize service processes in time, and the service quality of the enterprises is improved. The market application prospect is wide.

Description

Technical Field

[0001] The present invention relates to the technical field of call recording detection, and in particular to a call recording detection and analysis method based on AI. Background Art

[0002] With the rapid development of communication technology, call recording has become an important means for many companies and individuals to conduct business communication, customer service, dispute resolution, etc. However, the traditional call recording management method has many shortcomings, such as scattered storage of recording files, difficult retrieval, and inability to quickly obtain call key points. At the same time, with the continuous advancement of artificial intelligence technology, AI is increasingly being used in natural language processing, sentiment analysis, speech recognition, etc., which provides new possibilities for the intelligent management of call recordings.

[0003] At present, although there are some call recording management systems on the market, most of them have a single function and can only store and retrieve recording files, but cannot conduct in-depth analysis and comments on the content of the call. In addition, these systems often lack intelligence and automation, and require human participation to complete the sorting and analysis of recording files, which is inefficient and prone to errors.

[0004] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0005] In view of the problems in the related technology, the present invention proposes an AI-based call recording detection and analysis method to overcome the above-mentioned technical problems existing in the existing related technology.

[0006] The technical solution of the present invention is achieved in this way:

[0007] A call recording detection and analysis method based on AI, comprising the following steps:

[0008] Pre-configure the client to collect call audio and obtain recording files;

[0009] Encrypt the audio of the recording file and upload it to the server;

[0010] The server detects and analyzes the recording file, which at least includes: transcribing the call text of the recording file, and obtaining the subject category of the current call transcription text, which is used to represent the main content of the recording file, and obtaining the emotion category according to the current call transcription text, which is used to represent the voice emotion of the recording file;

[0011] Generate comment text based on the acquired topic category and sentiment category and in accordance with a pre-set comment template.

[0012] Furthermore, the collection of the call audio includes the following steps:

[0013] Pre-configure audio acquisition parameters and provide visual prompts on the interface;

[0014] Generate the corresponding audio file, add the audio file name, and save the audio file to the specified storage path.

[0015] Furthermore, the audio encryption of the recording file and uploading it to the server includes the following steps:

[0016] Perform network connection detection in advance and push network connection prompt information;

[0017] Uploading the unuploaded recording files according to the preset period includes: setting a period to check the unuploaded recording files in the storage, and starting the upload when detecting the recording files that meet the upload conditions.

[0018] The recording file is encrypted according to the symmetric encryption algorithm and uploaded to the server.

[0019] Furthermore, obtaining the subject category of the current call transcription text includes the following steps:

[0020] The frequency of occurrence of each word in a single call transcription is counted in advance, normalized, and the TF value is calculated, which is expressed as:

[0021] TF(word, document) = C / ZC;

[0022] Among them, C is the number of times the word appears in the document, and ZC is the total number of words in the document;

[0023] Calculating the IDF value includes: obtaining the document frequency DF and calculating the inverse document frequency, which is expressed as:

[0024] IDF(word)=log(ZW / (1+DF(word)));

[0025] Among them, the total number of ZW documents;

[0026] Calculate the TF-IDF value and filter keywords, expressed as:

[0027] TF-IDF(word, document) = TF(word, document) × IDF(word);

[0028] Among them, filter keywords, sort the words in descending order according to the TF-IDF value, and select the top several words as keywords, which are the predicted topic categories used to express product names, problem descriptions, and demand expressions.

[0029] Furthermore, the acquisition of the emotion category comprises the following steps:

[0030] The sentiment mapping corpus is pre-established, and the sentence vectors after word segmentation are calibrated based on the call transcription text, expressed as:

[0031] sk={wk1,wk2,...,wkm};

[0032] Where sk represents the vector corresponding to the k-th sentence, wki represents the word vector of the word in the sentence, (i=1, 2, ..., m), m represents the number of words in the sentence, (k=1, 2, ..., n), n represents the number of sentences;

[0033] Calculate the probability that sk is a positive sentiment, expressed as:

[0034]

[0035] Among them, P(C positive ∣s k ) is the probability that a given text sk belongs to positive sentiment, P(C positive ) is the ratio of the number of texts with positive sentiment to the total number of texts, N is the total number of texts, n positive For texts with positive emotions;

[0036] Among them, P(s k ∣C positive ) is the product of the probability of each word in the text appearing in the positive sentiment category, expressed as:

[0037]

[0038] Among them, P(wkm|C positive ) is the probability of word wkm appearing in the positive sentiment category.

