Instruction identification method based on large model

Abstract

The invention relates to the technical field of instruction recognition, and discloses an instruction recognition method based on a large model, and the method comprises the following steps: 1, collecting dialect voice data of different crowds in the same region to form a dialect data set, collecting voice data of a person who sends an instruction at present to form a current voice data set, the method comprises the following steps of: 1, acquiring a data set, 2, denoising the data set, 3, extracting dialect features and voice features, 4, calculating instruction features, 5, splitting the instruction features, 6, performing corresponding instruction recognition on the split instruction features, front-end instruction information, middle-end instruction information and tail-end instruction information, and 7, performing corresponding instruction recognition on the front-end instruction information, the middle-end instruction information and the tail-end instruction information. And step 7, the control operation module carries out retelling on the complete instruction and feeds back the complete instruction to a user for confirmation, so that the identification quality of the system on instruction characteristics is improved, misjudgment and interference caused by noise and dialects to instruction information are reduced, and an instructor is assisted to carry out correct instruction operation.

Description

Technical Field

[0001] The present invention relates to the technical field of instruction recognition, and in particular to an instruction recognition method based on a large model. Background Art

[0003] The large-model-based command recognition for elderly home companionship integrates advanced artificial intelligence technology to realize the understanding and response system for the needs of the elderly. The large-model-based command recognition method in the elderly home companionship system can provide accurate and personalized services through deep learning and natural language processing technology. It not only improves the accuracy of command recognition, but also enhances care and support for the elderly, helping them enjoy life. The core of this system is to use large-model technology to enable home companion robots or intelligent systems to more accurately recognize and execute the elderly's commands, thereby providing more intimate and effective services.

[0004] Large models are prone to challenges in recognition accuracy when processing voice data, especially in the presence of background noise, accents or dialects, which can easily lead to recognition failures or missed recognition. Summary of the invention

[0005] 1. Technical issues to be resolved In view of the shortcomings of the prior art, the present invention provides a command recognition method based on a large model, which enables the system to recognize dialects in speech and the voice of the person currently issuing the command, improves the system's recognition quality of command features, reduces the misjudgment and interference of command information caused by noise and dialects, assists the person issuing the command to perform correct command operations, reduces the problem of command recognition errors, and solves the above-mentioned problems.

[0006] (II) Technical solution To achieve the above object, the present invention provides the following technical solution: a large model-based instruction recognition method, comprising the following steps: Step 1: Use microphones and recording equipment to collect dialect speech data from different groups of people in the same area , and form a dialect data set, using microphones and recording equipment to collect the voice data of the person currently issuing the command , and constitute the current speech data set; Step 2: The data analysis and processing module performs denoising on the dialect data set and the current speech data set respectively to obtain a final dialect data set and a final current speech data set; Step 3: The data analysis and processing module extracts dialect features based on the final dialect dataset and the final current speech dataset. With voice features ; Step 4: Data analysis and processing module based on dialect characteristics With voice features Calculation instruction features ; Step 5: The data analysis and processing module converts the instruction features Split into front-end and issue instructions , the middle end issues instructions With terminal command , the data analysis and processing module has multiple sets of front-end instruction information preset inside , , ,,,Multiple sets of mid-range instruction information , , ,,,Multiple sets of terminal command information , , ,,,, Step 6: The data analysis and processing module identifies the corresponding instructions issued by the front end, the middle end, and the terminal instruction with the front end instruction information, the middle end instruction information, and the terminal instruction information, and forms a complete instruction to send to the control operation module; Step 7: The control operation module repeats the complete instruction and feeds it back to the user for confirmation.

[0007] Preferably, in step 1, the expression of the dialect data set is ,in, Corresponding to the first dialect speech data in the dialect data set, Corresponding to the last dialect speech data in the dialect data set, The representative dialect data set has Dialect voice data.

[0008] Preferably, in step 1, the expression of the current speech data set is ,in Corresponding to the voice data of the first person who currently issues the command in the current voice data set, Corresponding to the voice data of the last person who issued the command in the current voice data set, Indicates that there are The voice data of the person currently issuing the command.

