Multi-intention voice signal recognition method and device

Abstract

The invention provides a multi-intention voice signal recognition method and device, and the method comprises the following steps: S1, receiving a voice signal in real time through an equipment terminal, carrying out the real-time segmentation of voice data in the voice signal through a local voice recognition model, and sequentially obtaining a plurality of groups of audio segment data; s2, the equipment terminal obtains a plurality of groups of voice command recognition results corresponding to the plurality of groups of audio clip data through a local voice recognition model; and S3, the equipment terminal sequentially executes control actions based on the multiple groups of voice command recognition results. According to the multi-intention voice signal recognition method and device, the requirements of the multi-intention voice signal recognition method and device for own software and hardware performance can be effectively reduced, and a low-cost and high-efficiency multi-intention voice signal recognition function is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of voice control, and in particular relates to a method and device for recognizing multi-intention voice signals. Background Art

[0002] Voice control technology means that users can operate devices or applications to perform corresponding actions through verbal commands. In the actual voice interaction process, there may be multiple intentions or multiple control actions in a single voice signal. Under the existing technology, for complex multi-intent voice signals, the control terminal usually adopts a semantic understanding method to distinguish the content of multiple intents, that is, in the voice control system of the control terminal, natural language processing technology and deep learning technology are used to perform lexical analysis, syntactic analysis, semantic analysis, intent recognition, named entity recognition and other steps. In this process, the control terminal needs to complete data calculation through an online cloud server, or the control terminal has a high-computing power chip to realize data calculation. Therefore, the conventional multi-intent voice signal recognition method and device have high requirements for network and hardware use, and the overall construction cost is relatively high. Summary of the invention

[0003] The present invention aims to provide a multi-intent speech signal recognition method and device to solve the technical problems that the conventional multi-intent speech signal recognition method and device in the prior art have high requirements for network and hardware usage and high overall construction costs.

[0004] To solve the above problems, the technical solution of the present invention is: a method for recognizing multi-intention speech signals, comprising the following steps: S1: The device terminal receives a voice signal in real time, and segments the voice data in the voice signal in real time through a local voice recognition model to obtain several groups of audio segment data in sequence; S2: the device terminal obtains, through the local speech recognition model, several groups of speech command recognition results corresponding to several groups of audio segment data; S3: The device terminal executes control actions in sequence based on several groups of voice command recognition results.

[0005] Preferably, an audio comparison database is provided in the local speech recognition model, and a set of wake-up word comparison data and several sets of command word comparison data are pre-recorded in the audio data comparison database.

[0006] Preferably, in S1, the device terminal receives the voice signal in real time, further comprising the following steps: S11: The device terminal uses the local speech recognition model to accumulate the wake-up word audio data in the speech signal at fixed capture time intervals starting from the starting point of the speech signal, and compares the wake-up word audio data with the wake-up word comparison data in the audio comparison database in real time until the wake-up word audio data matches the wake-up word comparison data, then the device terminal enters the voice command recognition state.

[0007] Preferably, in S1, the device terminal obtains the audio segment data, further comprising the following steps: S12: the device terminal uses the local speech recognition model to remove the wake-up word audio data from the speech signal as a starting point, accumulates and obtains first audio data in the speech signal at fixed capture time intervals, and compares the first audio data with several groups of command word comparison data in the audio comparison database in real time until a group of command word comparison data matches the first audio data, and then marks the first audio data as first audio segment data; S13: If there is still remaining voice data after removing the wake-up word audio data and the first audio data from the voice signal, the device terminal uses the local voice recognition model to remove the wake-up word audio data and the first audio data from the voice signal as a starting point, and accumulates and obtains the second audio data in the remaining voice signal at fixed capture time intervals, and compares the second audio data with several groups of command word comparison data in the audio comparison database in real time until there is a group of command word comparison data that matches the second audio data, and then marks the second audio data as second audio segment data; S14: Repeat step S13 until the voice signal transmission is completed.

[0008] Preferably, limited to a single group of the voice signals, the device terminal can obtain at most four groups of the audio segment data based on the local voice recognition model. After the device terminal obtains four groups of the audio segment data based on the local voice recognition model, it automatically exits the current voice command recognition state.

[0009] Preferably, in steps S11-S14, the device terminal uses the local speech recognition model to compare the wake-up word audio data with the wake-up word comparison data, and compares the audio data with the command word comparison data. When at least part of the data in the wake-up word audio data has an overall time domain and frequency domain information similarity with the wake-up word comparison data that is higher than 80%, and at least part of the data in the audio data has an overall time domain and frequency domain information similarity with the command word comparison data that is higher than 80%, the local speech recognition model determines that the wake-up word audio data contains the wake-up word comparison data, and the audio data contains the command word comparison data.

