Method, device and storage medium for detecting dialogue text information

Abstract

The present application discloses a method, device and storage medium for detecting speech text information. The method comprises: obtaining speech text information; detecting the speech text information using at least one type of pre-set quality detection model to determine whether the speech text information has quality problems; and if the speech text information has quality problems, outputting alarm information corresponding to the quality problems.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to a method, device and storage medium for detecting dialogue text information. Background Art

[0002] At present, voice robot products and services based on AI technology are in the ascendant. Based on the technology of voice robot outbound call center, voice-related tasks in various business scenarios such as telemarketing, pre-sales, after-sales, return visits, reminders, collection, research, customer care, etc. can be completed.

[0003] Voice robot users configure voice robots according to their target tasks, and limit the content of the robot's speech by configuring fixed or semi-fixed words, which makes the task objectives clear and is very convenient for users to modify the configuration. However, voice robot users often set words according to their own business tasks or purposes, and may not take into account the user's psychological feelings and acceptance. Therefore, it is necessary to test the words of voice robots.

[0004] The existing method of detecting the speech of voice robots is to detect them during the robot's voice dialogue. This method cannot prevent speech risks, and this detection method is an overall assessment of the entire conversation. It only scores to mark the good and bad, and there is no way to point out the problems in detail. It has little guiding significance for dialogue improvement work.

[0005] The method of detecting the speech techniques of a voice robot in the above-mentioned prior art is to perform detection during the robot's voice conversation. This method cannot prevent speech techniques risks, and this detection method is an overall assessment of the entire conversation. It cannot point out the problem in detail. It is a technical problem with little guiding significance for conversation technique improvement work, and no effective solution has been proposed so far. Summary of the invention

[0006] The embodiments of the present disclosure provide a method, device and storage medium for detecting dialogue text information, so as to at least solve the technical problems that the method for detecting the speech of a voice robot in the prior art is to detect the speech during the robot voice dialogue work, which cannot prevent the speech risks, and this detection method is an overall assessment of the entire conversation, and there is no way to point out the problem in detail, and it has little guiding significance for the dialogue improvement work.

[0007] According to one aspect of an embodiment of the present disclosure, a method for detecting speech text information is provided, including: acquiring speech text information; detecting the speech text information using at least one type of pre-set quality detection model to determine whether the speech text information has quality problems; and in the event that the speech text information has quality problems, outputting alarm information corresponding to the quality problems.

[0008] According to another aspect of an embodiment of the present disclosure, a storage medium is further provided, the storage medium including a stored program, wherein when the program is running, a processor executes any one of the methods described above.

[0009] According to another aspect of an embodiment of the present disclosure, there is also provided a device for detecting speech text information, including: an acquisition module for acquiring speech text information; a detection module for detecting the speech text information using at least one type of pre-set quality detection model to determine whether the speech text information has quality problems; and an output module for outputting alarm information corresponding to the quality problem when the speech text information has quality problems.

[0010] According to another aspect of an embodiment of the present disclosure, there is also provided a device for detecting speech text information, including: a processor; and a memory connected to the processor, for providing the processor with instructions for processing the following processing steps: obtaining speech text information; detecting the speech text information using at least one type of pre-set quality detection model to determine whether the speech text information has quality problems; and in the event that the speech text information has quality problems, outputting alarm information corresponding to the quality problems.

[0011] In the disclosed embodiment, the detection of the speech text information of the voice robot is performed before the speech text information goes online, and the risks existing in the speech text information can be prevented in advance so as to improve the speech text information. In addition, the present invention detects the speech text information by detecting each detail in the speech text information, and outputs the warning information corresponding to the quality problem of the detail when a quality problem occurs in a certain detail. This reminds and assists the robot user to improve or optimize the configuration of the voice robot so that it can better complete the target task.

[0012] This solves the problem that the existing method of detecting the speech techniques of voice robots is to detect them during the robot's voice dialogue work. This method cannot prevent speech techniques risks, and this detection method is an overall assessment of the entire dialogue. It cannot point out the problem in detail and has little guiding significance for dialogue technique improvement work. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The drawings described herein are used to provide a further understanding of the present disclosure and constitute a part of the present application. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure. In the drawings:

[0014] Figure 1 is a hardware structure block diagram of a computer terminal for implementing the method according to Embodiment 1 of the present disclosure;

[0015] Figure 2 is a flowchart of a method for detecting dialogue text information according to the first aspect of Embodiment 1 of the present disclosure;

[0016] Figure 3 is a schematic diagram of the architecture of the method for detecting dialogue text information according to the first aspect of Embodiment 1 of the present disclosure;

[0017] Figure 4 is a schematic diagram of a device for detecting dialogue text information according to Embodiment 2 of the present disclosure; and

[0018] Figure 5 It is a schematic diagram of a device for detecting dialogue text information according to Embodiment 3 of the present disclosure. DETAILED DESCRIPTION

[0019] In order to enable those skilled in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only embodiments of a part of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present disclosure.

