A method and system for intelligent device AI voice control

Abstract

The present invention discloses an AI voice control method and system for smart devices, and relates to the field of voice control technology. The method comprises the following steps: collecting user's voice commands through a microphone, extracting features and converting text to the received voice through a voice recognition algorithm, recognizing the user's voice intention based on natural language understanding, and generating control commands; based on the user's voice commands, when unrecognizable voice information appears, triggering the unintelligible command through a speaker, and establishing a feature voice library at the same time, marking the unrecognizable voice information as incomprehensible voice, and recording the incomprehensible voice and the user's subsequent commands. The present invention establishes an incomprehensible voice archive based on the user's voice habits, so as to facilitate the recognition of control commands for incomprehensible voices that have appeared many times in the subsequent use of smart devices, thereby improving the device's response speed and efficiency to user needs and improving the adaptability to different user voice inputs.

Description

Technical Field

[0001] The present invention relates to the field of voice control technology, and in particular to an intelligent device AI voice control method and system. Background Art

[0002] Voice control technology is an artificial intelligence technology that controls devices or systems through voice input, converts voice into text through voice recognition, and processes and understands the text generated by voice recognition through natural language processing technology to analyze the user's intentions and instructions;

[0003] The Chinese patent with the announcement number "CN115631752B" discloses "a method and system for AI voice control of intelligent devices supporting machine learning, comprising the following steps: learning and analyzing the historical voice control information of the user to obtain a number of functional modes, and making each functional mode correspond to a mode name; collecting the voice control instructions issued by the user, and determining whether the voice control instructions exist in the functional mode". Although it can complete the recognition and intelligent control of the user's voice, the dialects in different parts of my country are different, and the speaking habits of people in different regions are also different. It is easy to fail to recognize non-standard Mandarin voices and the dialect cannot be recognized. The existing voice system cannot establish a feature library for the incomprehensible voices in different regions, and cannot associate the incomprehensible voices with control instructions, and has problems of low practicality and functionality;

[0004] In response to the above, this case proposes an AI voice control method and system for smart devices to solve the above technical problems. Summary of the invention

[0005] In view of the problems in the related art, the present invention proposes an AI voice control method and system for smart devices to overcome the above-mentioned technical problems existing in the existing related technology.

[0006] To this end, the specific technical solution adopted by the present invention is as follows:

[0007] A method and system for intelligent device AI voice control, the method comprising the following steps:

[0008] S1. Collect the user's voice commands through the microphone, extract features and convert the received voice into text through the voice recognition algorithm, recognize the user's voice intention based on natural language understanding, and generate control commands;

[0009] S2. Based on the user's voice command, when unrecognizable voice information appears, the incomprehensible command is triggered through the speaker, and a feature voice library is established at the same time, the unrecognizable voice information is marked as incomprehensible voice, the incomprehensible voice and the user's subsequent commands are recorded, and the subsequent commands and the incomprehensible voice are combined for judgment;

[0010] S3. For the data in the feature speech library, the feature group files are grouped according to the normal instruction content. Based on the incomprehensible speech formants in the group, the incomprehensible speech files represented by the same formant are divided into the same representative file. When there is a After the same formant is decoded, the current formant is directly associated with the control instruction to generate a feedback strategy;

[0011] S4. Combined with geographic location information, AI voice control devices in the same geographic area are promoted in an incomprehensible voice-related manner.

[0012] As a preferred embodiment, S2 includes the following sub-steps:

[0013] S21, based on the user voice command collected by the microphone, when the natural language understanding cannot perform semantic recognition, output the unintelligible command, and transmit the unintelligible command to the speaker;

[0014] S22. Record the sound of user instructions that cannot be semantically recognized, establish a feature voice database, store the voice in the feature voice database in wav format, mark the relevant voice as incomprehensible voice, record the user's subsequent normal instructions based on time period t, and make a correlation judgment based on the incomprehensible voice within time period t.

