Intelligent speech recognition method based on AI

Abstract

The present invention provides an AI-based intelligent speech recognition method, comprising the steps of: real-time monitoring of current hot spots, generating corresponding hot words, and adding the hot words to a hot data set; in the process of communicating with an outbound user, recognizing the real-time speech content of the outbound user, obtaining the real-time text content, judging whether the real-time hot words in the hot data set exist in the real-time text content, and if so, obtaining the emotional tendency of the outbound user for the real-time hot words according to the emotional analysis of the real-time text content, and performing hot spot interaction according to the preset speech corresponding to the emotional tendency; judging in real time whether the current interaction is a secondary process interaction, and if so, and the current interaction can be ended, returning to the preset main process interaction by jumping back to the speech, so as to complete product promotion according to the main process interaction. The present invention can ensure the efficiency of outbound calls while improving the accuracy of responses to outbound users.

Description

Technical Field

[0001] The present invention relates to the field of speech recognition technology, and in particular to an AI-based intelligent speech recognition method. Background Art

[0002] At present, telephone sales usually use manual customer service to make real-time calls, which has problems such as high manual customer service costs and unstable emotions of manual customer service. The unstable emotions of manual customer service will affect the user's communication experience, and ultimately affect sales performance. Therefore, the current telephone sales business needs AI voice robots to replace some manual customer service.

[0003] AI voice robots recognize the meaning of outbound users' voices and respond with pre-set corresponding words to promote products. However, this communication method is relatively rigid and cannot make appropriate responses according to the "sudden whims" of outbound users, which makes outbound users lose the desire to continue communicating. Therefore, there is an urgent need to solve the above problems. Summary of the invention

[0004] In order to solve the above problems in the prior art, the present invention provides an AI-based intelligent speech recognition method, which can improve the accuracy of responding to outbound calling users.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] In a first aspect, the present invention provides an AI-based intelligent speech recognition method, comprising the steps of:

[0007] Monitor current hot spots in real time, generate corresponding hot words, and add the hot words to the hot data set;

[0008] In the process of communicating with an outbound user, the real-time voice content of the outbound user is recognized to obtain the real-time text content, and it is determined whether the real-time hot words in the hot data set exist in the real-time text content. If so, the emotional tendency of the outbound user towards the real-time hot words is obtained according to the emotional analysis of the real-time text content, and hot spot interaction is performed according to the preset words corresponding to the emotional tendency;

[0009] Determine in real time whether the current interaction is a sub-process interaction. If so, determine whether the current interaction can be ended based on the latest voice content of the current interaction. If so, return to the pre-set main process interaction by jumping back to the main process interaction to complete product promotion based on the main process interaction. The sub-process interaction includes hotspot interaction.

[0010] The beneficial effect of the present invention is that: by monitoring and collecting the current hot spots, when hot words appear in the communication content of the outbound user, the hot words can be interacted with, so that the communication content of the outbound user can be responded to in time, so as to improve the accuracy of the response to the outbound user. And in the process of deviating from the main process such as hot spot interaction, by judging whether the interaction is over, it is decided to use the jump back speech technique, so that the main process can be returned to interact in time to complete the product promotion, thereby, the communication of the hot spot interaction will not affect the main process interaction, and the outbound call efficiency is improved, that is, the present invention can ensure the outbound call efficiency while improving the accuracy of the response to the outbound user.

[0011] Optionally, the real-time monitoring of the current hot spots, generating corresponding hot words, and adding the hot words to the hot data set comprises the following steps:

[0012] Monitor the current hot spots in real time, generate corresponding first hot words and second hot words, add the first hot words and the second hot words to the hot data set after adding corresponding labels, wherein the first hot words are hot words associated with other categories except the current category, and the second hot words are hot words associated with the current category;

[0013] The hotspot interaction according to the preset words corresponding to the emotional tendency comprises the steps of:

[0014] Determine whether the real-time hot words in the real-time text content belong to the first hot words. If so, make a voice reply based on the first jump back words associated with the first hot words and the content of the previous main process interaction. Otherwise, generate corresponding preset words for hot spot interaction based on the emotional tendency and the product to which the real-time hot words belong.

[0015] According to the above description, whether to jump back or conduct hotspot interaction is determined based on whether it is a hot word in the current category, thereby improving the accuracy of responding to outbound call users while ensuring outbound call efficiency.