[0020] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0021] Example 1

[0022] According to this embodiment, a method embodiment for detecting conversational text information is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0023] The method embodiment provided in this embodiment can be executed in a mobile terminal, a computer terminal, a server or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computing device for detecting conversational text information. Figure 1 As shown, the computing device may include one or more processors (the processor may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory for storing data, and a transmission device for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0024] It should be noted that the one or more processors and / or other data processing circuits described above may generally be referred to herein as "data processing circuitry". The data processing circuitry may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuitry may be a single independent processing module, or may be incorporated in whole or in part into any of the other components of the computing device. As involved in the embodiments of the present disclosure, the data processing circuitry acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0025] The memory can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the method for detecting the dialogue text information in the embodiment of the present disclosure. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, the method for detecting the dialogue text information of the above-mentioned application program is realized. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the computing device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0026] The transmission device is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computing device. In one example, the transmission device includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device can be a radio frequency (Radio Frequency, RF) module, which is used to communicate with the Internet wirelessly.

[0027] The display may be, for example, a touch screen liquid crystal display (LCD) that may enable a user to interact with a user interface of the computing device.

[0028] It should be noted that, in some optional embodiments, the above Figure 1 The computing device shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware elements and software elements. Figure 1 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computing devices described above.

[0029] According to a first aspect of the present embodiment, a method for detecting dialogue text information is provided. Figure 2 A schematic diagram showing the process of the method is shown in FIG. Figure 2 As shown, the method includes:

[0030] S202: Acquire speech text information;

[0031] S204: Using at least one type of pre-set quality detection model to detect the dialogue text information, and determine whether the dialogue text information has quality problems; and

[0032] S206: When there are quality problems in the speech text information, output alarm information corresponding to the quality problems.

[0033] Specifically, refer to Figure 3 As shown, first, the computing device obtains the speech text information of the voice robot, wherein the speech text information may be speech text information related to specific business scenarios such as pre-sales, after-sales, return visits, reminders, collection, surveys, customer care, etc. Then, the computing device uses at least one type of quality detection model to detect the speech text information to determine whether the speech text information has quality problems. The computing device detects the speech text information before the speech text information goes online. The quality detection model can be a quality detection model that detects the pronunciation duration of the speech text, or a quality detection model that detects whether the intention of the speech text information is clear, or a quality detection model that detects whether the speech text information contains some illegal words, or a quality detection model that detects other details in the speech text information.

[0034] Furthermore, in the case where there are quality problems with the speech text information, the computing device outputs warning information corresponding to the quality problem. For example, when the computing device detects that the duration of the speech text information exceeds a predetermined threshold, it outputs a duration warning message; when the computing device detects that the intention of the speech text information is not clear enough, it outputs an unclear intention warning message; when the computing device detects that the speech text information contains illegal words that make the user feel uncomfortable, it outputs illegal word warning information. Thus, by detecting each detail of the speech text information, it is beneficial to improve the speech text information later.

[0035] As described in the background technology, the existing method of detecting the speech of voice robots is to detect them during the robot's voice dialogue. This method cannot prevent speech risks, and this detection method is an overall assessment of the entire conversation. It only scores to mark the good and bad, and there is no way to point out the problems in detail. It has little guiding significance for dialogue improvement work.

[0036] In order to solve this technical problem, this embodiment detects the speech text information of the voice robot before the speech text information goes online, and the risks of the speech text information can be prevented in advance so as to improve the speech text information. In addition, this embodiment detects the speech text information by detecting each detail in the speech text information, and outputs warning information corresponding to the quality problem of the detail when a quality problem occurs in a certain detail. This reminds and assists the robot user to improve or optimize the configuration of the voice robot so that it can better complete the target task.

[0037] This solves the problem that the existing method of detecting the speech techniques of voice robots is to detect them during the robot's voice dialogue. This method cannot prevent speech risks, and this detection method is an overall assessment of the entire dialogue. It cannot point out the problem in detail and has little guiding significance for dialogue technique improvement work.

[0038] Optionally, the speech text information is detected using at least one type of pre-set quality detection model to determine whether the speech text information has quality problems, including: splitting the speech text information into word sequences, and determining the word pronunciation duration of the words in the word sequence; determining the total pronunciation duration of the speech text information and / or the single sentence pronunciation duration of the sentences in the speech text information based on the word pronunciation duration of the words; and determining whether the speech text information has quality problems related to pronunciation duration based on the determined total pronunciation duration and / or single sentence pronunciation duration.