[0015] As a preferred embodiment, the S22 includes the following sub-steps:

[0016] S221, through Establish a characteristic speech database, and establish an incomprehensible speech file in the database, which includes the original file of the incomprehensible speech;

[0017] S222, when semantic recognition fails, record the first speech after the incomprehensible speech within the time period t, and associate the incomprehensible speech based on the semantic recognition of the first speech, the specific steps are:

[0018] When the semantics of the first user speech in the time period t is correctly recognized, the control instruction text content is recorded and entered into the incomprehensible speech file established before the first user speech in the feature speech library;

[0019] When the first user speech within the time period t fails to correctly recognize the semantics, the incomprehensible speech file created before the first user speech is deleted, and the incomprehensible speech file that fails to correctly recognize the semantics this time is created, and the original file is recorded, and the first speech after the new incomprehensible speech within the new time period t is re-determined;

[0020] When no new user voice appears within the time period t, the most difficult-to-understand voice files in the feature voice database before the time period t are deleted. The specific steps are as follows:

[0021] The incomprehensible voice files before time period t are arranged in reverse chronological order based on the creation timestamp, and the incomprehensible voice file ranked first is deleted.

[0022] As a preferred embodiment, S3 includes the following sub-steps:

[0023] S31, based on the incomprehensible speech files in the characteristic speech library, a characteristic group file is established, and based on the control instruction text content inside the incomprehensible speech files, the incomprehensible speech files under the same control instruction text content are divided into characteristic group files;

[0024] S32, performing formant analysis on the original file of the incomprehensible speech file in the feature group file, and performing similarity analysis on the subordinate formants in the same feature group file, and associating control instructions based on the similarity analysis results.

[0025] As a preferred implementation, the S31 includes the following steps:

[0026] S311, based on the incomprehensible speech files in the characteristic speech library, each control instruction text is segmented. Similarity is used to determine the similarity of the control instruction text content. The specific algorithm formula is:

[0027] ; Where A and B represent two different control instruction text word sets, represents the size of the union of sets A and B, Represents the size of the intersection of sets A and B;

[0028] S312, for all control instruction texts, calculate the Similarity, according to the number of control instruction texts n, construct the similarity matrix S, where S is The matrix of Represents the difference between the i-th text and the j-th text Similarity;

[0029] S313, set the similarity threshold T, when the two texts The specific steps are as follows:

[0030] when , it means that the control instruction texts A and B are the same control instruction;

[0031] when , it means that the control instruction texts A and B are not the same control instruction;

[0032] S314, grouping similar control instruction texts, the steps are as follows:

[0033] Traverse each control instruction text in the file, and for each text i, check whether it already exists in a group. The steps are:

[0034] When text i exists in a group, skip the current text i;

[0035] When text i does not exist in any group, create a new group , add the text i middle;

[0036] For each ungrouped text j, calculate its relative Similarity, when , then add text j to the same group as text i .

[0037] As a preferred implementation, the S32 includes the following sub-steps:

[0038] S321, performing formant extraction on the original file of the incomprehensible speech file in the same feature group file, the specific steps are:

[0039] Analyze the speech signal in the original file through LPC and calculate the prediction coefficient, and calculate the formant frequency and bandwidth by solving the root of the LPC coefficient;

[0040] S322, the formant frequency and bandwidth in each voice file are combined into a feature vector, and the similar and identical original voice files in the file are determined by Euclidean distance, and the Euclidean distance threshold O is set:

[0041] when , it means the two original audio files are the same;

[0042] when , it means that the two original audio files are different;

[0043] Where d is the Euclidean distance between the formant frequency and bandwidth of any two original speech files in the feature group file.

[0044] S323, create in the feature group file The puzzling speech files represented by the same original speech file after Euclidean distance determination are divided into the same representative file.

