Vehicle-machine voice interaction method, vehicle-machine

Abstract

The present invention discloses a vehicle-computer voice interaction method and a vehicle-computer. The vehicle-computer voice interaction method includes the following steps: obtaining a voice signal to be recognized; when the vehicle-computer is in a wake-up-free scenario, decoding the voice signal using a wake-up-free finite state receiver (FSA) decoding method in a voice model; and when the vehicle-computer is in a non-wake-up-free scenario, decoding the voice signal using a weighted finite state transducer (WFST) decoding method and a non-wake-up-free finite state receiver (FSA) decoding method in the voice model. Thus, the vehicle-computer voice interaction method reduces crosstalk between the wake-up-free scenario and the non-wake-up-free scenario by enabling the vehicle-computer to independently decode the voice signal in the wake-up-free scenario and the non-wake-up-free scenario.

Description

Technical Field

[0001] The present invention relates to the field of voice technology, and in particular to a vehicle-computer voice interaction method and a vehicle-computer. Background Art

[0002] With the rapid development of AI (Artificial Intelligence) technology, cars are becoming increasingly intelligent. Since voice technology is a gateway to intelligence, creating a more convenient and intelligent voice recognition system is essential. Considering factors such as low memory usage, high recognition rate, high security, customization, and a good user experience, a voice recognition system using wake-up-free technology has been proposed in related technologies.

[0003] However, the speech recognition system in the above technology has the following problems:

[0004] 1) Most of them run on a full CPU (Central Processing Unit) platform environment, which has limited computing power, high power consumption, high cost, and cannot use parallel threads to complete specific tasks;

[0005] 2) Most implementations use a hybrid of WFST (Weighted Finite-State Transducer) and FSA (Finite State Acceptor). This leads to crosstalk in actual use, meaning it's impossible to tell whether the scenario is wake-up-free or not, impacting the user experience.

[0006] 3) Most of the speech recognition resources are loaded by directly reading them into the memory, which may result in low memory utilization. Summary of the Invention

[0007] The present invention aims to solve at least one of the technical problems in the related art to a certain extent. To this end, the first object of the present invention is to provide a vehicle-machine voice interaction method that reduces the crosstalk problem between wake-up-free scenarios and non-wake-up-free scenarios.

[0008] A second objective of the present invention is to provide a vehicle computer.

[0009] In order to achieve the above-mentioned objectives, an embodiment of the first aspect of the present invention proposes a vehicle-computer voice interaction method, which includes: obtaining a voice signal to be recognized; when the vehicle-computer is in a wake-up-free scenario, using the wake-up-free finite state receiver FSA decoding method in the voice model to decode the voice signal; when the vehicle-computer is in a non-wake-up-free scenario, using the weighted finite state transformer WFST decoding method and the non-wake-up-free finite state receiver FSA decoding method in the voice model to decode the voice signal.

[0010] According to an embodiment of the present invention, the vehicle-mounted computer voice interaction method first acquires a voice signal to be recognized. When the vehicle-mounted computer is in a wake-up-free scenario, the voice signal is decoded using the wake-up-free finite state receiver (FSA) decoding method in the voice model. When the vehicle-mounted computer is in a non-wake-up-free scenario, the voice signal is decoded using the weighted finite state transducer (WFST) decoding method and the non-wake-up-free finite state receiver (FSA) decoding method in the voice model. Thus, with this vehicle-mounted computer voice interaction method, the vehicle-mounted computer achieves independent decoding in both wake-up-free and non-wake-up-free scenarios, reducing crosstalk between the two scenarios.

[0011] In addition, the vehicle-machine voice interaction method proposed in the above embodiment of the present invention may also have the following additional technical features:

[0012] In one embodiment of the present invention, after decoding the voice signal using the wake-up-free FSA decoding method, the method further includes: performing semantic recognition on the wake-up-free FSA decoding result to obtain a first semantic recognition result; determining whether the first semantic recognition result is within a preset wake-up-free statement set; if so, generating a control instruction corresponding to the first semantic recognition result to control the vehicle accordingly and control the vehicle computer to switch to a non-wake-up-free scenario.