[0039] Specifically, the computing device splits the speech text information into word sequences, determines the word pronunciation durations of the words in the word sequences, and determines the total pronunciation duration of the speech text information and / or the single sentence pronunciation duration of the sentences in the speech text information based on the word pronunciation durations.

[0040] For example, in this embodiment, the computing device divides the speech text information Text into a sentence sequence according to punctuation: Text = (C1, C2, C3, ...), and performs text segmentation on each sentence Cx in the sequence, which is expressed as a word sequence: Cx = (W1, W2, W3, ...). The computing device inputs the word sequence Cx = (W1, W2, W3, ...) of each sentence Cx into a preset text-speech alignment model, determines the estimated pronunciation duration of each word Wx, sums the estimated pronunciation duration of each word Wx, and obtains the single-sentence pronunciation duration of each sentence Cx, and then sums the single-sentence pronunciation duration of each sentence Cx to obtain the total pronunciation duration of the speech text information Text. Among them, the text-speech alignment model can predict the estimated pronunciation duration of each word in the speech text information, thereby estimating the total pronunciation duration of the entire speech text.

[0041] Further, the computing device determines whether the speech text information has quality problems related to the pronunciation duration according to the determined total pronunciation duration and / or single sentence pronunciation duration. For example, when the total pronunciation duration is greater than the total pronunciation threshold duration and / or the single sentence pronunciation duration is greater than the single sentence pronunciation threshold duration, it is determined that the speech text information has quality problems related to the pronunciation duration.

[0042] Therefore, by detecting whether there are quality problems with the length of the conversation text information, the conversation text information is detected from the details of the length of the conversation text information so as to improve the length of the conversation text information, thereby avoiding users becoming impatient and interrupting the call due to the conversation text information being too long.

[0043] Optionally, the operation of determining whether the speech text information has quality problems related to the pronunciation duration based on the determined total pronunciation duration and / or single sentence pronunciation duration includes: comparing the total pronunciation duration of the speech text information with a preset first duration alarm threshold, and / or comparing the single sentence pronunciation duration of the speech text information with a preset second duration alarm threshold; and determining whether the speech text information has quality problems related to the pronunciation duration when the total pronunciation duration of the speech text information is greater than the preset first duration alarm threshold, and / or when the single sentence pronunciation duration of the sentence in the speech text information is greater than the preset second duration alarm threshold.

[0044] Specifically, the computing device compares the total pronunciation duration of the speech text information with a preset first duration alarm threshold, and / or compares the pronunciation duration of a single sentence in the speech text information with a preset second duration alarm threshold; and when the total pronunciation duration of the speech text information is greater than the preset first duration alarm threshold, and / or when the pronunciation duration of a single sentence in the speech text information is greater than the preset second duration alarm threshold, it determines whether the speech text information has quality problems related to the pronunciation duration.

[0045] For example, the total pronunciation duration of the speech text information is 8 minutes, and the preset first duration alarm threshold is 5 minutes. The computing device compares the total pronunciation duration of the speech text information of 8 minutes with the preset first duration alarm threshold of 5 minutes, and determines that the total pronunciation duration of the speech text information of 8 minutes is greater than the preset first duration alarm threshold of 5 minutes, and determines that the speech text information has quality problems related to the pronunciation duration. For example, the pronunciation duration of a single sentence in the speech text information is 50 seconds, and the preset second duration alarm threshold is 30 seconds. The computing device compares the single sentence pronunciation duration of 50 seconds of the sentence in the speech text information with the preset second duration alarm threshold of 30 seconds, and determines that the single sentence pronunciation duration of 50 seconds of the sentence in the speech text information is greater than the preset second duration alarm threshold of 30 seconds, and determines that the speech text information has quality problems related to the pronunciation duration.

[0046] In addition, regardless of whether the total pronunciation duration of the speech text information is greater than the preset first duration alarm threshold, or the pronunciation duration of a single sentence in the speech text information is greater than the preset second duration alarm threshold, or the total pronunciation duration of the speech text information is greater than the preset first duration alarm threshold and the pronunciation duration of a single sentence in the speech text information is greater than the preset second duration alarm threshold, the computing device will determine that the speech text information has quality problems related to the pronunciation duration.

[0047] Thus, the total pronunciation duration of the dialogue text information is compared with the preset first duration alarm threshold, and the single sentence pronunciation duration of the sentences in the dialogue text information is compared with the preset second alarm threshold to determine whether the dialogue text information has quality problems related to the pronunciation duration. The duration of the dialogue text information can be prevented and controlled so that the duration of the dialogue text information can be improved later.

