Method, device and computer equipment for recommending speech skills

Abstract

The present application belongs to the field of artificial intelligence technology, and particularly relates to a method, device and computer equipment for recommending speech. The method includes: obtaining the audio of the current call content; converting the audio into text information; parsing the customer's intention according to the text information; selecting at least one recommended speech that meets the preset proportion requirement from the speech library according to the customer's intention as the target recommended speech; pushing the customer's intention and the target recommended speech to the agent to guide the agent to talk to the customer according to the customer's intention and the target recommended speech; recording the feedback results of this call task, wherein the feedback results include sales success and sales failure; after a preset time, adjusting the proportion of the recommended speech in the speech library according to the results. The above method, device and computer equipment can improve call efficiency, sales success rate and the accuracy of recommended speech.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and more specifically, to a method, apparatus and computer device for recommending speech skills. Background Art

[0002] At present, agent sales has gradually become a widely used sales method in various fields, so companies are also paying more and more attention to the training and development of agents.

[0003] Although companies will provide irregular training for agents, new agents also need to spend a long time learning and accumulate a certain amount of experience to complete sales tasks well; in addition, since they need to face various types of customers, the content of the conversation may change, and there will be many unexpected conversation scenarios, which is very dependent on the agent's experience and on-the-spot reaction ability. Faced with the current complex and ever-changing sales scenarios, basic business training can no longer effectively cope with it, and the business capabilities of the agents are very high. Therefore, in many cases, the efficiency of agent sales calls is not high and the sales success rate is low. Summary of the invention

[0004] The main purpose of this application is to provide a method, device and computer equipment for recommending speech skills, aiming to solve the technical problems of low call efficiency and low sales success rate in the agent sales model.

[0005] In order to achieve the above-mentioned invention object, the present application provides a method for recommending speech skills, comprising:

[0006] Get the audio of the current call;

[0007] Converting the audio to text information;

[0008] Analyze customer intentions based on the text information;

[0009] According to the customer's intention, at least one recommended speech that meets a preset weight requirement is selected from the speech library as a target recommended speech;

[0010] Pushing the customer intention and the target recommended speech to the agent to guide the agent to talk to the customer according to the customer intention and the target recommended speech;

[0011] Record the feedback results of this call task, where the feedback results include sales success and sales failure;

[0012] After a preset time period, the weight of the recommended speech in the speech library is adjusted according to the result.

[0013] In one embodiment, before the step of obtaining the audio of the current call content, the following steps are included:

[0014] Get the sales success rate of each agent;

[0015] Obtain call records of agents whose sales success rate is greater than a preset threshold;

[0016] The speech in the call record is extracted and stored to obtain the speech library.

[0017] In one embodiment, the step of converting the audio into text information includes:

[0018] inputting the audio into an automatic speech recognition system;

[0019] The automatic speech recognition system generates corresponding text information according to the audio.

[0020] In one embodiment, the step of parsing the customer's intention according to the text information includes:

[0021] Inputting the text information into a natural language processing system;

[0022] The customer intention is analyzed according to the text information by the natural language system.

[0023] In one embodiment, the step of selecting at least one recommended speech that meets a preset weight requirement from a speech library according to the customer's intention includes:

[0024] Obtain keywords of the customer's intent;

[0025] Compare the keyword with the recommended speech words collected in the speech word library, and obtain the recommended speech words matching the keyword as the recommended speech words to be selected;

[0026] The candidate recommended speech that meets the preset weight requirement is selected as the target recommended speech.

[0027] In one embodiment, the speech library includes at least one sub-speech library, wherein the sub-speech library is named after a customer intent keyword, and the step of selecting at least one recommended speech that meets a preset weight requirement from the speech library according to the customer intent includes:

[0028] Obtain keywords of the customer's intent;

[0029] Compare the keyword with the name of the sub-speech library, and take the sub-speech library whose name matches the keyword as the target sub-speech library;

[0030] The recommended speech in the sub-speech library that meets the preset proportion requirement is selected as the target recommended speech.

[0031] In one embodiment, before the step of obtaining the audio of the current call content, the following steps are included:

[0032] Obtaining sales results of each agent, wherein the sales results include sales success and sales failure;

[0033] Obtain call records of agents whose sales results are successful sales;

[0034] The speech in the call record is extracted and stored to obtain the speech library.