[0013] In one embodiment of the present invention, the preset wake-up-free statement set includes personalized sentence patterns and non-personalized sentence patterns, and the method further includes: obtaining a wake-up-free hot word list, wherein the wake-up-free hot word list includes at least one wake-up-free hot word; based on the wake-up-free hot word list, constructing a wake-up-free hot word network; before using the wake-up-free FSA decoding method to decode the voice signal, loading the wake-up-free hot word network to the placeholder slot corresponding to the personalized sentence pattern.

[0014] In one embodiment of the present invention, when controlling the vehicle computer to switch to a non-wake-up-free scenario, the method further includes: uninstalling the wake-up-free hotword network.

[0015] In one embodiment of the present invention, after decoding the voice signal using the WFST decoding method and the non-wake-up-free FSA decoding method, the method further includes: performing semantic recognition on the WFST decoding result and the non-wake-up-free FSA decoding result to obtain a second semantic recognition result; and generating a control instruction corresponding to the second semantic recognition result to control the vehicle accordingly.

[0016] In one embodiment of the present invention, when the vehicle computer is in a non-wake-up-free scenario and enters the wake-up state, the method further includes: if the voice signal is not obtained after waiting for a preset time, controlling the vehicle computer to exit the wake-up state.

[0017] In one embodiment of the present invention, the method is executed based on a neural network processor NPU environment.

[0018] In one embodiment of the present invention, the method further includes: loading the speech model into the memory of the NPU, and establishing a mapping relationship between the virtual memory area and the file object in the memory; when it is necessary to load and read resources, data interaction is performed through the mapping relationship.

[0019] In one embodiment of the present invention, the memory includes a magnetic disk.

[0020] In order to achieve the above-mentioned purpose, the second embodiment of the present invention proposes a vehicle computer, including a neural network processor NPU and a storage medium, on which a computer program is stored. When the computer program is executed by the NPU, the above-mentioned vehicle computer voice interaction method is implemented.

[0021] According to an embodiment of the present invention, a vehicle computer includes a neural network processor (NPU) and a storage medium, the storage medium storing a computer program. When the computer program is executed by the NPU, it first obtains a speech signal to be recognized. When the vehicle computer is in a wake-up-free scenario, the speech signal is decoded using the wake-up-free finite state receiver (FSA) decoding method in the speech model. When the vehicle computer is in a non-wake-up-free scenario, the speech signal is decoded using the weighted finite state transformer (WFST) decoding method and the non-wake-up-free finite state receiver (FSA) decoding method in the speech model. Therefore, by implementing speech recognition in an NPU environment, CPU usage can be reduced and inference efficiency can be improved. Furthermore, the vehicle computer performs independent decoding in both wake-up-free and non-wake-up-free scenarios, reducing crosstalk between the wake-up-free and non-wake-up-free scenarios.

[0022] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flow chart of a vehicle-machine voice interaction method according to an embodiment of the present invention;

[0024] Figure 2 This is a schematic diagram of an FSA sentence structure for a wake-up-free scenario according to an example of the present invention;

[0025] Figure 3 This is a schematic diagram of a personalized sentence pattern in a preset wake-up-free statement set according to an example of the present invention;

[0026] Figure 4 This is a flow chart of a vehicle-machine voice interaction method according to an example of the present invention;

[0027] Figure 5 is a flow chart of another example of a vehicle-machine voice interaction method of the present invention;

[0028] Figure 6 1 is a schematic diagram of the structure of a neural network processor NPU according to an example of the present invention;

[0029] Figure 7 It is a flowchart of a vehicle-machine voice interaction method according to a specific embodiment of the present invention. DETAILED DESCRIPTION

[0030] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0031] The following describes a vehicle-machine voice interaction method and a vehicle-machine machine according to an embodiment of the present invention with reference to the accompanying drawings.

