Search

CN-121983054-A - Voice control method, system, storage medium and chip based on multi-core heterogeneous

CN121983054ACN 121983054 ACN121983054 ACN 121983054ACN-121983054-A

Abstract

The application discloses a voice control method, a voice control system, a voice control storage medium and a voice control chip based on multi-core heterogeneous, and belongs to the technical field of chips. The voice processing unit converts voice signals of a user into text instructions and sends the text instructions to the semantic reasoning unit, the semantic reasoning unit utilizes a semantic processing model to conduct semantic reasoning on the text instructions to generate control instructions, generates first check information according to the control instructions, generates first instructions containing the control instructions and the first check information and sends the first instructions to the safe trusted unit through the first safe channel, the safe trusted unit generates trusted identifications according to safe keys and the control instructions of the safe trusted unit after verifying the first instructions, generates second instructions containing the control instructions and the trusted identifications and sends the second instructions to the execution control unit through the second safe channel, and the execution control unit executes control operation according to the control instructions after verifying the second instructions. The application can improve the safety of voice control and reduce delay.

Inventors

  • WANG SHU

Assignees

  • 北京芯驰半导体科技股份有限公司

Dates

Publication Date
20260505
Application Date
20260407

Claims (10)

  1. 1. The voice control method based on multi-core heterogeneous is characterized by being used in a voice control system, wherein the voice control system comprises an execution control unit, a safe and reliable unit, a voice processing unit and a semantic reasoning unit, wherein the voice processing unit and the semantic reasoning unit operate in a non-safe domain, and the method comprises the following steps: The voice processing unit collects voice signals of a user, converts the voice signals into text instructions and sends the text instructions to the semantic reasoning unit; the semantic reasoning unit performs semantic reasoning on the text instruction by using a semantic processing model to generate a control instruction, performs integrity check processing on the control instruction to generate first check information, generates a first instruction containing the control instruction and the first check information, and sends the first instruction to the safe and trusted unit through a first safe channel; The safe and reliable unit verifies the first instruction, if the verification is passed, a reliable identifier is generated according to a safe key in the safe and reliable unit and the control instruction, a second instruction containing the control instruction and the reliable identifier is generated, and the second instruction is sent to the execution control unit through a second safe channel; And the execution control unit verifies the second instruction, and if the verification is passed, the execution control unit executes corresponding control operation on the control object according to the control instruction.
  2. 2. The voice control method based on multi-core heterogeneous according to claim 1, wherein, The voice processing unit generates voiceprint information according to the voice signal, detects whether the user is an authorized user according to the voiceprint information, converts the voice signal into a text instruction if the user is determined to be the authorized user, generates a third instruction containing the text instruction and the voiceprint information, and sends the third instruction to the semantic reasoning unit; The generation of the first instruction comprising the control instruction and the first verification information comprises the generation of the first instruction comprising the control instruction, the first verification information and the voiceprint information by the semantic reasoning unit; the method comprises the steps that the safe trusted unit generates a trusted identifier according to the safe secret key in the safe trusted unit and the control instruction, and generates a second instruction containing the control instruction and the trusted identifier, wherein the safe trusted unit generates the trusted identifier according to the safe secret key in the safe trusted unit, the control instruction and the voiceprint information, and generates the second instruction containing the control instruction, the voiceprint information and the trusted identifier.
  3. 3. The voice control method based on multi-core heterogeneous according to claim 2, wherein, The voice processing unit obtains the running state of the control object from the execution control unit, detects whether the text instruction is legal according to a preset safety strategy and the running state, and generates a third instruction containing the text instruction, the voiceprint information and the running state if the text instruction is determined to be legal; The generation of the first instruction comprising the control instruction and the first verification information comprises the generation of the first instruction comprising the control instruction, the first verification information, the voiceprint information and the running state by the semantic reasoning unit.
  4. 4. The multi-core heterogeneous based voice control method of claim 3, wherein the secure trusted unit verifies the first instruction comprising: The safe and trusted unit performs integrity check processing on the control instruction to generate second check information, and compares whether the first check information is identical with the second check information; The safe and trusted unit detects whether the user is an authorized user according to the voiceprint information; The safe and trusted unit detects whether the text instruction is legal or not according to a preset safety strategy and the running state; If the first verification information is determined to be the same as the second verification information, the user is an authorized user, and the text instruction is legal, the safe and trusted unit determines that the first instruction passes verification.
  5. 5. The multi-core heterogeneous based voice control method according to claim 2, wherein the execution control unit verifies the second instruction, comprising: the execution control unit performs signature verification on the trusted identification; if the verification sign passes, the execution control unit detects whether the user is an authorized user according to the voiceprint information; If the user is an authorized user, the execution control unit acquires the current running state of the control object, analyzes the control instruction into a machine code instruction, judges whether the machine code instruction is allowed to be executed in the running state based on a preset safety interlocking strategy table, and if the machine code instruction is allowed to be executed, determines that the second instruction passes verification.
  6. 6. The method of any one of claims 1 to 5, wherein the sending the first instruction to the secure trusted unit via a first secure channel comprises: The semantic reasoning unit writes the first instruction into a special receiving buffer area of the safe trusted unit, and sends a hardware interrupt signal to the safe trusted unit through a hardware interrupt line; The secure trusted unit reads the first instruction from the dedicated receive buffer in response to the hardware interrupt signal.
  7. 7. The multi-core heterogeneous based voice control method according to any one of claims 1 to 5, further comprising: after the trusted identifier is generated, the safe trusted unit writes a safe log in a safe storage, wherein the safe log at least comprises the control instruction and a log state, and the log state is configured to be a to-be-processed state; the execution control unit signs the execution result of the control operation to generate an execution receipt, and sends the execution receipt to the safe and trusted unit; and the safe and trusted unit verifies the execution receipt, and if the verification is passed, the log state is modified into the execution result.
  8. 8. The voice control system based on multi-core heterogeneous is characterized by comprising an execution control unit, a safe and reliable unit, a voice processing unit and a semantic reasoning unit, wherein the voice processing unit and the semantic reasoning unit are operated in a non-safe domain; the voice processing unit is used for collecting voice signals of a user, converting the voice signals into text instructions and then sending the text instructions to the semantic reasoning unit; The semantic reasoning unit is used for generating a control instruction after carrying out semantic reasoning on the text instruction by utilizing a semantic processing model, carrying out integrity check processing on the control instruction to generate first check information, generating a first instruction containing the control instruction and the first check information, and sending the first instruction to the safe and trusted unit through a first safe channel; The safe and trusted unit is used for verifying the first instruction, if the verification is passed, a trusted identifier is generated according to a safe key in the safe and trusted unit and the control instruction, a second instruction containing the control instruction and the trusted identifier is generated, and the second instruction is sent to the execution control unit through a second safe channel; and the execution control unit is used for verifying the second instruction, and if the verification is passed, the execution control unit executes corresponding control operation on the control object according to the control instruction.
  9. 9. A computer readable storage medium having stored therein at least one instruction that is loaded and executed by a processor to implement the multi-core heterogeneous based speech control method of any of claims 1 to 7.
  10. 10. A chip comprising the multi-core heterogeneous based voice control system of claim 8.