[0048] Optionally, the operation of determining whether the speech text information has quality problems related to the pronunciation duration based on the determined total pronunciation duration and / or single sentence pronunciation duration also includes: determining the position information of sentences in the speech text information in the speech text information; and determining whether the speech text information has quality problems related to the pronunciation duration based on the position information and the single sentence pronunciation duration.

[0049] Specifically, the computing device determines the position information of sentences in the speech text information; and determines whether the speech text information has quality problems related to the pronunciation duration based on the position information and the pronunciation duration of a single sentence.

[0050] For example, the computing device divides the speech text information Text into a sentence sequence according to punctuation: Text = (C1, C2, C3, ...), Cx represents the sentence in the speech text information Text, when x = 2, that is, C2 represents the second sentence in the speech text information Text. The computing device determines the probability of the sentence in the speech text information being interrupted through the probability model based on the fact that C2 represents the second sentence in the speech text information Text and the pronunciation duration of C2 is 1 minute. Therefore, when the probability of the sentence in the speech text information being interrupted is greater than the preset probability threshold, an alarm message of a quality problem related to the voice duration is output.

[0051] Optionally, the operation of determining whether the speech text information has quality problems related to the pronunciation duration based on the position information and the pronunciation duration of a single sentence includes: when the probability of sentences in the speech text information being interrupted is greater than a preset probability threshold, determining whether the speech text information has quality problems related to the pronunciation duration.

[0052] Specifically, when the probability of a sentence in the speech text information being interrupted is greater than a preset probability threshold, the computing device determines whether the speech text information has a quality problem related to the pronunciation duration. For example, the probability of a sentence Cx in the speech text information Text being interrupted is 80%, the preset probability threshold is 50%, and the probability of sentence Cx in the speech text information being interrupted is greater than the preset probability threshold, and the computing device determines that the speech text information Text has a quality problem related to the pronunciation duration.

[0053] Therefore, by detecting the probability of sentences in the dialogue text information being interrupted, the probability of the dialogue text information being interrupted can be controlled within a certain range, so as to provide users with the best calling experience and prevent them from interrupting the call due to irritability, allowing the voice robot to complete its task better.

[0054] Optionally, the operation of detecting the speech text information using at least one type of pre-set quality detection model to determine whether the speech text information has quality problems also includes: determining a speech feature vector corresponding to the speech text information; respectively determining the distance values ​​between the speech feature vector and multiple pre-set intent feature vectors; and determining whether the speech text information has quality problems related to the speech intent based on the distance values ​​between the speech feature vector and the multiple intent feature vectors.

[0055] Specifically, the computing device first determines the speech feature vector corresponding to the speech text information. For example, but not limited to, the speech feature vector of the speech text information Text is (x1, x2, x3, ...). Then, the computing device determines the distance value between the speech feature vector of the speech text information Text (x1, x2, x3, ...) and multiple pre-set intention feature vectors, for example, the distance value Q1 between the speech feature vector of the speech text information Text and the feature vector of the intention feature vector Text1 (for example, (y1, y2, y3, ...)) is: The distance value Q2 from the feature vector of the intended feature vector Text2 (e.g., (z1, z2, z3, ...)) is: Of course, the computing device can also determine the distance value Q3 between the feature vector of the speech text information Text and the intention feature vector Text3, etc., which will not be repeated here.

[0056] Furthermore, the computing device determines whether the speech text information has quality problems related to the speech intent based on the distance values ​​between the speech feature vector and multiple intent feature vectors, such as Q1 and Q2. Thus, it is determined whether the intent of the speech text information is clear and can be understood by the user, so as to better help the robot user achieve business goals.

[0057] Optionally, the operation of determining whether the speech text information has quality problems related to the speech intent based on the distance values ​​between the speech feature vector and multiple intention feature vectors includes: determining the number of intention feature vectors whose distance values ​​from the speech feature vector are less than a first predetermined threshold; and determining that the speech text information has quality problems related to the speech intent when it is determined that the number of intention feature vectors is greater than a second predetermined threshold.

[0058] Specifically, the computing device determines the number (e.g., M) of intention feature vectors whose distance values ​​from the speech feature vector are less than a first predetermined threshold (e.g., Q). When the number of intention feature vectors and the distance value from the speech feature vector are less than the first predetermined threshold (e.g., Q), it indicates that the speech feature vector contains the intention in the intention feature vector. For example, when the distance value Q1 between the speech feature vector of the speech text information Text and the feature vector of the intention feature vector Text1 is less than the first predetermined threshold (e.g., Q), the speech feature vector of the speech text information Text contains the intention of the intention feature vector Text1. For example, if the intention of the intention feature vector Text1 is "Happy Birthday", then the speech feature vector contains the intention of "Happy Birthday". The number (e.g., M) of intention feature vectors indicates that several intentions are contained in the speech text information.