[0010] Preferably, in S1, the device terminal obtains the audio segment data, further comprising the following steps: S15: After the device terminal obtains any group of the audio segment data through the local speech recognition model, it detects whether there is a new voice signal input within a unit time through the human voice recognition algorithm of the local speech recognition model. If there is a new input voice signal, the recognition process of the next audio data is executed. If there is no new input voice signal, the device terminal automatically exits the current voice command recognition state.

[0011] Preferably, a voice command mapping table is further provided in the local voice recognition model, and the voice command mapping table pre-records the voice command recognition results corresponding to several groups of command word comparison data.

[0012] Preferably, in S2, the device terminal obtains the voice command recognition result, further comprising the following steps: S21: The device terminal obtains several groups of command word comparison data corresponding to several groups of audio segment data, and obtains the voice command recognition results corresponding to the several groups of command word comparison data by querying the voice command mapping table in the local voice recognition model.

[0013] Based on the same concept, the present invention also provides a multi-intent speech signal recognition device, which executes a multi-intent speech signal recognition method as described in any one of the above, and is used to realize multi-intent recognition of speech information and multi-intent speech control functions of a device terminal through a local speech recognition model.

[0014] Due to the adoption of the above technical solution, the present invention has the following advantages and positive effects compared with the prior art: The present invention provides a multi-intention speech signal recognition method and device. In the process of speech signal recognition, a non-semantic understanding method is used to recognize command words, that is, a local speech recognition model is used to capture part of the audio data and compare it with the command word comparison data in the audio comparison database in real time, and the continuous speech signal is divided into several groups of audio segment data. The command word comparison data matching the audio segment data is further used to query and obtain the speech command recognition result corresponding to the audio segment data. In this process, no semantic understanding processing step is involved, that is, there is no need to perform a large amount of calculation on the speech signal data, and the chip computing power, hardware environment and online cloud server usage requirements of the device terminal itself are smaller, and the construction and use costs are lower. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The present invention provides a flowchart of a method for recognizing multi-intention speech signals. DETAILED DESCRIPTION

[0016] The following is a further detailed description of a multi-intention speech signal recognition method and device proposed by the present invention in conjunction with the accompanying drawings and specific embodiments. The advantages and features of the present invention will become more apparent from the following description and claims.

[0017] First embodiment See also Figure 1 This embodiment provides a multi-intention speech signal recognition method for realizing a low-cost recognition function of multi-intention commands in a speech signal, which specifically includes the following steps: S1: The device terminal receives a voice signal in real time, and segments the voice data in the voice signal in real time through a local voice recognition model, and obtains several groups of audio segment data in sequence. Any group of audio segment data represents a record of a group of voice command content.

[0018] S2: The device terminal obtains several groups of voice command recognition results corresponding to several groups of audio segment data through a local voice recognition model, and the voice command recognition results are used to instruct the device terminal to perform corresponding control actions.

[0019] S3: The device terminal executes control actions in sequence based on several groups of voice command recognition results to realize the voice control function of the device terminal.

[0020] Preferably, in one embodiment, an audio comparison database is provided in the local speech recognition model, and a set of wake-up word comparison data and several sets of command word comparison data are pre-recorded in the audio data comparison database.

[0021] In S1, the device terminal receives the voice signal in real time, further comprising the following steps: S11: The device terminal uses a local speech recognition model to accumulate the wake-up word audio data in the speech signal at fixed capture time intervals starting from the starting point of the speech signal, and compares the wake-up word audio data with the wake-up word comparison data in the audio comparison database in real time. If there is a match between the wake-up word audio data and the wake-up word comparison data, the device terminal enters the voice command recognition state.

[0022] Below, step S11 will be explained with a specific voice signal recognition process. For example, in the audio data comparison library, the wake-up word comparison data is recorded as the sound data of "hello device terminal". When the user continuously speaks the voice content of "xxxx hello device terminal xxxxxxxx" to the device terminal, the device terminal will use the local voice recognition model to start from the starting point of the voice signal. While the voice signal is input, it accumulates and obtains the audio data in the voice signal every 50ms. That is, at the 50th ms, the local voice recognition model will obtain the voice content of "xxx", and at the 100th ms, the local voice recognition model will accumulate and obtain the voice content of "hello xxxxx", until at 150ms, the local voice recognition model accumulates and obtains the complete voice content of "hello device terminal". At this time, the voice content of "hello device terminal" can be matched with the wake-up word comparison data "hello device terminal" in the audio comparison database, so the voice content of "hello device terminal" is the complete wake-up word audio data, and the device terminal is awakened and enters the voice command recognition state.