Description

Voice control method, system, storage medium and chip based on multi-core heterogeneous Technical Field The application relates to the technical field of chips, in particular to a voice control method, a voice control system, a voice control storage medium and a voice control chip based on multi-core heterogeneous. Background With the application of the vehicle-to-machine voice interaction and the large language model (Large Language Model, LLM), a driver can conveniently perform various control operations such as opening and closing of windows, adjustment of air conditioner, switching of driving modes and the like on a vehicle through natural voice interaction. However, while enjoying convenience, how to ensure that control instructions are highly safe and reliable in their full links of generation, transmission and execution has become a key challenge for intelligent automobile development. Currently, two types of technical architectures are mainly adopted in voice control systems. The first type is a voice recognition and application layer control architecture, which directly invokes control functions through an application programming interface of a vehicle machine operating system after voice recognition is completed. The scheme lacks a necessary security authentication and isolation mechanism, and control instructions are easy to intercept, forge or tamper at a software level, so that obvious security risks exist. The second type is a 'speech recognition + cloud model decision' architecture, which uploads speech data to a cloud server for semantic understanding and instruction generation, and then issues instructions to a vehicle end for execution. The scheme utilizes strong calculation power of the cloud, but has strong network dependence, high communication delay, risk of attack and theft or middle man tampering of an instruction transmission link from the cloud to the vehicle end, and can not meet the control requirement of part of the system under the scene of high real-time requirement or poor network condition. Disclosure of Invention The application provides a voice control method, a voice control system, a storage medium and a voice control chip based on multi-core heterogeneous, which are used for solving the problems that safety risks exist during voice control, network dependence is strong, communication delay is high, and real-time requirements cannot be met. The technical scheme is as follows: According to a first aspect of the present application, there is provided a multi-core heterogeneous-based voice control method for use in a voice control system, the voice control system including an execution control unit, a security trust unit, and a voice processing unit and a semantic reasoning unit operating in a non-security domain, the method comprising: The voice processing unit collects voice signals of a user, converts the voice signals into text instructions and sends the text instructions to the semantic reasoning unit; the semantic reasoning unit performs semantic reasoning on the text instruction by using a semantic processing model to generate a control instruction, performs integrity check processing on the control instruction to generate first check information, generates a first instruction containing the control instruction and the first check information, and sends the first instruction to the safe and trusted unit through a first safe channel; The safe and reliable unit verifies the first instruction, if the verification is passed, a reliable identifier is generated according to a safe key in the safe and reliable unit and the control instruction, a second instruction containing the control instruction and the reliable identifier is generated, and the second instruction is sent to the execution control unit through a second safe channel; And the execution control unit verifies the second instruction, and if the verification is passed, the execution control unit executes corresponding control operation on the control object according to the control instruction. In one possible implementation manner, the voice processing unit generates voiceprint information according to the voice signal, detects whether the user is an authorized user according to the voiceprint information, converts the voice signal into a text instruction if the user is determined to be the authorized user, generates a third instruction containing the text instruction and the voiceprint information, and sends the third instruction to the semantic reasoning unit; The generation of the first instruction comprising the control instruction and the first verification information comprises the generation of the first instruction comprising the control instruction, the first verification information and the voiceprint information by the semantic reasoning unit; the method comprises the steps that the safe trusted unit generates a trusted identifier according to the safe secret key in the safe trusted unit and the cont