[0032] Figure 1 This is a flow chart of a vehicle-machine voice interaction method according to an embodiment of the present invention.

[0033] S101: Acquire a speech signal to be recognized.

[0034] Specifically, the vehicle computer starts up, begins resource initialization, and acquires a voice signal to be recognized. The voice signal to be recognized can be a voice signal in any scenario and can be acquired through a voice acquisition device, such as a microphone. Furthermore, since noise may be present during the entire voice acquisition process by the vehicle computer, the acquired raw voice signal can also be subjected to noise filtering to obtain the voice signal to be recognized. This is not limited in the present invention.

[0035] It should be noted that when the car computer is started, the resources contained in the car computer will be mobilized. Since the resources are large, they need to be initialized after the car computer is started. For example, the resources may include VAD (Voice Activity Detection), acoustic CLDNN (Convolutional, Long Short-Term Memory, Fully Connected Deep Neural Networks, multi-model neural network) model, WFST (Weighted Finite State Transducer) network, etc. In particular, if the wake-up scenario needs to be enabled, the resources also need to include FSA (Finite State Acceptor) network.

[0036] S102 , when the vehicle computer is in a wake-up-free scenario, a wake-up-free finite state receiver (FSA) decoding method in the voice model is used to decode the voice signal.

[0037] The speech model includes a finite state receiver (FSA) decoding method, a non-FSA decoding method, and a weighted finite state transition (WFST) decoding method. Specifically, when the vehicle computer is in a wake-up-free scenario, only the FSA decoding method in the speech model is used for independent decoding. This means that the FSA decoding method in the acoustic CLDNN model and language model is used to obtain the FSA decoding result.

[0038] In some embodiments of the present invention, after decoding the voice signal using the wake-up-free FSA decoding method, the following steps may be further included:

[0039] S201 , performing semantic recognition on a wake-up-free FSA decoding result to obtain a first semantic recognition result.

[0040] Specifically, the voice signal is decoded by adopting the wake-up-free FSA decoding method to obtain the wake-up-free FSA decoding result. Further, it is necessary to perform semantic recognition on the wake-up-free FSA decoding result to obtain the first semantic recognition result. Figure 2 For example, each arc between nodes corresponds to a word, and the order in which the directed graph points is the sentence to be expressed. For the preset wake-up-free statement set in the wake-up-free scenario, similar expressions can be used. Figure 2 Therefore, you can set a special FSA sentence for the wake-up-free scenario to form a preset wake-up-free sentence set.

[0041] S202: Determine whether the first semantic recognition result is in a preset wake-up-free statement set.

[0042] Specifically, based on Figure 2 FSA sentences are generally divided into two types: personalized and non-personalized, depending on the scenario. Therefore, the preset wake-up-free sentence set includes both personalized and non-personalized sentences. For example, personalized sentences might be "Call XX," "Play XX song," or "Navigate to XX." Non-personalized sentences might be vehicle control commands, such as "Raise the temperature by one degree" or "Roll down the window halfway."

[0043] In some embodiments of the present invention, the personalized sentence patterns in the preset wake-up-free statement set may specifically include:

[0044] A1. Obtain a wake-up-free hotword list, where the wake-up-free hotword list includes at least one wake-up-free hotword.

[0045] The above-mentioned wake-up-free hotword list may include one or more wake-up-free hotwords, which can be imported according to the user's usage habits. For example, the wake-up-free hotwords can be songs that the user often listens to, places that the user often visits, and frequently used contacts. The present invention is not limited to this.

[0046] A2. Build a wake-up-free hotword network based on the wake-up-free hotword list.

[0047] It should be noted that the wake-up-free hotword network can be loaded or unloaded according to the switching of scenes.

[0048] A3. Before decoding the voice signal using the wake-up-free FSA decoding method, load the wake-up-free hotword network to the placeholder corresponding to the personalized sentence.