[0035] The present application also provides a speech recommendation device, comprising:

[0036] An audio acquisition module is used to acquire the audio of the current call content;

[0037] A conversion unit, used for converting the audio into text information;

[0038] A customer intention analysis module, used to analyze customer intention according to the text information;

[0039] A target speech selection module is used to select at least one recommended speech that meets a preset weight requirement from the speech library as a target recommended speech according to the customer's intention;

[0040] A target recommendation speech push module is used to push the customer intention and the target recommendation speech to the agent, so as to guide the agent to talk to the customer according to the customer intention and the target recommendation speech;

[0041] A feedback result recording unit, used to record the feedback result of this call task, wherein the feedback result includes sales success and sales failure;

[0042] A weight adjustment unit is used to adjust the weight of the recommended speech in the speech library according to the result after a preset time period.

[0043] The present application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the speech recommendation method provided in any of the above embodiments are implemented.

[0044] The present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for recommending speech provided in any of the above embodiments are implemented.

[0045] The present application provides a method, device and computer equipment for recommending speech, which obtains the audio of the current call content; converts the audio into text information; analyzes the customer intention according to the text information; selects at least one recommended speech that meets the preset proportion requirement from the speech library according to the customer intention as the target recommended speech; pushes the customer intention and the target recommended speech to the agent to guide the agent to talk to the customer according to the customer intention and the target recommended speech; records the feedback results of this call task, wherein the feedback results include sales success and sales failure; after a preset time, adjusts the proportion of the recommended speech in the speech library according to the result. Through the real-time speech recommendation method, not only can the tension of the agent be relieved, but also the problems of low service efficiency and reduced sales success rate caused by insufficient agent experience or slow on-the-spot reaction can be reduced; in addition, since the speech library will adjust the proportion of the recommended speech according to the call results fed back within the preset time, the accuracy of the target recommended speech push can also be continuously improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flowchart of a method for recommending speech techniques according to an embodiment of the present application;

[0047] Figure 2 This is a flowchart of step S20 in the speech recommendation method according to an embodiment of the present application;

[0048] Figure 3 This is a flowchart of step S30 in the speech recommendation method according to another embodiment of the present application;

[0049] Figure 4 This is a flowchart of step S40 in the speech recommendation method according to another embodiment of the present application;

[0050] Figure 5 This is a flowchart of step S40 in the speech recommendation method according to another embodiment of the present application;

[0051] Figure 6 This is a flowchart of a method for recommending speech techniques according to another embodiment of the present application;

[0052] Figure 7 This is a flowchart of a method for recommending speech techniques according to another embodiment of the present application;

[0053] Figure 8 This is a structural diagram of a speech recommendation device according to an embodiment of the present application;

[0054] Fig. 9 A schematic diagram of the structure of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0056] Please refer to Figure 1 , an embodiment of the present application provides a method for recommending speech techniques, including steps S10-S70, and a detailed description of each step of the method is as follows.

[0057] In one embodiment, the method for recommending a speech technique includes:

[0058] S10, obtaining the audio of the current call content;

[0059] S20, converting the audio into text information;

[0060] S30, analyzing the customer's intention according to the text information;

[0061] S40, selecting at least one recommended speech that meets a preset weight requirement from the speech library according to the customer's intention as a target recommended speech;

[0062] S50, pushing the customer intention and the target recommended speech to the agent, so as to guide the agent to talk with the customer according to the customer intention and the target recommended speech;

[0063] S60, recording the feedback result of this call task, wherein the feedback result includes sales success and sales failure;

[0064] S70. After a preset time period, adjust the weight of the recommended speech in the speech library according to the result.

[0065] As described in the above step S10, when the agent dials the customer's phone according to the customer list and the call is connected, the audio of the current call content is obtained in real time for subsequent use. The method of obtaining the audio of the call content can be to obtain it through the sound receiving equipment arranged on site and transmit it directly to the system; in some call scenarios with higher confidentiality requirements, for the sake of communication security, the audio of the call content can also be technically encrypted before transmission; or a communication transfer station can be set up in the telephone communication between the agent and the customer, and the communication transfer station can obtain the audio of the call content, etc., which are not limited here.

[0066] In one embodiment, a microphone array is used to collect the current call content. The advantages of microphone array sound collection mainly include echo suppression, reverberation removal, noise suppression, etc. For example, when an agent is talking to a customer in a studio, the voice signal collected by the microphone contains not only the signal propagated by the direct path, but also the delayed attenuated signal generated by the reflection of the room boundary and objects in the room. This multipath propagation effect introduces echo and spectral distortion into the observed signal, which becomes reverberation, which seriously affects the source signal and causes the signal quality and intelligibility to decrease. Therefore, removing reverberation plays a great role in improving the intelligibility of the voice signal. The purpose of the noise suppression algorithm is to estimate the target voice signal from the observed signal affected by additive noise. However, the main problem of all single-microphone noise suppression algorithms is that it will distort the voice signal. Although the voice quality is improved, the intelligibility is reduced. The use of a microphone array can suppress noise without affecting the voice signal too much.