[0016] Optionally, the hot words further include a third hot word that is unrelated to the product, and the hot word interaction according to the preset words corresponding to the emotional tendency further includes:

[0017] When it is the third hot word, a voice reply is made according to the second jump back speech associated with the third hot word and the content of the previous main process interaction.

[0018] Optionally, the method further comprises the steps of:

[0019] When the real-time hot word belongs to the first hot word, a category label is added to the outbound user according to the category to which the real-time hot word belongs, so as to promote products of the same category according to the category label.

[0020] According to the above description, the present invention distinguishes the first hot words and adds category labels, which can be used as a backup, so that when promoting products associated with these category labels, interested outbound users can be quickly and accurately obtained, thereby improving the outbound call success rate.

[0021] Optionally, generating corresponding preset words for hot spot interaction according to the sentiment tendency and the belonging product of the real-time hot spot words includes:

[0022] Determine whether the emotional tendency is negative. If so, make a voice reply to the hot spot interaction according to the first preset words. Otherwise, determine whether the emotional tendency is positive. If so, make a voice reply to the hot spot interaction according to the second preset words. Otherwise, if the emotional tendency is neutral, make a voice reply to the hot spot interaction according to the third preset words. The first preset words include the differences between the product to which the real-time hot spot words belong and the current product, the second preset words include the similarities between the product to which the real-time hot spot words belong and the current product, and the third preset words include the advantages of the current product.

[0023] According to the above description, the product introduction direction is determined according to the outbound user's emotional inclination towards hot topics, thereby increasing the outbound user's interest in the product.

[0024] Optionally, the real-time monitoring of current hot spots and generating corresponding hot spot words includes:

[0025] Monitor the current hot spots in real time, and generate hot words based on the keywords and synonyms of the current hot spots.

[0026] Optionally, the method further comprises the steps of:

[0027] When the real-time text content contains an expression of the need for manual customer service, manual customer service is accessed.

[0028] Optionally, accessing manual customer service includes:

[0029] Determine whether the access of the manual customer service needs to wait. If so, determine whether the outbound user has a purchase intention in the previous interaction process. If so, increase the queuing priority of the outbound user.

[0030] According to the above description, the priority of outbound users in accessing manual customer service is determined by whether they have purchase intention, thereby improving the communication experience of outbound users with purchase intention.

[0031] Optionally, the method further comprises the steps of:

[0032] When the overall purchase intention of the outbound user is neutral and the emotional tendency towards the AI ​​voice robot in the overall interaction content is negative, the outbound user is marked as a user to be manually dialed.

[0033] Optionally, the dialing order of the users to be manually dialed is determined based on the specific score of the outbound user's overall purchase intention and the specific score of the outbound user's emotional tendency towards the AI ​​voice robot, wherein, the closer the specific score of the outbound user's overall purchase intention is to the positive side, the higher the dialing order of the users to be manually dialed, and the lower the specific score of the outbound user's emotional tendency towards the AI ​​voice robot, the higher the dialing order of the users to be manually dialed. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a flow chart of an AI-based intelligent speech recognition method according to Embodiment 1 of the present invention;

[0035] Figure 2 This is a flow chart of an AI-based intelligent speech recognition method according to the second embodiment of the present invention. DETAILED DESCRIPTION

[0036] In order to better understand the above technical solution, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0037] Embodiment 1

[0038] This embodiment is applied in the scenario of using AI voice robots for sales promotion. When facing the interactive behaviors of outbound users that deviate from the main process business, AI voice robots usually only have fixed templates to handle these, resulting in a very poor communication experience for outbound users. This embodiment monitors and collects current hot spots, so that the AI ​​voice robot can adapt to most interactive behaviors, thereby greatly improving the accuracy of responding to outbound users. Please refer to the following for details.

[0039] Please refer to Figure 1 , an AI-based intelligent speech recognition method, comprising the steps of:

[0040] S1. Monitor the current hot spots in real time, generate corresponding hot words, and add the hot words to the hot data set;

[0041] In this embodiment, the current hot spot is a hot spot within a specified time range, such as seven days, fifteen days, etc., or even several years. It includes real-time hot spots, such as yesterday's hot search situation, seven days' hot search situation, etc.; it also includes long-term hot spots, such as a hot spot with a high average index of a product within a year, a hot spot with a high average index in the past history, etc.

[0042] In this embodiment, in step S1, the current hot spots are monitored in real time, and the corresponding hot spot words are generated, including:

[0043] Monitor current hot spots in real time and generate hot words based on the keywords and synonyms of current hot spots.