[0023] By setting the wake-up word, when the device terminal is in the non-voice command recognition state, some of its software processes and hardware are in a dormant state, which can reduce the power consumption of the device terminal and extend the service life of the device terminal. When the device terminal is in the non-voice command recognition state, it can reduce the continuous monitoring of the surrounding environment and reduce the potential risk of privacy leakage. By using specific wake-up words, it can reduce the false triggering of the device terminal caused by environmental noise or other sounds, thereby ensuring the accuracy of voice control of the device terminal.

[0024] In S1, the device terminal obtains the audio segment data, further comprising the following steps: S12: When the device terminal enters the voice command recognition state, the device terminal uses the local voice recognition model to remove the wake-up word audio data from the voice signal as the starting point, and accumulates and obtains the first audio data in the voice signal at a fixed capture time interval, and compares the first audio data with several groups of command word comparison data in the audio comparison database in real time, until there is a group of command word comparison data that matches the first audio data, then the first audio data is marked as the first audio segment data as a whole; S13: After the device terminal obtains the first audio segment data, if there is still remaining voice data after removing the wake-up word audio data and the first audio data from the voice signal, or there is still a new voice signal being input, the device terminal uses a local voice recognition model to remove the wake-up word audio data and the first audio data from the voice signal as a starting point, and accumulates and obtains the second audio data in the remaining voice signal at a fixed capture time interval, and compares the second audio data with several groups of command word comparison data in the audio comparison database in real time until there is a group of command word comparison data that matches the second audio data, then the second audio data is marked as the second audio segment data as a whole; S14: Repeat step S13 to segment the remaining voice data again, and obtain the third audio segment data, the fourth audio segment data, etc. in sequence until the voice signal transmission is completed.

[0025] Below, steps S12-S14 will be explained with a specific voice signal recognition process. For example, in the audio data comparison library, there are four groups of command word comparison data recorded as "open", "fan", "adjust", and "third gear" sound data respectively. When the user continuously says "Hello, device terminal xx, please turn on the fan xxx and adjust it to the third gear xxxx" to the device terminal, the voice content may contain some noise data. The device terminal will first recognize the wake-up word of the voice signal through the local voice recognition model. Refer to S11 above. When the device terminal enters the voice command recognition state, the remaining voice content provided by the user is "xx please turn on the fan xxx and adjust it to the third gear xxxx". Adjust to the third gear xxxx". At this time, the device terminal will further use the local voice recognition model to accumulate and obtain the audio data in the voice signal every 50ms, starting from the starting point of the remaining voice signal. That is, at the 50th ms, the local voice recognition model will obtain the voice content of "xx please", and at the 100th ms, the local voice recognition model will accumulate and obtain the voice content of "xx please open". At this time, the voice content of "xx please open" matches the sound data of "open" in the command word comparison data, so the voice content of "xx please open" is the complete first audio data, and the voice content of "xx please open" is further marked as the first audio segment data as a whole.

[0026] Subsequently, the device terminal recognizes the remaining voice signal again through the local voice recognition model, that is, it removes the wake-up word audio data and the first audio data from the voice signal as the starting point, and re-accumulates the audio data in the voice signal every 50ms, that is, at the 150ms, the local voice recognition model will obtain the voice content of "fan". At this time, the voice content of "fan" matches the sound data of "fan" in the command word comparison data, so the voice content of "fan" is the second audio data, and the voice content of "fan" is further marked as the second audio segment data.

[0027] Similarly, at 200ms, the local speech recognition model will obtain the speech content of "xxx and", and at 250ms, the local speech recognition model will accumulate and obtain the speech content of "xxx and adjust". At this time, the speech content of "xxx and adjust" matches the sound data of "adjust" in the command word comparison data, so the speech content of "xxx and adjust" is the complete third audio data, and the speech content of "xxx and adjust" is further marked as the third audio segment data. At 300ms, the local speech recognition model will obtain the speech content of "to the third gear". At this time, the speech content of "to the third gear" matches the sound data of "third gear" in the command word comparison data, so the speech content of "to the third gear" is the complete fourth audio data, and the speech content of "to the third gear" is further marked as the fourth audio segment data. At 350ms, the local speech recognition model will obtain the speech content of "xxxx" in the remaining speech signal, and no new speech signal will be input. At the same time, the speech content of "xxxx" cannot be matched with any set of command word comparison data, so the speech content of "xxxx" is discarded, and the voice signal transmission is confirmed to be completed, and then the device terminal exits the voice command recognition state.