[0045] As a preferred implementation, the S323 includes the following sub-steps:

[0046] S3231, record the number N of incomprehensible speech files in the same representative file, based on the number threshold , judge the representative files:

[0047] when When the current representative file is and bandwidth Control instructions for associating the characteristic voice library where the current file is located;

[0048] S3232. Calculate the formant frequency and bandwidth of the newly received user voice, match the representative file based on the Euclidean distance, and directly execute the control command when the matched representative file is associated with the control command.

[0049] As a preferred embodiment, the S4 includes the following sub-steps:

[0050] S41, collecting the GPS coordinates of the current smart device, and associating the corresponding coordinate county and district;

[0051] S42. For relevant smart devices in the same county or district, representative files associated with control instructions are collected based on period P and uploaded to the cloud-based total representative file via the Internet;

[0052] S43. In the characteristic voice database of each smart device in the current county and district, a cloud representative file is established, and based on the period P, the cloud total representative file content is synchronized into the cloud representative file.

[0053] As a preferred embodiment, the S1 includes the following sub-steps:

[0054] S11, based on the microphone device, capturing the user's voice input, converting the user's voice into an electrical signal, and converting the electrical signal captured by the microphone into a digital signal through an analog-to-digital converter;

[0055] S12, extracting acoustic features from the digital signal, including Mel frequency cepstral coefficients, and converting the user's voice into text based on the extracted acoustic features through ASR automatic speech recognition;

[0056] S13. Perform semantic analysis on the converted text through natural language understanding to extract entity information to generate control instructions, and transmit the generated control instructions to the corresponding smart device through wireless communication methods, including Wi-Fi and Bluetooth.

[0057] As a preferred implementation, an AI voice control system for intelligent devices includes a voice acquisition module, a voice recognition module, a device control module, a feature library building module, and a regional promotion module:

[0058] The voice acquisition module includes a microphone for acquiring the user's voice commands and transmitting them to the voice recognition module;

[0059] The speech recognition module includes a speaker, which performs semantic recognition on the user's speech based on the user's speech command collected by the speech collection module, and transmits the control command to the device control module when the semantic recognition is successful, and transmits the relevant speech data to the feature library establishment module when the semantic recognition fails;

[0060] The device control module transmits control instructions to the corresponding smart device via Wi-Fi and Bluetooth to control the smart device;

[0061] The feature library establishment module establishes a feature voice library based on voice instructions that fail semantic recognition, marks unrecognizable voice information as incomprehensible voice, records the incomprehensible voice and subsequent instructions of the user, makes judgments based on the subsequent instructions and the incomprehensible voice, groups the data in the feature voice library into feature group files according to the normal instruction content, and divides the incomprehensible voice files represented by the same resonance peak into the same representative file based on the resonance peaks of the incomprehensible voice in the group. When there is a After the same formant is decoded, the current formant is directly associated with the control instruction to generate a feedback strategy;

[0062] The regional promotion module combines the geographical location information of the smart device to perform voice-related promotion for AI voice-controlled devices in the same geographical area.

[0063] The beneficial effects of the present invention are:

[0064] The present invention establishes a characteristic voice library for the user's puzzling voice, associates the normal voice instructions after the puzzling voice is generated based on the time period, learns the correlation between the puzzling voice and the normal voice, learns the control instruction response mode according to the user's habits and needs, and adapts to the user's voice habits;

[0065] The present invention groups the control instructions of the puzzling speech files in the characteristic speech library to establish a representative file, associates the formants in the representative file with the control instructions based on the number of times the puzzling speech under the same representative file exists, and after receiving a new speech instruction from the user in the future, matches the representative file based on the formants and directly associates the control instructions, thereby improving the accuracy and functionality of the speech recognition system;

[0066] The present invention establishes a difficult-to-understand voice file based on the user's voice habits, so as to facilitate the recognition of control commands for difficult-to-understand voices that have appeared many times in the subsequent use of the smart device, thereby improving the response speed and efficiency of the device to user needs and improving the adaptability to voice inputs of different users;