[0059] Furthermore, when the computing device determines that the number of intent feature vectors (e.g., M) is greater than a second predetermined threshold (e.g., N), it determines that the speech text information has quality problems related to the speech intent. Referring to the above, the number of intent feature vectors (e.g., M) indicates that the speech text information contains several intents. When the number of intent feature vectors (e.g., M) is greater than the second predetermined threshold (e.g., N), the speech text information represents multiple intents, which exceeds the second predetermined threshold (e.g., N). This situation indicates that the number of intents contained in the speech text information has exceeded the number that can be understood by the user, and the intent of the speech text information is not clear enough, thereby determining that the speech text information has quality problems related to the speech intent.

[0060] When the number of intentions contained in the speech text information is greater than a preset value, the meaning expressed by the speech text information is not clear, which may easily lead to misunderstanding by the user, or even cause the user to be unable to understand the meaning of the speech text information. Therefore, the intention of the speech text information is detected so as to improve the intention of the speech text information.

[0061] Optionally, the operation of detecting the speech text information using at least one type of pre-set quality detection model to determine whether the speech text information has quality problems also includes: detecting illegal words contained in the speech text information; and determining whether the speech text information has quality problems related to the illegal words based on the detected illegal words.

[0062] Specifically, the computing device uses at least one type of pre-set quality detection model to detect illegal words contained in the speech text information, and determines whether the speech text information has quality problems related to the illegal words based on the detected illegal words. For example, the speech text information contains illegal words such as insult, intimidation, stimulation, anger, illegal words, immediate repayment, and inclusion in credit investigation. The computing device determines whether the speech text information has quality problems related to the illegal words based on these illegal words. Therefore, this embodiment determines whether the speech text information needs to be improved by detecting whether the speech text information contains illegal words, so that the voice robot can better complete the task.

[0063] Optionally, the operation of detecting illegal words contained in the speech text information includes: detecting whether the speech text information contains illegal words in a first illegal word list, and determining that the speech text information has quality problems related to the illegal words when the speech text information contains illegal words in the first illegal word list.

[0064] Specifically, the computing device first detects whether the speech text information contains illegal words in the first illegal word list. For example, the first illegal word list contains insults, threats, stimulation, anger, illegal and illegal words, etc. Then, in the case where the first illegal word is contained in the speech text information, the computing device determines that the speech text information has quality problems related to the illegal words. For example, the speech text information is a collection-type speech, "Pay back the money quickly", and the speech text information contains "Pay back the money quickly", which is a collection-type illegal word. Therefore, the computing device determines that the speech text information of "Pay back the money quickly" has quality problems related to illegal words. Thus, this embodiment detects illegal words in the speech text information so as to process the illegal words, thereby improving the speech text information.

[0065] Optionally, the operation of detecting illegal words contained in the speech text information also includes: detecting whether the speech text information contains illegal words in a second illegal word list; when the weight of the illegal words in the second illegal word list is greater than a predetermined weight threshold, determining that the speech text information has quality problems related to the illegal words.

[0066] Specifically, the computing device can also detect whether the speech text information contains illegal words in the second illegal word list, and when the weight of the illegal words in the second illegal word list is greater than a predetermined weight threshold, it is determined that the speech text information has quality problems related to illegal words. Among them, the second illegal word list varies according to the specific business, and each word has a different weight. For example, in the account collection robot speech, "immediate repayment", "included in the credit report", "pursuant to the liability for breach of contract", "affecting life", etc., these words can be used in normal speech, but because in the account collection scenario, it will cause negative psychological effects on the user who answers the call, so it is not suitable to use them in the same account collection speech text information in a concentrated, large, repeated, and continuous manner.

[0067] For example, taking the collection text information as an example, the second list of illegal words includes the following content: "repay the money" has a weight of 0.2, "hold accountable" has a weight of 0.15, and "record credit" has a weight of 0.1. For example, the predetermined weight threshold in the text information is set to 0.5, and the following two text information are taken as examples:

[0068] The weight of the illegal words in the second illegal word list of "If you don't pay back the money in time, this behavior may be recorded in the credit investigation." is: 0.2+0.1=0.3<0.5, so it is judged that the speech text information of "If you don't pay back the money in time, this behavior may be recorded in the credit investigation." does not have quality problems related to illegal words; the weight of the illegal words in the second illegal word list of "This time I talk to you, I will talk about three things: pay back, pay back, pay back." is: 0.2+0.2+0.2=0.6>0.5, so it is judged that the speech text information of "This time I talk to you, I will talk about three things: pay back, pay back, pay back." has quality problems related to illegal words. Therefore, the words that make users feel stressed in specific scenarios are controlled within a certain weight, so that users have the patience to listen to the speech text information.