[0049] As an example, see Figure 3 The personalized sentence is "Call XX," which includes the main sentence "Call" and the placeholder "XX." "XX" is the locally imported wake-up-free hotword, i.e., the placeholder content in the personalized FSA sentence. Users can update the placeholders in the personalized FSA sentence by locally importing the hotword network. The wake-up-free hotword network is built based on the acquired wake-up-free hotword list and pre-loaded into the placeholder corresponding to the personalized sentence to facilitate voice signal recognition.

[0050] It should be noted that before loading the wake-up-free FSA hotword network, if a non-wake-up-free FSA hotword network exists, it is also necessary to uninstall the non-wake-up-free FSA hotword network. This non-wake-up-free FSA hotword network exists in the non-wake-up-free FSA decoding method. The difference between the non-wake-up-free FSA decoding method and the wake-up-free FSA decoding method is that the sentence structure and hotword placeholder slots of the two decoding methods are different.

[0051] S203: If yes, generate a control instruction corresponding to the first semantic recognition result to control the vehicle accordingly and control the vehicle computer to switch to a non-wake-up-free scenario.

[0052] Specifically, if the first semantic recognition result is within the preset wake-up-free statement set, a corresponding control instruction is generated. For example, if the first semantic recognition result is "Call Xiao Zhang", the vehicle is controlled to complete the instruction "Call Xiao Zhang".

[0053] Furthermore, since the wake-up-free FSA sentence structure and other voice model resources have been loaded into the car computer, when controlling the car computer to switch to a non-wake-up-free scenario, it is only necessary to unload the hot word network in the personalized sentence structure, and the remaining main sentence structure and original placeholder content are sufficient.

[0054] In addition, if the first semantic recognition result is not in the above-mentioned preset wake-up-free statement set, the vehicle computer continues to be in the wake-up-free scenario and remains in the listening state until it is manually turned off or the first semantic recognition result is recognized to be in the above-mentioned preset wake-up-free statement set.

[0055] S103 , when the vehicle computer is in a non-wake-up-free scenario, a weighted finite state transformer (WFST) decoding method and a non-wake-up-free finite state receiver (FSA) decoding method in the speech model are used to decode the speech signal.

[0056] It should be noted that when the car computer switches to the non-wake-up-free scenario, if there is currently a wake-up-free FSA hotword network, it is necessary to uninstall the wake-up-free FSA hotword network and load the non-wake-up-free FSA hotword network.

[0057] In some embodiments of the present invention, after decoding the voice signal using the WFST decoding method and the non-wake-up-free FSA decoding method, the following steps may also be included:

[0058] S301 , performing semantic recognition on a WFST decoding result and a non-wake-up-free FSA decoding result to obtain a second semantic recognition result.

[0059] The speech signal is decoded by the acoustic CLDNN model, the WFST decoding method in the speech model, and the non-wake-up-free FSA decoding result to obtain the WFST decoding result and the non-wake-up-free FSA decoding result, and semantic recognition is performed on it to obtain the second semantic recognition result.

[0060] S302: Generate a control instruction corresponding to the second semantic recognition result to control the vehicle accordingly.

[0061] In a non-wake-up-free scenario, a wake-up word is required to control the car computer to enter the wake-up state, wherein the wake-up word can be set in advance according to the specific situation of the user or the car computer, and the present invention does not limit it. After the car computer enters the wake-up state, it waits for the acquisition of the voice signal. If the next voice signal is not obtained after waiting for the preset time, the car computer is controlled to exit the wake-up state; if a voice signal is obtained within the waiting preset time, and the voice signal is decoded by the WFST decoding method and the non-wake-up-free FSA decoding method, according to the semantic judgment result of the voice signal, the car computer is controlled to execute relevant instructions and then end the session. The above-mentioned waiting preset time can be set according to the actual situation. For example, the waiting preset time can be 20 seconds.