[0067] In the present invention, by recording and analyzing the user's incomprehensible speech, the system can provide personalized and customized control instructions according to the user's actual usage habits, thereby improving user satisfaction and adapting to changes in the user's speech habits, and improving the accuracy of speech recognition and user experience;

[0068] In the present invention, the geographical location of the smart device is recorded, the representative archives of the smart devices in the same county or district are integrated, and a cloud-based representative archive is established. The cloud-based representative archives of the smart devices in the same county or district are regularly updated synchronously. This can provide localized services that meet user habits for the dialects and language habits of the same area, thereby improving the adaptability of smart devices to local culture. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0070] Figure 1 is a flow chart of an AI voice control method for an intelligent device according to an embodiment of the present invention;

[0071] Figure 2 It is a block diagram of an AI voice control system for an intelligent device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0072] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in the field should be able to understand other possible implementations and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0073] According to an embodiment of the present invention, a smart device AI voice control method and system are provided.

[0074] The present invention is further described with reference to the accompanying drawings and specific embodiments:

[0075] Example 1: Figure 1 As shown, according to an embodiment of the present invention, a smart device AI voice control method comprises the following steps:

[0076] S1. Collect the user's voice commands through the microphone, extract features and convert the received voice into text through the voice recognition algorithm, recognize the user's voice intention based on natural language understanding, and generate control commands;

[0077] S11, based on the microphone device, capturing the user's voice input, converting the user's voice into an electrical signal, and converting the electrical signal captured by the microphone into a digital signal through an analog-to-digital converter;

[0078] S12, extracting acoustic features from the digital signal, including Mel frequency cepstral coefficients, and converting the user's voice into text based on the extracted acoustic features through ASR automatic speech recognition;

[0079] S13, performing semantic analysis on the converted text through natural language understanding to extract entity information to generate control instructions, and transmitting the generated control instructions to the corresponding smart device through wireless communication methods, including Wi-Fi and Bluetooth;

[0080] S2. Based on the user's voice command, when unrecognizable voice information appears, the incomprehensible command is triggered through the speaker, and a feature voice library is established at the same time, the unrecognizable voice information is marked as incomprehensible voice, the incomprehensible voice and the user's subsequent commands are recorded, and the subsequent commands and the incomprehensible voice are combined for judgment;

[0081] S21, based on the user voice command collected by the microphone, when the natural language understanding cannot perform semantic recognition, output the unintelligible command, and transmit the unintelligible command to the speaker;

[0082] S22. Record the sound of user instructions that cannot be semantically recognized, establish a feature voice database, store the voice in the feature voice database in wav format, mark the relevant voice as incomprehensible voice, record the user's subsequent normal instructions based on time period t, and make a correlation judgment based on the incomprehensible voice within time period t.

[0083] S221, through Establish a characteristic speech database, and establish an incomprehensible speech file in the database, which includes the original file of the incomprehensible speech;

[0084] S222, when semantic recognition fails, record the first speech after the incomprehensible speech within the time period t, and associate the incomprehensible speech based on the semantic recognition of the first speech, the specific steps are:

[0085] When the semantics of the first user speech in the time period t is correctly recognized, the control instruction text content is recorded and entered into the incomprehensible speech file established before the first user speech in the feature speech library;

[0086] When the first user speech within the time period t fails to correctly recognize the semantics, the incomprehensible speech file created before the first user speech is deleted, and the incomprehensible speech file that fails to correctly recognize the semantics this time is created, and the original file is recorded, and the first speech after the new incomprehensible speech within the new time period t is re-determined;

[0087] When no new user voice appears within the time period t, the most difficult-to-understand voice files in the feature voice database before the time period t are deleted. The specific steps are as follows:

[0088] The incomprehensible voice files before time period t are arranged in reverse chronological order based on the creation timestamp, and the incomprehensible voice file ranked first is deleted.