[0044] In this embodiment, aa is a certain electronic product consumption rate brand. When it releases a certain car, for example, the current hot spot is: "AAA" trademark is preemptively registered. Then the keyword of the current hot spot is "AAA", and its synonyms include "aa car" and so on. Therefore, hot words include "AAA" and "aa car". If a hot spot of a certain game is: BB start, then the keywords are "BB", "BB start", and this word has no synonyms. Therefore, in other embodiments, the hot words can also be just the keywords of the current hot spot. The two are selected according to the current category and the focus of product promotion.

[0045] In this embodiment, in step S1, real-time monitoring of current hot spots, generating corresponding hot words, and adding hot words to the hot data set include the following steps:

[0046] Monitor the current hot spots in real time, generate corresponding first hot words and second hot words, add the first hot words and second hot words to the hot data set after adding corresponding labels, the first hot words are hot words related to other categories except the current category, and the second hot words are hot words related to the current category;

[0047] In this embodiment, the hot words also include third hot words that are not related to the product.

[0048] Therefore, this embodiment divides hot words into three categories. The first two categories can be product words, or social event words that may be associated with products, such as product rate words: "BB start", "AAA" and "aa car", and social event words: "XX earthquake". Some products will be associated with earthquake scenes. Therefore, this part of social event words will also belong to the first hot words or the second hot words. The third hot words are all social event words and do not involve products, such as "YY visit" and the like. That is, the third hot words are the remaining hot words excluded after the first hot words and the second hot words determined by the user according to his own situation. The above hot word division facilitates subsequent adaptive interaction.

[0049] In this embodiment, the steps are also included:

[0050] When the real-time hot word belongs to the first hot word, a category label is added to the outbound user according to the category to which the real-time hot word belongs, so as to promote products of the same category according to the category label.

[0051] For example, the hot word "BB Start" belongs to the two-dimensional game category, "aa car" belongs to the car category, and so on. If the product promoted here is a mobile game, then "aa car" belongs to the second hot word and "AAA" belongs to the first hot word. After labeling, if there are subsequent products in the car category, you can consider calling the number of this outbound user to communicate.

[0052] S2. In the process of communicating with the outbound user, the real-time voice content of the outbound user is recognized to obtain the real-time text content, and it is determined whether the real-time hot words in the hot data set exist in the real-time text content. If so, the emotional tendency of the outbound user for the real-time hot words is obtained according to the emotional analysis of the real-time text content, and the hot interaction is performed according to the preset words corresponding to the emotional tendency;

[0053] Among them, the existing technologies can be used for speech transcription technology and text sentiment analysis.

[0054] When the outbound user's emotional tendency for the real-time hot words is obtained in step S2, hot interaction is performed according to the preset words corresponding to the emotional tendency, including the following steps:

[0055] Determine whether the real-time hot words in the real-time text content belong to the first hot words. If so, make a voice reply based on the first jump back speech associated with the first hot words and the content of the previous main process interaction. Otherwise, generate the corresponding preset speech for hot interaction based on the emotional tendency and the belonging product of the real-time hot words.

[0056] When it is the third hot word, a voice reply is made according to the second jump back speech associated with the third hot word and the content of the previous main process interaction.

[0057] Specifically, in the above steps, generating corresponding preset words for hot spot interaction according to the sentiment tendency and the belonging product of the real-time hot words includes:

[0058] Determine whether the emotional tendency is negative. If so, make a voice reply to the hot spot interaction according to the first preset words. Otherwise, determine whether the emotional tendency is positive. If so, make a voice reply to the hot spot interaction according to the second preset words. Otherwise, if the emotional tendency is neutral, make a voice reply to the hot spot interaction according to the third preset words. The first preset words include the differences between the product to which the real-time hot spot words belong and the current product, the second preset words include the similarities between the product to which the real-time hot spot words belong and the current product, and the third preset words include the advantages of the current product.

[0059] Therefore, since the first hot word and the third hot word have no connection with the currently promoted product, both can be transitioned to the original process for interaction through the corresponding jump back words.

[0060] When it is the second hot word, there are three situations:

[0061] (1) The real-time text content is "BB, have you heard of it?" At this time, the outbound user is neutral about "BB", and the third preset script can be used to introduce the advantages of this mobile game.