[0039] Furthermore, the step of generating the comment text according to the preset comment template further includes:

[0040] Add an annotation tag to the comment text to indicate the review status.

[0041] Beneficial effects of the present invention:

[0042] The call recording detection and analysis method based on AI of the present invention collects call audio through pre-configured client, obtains recording file, encrypts the audio of recording file and uploads it to server, the server detects and analyzes the recording file, transcribes the call to the recording file, obtains the subject category of the current call transcription text, which is used to represent the main content of the recording file, and obtains the emotion category according to the current call transcription text, which is used to represent the voice emotion of the recording file, and generates comment text according to the obtained subject category and emotion category and according to the pre-set comment template, which can not only automatically analyze a large number of call recordings without manual listening and evaluation one by one, but also greatly save time and manpower. It reduces costs, realizes the automation and batching of quality inspection processes, improves work efficiency, and can analyze calls during or immediately after a call, so as to better understand and analyze the content of the call, thereby more accurately judging the call quality, customer needs and emotional tendencies, discovering problems and giving feedback in a timely manner, enabling companies to respond quickly, shorten the problem-solving cycle, and improve customer satisfaction. It also analyzes call recordings from multiple dimensions such as topic categories and voice emotions, which can more comprehensively and accurately grasp the call content and related information, avoiding omissions and subjective judgment errors that may occur in manual quality inspections, providing companies with valuable business insights, and helping them to adjust strategies and optimize business processes in a timely manner. It has broad market application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0044] Figure 1 It is a flowchart of a call recording detection and analysis method based on AI according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.

[0046] According to an embodiment of the present invention, a call recording detection and analysis method based on AI is provided.

[0047] like Figure 1As shown, the call recording detection and analysis method based on AI according to an embodiment of the present invention includes the following steps:

[0048] Step S1, pre-configure the client to collect call audio and obtain the recording file;

[0049] The collection of the call audio includes the following steps:

[0050] Pre-configure audio acquisition parameters and provide visual prompts on the interface;

[0051] Generate the corresponding audio file, add the audio file name, and save the audio file to the specified storage path.

[0052] In this technical solution, the audio acquisition parameters can be set, the sampling rate is generally 44100Hz to ensure good sound quality, the number of channels can be selected as mono or stereo, etc., and the visual prompts include: displaying the recording time. The naming rules can be based on the date and time stamp plus some custom identifiers (such as user ID, etc.) to facilitate subsequent search and management, and the storage path of the specified book, such as the application-specific storage directory of the mobile phone or a specific folder on the computer.

[0053] When this technical solution is applied, the client is configured, including: configuring the mobile terminal (such as a mobile phone application): you can use the platform's native audio recording API, for example, in the Android system, use the MediaRecorder class to implement the audio recording function. Configure the audio source (such as a microphone), audio format (common ones such as MP3, AAC, etc.), encoding rate and other parameters, and start, stop and save the recorded audio file through simple code logic. In the iOS system, you can use the AVFoundation framework to create an AVAudioRecorder instance, configure the relevant audio session and recording settings, and then perform the recording operation.

[0054] Among them, configure the desktop (such as computer software): For Windows platform, use Windows Media API or audio recording library, such as NAudio, to implement the recording function; on macOS platform, you can use the Core Audio framework to collect and record audio.

[0055] Step S2, encrypting the audio of the recording file and uploading it to the server, specifically includes the following steps:

[0056] Perform network connection detection in advance and push network connection prompt information;

[0057] This technical solution first detects the network connection status of the current client before uploading to ensure that it is connected to the Internet. Specifically, the network connection API is used to determine whether it is a Wi-Fi connection, a mobile data connection, etc. If the network is unavailable, the file can be temporarily stored and uploaded after the network is restored, and a network connection prompt message is pushed to the client;

[0058] Upload unuploaded recording files according to the preset period;

[0059] Specifically, a periodic check is performed to check whether there are any recording files that have not been uploaded in the storage of the device. When a recording file that meets the upload conditions is detected, the upload is started.

[0060] The recording file is encrypted according to the symmetric encryption algorithm and uploaded to the server.

[0061] Specifically, for the recording file to be uploaded, read its byte stream data, generate a key according to the specification of the selected symmetric encryption algorithm, and then encrypt the file data, and use the generated encrypted data as the main content of the upload.