[0009] Preferably, in step 2, the final dialect data set is calculated as follows:

[0010] In the above formula, Represents the final dialect data, Represents the index of the data group, from arrive , represents the number of sample points, Represents the index of the sample point. In each set of data, from arrive , represents each specific sample point in this set of data, Expressing the All in the group data The noise components of each sample point are summed; The final dialect dataset is expressed as follows: ,in, Represents the first dialect data in the dialect data set After denoising, Represents the last dialect data in the dialect data set The denoised data.

[0011] Preferably, in step 2, the final current speech data set is calculated as follows:

[0012] In the above formula, Represents the final current voice data, Represents the index of the data group, from arrive , represents the number of sample points, Represents the index of the sample point. In each set of data, from arrive , represents each specific sample point in this set of data, Expressing the All in the group data The noise components of each sample point are summed; Finally, the current speech data set is expressed as follows: ,in, Represents the first current voice data in the current voice data set After denoising, Represents the last current voice data in the current voice data set The denoised data.

[0013] Preferably, in step 3, the dialect feature The calculation method is:

[0014] In the above formula, Represents the index of the final dialect dataset, from arrive , Representative The final dialect data.

[0015] Preferably, in step 3, the voice feature The calculation method is:

[0016] In the above formula, Indicates the index of the final current speech dataset, from arrive , Representative The final current voice.

[0017] Preferably, in step 4, the instruction feature The calculation method is:

[0018] Preferably, in step 5, the specific corresponding instruction identification method is as follows: for example, the issued instruction Split into , , , respectively With multiple sets of front-end instruction information , , ,,, and compare them, and compare the specific front-end instruction information with higher compliance from the front-end instruction information. With multiple sets of mid-range instruction information , , ,,, and compare them, and compare the specific mid-range instruction information with higher compliance from the mid-range instruction information. With multiple sets of terminal command information , , ,,, for comparison, and specific terminal command information with higher compliance is compared from multiple groups of terminal command information, and the compared higher specific front-end command information, higher specific mid-end command information, and higher specific terminal command information are combined into specific command information and sent to the control operation module.

[0019] Preferably, in step seven, the control operation module issues a voice confirmation to the user, performs a corresponding instruction operation after confirming that the recognition is correct, and returns to step two for re-analysis and processing after confirming that the recognition is wrong.

[0020] Compared with the prior art, the present invention provides an instruction recognition method based on a large model, which has the following beneficial effects: 1. The present invention forms a dialect data set by combining the dialect speech data of different groups of people in the same area, and then forms a current speech data set by combining the speech data of the person who currently issues the command, and then denoises the dialect data set and the current speech data set to obtain a final dialect data set and a final current speech data set, thereby improving the quality and accuracy of the dialect speech data and the speech data of the person who currently issues the command, and then extracts dialect features according to the final dialect data set and the final current speech data set. With voice features Computation instruction characteristics , enabling the system to recognize dialects in speech and the voice of the person currently issuing the command, improving the system's recognition quality of command features and reducing misjudgment and interference in command information caused by noise and dialects.

[0021] 2. The present invention can better identify and analyze the characteristics of the issued instructions by splitting the issued instructions into front-end, middle-end and end-end, and then respectively performing corresponding instruction identification with the front-end instruction information, middle-end instruction information and end-end instruction information, and analyze the true intention of the instructions from the front-end, middle-end and end respectively, to assist the person giving the instructions to perform correct instruction operations and reduce the problem of instruction identification errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram of the steps of the method of the present invention; DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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 obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] See also Figure 1 , an instruction recognition method based on a large model, comprising the following steps: Step 1: Use microphones and recording equipment to collect dialect speech data from different groups of people in the same area , and form a dialect data set, using microphones and recording equipment to collect the voice data of the person currently issuing the command , and constitute the current speech data set; The expression of the dialect dataset is ,in, Corresponding to the first dialect speech data in the dialect data set, Corresponding to the last dialect speech data in the dialect data set, The representative dialect data set has Dialect speech data; The expression of the current speech dataset is ,in Corresponding to the voice data of the first person who currently issues the command in the current voice data set, Corresponding to the voice data of the last person who issued the command in the current voice data set, Indicates that there are Voice data of the person currently issuing the command; Step 2: The data analysis and processing module performs denoising on the dialect data set and the current speech data set respectively to obtain a final dialect data set and a final current speech data set; The final dialect dataset is calculated as follows:

[0025] In the above formula, Represents the final dialect data, Represents the index of the data group, from arrive , represents the number of sample points, Represents the index of the sample point. In each set of data, from arrive , represents each specific sample point in this set of data, Expressing the All in the group data The noise components of each sample point are summed; The final dialect dataset is expressed as follows: ,in, Represents the first dialect data in the dialect data set After denoising, Represents the last dialect data in the dialect data set Denoised data; Finally, the current voice data set is calculated as follows:

[0026] In the above formula, Represents the final current voice data, Represents the index of the data group, from arrive , represents the number of sample points, Represents the index of the sample point. In each set of data, from arrive , represents each specific sample point in this set of data, Indicates All in the group data The noise components of each sample point are summed; Finally, the current speech data set is expressed as follows: ,in, Represents the first current voice data in the current voice data set After denoising, Represents the last current voice data in the current voice data set Denoised data Step 3: The data analysis and processing module extracts dialect features based on the final dialect dataset and the final current speech dataset. With voice features ; Dialect features The calculation method is:

[0027] In the above formula, Represents the index of the final dialect dataset, from arrive , Representative Final dialect data; Voice Features The calculation method is:

[0028] In the above formula, Indicates the index of the final current speech dataset, from arrive , Representative The final current voice; The dialect speech data of different groups of people in the same area are formed into a dialect data set, and the speech data of the current person issuing the command is formed into a current speech data set. Then, the dialect data set and the current speech data set are denoised to obtain the final dialect data set and the final current speech data set, thereby improving the quality and accuracy of the dialect speech data and the speech data of the current person issuing the command. Then, dialect features are extracted according to the final dialect data set and the final current speech data set. With voice features Computation instruction characteristics , enabling the system to recognize dialects in speech and the voice of the person currently issuing the command, improving the system's recognition quality of command features and reducing misjudgment and interference in command information caused by noise and dialects.

[0029] Step 4: Data analysis and processing module based on dialect characteristics With voice features Calculation instruction features ; Instruction characteristics The calculation method is:

[0030] Step 5: The data analysis and processing module converts the instruction features Split into front-end and issue instructions , the middle end issues instructions With terminal command , the data analysis and processing module has multiple sets of front-end instruction information preset inside , , ,,,Multiple sets of mid-range instruction information , , ,,,Multiple sets of terminal command information , , ,,,, Step 6: The data analysis and processing module identifies the corresponding instructions issued by the front end, the middle end, and the terminal instruction with the front end instruction information, the middle end instruction information, and the terminal instruction information, and forms a complete instruction to send to the control operation module; For example, the command to issue Split into , , , respectively With multiple sets of front-end instruction information , , ,,, and compare them, and compare the specific front-end instruction information with higher compliance from the front-end instruction information. With multiple sets of mid-range instruction information , , ,,, and compare them, and compare the specific mid-range instruction information with higher compliance from the mid-range instruction information. With multiple sets of terminal command information , , ,,, compare, compare the specific terminal instruction information with higher conformity from the multiple groups of terminal instruction information, combine the compared higher specific front-end instruction information, higher specific mid-end instruction information, and higher specific terminal instruction information into specific instruction information and send it to the control operation module; Step 7: The control operation module repeats the complete instruction and feeds it back to the user for confirmation. After confirming that the recognition is correct, the corresponding instruction operation is performed. After confirming that the recognition is wrong, it returns to step 2 for re-analysis and processing; The issued instructions are split into front-end, middle-end and end-end, and then the corresponding instruction recognition is performed with the front-end instruction information, middle-end instruction information and end-end instruction information respectively. This can better identify and analyze the characteristics of the issued instructions, analyze the true intention of the instructions from the front-end, middle-end and end respectively, assist the person giving the instructions to perform the correct instruction operations, and reduce the problem of instruction recognition errors.