[0062] Based on the switching between the above-mentioned wake-up-free scenario and the non-wake-up-free scenario, refer to Figure 4 After the vehicle computer is powered on, it is globally initialized and its resources are loaded. As the vehicle computer switches between wake-up-free and non-wake-up-free scenarios, the corresponding decoding method also switches. In the wake-up-free scenario, independent decoding is performed using the wake-up-free FSA decoding method, while in the non-wake-up-free scenario, WFST decoding and non-wake-up-free FSA decoding methods are used. This involves a serial decoding process using the acoustic CLDNN model and language model. When the first semantic recognition result in the wake-up-free scenario falls within the preset wake-up-free statement set, it is considered a semantic hit and speech signal reception is stopped. Therefore, switching between scenarios by updating the decoding method in the language model can reduce false triggering of the decoding methods in both wake-up-free and non-wake-up-free scenarios.

[0063] In some embodiments of the present invention, the vehicle-computer voice interaction method is executed based on a neural network processor NPU environment.

[0064] In the process of voice signal recognition, to improve generalization and accuracy, the trained model structure is relatively complex and has a large number of parameters, which places high demands on computing power. Therefore, in some embodiments of the present invention, the vehicle-machine voice interaction method is implemented based on a neural network processor (NPU) environment to better utilize the NPU computing power. As an example, the Qualcomm 8155 chip can be selected to implement the NPU environment.

[0065] Specifically, refer to Figure 5The CPU architecture is usually a von Neumann structure, in which the main body is the storage unit and the control unit, and the computing units only occupy a minority. The CPU can perform well when there are relatively few computing units. Then, for the acoustic CLDNN model in the embodiment of the present invention, if the CPU is used to execute it, the computing power cannot be met, which will cause the CPU occupancy to be very high. On the other hand, the advantage of the NPU is that it can run parallel threads. It is fully called SA8155P and has 8 core units. Its computing power can reach 8TOPS, and when running the acoustic CLDNN model, it can reach 8 trillion times / second. Since the acoustic CLDNN model uses the Pytorch framework to export specific model resources, this format can make corresponding operator library support and optimization based on the NPU computing power environment, and can also migrate the voice model involved in the car-machine voice interaction method from the CPU to the NPU environment, which can reduce the computing use of the car-machine CPU and allow the CPU resources to be more inclined to its own advantages.

[0066] In some embodiments of the present invention, the vehicle-machine voice interaction method further includes:

[0067] S401, loading the speech model into the memory of the NPU, and establishing a mapping relationship between the virtual memory area and the file object in the memory.

[0068] Specifically, the above-mentioned memory may include a magnetic disk.

[0069] S402: When resources need to be loaded and read, data interaction is performed through mapping relationships.

[0070] Since the WFST and FSA resources in the speech model are large, and memory is a very valuable resource for offline devices, it is necessary to process the speech model resources through memory mapping. Figure 6 First, the speech model resources are loaded into the storage, such as the disk, and a virtual memory area is established, which is mapped to the file object on the disk. When the operation object needs to load and read the speech model resources, the data exchange between the disk and the virtual memory area can be directly carried out through the mapping relationship, which reduces the number of copies, improves the efficiency of data reading, and improves memory utilization.

[0071] In order to better understand the vehicle-machine voice interaction method according to an embodiment of the present invention, as an example, Figure 7 A flow chart of a vehicle-machine voice interaction method according to an example of the present invention.

[0072] like Figure 7As shown, after the car computer is started and the resources are initialized, if the user turns on the wake-up-free scenario, it will always be in a listening state. At this time, only the wake-up-free FSA decoding method in the speech model is used to decode the voice signal to be recognized, and the decoding path will not enter the WFST decoding, and the wake-up-free FSA hot word network will be loaded, and the non-wake-up-free FSA hot word network will be unloaded at the same time; first, it is determined whether there is a voice signal. If not, it will always be in a listening state, and the acoustic CLDNN model and speech model will not be run. If the car computer detects a voice signal, the wake-up-free FSA recognition result is obtained through the acoustic CLDNN model and the wake-up-free FSA decoding method in the speech model, and the first semantic recognition result is obtained through semantic recognition. If the first semantic recognition result is not in the preset wake-up-free statement set, the wake-up-free scenario will continue to run until the user manually turns off the wake-up-free scenario or the first semantic recognition result is in the preset wake-up-free statement set.