[0089] Embodiment 2: S3, for the data in the feature speech library, the feature group files are grouped according to the normal instruction content, and based on the incomprehensible speech formants in the group, the incomprehensible speech files represented by the same formant are divided into the same representative file. When there is After the same formant is decoded, the current formant is directly associated with the control instruction to generate a feedback strategy;

[0090] S31, based on the incomprehensible speech files in the characteristic speech library, a characteristic group file is established, and based on the control instruction text content inside the incomprehensible speech files, the incomprehensible speech files under the same control instruction text content are divided into characteristic group files;

[0091] S311, based on the incomprehensible speech files in the characteristic speech library, each control instruction text is segmented. Similarity is used to determine the similarity of the control instruction text content. The specific algorithm formula is:

[0092] ; Where A and B represent two different control instruction text word sets, represents the size of the union of sets A and B, Represents the size of the intersection of sets A and B;

[0093] It should be noted that if the two texts If the similarity is close to 1, it means that their vocabulary overlaps a lot and they are likely to be the same control instruction. The similarity is close to 0, which means that the vocabulary of these two control instructions has almost no overlap and they may be different control instructions.

[0094] S312, for all control instruction texts, calculate the Similarity, according to the number of control instruction texts n, construct the similarity matrix S, where S is The matrix of Represents the difference between the i-th text and the j-th text Similarity;

[0095] S313, set the similarity threshold T, when the two texts The specific steps are as follows:

[0096] when , it means that the control instruction texts A and B are the same control instruction;

[0097] when , it means that the control instruction texts A and B are not the same control instruction;

[0098] S314, grouping similar control instruction texts, the steps are as follows:

[0099] Traverse each control instruction text in the file, and for each text i, check whether it already exists in a group. The steps are:

[0100] When text i exists in a group, skip the current text i;

[0101] When text i does not exist in any group, create a new group , add the text i middle;

[0102] For each ungrouped text j, calculate its relative Similarity, when , then add text j to the same group as text i .

[0103] It should be noted that when constructing the similarity matrix, given n control instruction texts, each text corresponds to an index, and i can be used to represent the index of the current text. The elements in the matrix are Represents the difference between text i and text j Similarity, by calculating the similarity of the control instructions of the puzzling voice files in the current feature voice library, the puzzling voice files of the same control instructions can be classified into the same feature group file. At the same time, for the newly generated puzzling voice files, by comparing them with the grouped files Similarity is used to facilitate the creation of new feature group files or the classification of old feature group files for newly generated incomprehensible speech files. The similarity threshold T is usually set to 0.75, and can also be flexibly set according to specific requirements by consulting experts in related fields.

[0104] S32, performing formant analysis on the original file of the incomprehensible speech file in the feature group file, and performing similarity analysis on the subordinate formants in the same feature group file, and associating control instructions based on the similarity analysis results;

[0105] S321, performing formant extraction on the original file of the incomprehensible speech file in the same feature group file, the specific steps are:

[0106] Analyze the speech signal in the original file through LPC and calculate the prediction coefficient, and calculate the formant frequency and bandwidth by solving the root of the LPC coefficient;

[0107] S322, the formant frequency and bandwidth in each voice file are combined into a feature vector, and the similar and identical original voice files in the file are determined by Euclidean distance, and the Euclidean distance threshold O is set:

[0108] when , it means the two original audio files are the same;

[0109] when , it means that the two original voice files are different;

[0110] Where d is the Euclidean distance between the feature vectors composed of the formant frequency and bandwidth of any two original speech files in the feature group file.

[0111] S323, create in the feature group file The puzzling speech files represented by the same original speech file after Euclidean distance determination are divided into the same representative file.

[0112] It should be noted that the representative archives The number of needs to be determined according to the number of the same incomprehensible speech files in the feature group file, the formant frequency and bandwidth The calculation formula is:

[0113] ;in, represents the root calculated by the LPC model, Represents the root The imaginary and real parts of is the sampling frequency of the audio signal, is the constant that converts frequency from radians to Hertz, Representative root The modulus of .