[0031] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A large model-based instruction recognition method, characterized in that: The following steps are involved: Step 1: Use microphones and recording equipment to collect dialect speech data from different groups of people in the same area , and form a dialect data set, using microphones and recording equipment to collect the voice data of the person currently issuing the command , and constitute the current speech data set; Step 2: The data analysis and processing module performs denoising on the dialect data set and the current speech data set respectively to obtain a final dialect data set and a final current speech data set; Step 3: The data analysis and processing module extracts dialect features based on the final dialect dataset and the final current speech dataset. With voice features ; Step 4: Data analysis and processing module based on dialect characteristics With voice features Calculation instruction features ; Step 5: The data analysis and processing module converts the instruction features Split into front-end and issue instructions , the middle end issues instructions With terminal command , the data analysis and processing module has multiple sets of front-end instruction information preset inside , , ,,,Multiple sets of mid-range instruction information , , ,,,Multiple sets of terminal command information , , ,,,, Step 6: The data analysis and processing module identifies the corresponding instructions issued by the front end, the middle end, and the terminal instruction with the front end instruction information, the middle end instruction information, and the terminal instruction information, and forms a complete instruction to send to the control operation module; Step 7: The control operation module repeats the complete instruction and feeds it back to the user for confirmation.

2. The instruction recognition method based on a large model according to claim 1, characterized in that: In step 1, the expression of the dialect data set is: ,in, Corresponding to the first dialect speech data in the dialect data set, Corresponding to the last dialect speech data in the dialect data set, The representative dialect data set has Dialect voice data.

3. The instruction recognition method based on a large model according to claim 1, characterized in that: In step 1, the expression of the current speech data set is: ,in Corresponding to the voice data of the first person who currently issues the command in the current voice data set, Corresponding to the voice data of the last person who issued the command in the current voice data set, Indicates that there are The voice data of the person currently issuing the command.

4. The instruction recognition method based on a large model according to claim 2, characterized in that: In step 2, the final dialect data set is calculated as follows: In the above formula, Represents the final dialect data, Represents the index of the data group, from arrive , represents the number of sample points, Represents the index of the sample point. In each set of data, from arrive , represents each specific sample point in this set of data, Indicates All in the group data The noise components of each sample point are summed; The final dialect dataset is expressed as follows: ,in, Represents the first dialect data in the dialect data set After denoising, Represents the last dialect data in the dialect data set The denoised data.

5. The instruction recognition method based on a large model according to claim 3, characterized in that: In step 2, the final calculation method of the current speech data set is: In the above formula, Represents the final current voice data, Represents the index of the data group, from arrive , represents the number of sample points, Represents the index of the sample point. In each set of data, from arrive , represents each specific sample point in this set of data, Indicates All in the group data The noise components of each sample point are summed; Finally, the current speech data set is expressed as follows: ,in, Represents the first current voice data in the current voice data set After denoising, Represents the last current voice data in the current voice data set The denoised data.

6. The instruction recognition method based on a large model according to claim 4, characterized in that: In step 3, the dialect features The calculation method is: In the above formula, Represents the index of the final dialect dataset, from arrive , Representative The final dialect data.

7. The instruction recognition method based on a large model according to claim 6, characterized in that: In step 3, the voice feature The calculation method is: In the above formula, Indicates the index of the final current speech dataset, from arrive , Representative The final current voice.

8. A large model-based instruction recognition method according to claims 6 and 7, characterized in that: In step 4, the instruction feature The calculation method is: 。 9. The instruction recognition method based on a large model according to claim 8, characterized in that: In step 5, the specific method of corresponding instruction recognition is as follows: for example, the instruction issued Split into , , , respectively With multiple sets of front-end instruction information , , ,,, and compare them, and compare the specific front-end instruction information with higher compliance from the front-end instruction information. With multiple sets of mid-range instruction information , , ,,, and compare them, and compare the specific mid-range instruction information with higher compliance from the mid-range instruction information. With multiple sets of terminal command information , , ,,, for comparison, and specific terminal command information with higher compliance is compared from multiple groups of terminal command information, and the compared higher specific front-end command information, higher specific mid-end command information, and higher specific terminal command information are combined into specific command information and sent to the control operation module.

10. The instruction recognition method based on a large model according to claim 9, characterized in that: In the step seven, the control operation module sends a voice confirmation to the user, performs a corresponding instruction operation after confirming that the recognition is correct, and returns to step two for re-analysis and processing after confirming that the recognition is wrong.

Images