[0067] As described in step S20 above, please refer to Figure 2 In one embodiment, step S20, i.e., the step of converting the audio into text information, comprises:

[0068] S201, inputting the audio into an automatic speech recognition system;

[0069] S202: Generate corresponding text information according to the audio through the automatic speech recognition system.

[0070] As described in the above steps S201-S202, the technology that the automatic speech recognition system relies on is the automatic speech recognition technology (ASR). The automatic speech recognition technology is mainly used to convert human speech into text, and its principle process is "speech input→encoding (feature extraction)→decoding→text output". The acoustic model and language model of the automatic speech recognition system in this embodiment are pre-trained models. During the operation of the automatic speech recognition system, the continuously acquired speech information can also be used as training samples to synchronously train the acoustic model and the speech model, so that its output performance can be continuously optimized.

[0071] In one embodiment, in order to achieve better speech recognition effect, the audio is preprocessed before entering the automatic speech recognition system. The process of preprocessing the audio may include the following steps: pre-filtering → A / D conversion → pre-emphasis → framing → windowing → endpoint detection. Among them, the purpose of pre-emphasis is to emphasize the high-frequency part of the speech, remove the influence of lip radiation, and increase the high-frequency resolution of the speech, because the high-frequency end is attenuated at 6dB / oct (octave) above about 800Hz, and the higher the frequency, the smaller the corresponding component. Therefore, the high-frequency part of the speech signal should be enhanced before analyzing it; although the speech signal has time-varying characteristics, its characteristics remain basically unchanged within a short time range (generally considered to be 10 to 30ms), that is, it is relatively stable, so it can be regarded as a quasi-steady-state process, that is, the speech signal has short-term stability, so the speech signal needs to be framed; the purpose of windowing is to emphasize the speech waveform near a certain speech signal sampling and weaken the rest of the waveform; and the purpose of endpoint detection is to distinguish between speech and non-speech areas, that is, to accurately locate the starting point and ending point of the speech from the noisy speech, and then remove the silence and noise parts to find the truly effective content of a speech.

[0072] In one embodiment, in order to further optimize the effect of speech recognition, an automatic error correction mechanism is introduced for the automatic speech recognition system. For example, when one of the customers that the agent wants to contact is named Guo Jia, if the voice content of "Hello, are you Mr. Guo Jia Guo?" appears in the conversation of the agent during the call, the general speech recognition will most likely recognize the content as "Hello, are you Mr. Guo Guo?", but in this embodiment, the customer list can be stored in the automatic speech recognition system in advance, and the automatic speech recognition system combines the customer list and audio information to obtain the text "Guo Jia" that better matches the audio information of "Guo Jia", so that the final output text is "Hello, are you Mr. Guo Jia Guo?", thereby obtaining a more accurate speech recognition result.

[0073] As described in step S30 above, please refer to Figure 3 In one embodiment, step S30, i.e., the step of parsing the customer's intention according to the text information, includes:

[0074] S301, inputting the text information into a natural language processing system;

[0075] S302: Analyze the customer's intention according to the text information through the natural language system.

[0076] As described in the above steps S301-S302, Natural Language Processing (NLP) is an important direction in the fields of computer science and artificial intelligence. It studies various theories and methods that can achieve effective communication between people and computers using natural language. Natural language processing works through machine learning. Phrases, sentences, and sometimes even the contents of entire books are input into the machine learning engine and processed using grammatical rules and / or people's real language habits. The computer then uses this data to find patterns and infer the next results.

[0077] Currently, the more commonly used natural language systems include NLTK, SpaCy, TextBlob, etc. In this embodiment, TextBlob is used as a natural language system to analyze customer intent. TextBlob is an open source text processing library. It can be used to perform many natural language processing tasks, such as part-of-speech tagging, noun component extraction, sentiment analysis, text translation, etc.

[0078] As described in step S40 above, please refer to Figure 4 In one embodiment, step S40, i.e., the step of selecting at least one recommended speech that meets a preset weight requirement from the speech library as a target recommended speech according to the customer's intention, includes:

[0079] S401, obtaining keywords of the customer's intention;

[0080] S402, comparing the keyword with the recommended speech words collected in the speech word library, and obtaining the recommended speech words matching the keyword as the recommended speech words to be selected;

[0081] S403: Select the candidate recommended speech that meets the preset weight requirement as the target recommended speech.