[0114] S3231, record the number N of incomprehensible speech files in the same representative file, based on the number threshold , judge the representative files:

[0115] when When the current representative file is and bandwidth Control instructions for associating the characteristic voice library where the current file is located;

[0116] S3232. Calculate the formant frequency and bandwidth of the newly received user voice, match the representative file based on the Euclidean distance, and directly execute the control command when the matched representative file is associated with the control command.

[0117] S4. Combined with geographic location information, AI voice control devices in the same geographic area are promoted in association with incomprehensible voices;

[0118] S41, collecting the GPS coordinates of the current smart device, and associating the corresponding coordinate county and district;

[0119] S42. For relevant smart devices in the same county or district, representative files associated with control instructions are collected based on period P and uploaded to the cloud-based total representative file via the Internet;

[0120] S43. In the characteristic voice database of each smart device in the current county and district, a cloud representative file is established, and based on the period P, the cloud total representative file content is synchronized into the cloud representative file.

[0121] It should be noted that the period P is usually set to 7 days. By synchronizing the cloud representative files of each smart device, possible predictions can be made in advance for the incomprehensible speech in the same district or county. By calculating the resonance peak frequency and bandwidth of the new user voice received by each smart device, the cloud representative files are matched based on the Euclidean distance. When the matched cloud representative files are associated with control instructions, the control instructions are directly executed.

[0122] Example 3: Figure 2 As shown, an AI voice control system for intelligent devices includes a voice acquisition module, a voice recognition module, a device control module, a feature library building module, and a regional promotion module:

[0123] The voice acquisition module includes a microphone, which is used to collect the user's voice commands and transmit them to the voice recognition module;

[0124] The voice recognition module includes a speaker, which performs semantic recognition on the user's voice based on the user's voice command collected by the voice collection module. When the semantic recognition is successful, the control command is transmitted to the device control module. When the semantic recognition fails, the relevant voice data is transmitted to the feature library establishment module.

[0125] The device control module transmits control instructions to the corresponding smart device via Wi-Fi and Bluetooth to control the smart device;

[0126] The feature library establishment module establishes a feature voice library based on voice commands that fail semantic recognition, marks unrecognizable voice information as incomprehensible voice, records incomprehensible voice and subsequent user commands, and makes judgments based on subsequent commands and incomprehensible voice. The data in the feature voice library is grouped into feature group files according to the normal command content. Based on the incomprehensible voice resonance peaks in the group, the incomprehensible voice files represented by the same resonance peaks are divided into the same representative file. When there is a problem in the representative file, After the same formant is decoded, the current formant is directly associated with the control instruction to generate a feedback strategy;

[0127] The regional promotion module combines the geographical location information of smart devices to carry out voice-related promotion for AI voice-controlled devices in the same geographical area.

[0128] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for AI voice control of intelligent devices, characterized in that: The method comprises the following steps: S1. Collect the user's voice commands through the microphone, extract features and convert the received voice into text through the voice recognition algorithm, recognize the user's voice intention based on natural language understanding, and generate control commands; S2. Based on the user's voice command, when unrecognizable voice information appears, the incomprehensible command is triggered through the speaker, and a feature voice library is established at the same time, the unrecognizable voice information is marked as incomprehensible voice, the incomprehensible voice and the user's subsequent normal commands are recorded, and the correlation is determined by combining the subsequent normal commands with the incomprehensible voice; S3. For the data in the feature speech library, the feature group files are grouped according to the normal command content, and the incomprehensible speech files under the same control command text content are divided into the feature group files. Based on the incomprehensible speech formants in the group, the incomprehensible speech files represented by the same formant are divided into the same representative file. When there is a After finding the same formant in the speech, directly associate the formant in the current representative file with the control instruction, calculate the formant of the newly received user's speech, match the representative file based on the Euclidean distance, and directly execute the control instruction when the matched representative file is associated with the control instruction; S4. Combined with geographic location information, AI voice control devices in the same geographic area are promoted in an incomprehensible voice-related manner.