[0062] Step S3, the server detects and analyzes the recording file, including the following steps:

[0063] Transcribe the recorded call and obtain the subject category of the current transcribed call to indicate the main content of the recorded call.

[0064] The process of obtaining the subject category of the current call transcription text includes the following steps:

[0065] The frequency of occurrence of each word in a single call transcription is counted in advance, and normalized so that the word frequencies in documents of different lengths are comparable. The TF value is calculated and expressed as:

[0066] TF(word, document) = C / ZC;

[0067] Among them, C is the number of times the word appears in the document, and ZC is the total number of words in the document.

[0068] In this technical solution, for example, a call transcription text has a total of 100 words, and the word "product" appears 5 times, then the TF value of "product" is 5 / 100=0.05.

[0069] Calculating the IDF (Inverse Document Frequency) value includes the following steps:

[0070] The document frequency DF is calculated to count the number of documents containing a certain word. Assuming there are 10 call transcription texts in total, 3 of which contain the word "product", then the document frequency DF (product) of "product" = 3.

[0071] Calculate the inverse document frequency, expressed as:

[0072] IDF(word)=log(ZW / (1+DF(word))), where ZW is the total number of documents;

[0073] Where, for the above example, IDF(product)=log(10 / (1+3))=log(2.5).

[0074] Calculate TF-IDF values ​​and filter keywords, including:

[0075] Calculate TF-IDF: Multiply the TF value of each word by the IDF value to get the TF-IDF value, expressed as:

[0076] TF-IDF(word, document) = TF(word, document) × IDF(word);

[0077] For example, the TF-IDF value of the word "product" in a document is 0.05×log(2.5).

[0078] Among them, keywords are filtered, and the words are sorted in descending order according to the TF-IDF value. The top several words are selected as keywords, which are the predicted topic categories, which can be used to express product names, problem descriptions, demand expressions, etc.

[0079] This technical solution can better highlight more representative keywords that appear frequently in specific documents but less frequently in other documents, because the TF-IDF value comprehensively considers the frequency of the word in a single document and the prevalence of the word in the entire document collection.

[0080] Among them, according to the current call transcription text, the emotion category is obtained to represent the voice emotion of the recording file;

[0081] The sentiment mapping corpus is pre-established, and the sentence vectors after word segmentation are calibrated based on the call transcription text, expressed as:

[0082] sk={wk1,wk2,...,wkm};

[0083] Where sk represents the vector corresponding to the k-th sentence, wki represents the word vector of the word in the sentence, (i=1, 2, ..., m), m represents the number of words in the sentence, (k=1, 2, ..., n), n represents the number of sentences;

[0084] Calculate the probability that sk is a positive sentiment, expressed as:

[0085]

[0086] Among them, P(C positive ∣sk ) is the probability that a given text sk belongs to positive sentiment, P(C positive ) is the ratio of the number of texts with positive sentiment to the total number of texts, N is the total number of texts, n positive For texts with positive emotions;

[0087] Among them, P(s k ∣C positive ) is the product of the probability of each word in the text appearing in the positive sentiment category, expressed as:

[0088]

[0089] Among them, P(wkm|C positive ) is the probability of word wkm appearing in the positive sentiment category.

[0090] Step S4, generating a comment text according to the acquired subject category and sentiment category and a pre-set comment template;

[0091] Specifically, when applied, for example, important cooperation terms mentioned in business negotiation calls, core issues reported by customers in customer service calls, etc., it generates comment text, such as "In this customer service call, the customer focused on the problem of poor product quality and was in a negative mood. We need to follow up and deal with this quality issue as soon as possible to improve customer satisfaction."

[0092] In addition, an annotation tag is added to the comment text to indicate the review status.