[0062] (2) The real-time text content is "Isn't your mobile game more casual than BB?" At this time, the outbound user has a negative impression of "BB". The first preset script is used to introduce the differences between this mobile game and the game BB.

[0063] (3) The real-time text content is "Is your mobile game as fun as BB?" At this time, the outbound user has a positive attitude towards "BB". The second preset script is used to introduce the similarities between this mobile game and the game BB.

[0064] S3. Determine in real time whether the current interaction is a sub-process interaction. If so, determine whether the current interaction can be ended based on the latest voice content of the current interaction. If so, return to the pre-set main process interaction by jumping back to the main process interaction to complete product promotion based on the main process interaction. Sub-process interactions include hotspot interactions.

[0065] Among them, the main process interactions are pre-set, such as opening communication, introducing product information, asking for customer opinions, and ending communication, and other common main process interactions.

[0066] In this embodiment, since the hotspot interaction belongs to the sub-process interaction separated from the original main process interaction, the communication efficiency needs to be considered on the basis of considering improving the response accuracy. Therefore, in this embodiment, the content and context of the conversation will be analyzed to determine whether a conversation can be ended. It uses deep learning and natural language processing technology. The existing technology can be used here. This embodiment mainly returns to the main process interaction by jumping back to the speech on the basis of judging whether the interaction can be ended. In this case, first, since there is a lot of workload and uncertainty in the sub-process interaction, timely termination and returning to the main process interaction can reduce the workload of the staff; second, in the telemarketing scenario, too much sub-process interaction is involved, and its final purchase intention rate is basically the same as the appropriate response; third, the communication time corresponds to the communication cost, that is, the outbound users that the AI ​​voice robot with a single computing amount can communicate in a unit time will be reduced due to too many sub-process interactions. Therefore, this embodiment can take into account the outbound call efficiency while improving the response accuracy of outbound users.

[0067] Embodiment 2

[0068] This embodiment is based on the application scenario described in the first embodiment. When there is manual customer service, please refer to Figure 2 , further comprising the steps of:

[0069] S4. When the real-time text content contains the intention to require manual customer service, manual customer service is connected;

[0070] Among them, access to manual customer service includes:

[0071] Determine whether the access to manual customer service requires waiting. If so, determine whether the outbound user has purchase intentions in the previous interaction process. If so, increase the queuing priority of the outbound user.

[0072] The above is a real-time process of voice interaction, which usually requires access to manual customer service, and the probability of interest is relatively high. Therefore, when the manual customer service is idle, directly access the manual customer service for communication. If the manual customer service is busy, it is necessary to judge the purchase intention of the outbound user to determine the queue priority, so as to ensure that the outbound users with purchase intention can access the manual customer service faster, thereby improving the communication experience of the outbound users with purchase intention.

[0073] S5. When the overall purchase intention of the outbound user is neutral and the emotional tendency towards the AI ​​voice robot in the overall interaction content is negative, the outbound user is marked as a user to be manually dialed.

[0074] Specifically, after hanging up the phone, those with a neutral overall purchase intention can make another call through a manual customer service. At this time, when the emotional tendency of the AI ​​voice robot is negative, the outbound user with a neutral overall purchase intention usually shows a high level of interest. Therefore, he or she will be included in the list of users to be dialed manually.

[0075] Of course, the dialing order of users to be manually dialed needs to be comprehensively considered through weight values. In this embodiment, the dialing order of users to be manually dialed is comprehensively considered based on the specific score of the outbound user's overall purchase intention and the specific score of the outbound user's emotional tendency towards the AI ​​voice robot. Among them, the closer the specific score of the outbound user's overall purchase intention is to the positive side, the higher the dialing order of the users to be manually dialed, and the lower the specific score of the outbound user's emotional tendency towards the AI ​​voice robot, the higher the dialing order of the users to be manually dialed. The weight values ​​of the two can be determined according to the actual situation of different categories.

[0076] It should be understood by those skilled in the art that embodiments of the present invention may be provided as methods, apparatuses or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0077] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (apparatus) and computer program products according to embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions.

[0078] It should be noted that in the claims, any reference numerals placed between brackets shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention may be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In the claims enumerating several means, several of these means may be embodied by the same hardware. The use of the words first, second, third, etc., is for convenience of expression only and does not indicate any order. These words may be understood as part of the component name.

[0079] In addition, it should be noted that, in the description of this specification, the description of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0080] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments after knowing the basic creative concept. Therefore, the claims should be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0081] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention should also include these modifications and variations.