[0073] Furthermore, if the first semantic recognition result is in the preset wake-free statement set, it is defined as hitting the wake-free semantics, and this round of recognition ends. At this time, the scene is switched to the non-wake-free scene, and the decoding method is changed to WFST decoding and non-wake-free FSA decoding. At the same time, it is necessary to unload the wake-free FSA hot word network and load the non-wake-free FSA hot word network. The remaining sentence network and the original placeholder content can be referred to Figure 3 ; The voice signal is decoded by the WFST decoding method and the non-wake-up FSA decoding method to obtain the WFST decoding result and the non-wake-up FSA decoding result, and the second semantic recognition result is obtained through semantic recognition. If the second semantic recognition result is used, the vehicle is controlled accordingly.

[0074] In summary, the vehicle-machine voice interaction method of an embodiment of the present invention first obtains a voice signal to be recognized. When the vehicle-machine is in a wake-up-free scenario, the voice signal is decoded using the finite state receiver wake-up-free FSA decoding method in the voice model. When the vehicle-machine is in a non-wake-up-free scenario, the voice signal is decoded using the weighted finite state transducer WFST decoding method and the finite state receiver non-wake-up-free FSA decoding method in the voice model. Therefore, the vehicle-machine voice interaction method reduces the problem of false triggering between the wake-up-free scenario and the non-wake-up-free scenario by independently decoding the vehicle-machine in the wake-up-free scenario and the non-wake-up-free scenario. In addition, based on the Qualcomm 8155 chip, the voice model resources of the vehicle-machine voice interaction method are migrated to the neural network processor NPU environment, which reduces the CPU occupancy while improving the model's computational efficiency. In particular, in the scenario of offline voice interaction, the mapping relationship between the disk and the virtual memory area is used to improve memory utilization and data interaction efficiency.

[0075] Furthermore, the present invention provides a vehicle computer.

[0076] In an embodiment of the present invention, the vehicle computer includes a neural network processor NPU and a storage medium, and a computer program is stored on the storage medium. When the computer program is executed by the NPU, the above-mentioned vehicle computer voice interaction method is implemented.

[0077] An in-vehicle computer according to an embodiment of the present invention includes a neural network processor (NPU) and a storage medium storing a computer program. When the computer program is executed by the NPU, it first acquires a speech signal to be recognized. When the computer is in a wake-up-free scenario, the speech signal is decoded using the finite state receiver (FSA) decoding method in the speech model. When the computer is in a non-wake-up-free scenario, the speech signal is decoded using the weighted finite state transducer (WFST) decoding method and the finite state receiver (NSA) non-wake-up-free decoding method in the speech model. Thus, the in-vehicle voice interaction method reduces false triggering issues between wake-up-free and non-wake-up-free scenarios by independently decoding the computer in both wake-up-free and non-wake-up-free scenarios. Furthermore, based on the Qualcomm 8155 chip, the speech model resources of the in-vehicle voice interaction method are migrated to the neural network processor (NPU) environment, reducing CPU usage while improving model computational efficiency. Specifically, in offline voice interaction scenarios, the mapping between disk and virtual memory areas improves memory utilization and data interaction efficiency.

[0078] It should be noted that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device), or in conjunction with such instruction execution system, apparatus, or device. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transmit a program for use by an instruction execution system, apparatus, or device, or in conjunction with such instruction execution system, apparatus, or device. More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection portion having one or more wires (electronic device), a portable computer disk cartridge (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and portable compact disc read-only memory (CDROM). Furthermore, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing it in another suitable manner if necessary, and then storing it in a computer memory.