[0082] As described in the above steps S401-S403, when the customer's intention is parsed by the automatic speech recognition system, the keyword of the customer's intention is obtained. For example, if the customer's intention is "consulting about credit-related issues", "credit" is used as the keyword of the customer's intention. The recommended words collected in the speech library can include "gold medal sales words" of various industries, words specially customized for a certain business or business type, historical words with a high transaction success rate, etc. In this embodiment, the recommended words collected in the speech library can be obtained from the historical call content of the excellent performance seat, that is, the historical words with a high transaction success rate. "Credit" is used as the customer intention keyword to compare with the recommended words collected in the speech library, and the recommended words containing the text "credit" in the text content are regarded as recommended words matching the keyword "credit", and are used as words to be recommended. Since the speech library may contain multiple (more than three) words to be recommended, in order to achieve more efficient and targeted speech recommendation, the recommended words in the speech library can be weighted. Please refer to Table 1, which shows a table of recommended speech matching the keyword "credit investigation", where there are five recommended speech, the weight of recommended speech 1 is 30%, the weight of recommended speech 2 is 25%, the weight of recommended speech 3 is 25%, the weight of recommended speech 4 is 12%, and the weight of recommended speech 4 is 8%. If the preset weight requirement is to select the top three recommended speech as the target recommended speech, then recommended speech 1, recommended speech 2 and recommended speech 3 will be pushed to the agent. It should be noted that the preset weight requirement can be adjusted according to actual needs, for example, the recommended speech ranked first in weight can be used as the target recommended speech, or the recommended speech ranked first in weight can be used as the target recommended speech, etc., which is not limited here.

[0083] Table 1: Recommended scripts matching the keyword "credit investigation"

[0084]

[0085] As described in step S50 above, after the target recommendation words are selected, the analyzed customer intent and the selected target recommendation words are pushed to the agent to guide the agent to communicate with the customer according to the customer intent and the target recommendation words. In one embodiment, by installing a large screen panel in the agent's studio, the customer intent and the target recommendation words are displayed on the large screen to guide the agent to communicate with the customer.

[0086] As described in the above steps S60-S70, the agent communicates with the customer through the analyzed customer intention and target recommendation speech, and records the feedback result of this call task after the call, that is, whether this sale is successful or failed. And after a preset time (which can be set to 1 hour), the proportion of recommended speech in the speech library is adjusted according to the feedback result. Specifically, when the feedback result of this call task is a successful sale, the proportion of the target recommended speech is increased, and when the feedback result of this call task is a failed sale, the proportion of the target recommended speech is reduced. The proportion of increasing or decreasing the proportion can be set according to actual needs and is not limited here.

[0087] The present application provides a method for recommending a speech, which obtains the audio of the current call content; converts the audio into text information; analyzes the customer intention according to the text information; selects at least one recommended speech that meets the preset weight requirement from the speech library according to the customer intention as the target recommended speech; pushes the customer intention and the target recommended speech to the agent to guide the agent to talk to the customer according to the customer intention and the target recommended speech; records the feedback results of this call task, wherein the feedback results include sales success and sales failure; after a preset time, adjusts the weight of the recommended speech in the speech library according to the result. Through the real-time speech recommendation method, not only can the tension of the agent be relieved, but also the problems of low service efficiency and reduced sales success rate caused by insufficient agent experience or slow on-the-spot reaction can be reduced; in addition, since the speech library will adjust the weight of the recommended speech according to the call results fed back within the preset time, the accuracy of the target recommended speech push can also be continuously improved.

[0088] In one embodiment, the speech library includes at least one sub-speech library, wherein the sub-speech library is named after the customer's intended keyword. Exemplarily, the speech library includes three sub-speech libraries, specifically "credit sub-speech library", "insurance sub-speech library" and "loan sub-speech library". Among them, the recommended speech collected in the credit sub-speech library is speech related to the credit business, the recommended speech collected in the insurance sub-speech library is speech related to the insurance business, and the recommended speech collected in the loan sub-speech library is speech related to the loan business.

[0089] In one embodiment, see Figure 5 , according to the customer's intention, the step of selecting at least one recommended speech that meets a preset weight requirement from the speech library as a target recommended speech includes:

[0090] S501, obtaining keywords of the customer's intention;

[0091] S502, comparing the keyword with the name of the sub-talk library, and taking the sub-talk library whose name matches the keyword as the target sub-talk library;

[0092] S503: Select the recommended speech that meets the preset weight requirement in the sub-speech library as the target recommended speech.

[0093] As described in the above steps S501-S503, when the customer's intention is analyzed by the automatic speech recognition system, the keywords of the customer's intention are obtained. For example, if the customer's intention is "to consult about credit-related issues," "credit" is used as the keyword of the customer's intention. The recommended scripts collected in the script library can include "gold medal sales scripts" of various industries, scripts specially customized for a certain business or business type, and historical scripts with high transaction success rates.