2. The AI ​​voice control method for smart devices according to claim 1, characterized in that: The S2 comprises the following sub-steps: S21, based on the user voice command collected by the microphone, when the natural language understanding cannot perform semantic recognition, output the unintelligible command, and transmit the unintelligible command to the speaker; S22. Record the sound of user instructions that cannot be semantically recognized, establish a feature voice database, store the voice in the feature voice database in wav format, mark the relevant voice as incomprehensible voice, record the user's subsequent normal instructions based on a time period, and make a correlation judgment based on the incomprehensible voice within the time period.

3. The AI ​​voice control method for smart devices according to claim 2, characterized in that: The S22 comprises the following sub-steps: S221, establishing a characteristic speech database through MySQL, and establishing an incomprehensible speech archive in the database, wherein the archive includes the original file of the incomprehensible speech; S222, when semantic recognition fails, record the first speech after the incomprehensible speech within the time period t, and associate the incomprehensible speech based on the semantic recognition of the first speech, the specific steps are: When the semantics of the first user speech in the time period t is correctly recognized, the control instruction text content is recorded and entered into the incomprehensible speech file established before the first user speech in the feature speech library; When the first user speech within a time period t fails to correctly recognize the semantics, the incomprehensible speech file created before the first user speech is deleted, and the incomprehensible speech file that fails to correctly recognize the semantics this time is created, and the original file is recorded, and the first speech after the new incomprehensible speech within a new time period t is re-determined; When no new user voice appears within the time period t, the most difficult-to-understand voice files in the feature voice database before the time period t are deleted. The specific steps are as follows: The incomprehensible voice files before time period t are arranged in reverse chronological order based on the creation timestamp, and the incomprehensible voice file ranked first is deleted.

4. The AI ​​voice control method for smart devices according to claim 1, characterized in that: The S3 comprises the following sub-steps: S31, based on the incomprehensible speech files in the characteristic speech library, a characteristic group file is established, and based on the control instruction text content inside the incomprehensible speech files, the incomprehensible speech files under the same control instruction text content are divided into characteristic group files; S32, performing formant analysis on the original file of the incomprehensible speech file in the feature group file, and performing similarity analysis on the subordinate formants in the same feature group file, and associating control instructions based on the similarity analysis results.

5. The AI ​​voice control method for smart devices according to claim 4, characterized in that: The S31 comprises the following steps: S311, based on the incomprehensible speech files in the characteristic speech library, each control instruction text is segmented, and the similarity of the control instruction text content is determined by jaccard similarity. The specific algorithm formula is: ; Among them, A and B represent two different control instruction text word sets, represents the size of the union of sets A and B, Represents the size of the intersection of sets A and B; S312, for all control instruction texts, calculate the jaccard similarity between each pair of texts, and construct a similarity matrix S according to the number of control instruction texts n, where S is The matrix of Represents the jaccard similarity between the i-th text and the j-th text; S313, setting a similarity threshold T, when the jaccard similarity of two texts is determined, the specific steps are: when , it means that the control instruction texts A and B are the same control instruction; when , it means that the control instruction texts A and B are not the same control instruction; S314, grouping similar control instruction texts, the steps are as follows: Traverse each control instruction text in the file, and for each text i, check whether it already exists in a group. The steps are: When text i exists in a group, skip the current text i; When text i does not exist in any group, create a new group , add the text i to ; For each ungrouped text j, calculate its jaccard similarity with the grouped text. , then add text j to the same group as text i .

