Estimation method of natural language understanding confidence

Abstract

A natural language understanding confidence estimation method mainly comprises the following steps: acquiring voice information of a user, and performing voice recognition on the voice information to obtain recognition text information; natural language understanding decoding is carried out on the recognition text information, and decoding path confidence coefficients of decoding paths are given for the recognition text information by combining sentence components according to the decoding paths; obtaining a confidence coefficient corresponding to each recognition fragment in the speech recognition process; and fusing the confidence corresponding to each identification fragment with the decoding path confidence on each decoding path understood by the natural language to obtain the final confidence of each decoding path. According to the method, the confidence degree information of each recognition result segment in the speech recognition process is effectively used, meanwhile, it is considered that the influence of key parts in a sentence on the confidence degree is larger, the intention of speech can be grasped more accurately, the semantic accuracy is improved, and then the success rate of human-computer interaction is improved.

Description

Technical Field

[0001] This technology belongs to the field of artificial intelligence, and particularly relates to a method for estimating the confidence of natural language understanding. Background Art

[0002] Natural Language Understanding (NLU) is a general term for all methods, models or tasks that support machines to understand text content. Natural language understanding plays a very important role in text information processing systems and is an essential module for recommendation, question-answering, search and other systems. In the entire human-computer interaction process, natural language understanding is mainly about understanding the content of text data, and natural language generation (NLG) is mainly about machines generating text that humans can understand, and finally displaying it through application layer control (UI display, voice broadcast, etc.).

[0003] The confidence of the understanding result is a very important indicator to measure the accuracy of the natural language understanding result. At present, natural language understanding is mainly divided into engines based on grammar rules and engines based on neural network models. Natural language understanding based on grammar rules is mainly used in embedded devices on the terminal side, because the system memory and computing power on the terminal side are relatively limited; natural language understanding engines based on neural network models are basically deployed on servers. The process of natural language understanding involves word segmentation, part-of-speech tagging, and grammatical analysis. The estimation of confidence in the existing technology is also based on the current text input, and does not consider the confidence of some pronunciation fragments in the recognition process and the tone, volume, emotion and other information of the speech. At the same time, it does not combine the proportion of sentence components (prefix / suffix, modal particles, action intentions, etc.) to uniformly measure confidence. Summary of the invention

[0004] In order to solve the problems existing in the prior art, the present invention proposes a method for estimating the confidence of understanding results by combining the confidence results of recognition fragments and the proportion of sentence components (prefix / suffix, modal particles, action intentions, etc.) in the speech recognition process.

[0005] The present invention provides a method for estimating the confidence of natural language understanding, which mainly includes the following steps:

[0006] Acquire the user's voice information, and perform voice recognition on the voice information to obtain recognized text information; perform natural language understanding decoding on the recognized text information, and for each decoding path, combine sentence components, and provide a decoding path confidence for each decoding path for the recognized text information;

[0007] Obtain the confidence level corresponding to each recognition segment during the speech recognition process;

[0008] The confidence corresponding to each recognition segment is merged with the decoding path confidence on each decoding path of natural language understanding to obtain the final confidence of each decoding path.

[0009] Furthermore, the sentence components include prefixes, suffixes, modal particles, or main actions in a sentence.

[0010] Furthermore, the confidence level corresponding to each recognition segment in the speech recognition process is affected by the clarity, speaking speed, and / or volume of the pronunciation in the user's speech information.

[0011] Furthermore, during the fusion, each recognition segment is first aligned with the recognition text in each decoding path; then, a weighted average is taken of the decoding path confidence on each decoding path for natural language understanding and the confidence corresponding to each recognition segment to obtain the final confidence of each decoding path.

[0012] The present invention provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of implementing the method of the present invention are realized.

[0013] The present invention provides an electronic device, comprising:

[0014] one or more processors; and

[0015] A memory associated with the one or more processors, the memory is used to store program instructions, and when the program instructions are read and executed by the one or more processors, the steps of the method of the present invention are executed.

[0016] The present invention provides a computer program product, comprising computer program instructions, which are used to implement the method of the present invention when the instructions are executed by a processor.