[0094] In this embodiment, the recommended speech collected in the speech library is obtained from the call content of the seats with excellent performance, that is, the historical speech with a high transaction success rate. The speech library includes three sub-speech libraries, specifically "credit sub-speech library", "insurance sub-speech library" and "loan sub-speech library". Among them, the recommended speech collected in the credit sub-speech library is speech related to the credit business, the recommended speech collected in the insurance sub-speech library is speech related to the insurance business, and the recommended speech collected in the loan sub-speech library is speech related to the loan business.

[0095] Take "credit" as the customer intent keyword and compare it with the name of the sub-speech library in the speech library. The "credit sub-speech library" is the sub-speech library that matches the keyword "credit", and it is used as the target sub-speech. Since the speech library may contain multiple (more than three) speech to be recommended, in order to achieve more efficient and targeted speech recommendation, the recommended speech in the speech library can be weighted. The principle is the same as above. Figure 4 The principle of setting the speech weight in the illustrated embodiment is the same and will not be elaborated here.

[0096] In one embodiment, please refer to Figure 6 , before step S10, that is, the step of obtaining the audio of the current call content, comprising:

[0097] S00. Obtain the sales success rate of each agent;

[0098] S01. Obtain call records of agents whose sales success rate is greater than a preset threshold;

[0099] S02. Extract and store the speech in the call record to obtain the speech library.

[0100] As described in the above steps S00-S02, in order to obtain high-quality and targeted recommendation scripts, they can be obtained from the historical call records of agents whose sales success rate is greater than a preset threshold (such as 80%). Specifically, first, the sales success rate of each agent is obtained; secondly, the call records of agents whose sales rate is greater than the preset threshold are obtained; finally, the scripts in the above call records are extracted and stored to obtain a script library.

[0101] In one embodiment, please refer to Figure 7 , before step S10, that is, the step of obtaining the audio of the current call content, comprising:

[0102] S03, obtaining the sales results of each seat, wherein the sales results include sales success and sales failure;

[0103] S04. Acquire the call records of the agents whose sales results are successful sales;

[0104] S05. Extract and store the speech in the call record to obtain the speech library.

[0105] As described in the above steps S03-S05, in order to obtain higher quality recommendation scripts to improve the sales success rate, the recommendation scripts can be obtained from the historical call records of the agents whose sales results are successful sales. Specifically, first, the sales results of each agent are obtained, where the sales results include successful sales and failed sales; secondly, the call records of the agents whose sales results are successful sales are obtained; finally, the scripts in the above call records are extracted and stored to obtain a script library.

[0106] Please refer to Figure 8 In an embodiment of the present application, a speech recommendation device is also provided, including:

[0107] The audio acquisition module 801 is used to acquire the audio of the current call content;

[0108] A conversion module 802, used to convert the audio into text information;

[0109] A customer intention analysis module 803 is used to analyze customer intention according to the text information;

[0110] A target speech selection module 804 is used to select at least one recommended speech that meets a preset weight requirement from the speech library as a target recommended speech according to the customer's intention;

[0111] The target recommendation speech push module 805 is used to push the customer intention and the target recommendation speech to the agent, so as to guide the agent to talk to the customer according to the customer intention and the target recommendation speech;

[0112] Feedback result recording module 806, used to record the feedback result of this call task, wherein the feedback result includes sales success and sales failure;

[0113] The weight adjustment module 807 is used to adjust the weight of the recommended speech in the speech library according to the result after a preset time period.

[0114] In this embodiment, when the agent dials a customer's phone according to the customer list and the call is connected, the audio of the current call content is obtained in real time for subsequent use. The method of obtaining the audio of the call content can be to obtain it through the sound receiving equipment arranged on site and transmit it directly to the system; in some call scenarios with higher confidentiality requirements, for the sake of communication security, the audio of the call content can also be technically encrypted before transmission; or a communication transfer station can be set up in the telephone communication between the agent and the customer, and the communication transfer station can obtain the audio of the call content, etc., which are not limited here.

[0115] In one embodiment, a microphone array is used to collect the current call content. The advantages of microphone array sound collection mainly include echo suppression, reverberation removal, noise suppression, etc. For example, when an agent is talking to a customer in a studio, the voice signal collected by the microphone contains not only the signal propagated by the direct path, but also the delayed attenuated signal generated by the reflection of the room boundary and objects in the room. This multipath propagation effect introduces echo and spectral distortion into the observed signal, which becomes reverberation, which seriously affects the source signal and causes the signal quality and intelligibility to decrease. Therefore, removing reverberation plays a great role in improving the intelligibility of the voice signal. The purpose of the noise suppression algorithm is to estimate the target voice signal from the observed signal affected by additive noise. However, the main problem of all single-microphone noise suppression algorithms is that it will distort the voice signal. Although the voice quality is improved, the intelligibility is reduced. The use of a microphone array can suppress noise without affecting the voice signal too much.