[0069] In addition, reference Figure 1 As shown, according to the second aspect of this embodiment, a storage medium 104 is provided. The storage medium 104 includes a stored program, wherein when the program is running, a processor executes any one of the above methods.

[0070] Therefore, in this embodiment, the detection of the speech text information of the voice robot is performed before the speech text information goes online, and the risks existing in the speech text information can be prevented in advance so as to improve the speech text information. In addition, in this embodiment, the detection of the speech text information is to detect each detail in the speech text information, and when a quality problem occurs in a certain detail, an alarm message corresponding to the quality problem of the detail is output. Thus, the robot user is reminded and assisted to improve or optimize the configuration of the voice robot so that it can better complete the target task.

[0071] This solves the problem that the existing method of detecting the speech techniques of voice robots is to detect them during the robot's voice dialogue work. This method cannot prevent speech techniques risks, and this detection method is an overall assessment of the entire dialogue. It cannot point out the problem in detail and has little guiding significance for dialogue technique improvement work.

[0072] In addition, the quality detection models in this embodiment are all for warning and analyzing possible problems in the dialogue text information, so as to improve the rationality and effectiveness of the voice robot's dialogue text information. The quality detection model in this embodiment can arbitrarily adjust the working order or priority of the module according to the application scenario. The quality detection model, threshold training method, and calculation method involved in this embodiment are not limited, and they can all be replaced and optimized. The calculation method of the text feature vector in this embodiment does not necessarily have to be TF-IDF. For example, word embedding calculation, BERT, etc. can also be used.

[0073] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0074] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0075] Example 2

[0076] Figure 4 The device 400 for detecting the dialogue text information according to this embodiment is shown, and the device 400 corresponds to the method according to the first aspect of the embodiment 1. Figure 4 As shown, the device 400 includes: an acquisition module 410, used to acquire speech text information; a detection module 420, used to detect the speech text information using at least one type of pre-set quality detection model to determine whether the speech text information has quality problems; and an output module 430, used to output alarm information corresponding to the quality problem when the speech text information has quality problems.

[0077] Optionally, the detection module 420 includes: a word pronunciation determination submodule, which is used to split the speech text information into word sequences and determine the word pronunciation duration of the words in the word sequences; a pronunciation duration determination submodule, which is used to determine the total pronunciation duration of the speech text information and / or the single sentence pronunciation duration of the sentences in the speech text information according to the word pronunciation duration of the words; and a duration determination submodule, which is used to determine whether the speech text information has quality problems related to pronunciation duration based on the determined total pronunciation time and / or single sentence pronunciation time.

[0078] Optionally, the duration determination submodule includes: a comparison unit, used to compare the total pronunciation duration of the speech text information with a preset first duration alarm threshold, and / or compare the pronunciation duration of a single sentence in the speech text information with a preset second duration alarm threshold; and a first duration determination unit, used to determine whether the speech text information has quality problems related to pronunciation duration when the total pronunciation duration of the speech text information is greater than the preset first duration alarm threshold, and / or when the pronunciation duration of a single sentence in the speech text information is greater than the preset second duration alarm threshold.

[0079] Optionally, the duration determination submodule also includes: a position information determination unit, used to determine the position information of a sentence in the speech text information; and a second duration determination unit, used to determine whether the speech text information has quality problems related to the pronunciation duration based on the position information and the pronunciation duration of a single sentence.

[0080] Optionally, the second duration determination unit includes: a determination subunit, used to determine whether the speech text information has quality problems related to pronunciation duration when the probability of sentences in the speech text information being interrupted is greater than a preset probability threshold.

[0081] Optionally, the detection module 420 also includes: a feature vector determination submodule, which is used to determine the speech feature vector corresponding to the speech text information; a distance value determination submodule, which is used to respectively determine the distance values ​​between the speech feature vector and multiple pre-set intention feature vectors; and an intention determination submodule, which is used to determine whether the speech text information has quality problems related to the speech intention based on the distance values ​​between the speech feature vector and multiple intention feature vectors.

[0082] Optionally, the intention determination submodule includes: a quantity determination unit, used to determine the number of intention feature vectors whose distance values ​​from the speech feature vector are less than a first predetermined threshold; and an intention determination unit, used to determine that the speech text information has quality problems related to the speech intention when the number of intention feature vectors is greater than a second predetermined threshold.

[0083] Optionally, the detection module 420 also includes: a detection submodule for detecting illegal words contained in the speech text information; and a submodule for determining illegal words for determining whether the speech text information has quality problems related to illegal words based on the detected illegal words.

[0084] Optionally, the detection submodule includes: a first detection unit, used to detect whether the speech text information contains illegal words in the first illegal word list, and a first illegal word determination unit, used to determine that the speech text information has quality problems related to illegal words when the speech text information contains illegal words in the first illegal word list.