6. The AI ​​voice control method for smart devices according to claim 4, characterized in that: The S32 includes the following sub-steps: S321, performing formant extraction on the original file of the incomprehensible speech file in the same feature group file, the specific steps are: Analyze the speech signal in the original file through LPC and calculate the prediction coefficient, and calculate the formant frequency and bandwidth by solving the root of the LPC coefficient; S322, the formant frequency and bandwidth in each voice file are combined into a feature vector, and the similar and identical original voice files in the file are determined by Euclidean distance, and the Euclidean distance threshold O is set: when , it means the two original audio files are the same; when , it means that the two original audio files are different; Where d is the Euclidean distance between the formant frequency and bandwidth of any two original speech files in the feature group file. S323. Establish I representative files in the feature group file, and divide the incomprehensible speech files represented by the same original speech file after Euclidean distance judgment into the same representative file.

7. The method for AI voice control of intelligent devices according to claim 6, characterized in that: The S323 includes the following sub-steps: S3231, record the number N of incomprehensible speech files in the same representative file, based on the number threshold , judge the representative files: when When the current representative file is and bandwidth Control instructions for associating the characteristic voice library where the current file is located; S3232. Calculate the formant frequency and bandwidth of the newly received user voice, match the representative file based on the Euclidean distance, and directly execute the control command when the matched representative file is associated with the control command.

8. The method for AI voice control of intelligent devices according to claim 1, characterized in that: The S4 comprises the following sub-steps: S41, collecting the GPS coordinates of the current smart device, and associating the corresponding coordinate county and district; S42. For relevant smart devices in the same county or district, representative files associated with control instructions are collected based on period P and uploaded to the cloud-based total representative file via the Internet; S43. In the characteristic voice database of each smart device in the current county and district, a cloud representative file is established, and based on the period P, the cloud total representative file content is synchronized into the cloud representative file.

9. The method for AI voice control of intelligent devices according to claim 1, characterized in that: The S1 comprises the following sub-steps: S11, based on the microphone device, capturing the user's voice input, converting the user's voice into an electrical signal, and converting the electrical signal captured by the microphone into a digital signal through an analog-to-digital converter; S12, extracting acoustic features from the digital signal, including Mel frequency cepstral coefficients, and converting the user's voice into text based on the extracted acoustic features through ASR automatic speech recognition; S13. Perform semantic analysis on the converted text through natural language understanding to extract entity information to generate control instructions, and transmit the generated control instructions to the corresponding smart device through wireless communication methods, including Wi-Fi and Bluetooth.

10. An AI voice control system for smart devices, characterized in that: The system adopts the intelligent device AI voice control method as described in any one of claims 1 to 9, including a voice acquisition module, a voice recognition module, a device control module, a feature library establishment module, and a regional promotion module: The voice acquisition module includes a microphone for acquiring the user's voice commands and transmitting them to the voice recognition module; The speech recognition module includes a speaker, which performs semantic recognition on the user's speech based on the user's speech command collected by the speech collection module, and transmits the control command to the device control module when the semantic recognition is successful, and transmits the relevant speech data to the feature library establishment module when the semantic recognition fails; The device control module transmits control instructions to the corresponding smart device via Wi-Fi and Bluetooth to control the smart device; The feature library establishment module establishes a feature voice library based on voice instructions that fail semantic recognition, marks unrecognizable voice information as incomprehensible voice, records the incomprehensible voice and subsequent normal instructions of the user, makes a correlation determination based on the subsequent normal instructions and the incomprehensible voice, groups the data in the feature voice library into feature group files according to the content of the normal instructions, divides the incomprehensible voice files under the same control instruction text content into feature group files, divides the incomprehensible voice files represented by the same resonance peak into the same representative file based on the resonance peaks of the incomprehensible voice in the group, and divides the incomprehensible voice files represented by the same resonance peak into the same representative file when there is a resonance peak in the representative file. After finding the same formant in the speech, directly associate the current formant with the control instruction, calculate the formant of the newly received user's speech, match the representative file based on the Euclidean distance, and directly execute the control instruction when the matched representative file is associated with the control instruction; The regional promotion module combines the geographical location information of the smart device to perform voice-related promotion for AI voice-controlled devices in the same geographical area.

Images