[0028] It is worth noting that in this embodiment, the capture time is set to 50ms for example. In order to improve the recognition accuracy of the voice signal, the capture time can be further shortened in other embodiments. At the same time, it should be clear that in this embodiment, the input process of the voice signal and the recognition process of the voice signal are carried out synchronously, and there is only a certain processing time difference between the two. For example, the recognition process of the voice signal will be carried out 100ms after the voice signal starts to be input, thereby ensuring that the voice control of the device terminal can respond in time. At the same time, in order to further improve the flexibility of the voice command recognition of the device terminal, when the device terminal obtains the previous set of audio segment data, if the next set of audio segment data cannot be recognized within 1000ms, the device terminal automatically exits the voice command recognition state to avoid continuous irrelevant voice signal input, causing the voice recognition process of the device terminal to continue.

[0029] In summary, through steps S12-S14, the device terminal uses the local speech recognition model to accumulate the audio data in the voice signal, and compares the accumulated audio data with the command word comparison data in real time. When the accumulated audio data matches any set of command word comparison data, it means that the current accumulated audio data contains a set of voice command content. The above steps are repeated to achieve the recognition of multiple intention contents in a single continuous voice signal. In this recognition process, there is no need to perform semantic understanding processing on the voice signal, which reduces a large amount of data calculation operations and can effectively reduce the device terminal's demand for computing chips, hardware environment and online cloud servers.

[0030] Preferably, in one embodiment, it is limited that in a single group of voice signals, the device terminal can obtain at most four groups of audio segment data based on the local voice recognition model. After the device terminal obtains four groups of audio segment data based on the local voice recognition model, it automatically exits the current voice command recognition state. In the conventional single device terminal voice control process, four groups of audio segment data and four groups of voice command content can basically complete the entire control of a certain execution action of the device terminal, such as "open", "fan", "adjust", and "third gear". By limiting the number of audio segment data that the device terminal can obtain based on the local voice recognition model in a single group of voice signals, the performance requirements for the device terminal are reduced. When the device terminal completes the control action based on the current voice signal, the cache of the current voice signal will be deleted, ensuring that low-cost, low-performance device terminals can also quickly and efficiently perform voice signal processing operations.

[0031] Preferably, in one embodiment, in steps S11-S14, the device terminal uses the local speech recognition model to compare the wake-up word audio data with the wake-up word comparison data, and in the process of comparing the audio data with the command word comparison data, when at least part of the wake-up word audio data has an overall time domain and frequency domain information similarity with the wake-up word comparison data that is higher than 80%, and at least part of the audio data has an overall time domain and frequency domain information similarity with the command word comparison data that is higher than 80%, the local speech recognition model determines that the wake-up word audio data contains the wake-up word comparison data, and the audio data contains the command word comparison data, that is, there is a match between the wake-up word audio data and the wake-up word comparison data, and there is a match between the audio data and the command word comparison data.

[0032] For example, in the above steps S12-S14, a certain voice content "xx please open" is obtained, and the voice content of "xx please open" contains the voice content of "open", and the time domain and frequency domain information of the voice content of "open" are compared with the overall time domain and frequency domain information of the sound data of "open" in the pre-recorded command word comparison data. If the similarity between the two is higher than 80%, it is determined that the voice content of "xx please open" matches the sound data of "open" in the command word comparison data. It is worth noting that the judgment standard of similarity can be adjusted based on the actual accuracy required for use. At the same time, this embodiment only provides a relatively simple matching method with less requirements on the performance of the device terminal. In other embodiments, it can also be implemented by methods including audio fingerprint algorithm, perceptual hash comparison, etc.

[0033] Preferably, in one embodiment, the device terminal obtains the audio segment data in S1, further comprising the following steps: S15: After the device terminal obtains any set of audio segment data through the local voice recognition model, it detects whether there is a new voice signal input within a unit time through the human voice recognition algorithm of the local voice recognition model. If there is a new voice signal input, the recognition process of the next audio data is executed. If there is no new voice signal input, the device terminal automatically exits the current voice command recognition state. The unit time is limited to 200ms.