[0085] Optionally, the detection submodule also includes: a second detection unit, used to detect whether the speech text information contains illegal words in the second illegal word list; a second illegal word determination unit, used to determine that the speech text information has quality problems related to illegal words when the weight of the illegal words in the second illegal word list is greater than a predetermined weight threshold.

[0086] Therefore, according to this embodiment, the detection of the speech text information of the voice robot is performed before the speech text information goes online through the device 400 for detecting the speech text information, and the risks existing in the speech text information can be prevented in advance so as to improve the speech text information. In addition, the detection of the speech text information in this embodiment is to detect each detail in the speech text information, and when a quality problem occurs in a certain detail, an alarm message corresponding to the quality problem of the detail is output. Thus, the robot user is reminded and assisted to improve or optimize the configuration of the voice robot so that it can better complete the target task.

[0087] This solves the problem that the existing method of detecting the speech techniques of voice robots is to detect them during the robot's voice dialogue work. This method cannot prevent speech techniques risks, and this detection method is an overall assessment of the entire dialogue. It cannot point out the problem in detail and has little guiding significance for dialogue technique improvement work.

[0088] Example 3

[0089] Figure 5 The apparatus 500 for detecting the dialogue text information according to this embodiment is shown, and the apparatus 500 corresponds to the method according to the first aspect of the embodiment 1. Figure 5 As shown, the device 500 includes: a processor 510; and a memory 520, which is connected to the processor 510 and is used to provide the processor 510 with instructions for processing the following processing steps: obtaining speech text information; using at least one type of pre-set quality detection model to detect the speech text information to determine whether the speech text information has quality problems; and in the case that the speech text information has quality problems, outputting alarm information corresponding to the quality problems.

[0090] Optionally, the speech text information is detected using at least one type of pre-set quality detection model to determine whether the speech text information has quality problems, including: splitting the speech text information into word sequences, and determining the word pronunciation duration of the words in the word sequence; determining the total pronunciation duration of the speech text information and / or the single sentence pronunciation duration of the sentences in the speech text information based on the word pronunciation duration of the words; and determining whether the speech text information has quality problems related to pronunciation duration based on the determined total pronunciation duration and / or single sentence pronunciation duration.

[0091] Optionally, the operation of determining whether the speech text information has quality problems related to the pronunciation duration based on the determined total pronunciation duration and / or single sentence pronunciation duration includes: comparing the total pronunciation duration of the speech text information with a preset first duration alarm threshold, and / or comparing the single sentence pronunciation duration of the speech text information with a preset second duration alarm threshold; and determining whether the speech text information has quality problems related to the pronunciation duration when the total pronunciation duration of the speech text information is greater than the preset first duration alarm threshold, and / or when the single sentence pronunciation duration of the sentence in the speech text information is greater than the preset second duration alarm threshold.

[0092] Optionally, the operation of determining whether the speech text information has quality problems related to the pronunciation duration based on the determined total pronunciation duration and / or single sentence pronunciation duration also includes: determining the position information of sentences in the speech text information in the speech text information; and determining whether the speech text information has quality problems related to the pronunciation duration based on the position information and the single sentence pronunciation duration.

[0093] Optionally, the operation of determining whether the speech text information has quality problems related to the pronunciation duration based on the position information and the pronunciation duration of a single sentence includes: when the probability of sentences in the speech text information being interrupted is greater than a preset probability threshold, determining whether the speech text information has quality problems related to the pronunciation duration.

[0094] Optionally, the operation of detecting the speech text information using at least one type of pre-set quality detection model to determine whether the speech text information has quality problems also includes: determining a speech feature vector corresponding to the speech text information; respectively determining the distance values ​​between the speech feature vector and multiple pre-set intent feature vectors; and determining whether the speech text information has quality problems related to the speech intent based on the distance values ​​between the speech feature vector and the multiple intent feature vectors.

[0095] Optionally, the operation of determining whether the speech text information has quality problems related to the speech intent based on the distance values ​​between the speech feature vector and multiple intention feature vectors includes: determining the number of intention feature vectors whose distance values ​​from the speech feature vector are less than a first predetermined threshold; and determining that the speech text information has quality problems related to the speech intent when it is determined that the number of intention feature vectors is greater than a second predetermined threshold.

[0096] Optionally, the operation of detecting the speech text information using at least one type of pre-set quality detection model to determine whether the speech text information has quality problems also includes: detecting illegal words contained in the speech text information; and determining whether the speech text information has quality problems related to the illegal words based on the detected illegal words.

[0097] Optionally, the operation of detecting illegal words contained in the speech text information includes: detecting whether the speech text information contains illegal words in a first illegal word list, and determining that the speech text information has quality problems related to the illegal words when the speech text information contains illegal words in the first illegal word list.