[0017] The present invention effectively uses the confidence information of each recognition result segment in the speech recognition process, fully considering that the understanding confidence results can be differentiated under different speaking tones, volumes, and emotions of the speaker; at the same time, it considers that the key parts of a sentence have a greater impact on the confidence. The natural language understanding confidence estimation method described in the present invention can more accurately grasp the intention of the speech, improve the accuracy of semantics, and thus improve the success rate of human-computer interaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] For a more complete understanding of the present invention, reference will now be made to the following description taken in conjunction with the accompanying drawings, in which:

[0019] Figure 1 A schematic diagram of a flow chart of a method for estimating confidence in natural language understanding provided by the present invention;

[0020] Figure 2 This is a schematic diagram of the decoding path during natural language understanding in a specific implementation manner of the present invention. DETAILED DESCRIPTION

[0021] In order to make the problems, technical solutions and technical effects to be solved by the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments and obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0022] Before describing the exemplary embodiments of the present invention in more detail, it should be noted that although the flowchart of the present invention describes the operations as a sequential processing mode, many of the operation steps can be implemented in parallel or simultaneously, and the order of the operations can also be rearranged as needed. In addition, the processing can be terminated when the operation is completed, but there can also be additional steps not included in the drawings.

[0023] In the following description, the terms "first, second, third", etc. are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It is understandable that "first, second, third", etc. can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0024] The present invention provides a method for estimating the confidence level of natural language understanding, see the attached drawings of the specification Figure 1 , mainly including the following steps.

[0025] Step S01, obtaining the user's voice information, and performing voice recognition on the voice information to obtain recognized text information; performing natural language understanding decoding on the recognized text information, and for each decoding path, combining sentence components, providing a decoding path confidence for each decoding path for the recognized text information.

[0026] For example, the user's voice signal obtained is "Please help me turn on the music." First, the obtained voice signal is voice recognized and the recognized text information is obtained.

[0027] Among them, the speech recognition process may include the following steps: obtaining the user's voice information, such as capturing the user's oral language through a microphone or other recording equipment; performing noise elimination, echo suppression and other processing on the collected sound signal; dividing the pre-processed continuous speech signal into shorter frames or segments; extracting features from each speech frame to characterize the key information of the speech; using an acoustic model to identify the phonemes or letters corresponding to the speech frame; decoding the phoneme sequence in combination with a language model to form a sequence of words or phrases; finding the most likely word sequence among multiple possibilities through a search algorithm; and correcting and optimizing the recognition results to form the final recognized text information.

[0028] After obtaining the recognized text information, it is necessary to perform natural language understanding and decoding on the recognized text information. In this step, it is required that when constructing a natural language understanding and decoding network, the weight of each word in a sentence can be given according to the prefix, suffix, modal particle, main action, etc. in the sentence.

[0029] For example, if the received text is "Please help me turn on the music", the decoding path is as shown in the figure of the manual. Figure 2 shown.

[0030] Among them, "please help me" and "next" are the prefix and suffix of a sentence, which do not play a big role in expressing the intention of the sentence itself, so the confidence score corresponding to each word is set to a smaller 0.05; while "turn on the music" is the main part of the intention, so it plays a bigger role, and the confidence score corresponding to each word is set to a larger 0.2.

[0031] The confidence of the decoding path obtained is: "Please: 0.05; Help: 0.05; Me: 0.05; Fight: 0.2; Open: 0.2; Sound: 0.2; Music: 0.2; Next: 0.05".

[0032] Step S02: Obtain the confidence level corresponding to each recognition segment in the speech recognition process.

[0033] When users express their meanings with voice, the sound signal contains a lot of voice information. For example, for content that users pay more attention to and emphasize more, users may express it with clearer pronunciation, slower speaking speed, and louder volume.

[0034] During speech recognition, an overall confidence level can be output for the recognition result of the entire sentence, or the confidence level corresponding to each recognized segment can be output for each individual segment in the sentence, such as a word, phoneme, or phrase. The confidence level corresponding to each recognized segment reflects the information that the user can contain in the speech but may not be reflected in the text.

[0035] For example, in response to the user's voice signal "Please help me turn on the music", the voice recognition process outputs the confidence level for each recognition segment, score1, score2...: "Please: score1; Help: score2; Me: score3; Play: score4; Turn on: score5; Sound: score6; Music: score7".

[0036] Obtaining the confidence level corresponding to each recognition segment during the speech recognition process can reflect the information of the user when expressing his meaning with voice.

[0037] Step S03: The confidence corresponding to each recognition segment is integrated with the confidence of each decoding path on the natural language understanding decoding path, so that the confidence score of the key part of the sentence has a higher weight, so as to obtain the final confidence of each decoding path.