Claims

1. An AI-based intelligent speech recognition method, characterized in that: Includes steps: Monitor current hot spots in real time, generate corresponding hot words, and add the hot words to the hot data set; In the process of communicating with an outbound user, the real-time voice content of the outbound user is recognized to obtain the real-time text content, and it is determined whether the real-time hot words in the hot data set exist in the real-time text content. If so, the emotional tendency of the outbound user towards the real-time hot words is obtained according to the emotional analysis of the real-time text content, and hot spot interaction is performed according to the preset words corresponding to the emotional tendency; Determine in real time whether the current interaction is a secondary process interaction. If so, determine whether the current interaction can be ended based on the latest voice content of the current interaction. If so, return to the pre-set main process interaction by jumping back to the speech to complete product promotion according to the main process interaction. The secondary process interaction includes hot spot interaction. The real-time monitoring of the current hot spots, generating corresponding hot words, and adding the hot words to the hot data set comprises the following steps: Monitor the current hot spots in real time, generate corresponding first hot words and second hot words, add the first hot words and the second hot words to the hot data set after adding corresponding labels, wherein the first hot words are hot words associated with other categories except the current category, and the second hot words are hot words associated with the current category; The hotspot interaction according to the preset words corresponding to the emotional tendency comprises the steps of: Determine whether the real-time hot words in the real-time text content belong to the first hot words. If so, make a voice reply based on the first jump back words associated with the first hot words and the content of the previous main process interaction. Otherwise, generate corresponding preset words for hot spot interaction based on the emotional tendency and the product to which the real-time hot words belong.

2. The AI-based intelligent speech recognition method according to claim 1, characterized in that: The hot words also include third hot words that are not related to the product, and the hot words interaction according to the preset words corresponding to the emotional tendency also includes: When it is the third hot word, a voice reply is made according to the second jump back speech associated with the third hot word and the content of the previous main process interaction.

3. The AI-based intelligent speech recognition method according to claim 1, characterized in that: Also includes the steps: When the real-time hot word belongs to the first hot word, a category label is added to the outbound user according to the category to which the real-time hot word belongs, so as to promote products of the same category according to the category label.

4. The AI-based intelligent speech recognition method according to claim 1, characterized in that: Generating corresponding preset words for hot spot interaction according to the sentiment tendency and the belonging product of the real-time hot spot words includes: Determine whether the emotional tendency is negative. If so, make a voice reply to the hot spot interaction according to the first preset words. Otherwise, determine whether the emotional tendency is positive. If so, make a voice reply to the hot spot interaction according to the second preset words. Otherwise, if the emotional tendency is neutral, make a voice reply to the hot spot interaction according to the third preset words. The first preset words include the differences between the product to which the real-time hot spot words belong and the current product, the second preset words include the similarities between the product to which the real-time hot spot words belong and the current product, and the third preset words include the advantages of the current product.

5. The AI-based intelligent speech recognition method according to claim 1, characterized in that: The real-time monitoring of the current hot spots and the generation of corresponding hot words include: Monitor the current hot spots in real time, and generate hot words based on the keywords and synonyms of the current hot spots.

6. The AI-based intelligent speech recognition method according to any one of claims 1 to 5, characterized in that: Also includes the steps: When the real-time text content contains an expression of the need for manual customer service, manual customer service is accessed.

7. The AI-based intelligent speech recognition method according to claim 6, characterized in that: The access to manual customer service includes: Determine whether the access of the manual customer service needs to wait. If so, determine whether the outbound user has a purchase intention in the previous interaction process. If so, increase the queuing priority of the outbound user.

8. The AI-based intelligent speech recognition method according to any one of claims 1 to 5, characterized in that: Also includes the steps: When the overall purchase intention of the outbound user is neutral and the emotional tendency towards the AI ​​voice robot in the overall interaction content is negative, the outbound user is marked as a user to be manually dialed.

9. The AI-based intelligent speech recognition method according to claim 8, characterized in that: The dialing order of the users to be manually dialed is determined based on the specific score of the outbound user's overall purchase intention and the specific score of the outbound user's emotional tendency towards the AI ​​voice robot, wherein, the closer the specific score of the outbound user's overall purchase intention is to the positive side, the higher the dialing order of the users to be manually dialed; and the lower the specific score of the outbound user's emotional tendency towards the AI ​​voice robot, the higher the dialing order of the users to be manually dialed.

Images