[0116] In one embodiment, the conversion module 802 includes an audio input unit and a text information generation unit. The audio input unit is used to input the audio into the automatic speech recognition system, and the text information generation unit is used to generate corresponding text information according to the audio through the automatic speech recognition system.

[0117] In this embodiment, the technology that the automatic speech recognition system relies on is automatic speech recognition technology (ASR). Automatic speech recognition technology is mainly used to convert human speech into text, and its principle process is "speech input→encoding (feature extraction)→decoding→text output". The acoustic model and language model of the automatic speech recognition system in this embodiment are pre-trained models. During the operation of the automatic speech recognition system, the continuously acquired speech information can also be used as training samples to synchronously train the acoustic model and the speech model, so that its output performance can be continuously optimized.

[0118] In one embodiment, in order to achieve better speech recognition effect, the audio is preprocessed before entering the automatic speech recognition system. The process of preprocessing the audio may include the following steps: pre-filtering → A / D conversion → pre-emphasis → framing → windowing → endpoint detection. Among them, the purpose of pre-emphasis is to emphasize the high-frequency part of the speech, remove the influence of lip radiation, and increase the high-frequency resolution of the speech, because the high-frequency end is attenuated at 6dB / oct (octave) above about 800Hz, and the higher the frequency, the smaller the corresponding component. Therefore, the high-frequency part of the speech signal should be enhanced before analyzing it; although the speech signal has time-varying characteristics, its characteristics remain basically unchanged within a short time range (generally considered to be 10 to 30ms), that is, it is relatively stable, so it can be regarded as a quasi-steady-state process, that is, the speech signal has short-term stability, so the speech signal needs to be framed; the purpose of windowing is to emphasize the speech waveform near a certain speech signal sampling and weaken the rest of the waveform; and the purpose of endpoint detection is to distinguish between speech and non-speech areas, that is, to accurately locate the starting point and ending point of the speech from the noisy speech, and then remove the silence and noise parts to find the truly effective content of a speech.

[0119] In one embodiment, in order to further optimize the effect of speech recognition, an automatic error correction mechanism is introduced for the automatic speech recognition system. For example, when one of the customers that the agent wants to contact is named Guo Jia, if the voice content of "Hello, are you Mr. Guo Jia Guo?" appears in the conversation of the agent during the call, the general speech recognition will most likely recognize the content as "Hello, are you Mr. Guo Guo?", but in this embodiment, the customer list can be stored in the automatic speech recognition system in advance, and the automatic speech recognition system combines the customer list and audio information to obtain the text "Guo Jia" that better matches the audio information of "Guo Jia", so that the final output text is "Hello, are you Mr. Guo Jia Guo?", thereby obtaining a more accurate speech recognition result.

[0120] In one embodiment, the customer intention analysis module 803 includes a text information input unit and a customer intention analysis unit. The text information input unit is used to input the text information into a natural language processing system, and the customer intention analysis unit is used to analyze the customer intention according to the text information through the natural language system.

[0121] In this embodiment, Natural Language Processing (NLP) is an important direction in the fields of computer science and artificial intelligence, which studies various theories and methods that can achieve effective communication between people and computers using natural language. Natural language processing works through machine learning, where phrases, sentences, and sometimes even the contents of entire books are input into the machine learning engine and processed using grammatical rules and / or people's real language habits, and then the computer uses this data to find patterns and infer the next results.

[0122] Currently, the more commonly used natural language systems include NLTK, SpaCy, TextBlob, etc. In this embodiment, TextBlob is used as a natural language system to analyze customer intent. TextBlob is an open source text processing library. It can be used to perform many natural language processing tasks, such as part-of-speech tagging, noun component extraction, sentiment analysis, text translation, etc.

[0123] In one embodiment, the target speech selection module 804 includes a keyword acquisition unit, a candidate recommended speech acquisition unit and a target recommended speech selection unit. The keyword acquisition unit is used to acquire the keyword of the customer's intention; the candidate recommended speech acquisition unit is used to compare the keyword with the recommended speech collected in the speech library, and acquire the recommended speech matching the keyword as the candidate recommended speech; the target recommended speech selection unit is used to select the candidate recommended speech that meets the preset weight requirement as the target recommended speech.

