CN-121983053-A - Whole house intelligent voice control method based on semantic compression and dynamic priority

Abstract

The invention belongs to the technical field of voice recognition and processing, and relates to a whole house intelligent voice control method based on semantic compression and dynamic priority, which is used for analyzing a user voice instruction based on a large language model to obtain control intention, target equipment and adjustment parameters; the method comprises the steps of analyzing control intention and adjustment parameters to determine instruction priority, encoding the control intention and the adjustment parameters to generate compression codes, mapping the instruction priority to channel access parameters of a medium access control layer when the load of a whole house network exceeds a set threshold, dynamically adjusting channel competition priority of compressed data packets, determining transmission time sequence according to the adjusted priority, sending the data packets to target equipment according to a 2.4G wireless protocol according to the time sequence, and if the response indicates that the data packets are lost, improving the instruction priority and remapping the channel access parameters. Through the cooperation of semantic compression and dynamic priority, the transmission light weight and instantaneity are realized while the semantic understanding depth is maintained.

Inventors

• WANG XIANG
• LIU XINYAO

Assignees

• 苏州智为微电子有限公司

Dates

Publication Date

20260505

Application Date

20260402

Claims (10)

1. 1. The whole house intelligent voice control method based on semantic compression and dynamic priority is characterized by comprising the following steps: analyzing a user voice instruction based on the large language model to obtain a control intention, target equipment and adjustment parameters; analyzing the control intention and the adjustment parameters to obtain the emergency degree characteristics of the voice command so as to determine the command priority; When the load of the whole house network exceeds a threshold, mapping the instruction priority to a channel access parameter of a medium access control layer so as to adjust the channel competition priority of the compressed data packet; determining a transmission time sequence according to the adjusted channel competition priority, and transmitting the compressed data packet to target equipment based on a 2.4G wireless protocol according to the transmission time sequence; And receiving an acknowledgement response of the target equipment, if the acknowledgement response indicates that the data packet is lost, increasing the instruction priority, and remapping channel access parameters according to the increased instruction priority.
2. 2. The whole house intelligent voice control method based on semantic compression and dynamic priority according to claim 1, wherein analyzing the control intention and adjustment parameters to obtain the urgency characteristic of the voice command to determine command priority comprises: extracting instruction type characteristics in the control intention and variation amplitude characteristics in the adjustment parameters; Inputting the instruction type features and the variation amplitude features into a decision tree model to obtain corresponding instruction priorities; The decision tree model takes a historical voice instruction and the corresponding actual transmission delay as a training sample, and is constructed through a gradient lifting algorithm, so that the instruction priority can reflect the real-time requirement of the voice instruction in a historical network environment.
3. 3. The full house intelligent voice control method based on semantic compression and dynamic priority according to claim 2, further comprising incremental learning after transmitting the compressed data packet to a target device via a 2.4G wireless protocol: the method comprises the steps of obtaining a sending log recorded in a preset period, wherein the sending log comprises an instruction priority, sending time, retransmission times and transmission delay of each sending; Extracting training samples from the transmission log, wherein each training sample comprises an input characteristic and a label, the input characteristic is an instruction priority and a network load during transmission, and the label is an actually-occurring transmission delay; combining training samples in the current period with historical training samples to form an incremental training set; Performing incremental training on the decision tree model on the incremental training set by adopting a gradient lifting algorithm, and updating parameters of the decision tree model; And deploying the updated decision tree model to a priority evaluation module for determining the instruction priority, so that the decision tree model can adapt to different home network environments and the change of the use habit of a user.
4. 4. The whole house intelligent voice control method based on semantic compression and dynamic priority according to claim 1, wherein the control intention and adjustment parameters are encoded to obtain a compression code, comprising: Inputting the control intention into a semantic embedding model to obtain a semantic feature vector; carrying out hash mapping on the semantic feature vector based on a local sensitive hash algorithm to obtain an intention hash value; hash mapping is carried out on the adjustment parameters based on a cyclic redundancy check algorithm, and parameter hash values are obtained; And splicing the intention hash value and the parameter hash value to obtain the compressed code.
5. 5. The whole house intelligent voice control method based on semantic compression and dynamic priority according to claim 4, wherein the local sensitive hash algorithm enables semantic feature vectors corresponding to semantically similar control intents to fall into the same hash bucket with high probability, so that the same intent hash value is generated.
6. 6. The full house intelligent voice control method based on semantic compression and dynamic priority according to claim 1, wherein mapping the instruction priority to the channel access parameter of the medium access control layer comprises mapping the instruction priority to the arbitration inter-frame space number and the minimum contention window number of the medium access control layer, wherein the higher the instruction priority is, the smaller the mapped arbitration inter-frame space number is, and the smaller the minimum contention window number is.
7. 7. The method for intelligent voice control of whole house based on semantic compression and dynamic priority according to claim 6, wherein determining transmission time sequence according to the adjusted channel contention priority comprises randomly generating a backoff count value within the minimum contention window number range according to the arbitration inter-frame interval waiting channel idle, and determining channel access time of the compressed data packet according to a backoff count mechanism as the transmission time sequence.
8. 8. The whole house intelligent voice control method based on semantic compression and dynamic priority according to claim 1, wherein analyzing user voice instructions based on a large language model to obtain control intents, target devices and adjustment parameters comprises: An encoder based on the large language model converts the user voice instruction into a semantic vector sequence; The semantic vector sequence is input into an intention classification head, and the control intention is calculated and output through a softmax function; The semantic vector sequence is input to an entity identification head, and each position in the semantic vector sequence is subjected to label decoding through a conditional random field layer to obtain the target equipment and the adjustment parameters.
9. 9. The whole house intelligent voice control method based on semantic compression and dynamic priority according to claim 8, wherein after analyzing the user voice command based on a large language model to obtain the control intention, the target device and the adjustment parameters, further comprising: If the verification indicates that the target equipment is not in the whole house equipment list or the adjustment parameter exceeds the equipment support range, generating error prompt information; And determining a feedback mode of the error prompt message according to the instruction priority, adopting voice instant broadcasting feedback for the high-priority instruction, and adopting mobile terminal pushing feedback for the low-priority instruction.
10. 10. The intelligent voice control method for the whole house based on semantic compression and dynamic priority according to claim 1, further comprising monitoring channel occupancy and collision times of the whole house network, and judging that network load exceeds a threshold when the channel occupancy exceeds the threshold or the collision times exceed the threshold.