[0079] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0080] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0081] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like to indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0082] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0083] In the present invention, unless otherwise specified or limited, the terms "installed," "connected," "connect," "fixed," etc. should be understood in a broad sense. For example, they can refer to fixed connection, detachable connection, or integration; mechanical connection, electrical connection; direct connection, or indirect connection through an intermediate medium; internal communication between two components, or interaction between two components, unless otherwise specified. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0084] In the present invention, unless otherwise expressly specified or limited, when a first feature is "above" or "below" a second feature, it may mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediary. Furthermore, when a first feature is "above," "above," or "above" a second feature, it may mean that the first feature is directly above or diagonally above the second feature, or simply means that the first feature is at a higher level than the second feature. When a first feature is "below," "below," or "below" a second feature, it may mean that the first feature is directly below or diagonally below the second feature, or simply means that the first feature is at a lower level than the second feature.

[0085] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A vehicle-machine voice interaction method, characterized in that: The method comprises: Obtaining a speech signal to be recognized; When the vehicle computer is in a wake-up-free scenario, the voice signal is decoded using the wake-up-free finite state receiver (FSA) decoding method in the voice model; When the vehicle computer is in a non-wake-up-free scenario, the voice signal is decoded using a weighted finite state converter (WFST) decoding method and a non-wake-up-free finite state receiver (FSA) decoding method in the voice model; After decoding the voice signal using the wake-up-free FSA decoding method, the method further includes: Performing semantic recognition on the wake-up-free FSA decoding result to obtain a first semantic recognition result; Determining whether the first semantic recognition result is within a preset wake-up-free statement set; If so, a control instruction corresponding to the first semantic recognition result is generated to control the vehicle accordingly and control the vehicle computer to switch to a non-wake-up-free scenario.

2. The vehicle-machine voice interaction method according to claim 1, characterized in that: The preset wake-up-free sentence set includes personalized sentence patterns and non-personalized sentence patterns, and the method further includes: Obtaining a wake-up-free hotword list, wherein the wake-up-free hotword list includes at least one wake-up-free hotword; Based on the wake-up-free hotword list, construct a wake-up-free hotword network; Before adopting the wake-up-free FSA decoding method to decode the voice signal, the wake-up-free hot word network is loaded into the placeholder slot corresponding to the personalized sentence.

3. The vehicle-machine voice interaction method according to claim 2, characterized in that: When controlling the vehicle computer to switch to a non-wake-up-free scenario, the method further includes: Uninstall the wake-up-free hotword network.

4. The vehicle-machine voice interaction method according to claim 1, characterized in that: After decoding the voice signal using the WFST decoding method and the non-wake-up-free FSA decoding method, the method further includes: Perform semantic recognition on the WFST decoding result and the non-wake-free FSA decoding result to obtain a second semantic recognition result; Generate a control instruction corresponding to the second semantic recognition result to control the vehicle accordingly.

5. The vehicle-machine voice interaction method according to claim 4, characterized in that: When the vehicle computer is in a non-wake-up-free scenario and enters a wake-up state, the method further includes: If the voice signal is not obtained after waiting for a preset time, the vehicle computer is controlled to exit the awakening state.

6. The vehicle-machine voice interaction method according to any one of claims 1 to 5, characterized in that: The method is executed based on a neural network processor NPU environment.

7. The vehicle-machine voice interaction method according to claim 6, characterized in that: The method further comprises: Loading the speech model into the memory of the NPU and establishing a mapping relationship between a virtual memory area and a file object in the memory; When resources need to be loaded and read, data interaction is performed through the mapping relationship.

8. The vehicle-machine voice interaction method according to claim 7, characterized in that: The memory includes a magnetic disk.

9. A vehicle computer, characterized in that: It includes a neural network processor NPU and a storage medium, the storage medium stores a computer program, and when the computer program is executed by the NPU, the vehicle-machine voice interaction method as described in any one of claims 1 to 8 is implemented.

Images