[0124] In this embodiment, when the customer's customer intent is parsed by the automatic speech recognition system, the keyword of the customer intent is obtained. For example, if the customer's customer intent is "consulting about credit-related issues," "credit" is used as the keyword of the customer intent. The recommended speech collected in the speech library may include "gold medal sales speech" in various industries, speech specially customized for a certain business or business type, historical speech with a high transaction success rate, etc. In this embodiment, the recommended speech collected in the speech library is obtained from the call content of the excellent performance seat, that is, the historical speech with a high transaction success rate. "Credit" is used as the customer intent keyword to compare with the recommended speech collected in the speech library, and the recommended speech containing the text "credit" in the text content is regarded as the recommended speech matching the keyword "credit", and it is used as the speech to be recommended. Since the speech library may contain multiple (more than three) speech to be recommended, in order to achieve more efficient and targeted speech recommendation, the recommended speech in the speech library can be weighted. For example, a table of recommended words matching the keyword "credit investigation" contains recommended words 1, 2, 3, 4 and 5, among which the weight of recommended words 1 is 30%, the weight of recommended words 2 is 25%, the weight of recommended words 3 is 25%, the weight of recommended words 4 is 12% and the weight of recommended words 4 is 8%. If the preset weight requirement is to select the top three recommended words as the target recommended words, then recommended words 1, 2 and 3 are pushed to the agent. It should be noted that the preset weight requirement can be adjusted according to actual needs, for example, the recommended words ranked first in weight are used as the target recommended words, or the recommended words ranked first in weight are used as the target recommended words, etc., which are not limited here.

[0125] In one embodiment, the target speech selection module 804 includes a keyword acquisition unit, a target sub-speech library acquisition unit, and a target recommended speech selection unit. The keyword acquisition unit is used to acquire the keyword of the customer's intention; the target sub-speech library acquisition unit is used to compare the keyword with the name of the sub-speech library, and use the sub-speech library whose name matches the keyword as the target sub-speech library; the target recommended speech selection unit is used to select the candidate recommended speech that meets the preset weight requirement as the target recommended speech.

[0126] In one embodiment, the speech recommendation device further includes a keyword acquisition module, a candidate recommended speech acquisition module and a target recommended speech acquisition module. The keyword acquisition module is used to acquire the keyword of the customer's intention; the candidate recommended speech acquisition module is used to compare the keyword with the recommended speech collected in the speech library, and acquire the recommended speech matching the keyword as the candidate recommended speech; the target recommended speech acquisition module is used to select the candidate recommended speech that meets the preset weight requirement as the target recommended speech.

[0127] In one embodiment, the speech recommendation device further includes a keyword acquisition module, a target sub-speech library acquisition module and a target recommended speech acquisition module. The keyword acquisition module is used to acquire the keyword of the customer's intention; the target sub-speech library acquisition module is used to compare the keyword with the name of the sub-speech library, and use the sub-speech library whose name matches the keyword as the target sub-speech library; the target recommended speech acquisition module is used to select the recommended speech in the sub-speech library that meets the preset weight requirement as the target recommended speech.

[0128] In this embodiment, when the customer's intention is analyzed by the automatic speech recognition system, the keyword of the customer's intention is obtained. For example, if the customer's intention is "consulting about credit-related issues," "credit" is used as the keyword of the customer's intention. The recommended scripts collected in the script library can include "gold medal sales scripts" of various industries, scripts specially customized for a certain business or business type, and historical scripts with a high transaction success rate.

[0129] In this implementation, the recommended scripts collected in the script library can be obtained from the call content of the seats with excellent performance, that is, the historical scripts with a high transaction success rate. Exemplarily, the script library includes three sub-script libraries, specifically "credit sub-script library", "insurance sub-script library" and "loan sub-script library". Among them, the recommended scripts collected in the credit sub-script library are scripts related to the credit business, the recommended scripts collected in the insurance sub-script library are scripts related to the insurance business, and the recommended scripts collected in the loan sub-script library are scripts related to the loan business.

[0130] Take "credit" as the customer intent keyword and compare it with the name of the sub-speech library in the speech library. The "credit sub-speech library" is the sub-speech library that matches the keyword "credit", and it is used as the target sub-speech. Since the speech library may contain multiple (more than three) speech to be recommended, in order to achieve more efficient and targeted speech recommendations, the recommended speech in the speech library can be weighted.

[0131] Please refer to Fig. 9 In the embodiment of the present application, a computer device is also provided, whose internal structure can be as follows Fig. 9As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a storage medium and an internal memory. The storage medium stores an operating system, a computer program and a database. The memory provides an environment for the operation of the operating system and the computer program in the storage medium. The database of the computer device is used to store relevant data of a speech recommendation method. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the speech recommendation method provided in any of the above embodiments is implemented.