[0093] To sum up, with the help of the above-mentioned technical scheme of the present invention, by pre-configuring the client to collect call audio, obtain the recording file, encrypt the audio of the recording file and upload it to the server, the server detects and analyzes the recording file, transcribes the call text of the recording file, and obtains the subject category of the current call transcription text, which is used to represent the main content of the recording file, and obtains the emotion category based on the current call transcription text, which is used to represent the voice emotion of the recording file, and generates comment text based on the obtained subject category and emotion category and according to the pre-set comment template. It can not only automatically analyze a large number of call recordings without manual listening and evaluation one by one, but also greatly save time and manpower. It reduces costs, realizes the automation and batching of quality inspection processes, improves work efficiency, and can analyze calls during or immediately after a call, so as to better understand and analyze the content of the call, thereby more accurately judging the call quality, customer needs and emotional tendencies, discovering problems and giving feedback in a timely manner, enabling companies to respond quickly, shorten the problem-solving cycle, and improve customer satisfaction. It also analyzes call recordings from multiple dimensions such as topic categories and voice emotions, which can more comprehensively and accurately grasp the call content and related information, avoiding omissions and subjective judgment errors that may occur in manual quality inspections, providing companies with valuable business insights, and helping them to adjust strategies and optimize business processes in a timely manner. It has broad market application prospects.

[0094] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. After considering the disclosure of the specification and the examples, those skilled in the art will easily think of other embodiments of the present disclosure. This application is intended to cover any modification, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The specification and the examples are only regarded as exemplary, and the true scope and spirit of the present disclosure are indicated by the claims.

[0095] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A call recording detection and analysis method based on AI, characterized in that: The following steps are involved: Pre-configure the client to collect call audio and obtain recording files; Encrypt the audio of the recording file and upload it to the server; The server detects and analyzes the recording file, which at least includes: transcribing the call text of the recording file, and obtaining the subject category of the current call transcription text, which is used to represent the main content of the recording file, and obtaining the emotion category according to the current call transcription text, which is used to represent the voice emotion of the recording file; Generate comment text based on the acquired topic category and sentiment category and in accordance with a pre-set comment template.

2. The call recording detection and analysis method based on AI according to claim 1 is characterized in that: The collection of the call audio includes the following steps: Pre-configure audio acquisition parameters and provide visual prompts on the interface; Generate the corresponding audio file, add the audio file name, and save the audio file to the specified storage path.

3. The call recording detection and analysis method based on AI according to claim 1, characterized in that: The method of encrypting the audio of the recording file and uploading it to the server includes the following steps: Perform network connection detection in advance and push network connection prompt information; Uploading the unuploaded recording files according to the preset period includes: setting a period to check the unuploaded recording files in the storage, and starting the upload when detecting the recording files that meet the upload conditions. The recording file is encrypted according to the symmetric encryption algorithm and uploaded to the server.

4. The AI-based call recording detection and analysis method according to claim 1, characterized in that: The method of obtaining the subject category of the current call transcription text includes the following steps: The frequency of occurrence of each word in a single call transcription is counted in advance, normalized, and the TF value is calculated, which is expressed as: TF(word, document) = C / ZC; Among them, C is the number of times the word appears in the document, and ZC is the total number of words in the document; Calculating the IDF value includes: obtaining the document frequency DF and calculating the inverse document frequency, which is expressed as: IDF(word)=log(ZW / (1+DF(word))); Among them, the total number of ZW documents; Calculate the TF-IDF value and filter keywords, expressed as: TF-IDF(word, document) = TF(word, document) × IDF(word); Among them, filter keywords, sort the words in descending order according to the TF-IDF value, and select the top several words as keywords, which are the predicted topic categories used to express product names, problem descriptions, and demand expressions.

5. The AI-based call recording detection and analysis method according to claim 4 is characterized in that: The step of obtaining the emotion category comprises the following steps: The sentiment mapping corpus is pre-established, and the sentence vectors after word segmentation are calibrated based on the call transcription text, expressed as: sk={wk1,wk2,...,wkm}; Where sk represents the vector corresponding to the k-th sentence, wki represents the word vector of the word in the sentence, (i=1, 2, ..., m), m represents the number of words in the sentence, (k=1, 2, ..., n), n represents the number of sentences; Calculate the probability that sk is a positive sentiment, expressed as: Among them, P(C positive ∣s k ) is the probability that a given text sk belongs to positive sentiment, P(C positive ) is the ratio of the number of texts with positive sentiment to the total number of texts, N is the total number of texts, n positive For texts with positive emotions; Among them, P(s k ∣C positive ) is the product of the probability of each word in the text appearing in the positive sentiment category, expressed as: Among them, P(wkm|C positive ) is the probability of word wkm appearing in the positive sentiment category.

6. The AI-based call recording detection and analysis method according to claim 1, characterized in that: The step of generating the comment text according to the preset comment template also includes: Add an annotation tag to the comment text to indicate the review status.

Images