[0038] In the fusion process, the recognized text in the natural language understanding decoding path in step S01 is first aligned with the recognized segment in the speech recognition process in step S02. The alignment method may adopt a finite state transducer (FST) or the like.

[0039] For example, a decoding path of “Please help me turn on the music” “Please: 0.05; Help: 0.05; Me: 0.05; Play: 0.2; On: 0.2; Sound: 0.2; Music: 0.2; Down: 0.05” is aligned with the recognition segments in the speech recognition process “Please: score1; Help: score2; Me: score3; Play: score4; On: score5; Sound: score6; Music: score7”.

[0040] Secondly, the decoding path confidence on each decoding path of natural language understanding is weighted averaged with the confidence corresponding to each recognition segment, and the confidence score of the key part of the sentence is given a higher weight, so as to finally obtain the final confidence of each decoding path.

[0041] For example, for “Please help me turn on the music”, the 0.05 of “please” and score1 are weighted averaged, the 0.05 of “help” and score2 are weighted averaged, and so on, until the confidence of each recognition segment and the confidence of the decoding path corresponding to the recognition segment on each decoding path are weighted averaged to obtain the final confidence of the recognition segment. In this way, the confidence score of the key part of the sentence will have a higher weight.

[0042] In addition, the present invention also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the method described in the aforementioned method embodiment are implemented.

[0043] The present invention also provides an electronic device, comprising:

[0044] one or more processors; and

[0045] A memory associated with the one or more processors, the memory is used to store program instructions, and when the program instructions are read and executed by the one or more processors, the steps of the method described in the above method embodiment are executed.

[0046] The present invention also provides a computer program product, comprising computer program instructions, which are used to implement the method described in the aforementioned method embodiment when the instructions are executed by a processor.

[0047] It should be noted that a part of the present invention can be applied as a computer program product, such as a computer program instruction, when it is executed by an intelligent electronic device (such as a smart mobile phone or a tablet computer, etc.), the method and / or technical solution according to the present invention can be called or provided through the operation of the intelligent electronic device. The program instruction for calling the method of the present invention may be stored in a fixed or removable recording medium, and / or transmitted through a data stream in a broadcast or other signal-bearing medium, and / or stored in a working memory of an intelligent electronic device that operates according to the program instruction. Here, according to an embodiment of the present invention, a device is included, the device including a memory for storing computer program instructions and a processor for executing program instructions, wherein when the computer program instruction is executed by the processor, the device is triggered to run the method and / or technical solution based on the aforementioned multiple embodiments of the present invention.

[0048] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, no matter from which aspect, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the present invention is defined by the appended claims rather than the description of the above embodiments, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any figure mark in the claims should not be regarded as limiting the claims involved. In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claim can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.

Claims

1. A method for estimating confidence of natural language understanding, mainly comprising the following steps: Acquire the user's voice information, and perform voice recognition on the voice information to obtain recognized text information; perform natural language understanding decoding on the recognized text information, and for each decoding path, combine sentence components, and provide a decoding path confidence for each decoding path for the recognized text information; Obtain the confidence level corresponding to each recognition segment during the speech recognition process; The confidence corresponding to each recognition segment is merged with the decoding path confidence on each decoding path of natural language understanding to obtain the final confidence of each decoding path.

2. The method according to claim 1, characterized in that The sentence components include prefixes, suffixes, modal particles, or main actions in a sentence.

3. The method according to claim 1, characterized in that The confidence level corresponding to each recognition segment in the speech recognition process is affected by the clarity, speaking speed, and / or volume of the pronunciation in the user's speech information.

4. The method according to claim 1, characterized in that: During the fusion, each recognition segment is first aligned with the recognition text in each decoding path; then the decoding path confidence on each decoding path for natural language understanding and the confidence corresponding to each recognition segment are weighted averaged to obtain the final confidence of each decoding path.

5. A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the steps of implementing the method according to any one of claims 1 to 4 are implemented.

6. An electronic device comprising: one or more processors; as well as A memory associated with the one or more processors, the memory being used to store program instructions, wherein the program instructions, when read and executed by the one or more processors, execute the steps of the method according to any one of claims 1 to 4.

7. A computer program product, comprising computer program instructions, which, when executed by a processor, are used to implement a method for estimating natural language understanding confidence as described in any one of claims 1 to 4.

Images