[0132] A computer-readable storage medium is also provided in an embodiment of the present application. The computer-readable storage medium may be non-volatile or volatile, and a computer program is stored thereon. When the computer program is executed by a processor, the method for recommending speech provided in any of the above embodiments is implemented.

[0133] It will be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided in the present application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), expanded SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0134] The present application provides a method, device and computer equipment for recommending speech, which obtains the audio of the current call content; converts the audio into text information; analyzes the customer intention according to the text information; selects at least one recommended speech that meets the preset proportion requirement from the speech library according to the customer intention as the target recommended speech; pushes the customer intention and the target recommended speech to the agent to guide the agent to talk to the customer according to the customer intention and the target recommended speech; records the feedback results of this call task, wherein the feedback results include sales success and sales failure; after a preset time, adjusts the proportion of the recommended speech in the speech library according to the result. Through the real-time speech recommendation method, not only can the tension of the agent be relieved, but also the problems of low service efficiency and reduced sales success rate caused by insufficient agent experience or slow on-the-spot reaction can be reduced; in addition, since the speech library will adjust the proportion of the recommended speech according to the call results fed back within the preset time, the accuracy of the target recommended speech push can also be continuously improved.

[0135] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the existence of other identical elements in the process, device, article or method including the element.

[0136] The above description is only a preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for recommending speech skills, characterized in that: include: Get the audio of the current call; Converting the audio to text information; Analyze customer intentions based on the text information; According to the customer's intention, at least one recommended speech that meets a preset weight requirement is selected from the speech library as a target recommended speech; Pushing the customer intention and the target recommended speech to the agent to guide the agent to talk to the customer according to the customer intention and the target recommended speech; Record the feedback results of this call task, where the feedback results include sales success and sales failure; After a preset time period, the weight of the recommended speech in the speech library is adjusted according to the result.

2. The method for recommending words according to claim 1, characterized in that: Before the step of obtaining the audio of the current call content, the method includes: Get the sales success rate of each agent; Obtain call records of agents whose sales success rate is greater than a preset threshold; The speech in the call record is extracted and stored to obtain the speech library.

3. The method for recommending words according to claim 1, characterized in that: The step of converting the audio into text information comprises: inputting the audio into an automatic speech recognition system; The automatic speech recognition system generates corresponding text information according to the audio.

4. The method for recommending words according to claim 1, characterized in that: The step of analyzing the customer's intention according to the text information includes: Inputting the text information into a natural language processing system; The customer intention is analyzed according to the text information by the natural language system.

5. The method for recommending words according to claim 1, characterized in that: The step of selecting at least one recommended speech that meets a preset weight requirement from a speech library according to the customer's intention includes: Obtain keywords of the customer's intent; Compare the keyword with the recommended speech words collected in the speech word library, and obtain the recommended speech words matching the keyword as the recommended speech words to be selected; The candidate recommended speech that meets the preset weight requirement is selected as the target recommended speech.

6. The method for recommending words according to claim 1, characterized in that: The speech library includes at least one sub-speech library, wherein the sub-speech library is named after a customer intent keyword, and the step of selecting at least one recommended speech that meets a preset weight requirement from the speech library according to the customer intent includes: Obtain keywords of the customer's intent; Compare the keyword with the name of the sub-speech library, and take the sub-speech library whose name matches the keyword as the target sub-speech library; The recommended speech in the sub-speech library that meets the preset proportion requirement is selected as the target recommended speech.

7. The method for recommending speech techniques according to claim 1, characterized in that: Before the step of obtaining the audio of the current call content, the method includes: Obtaining sales results of each agent, wherein the sales results include sales success and sales failure; Get the call records of agents whose sales results are successful; The speech in the call record is extracted and stored to obtain the speech library.

8. A speech recommendation device, characterized in that: include: An audio acquisition module is used to acquire the audio of the current call content; A conversion module, used for converting the audio into text information; A customer intention analysis module, used to analyze customer intention according to the text information; A target speech selection module is used to select at least one recommended speech that meets a preset weight requirement from the speech library as a target recommended speech according to the customer's intention; A target recommendation speech push module is used to push the customer intention and the target recommendation speech to the agent, so as to guide the agent to talk to the customer according to the customer intention and the target recommendation speech; A feedback result recording module is used to record the feedback result of this call task, wherein the feedback result includes sales success and sales failure; The weight adjustment module is used to adjust the weight of the recommended speech in the speech library according to the results after a preset period of time.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the speech recommendation method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the speech recommendation method described in any one of claims 1 to 7 are implemented.

Images