[0098] Optionally, the operation of detecting illegal words contained in the speech text information also includes: detecting whether the speech text information contains illegal words in a second illegal word list; and determining that the speech text information has quality problems related to illegal words when the weight of the illegal words in the second illegal word list is greater than a predetermined weight threshold.

[0099] Therefore, according to this embodiment, the detection of the speech text information of the voice robot is performed before the speech text information goes online through the device 500 for detecting the speech text information, and the risks existing in the speech text information can be prevented in advance so as to improve the speech text information. In addition, the detection of the speech text information in this embodiment is to detect each detail in the speech text information, and when a quality problem occurs in a certain detail, an alarm message corresponding to the quality problem of the detail is output. Thus, the robot user is reminded and assisted to improve or optimize the configuration of the voice robot so that it can better complete the target task.

[0100] This solves the problem that the existing method of detecting the speech techniques of voice robots is to detect them during the robot's voice dialogue work. This method cannot prevent speech techniques risks, and this detection method is an overall assessment of the entire dialogue. It cannot point out the problem in detail and has little guiding significance for dialogue technique improvement work.

[0101] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0102] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0103] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0104] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0105] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0106] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0107] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for detecting dialogue text information, characterized in that: include: Get the text information of the speech; Using at least one type of pre-set quality detection model to detect the speech text information, and determine whether the speech text information has quality problems; as well as In the case where there is a quality problem with the speech text information, outputting an alarm message corresponding to the quality problem; The operation of detecting the speech text information by using at least one type of pre-set quality detection model to determine whether the speech text information has quality problems includes: Splitting the speech text information into word sequences, and determining the pronunciation duration of the words in the word sequences; Determine the total pronunciation duration of the speech text information and / or the single sentence pronunciation duration of the sentences in the speech text information according to the word pronunciation duration of the word; and Determine whether the speech text information has quality problems related to the pronunciation duration according to the determined total pronunciation duration and / or single sentence pronunciation duration; The operation of determining whether the speech text information has quality problems related to the pronunciation duration according to the determined total pronunciation duration and / or single sentence pronunciation duration includes: Comparing the total pronunciation duration of the speech text information with a preset first duration alarm threshold, and / or comparing the pronunciation duration of a single sentence in the speech text information with a preset second duration alarm threshold; and In the case where the total pronunciation duration of the speech text information is greater than the preset first duration alarm threshold, and / or in the case where the pronunciation duration of a single sentence in the speech text information is greater than the preset second duration alarm threshold, determine whether the speech text information has quality problems related to pronunciation duration; The operation of determining whether the speech text information has quality problems related to the pronunciation duration according to the determined total pronunciation duration and / or single sentence pronunciation duration also includes: Determining position information of sentences in the speech text information in the speech text information; and Determine whether the speech text information has quality problems related to the pronunciation duration according to the position information and the pronunciation duration of the single sentence; The operation of determining whether the speech text information has a quality problem related to the pronunciation duration according to the position information and the pronunciation duration of the single sentence includes: When the probability that the sentences in the speech text information are interrupted is greater than a preset probability threshold, determining whether the speech text information has a quality problem related to the pronunciation duration; The operation of detecting the speech text information by using at least one type of pre-set quality detection model to determine whether the speech text information has quality problems also includes: Determine a speech feature vector corresponding to the speech text information; respectively determining the distance values ​​between the speech feature vector and a plurality of preset intention feature vectors; and Determine whether the speech text information has quality problems related to speech intent according to the distance value between the speech feature vector and the multiple intent feature vectors; The operation of determining whether the speech text information has a quality problem related to the speech intent according to the distance value between the speech feature vector and the multiple intent feature vectors includes: Determining the number of intention feature vectors whose distance values ​​from the speech feature vector are less than a first predetermined threshold; and When it is determined that the number of intention feature vectors is greater than a second predetermined threshold, it is determined that the speech text information has a quality problem related to the speech intention; The operation of detecting the speech text information by using at least one type of pre-set quality detection model to determine whether the speech text information has quality problems also includes: Detecting illegal words contained in the speech text information; and Determine, based on the detected illegal words, whether the speech text information has quality problems related to the illegal words; The operation of detecting illegal words contained in the speech text information includes: Detecting whether the speech text information contains illegal words in the first illegal word list, and In a case where the speech text information includes illegal words in the first illegal word list, determining that the speech text information has quality problems related to the illegal words; The operation of detecting illegal words contained in the speech text information also includes: Detecting whether the speech text information contains illegal words in the second illegal word list; and When the weight of the illegal words in the second illegal word list is greater than a predetermined weight threshold, it is determined that the speech text information has quality problems related to the illegal words.

Images