Description

Whole house intelligent voice control method based on semantic compression and dynamic priority Technical Field The invention relates to the technical field of voice recognition and processing, in particular to a whole house intelligent voice control method based on semantic compression and dynamic priority. Background With the rapid development of intelligent home technology, full-house intelligent voice control has become a core interaction mode for improving living experience. The user expects to have a seamless conversation with the home device through natural language, which requires the system to recognize not only the literal meaning of the instruction, but also the real intention behind the complex sentence. However, the current mainstream voice control scheme depends on preset fixed command words or simple semantic templates, and when faced with the user diversified spoken language expressions and the compound instructions containing implicit conditions, the current mainstream voice control scheme is often stiff and clumsy, and is difficult to accurately capture the core targets and specific parameters of the instructions, so that control deviation is caused or repeated correction is required by the user, and the natural smoothness of interaction is seriously affected. A further reason for this limitation is that there are significant technical faults in adapting the results of deep semantic understanding to the actual network environment of the home. Specifically, although the large language model can finely analyze the user instruction and output the structured content containing rich semantic information, the 2.4G wireless network widely used in the home environment has limited bandwidth and is easily disturbed. If the understanding result with huge information output by the model is directly transmitted, excessive channel resources are occupied, transmission delay and even packet loss are caused, and accurate semantics are lost or deformed before reaching the equipment. Aiming at the problems, an attempt is made to side a lightweight semantic understanding model at a terminal device, instruction analysis is completed locally to reduce data transmission quantity, the method is limited by calculation and storage resources of intelligent home equipment, the understanding precision of the lightweight model is far lower than that of a cloud large language model, and complex instructions are difficult to process. Disclosure of Invention The invention aims to solve the problems that a deep semantic understanding result is difficult to be efficiently transmitted in a 2.4G home network with limited bandwidth due to large data volume and a key instruction is easy to be blocked by non-key data to cause control delay, and provides a full-house intelligent voice control method based on semantic compression and dynamic priority. In order to solve the technical problems, the invention provides a whole house intelligent voice control method based on semantic compression and dynamic priority, which comprises the following steps: analyzing a user voice instruction based on the large language model to obtain a control intention, target equipment and adjustment parameters; analyzing the control intention and the adjustment parameters to obtain the emergency degree characteristics of the voice command so as to determine the command priority; When the load of the whole house network exceeds a threshold, mapping the instruction priority to a channel access parameter of a medium access control layer so as to adjust the channel competition priority of the compressed data packet; determining a transmission time sequence according to the adjusted channel competition priority, and transmitting the compressed data packet to target equipment based on a 2.4G wireless protocol according to the transmission time sequence; And receiving an acknowledgement response of the target equipment, if the acknowledgement response indicates that the data packet is lost, increasing the instruction priority, and remapping channel access parameters according to the increased instruction priority. The method comprises the steps of analyzing control intention and adjusting parameters to obtain emergency degree characteristics of voice instructions to determine instruction priorities, wherein the method comprises the steps of extracting instruction type characteristics in the control intention and change amplitude characteristics in the adjusting parameters, inputting the instruction type characteristics and the change amplitude characteristics into a decision tree model to obtain corresponding instruction priorities, and constructing the decision tree model by taking historical voice instructions and corresponding actual transmission delays thereof as training samples through a gradient lifting algorithm, so that the instruction priorities can reflect real-time requirements of the voice instructions in a historical network environment. Prefera