[0034] It is worth noting that, in this embodiment, the voice signals must be acquired through the human voice recognition algorithm of the local voice recognition model, thereby eliminating irrelevant noise in the ambient sound and improving the accuracy of voice recognition of the device terminal.

[0035] Preferably, in one embodiment, a voice command mapping table is further provided in the local voice recognition model, in which voice command mapping table are pre-recorded voice command recognition results corresponding to several groups of command word comparison data, i.e., the sound data of the four groups of command word comparison data of "open", "fan", "adjust", and "third gear" as described in steps S12-S14 above, respectively correspond to the machine language of the four groups of voice command recognition results of open, fan, adjust, and third gear that can be understood by the device terminal.

[0036] In S2, the device terminal obtains the voice command recognition result, further comprising the following steps: S21: The device terminal obtains several groups of command word comparison data corresponding to the several groups of audio segment data, and obtains voice command recognition results corresponding to the several groups of command word comparison data by querying the voice command mapping table in the local voice recognition model.

[0037] It is worth noting that there is a directional relationship between the voice command recognition results. For example, in a single voice message, the device terminal obtains four groups of voice command recognition results, namely, open, fan, adjust, and third gear, in sequence through the local voice recognition model. According to the order in which the audio fragment data appear in the voice signal, the device terminal can automatically know through the local voice recognition model that the first voice command recognition result "open" points to the "fan", the second voice command recognition result "fan" has a further target action of "adjusting" the gear, and the third voice command recognition result "adjusting" has a further target result of adjusting to the "third gear", thereby realizing multi-intention control of the device terminal.

[0038] In another embodiment, the voice command recognition results may be set with priorities. For example, among the four groups of voice command recognition results of open, fan, adjust, and three gears obtained by the device terminal through the local voice recognition model, the voice command recognition result "open" for device switch control has a first-level priority, the voice command recognition result "fan" for the device body has a second-level priority, the voice command recognition result "adjust" for the device target action has a third-level priority, and the voice command recognition result "three gears" for the device target action mode has a fourth-level priority. When the device terminal executes multi-intent tasks, the control actions can be executed in sequence from high priority to low priority according to the priority of the voice command recognition results. Therefore, in this embodiment, the device terminal can independently determine the execution order of multi-intent control commands through priority setting, and has lower requirements for the order of providing multiple intent commands in the voice data provided by the user. Among them, if voice command recognition results of the same priority appear in a single voice information, such as when the device terminal obtains four groups of voice command recognition results of turn on, fan, turn off, and air conditioning through the local voice recognition model, the device terminal will follow the order of appearance of the audio segment data in the voice signal and the principle of independently controlling the low-priority voice command recognition results with the same high-priority voice command recognition results, that is, the device terminal executes the control actions of turning on the fan and turning off the air conditioner in turn.

[0039] Therefore, in the present embodiment, during the voice recognition process of the device terminal, it is only necessary to obtain several groups of command word comparison data corresponding to several groups of audio segment data, and then the voice command recognition results corresponding to the several groups of command word comparison data can be obtained by querying the pre-recorded voice command mapping table. There is no need to make the device terminal understand the specific meaning of the audio segment data or the command word comparison data, and the output of the voice command recognition results can be completed, which places lower requirements on the software and hardware performance of the device terminal.

[0040] In summary, this embodiment provides a multi-intent speech signal recognition method. In the process of speech signal recognition, a non-semantic understanding method is used to recognize command words, that is, a local speech recognition model is used to capture part of the audio data and compare it with the command word comparison data in the audio comparison database in real time, and the continuous speech signal is divided into several groups of audio segment data, and further the command word comparison data matching the audio segment data is used to query and obtain the speech command recognition result corresponding to the audio segment data. In this process, no semantic understanding processing step is involved, that is, there is no need to perform a large amount of calculation on the speech signal data, and the chip computing power, hardware environment and online cloud server usage requirements of the device terminal itself are smaller, and the construction and use costs are lower.

[0041] Second embodiment Based on the same concept, this embodiment provides a multi-intent speech signal recognition device, which executes a multi-intent speech signal recognition method as described in the first embodiment, and is used to realize multi-intent recognition of speech signals and multi-intent speech control functions of device terminals through a local speech recognition model. On the premise of realizing efficient recognition of multi-intent speech signals, the overall construction cost of the device terminal and the voice control implementation cost are effectively reduced.

[0042] The embodiments of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments. Even if various changes are made to the present invention, if these changes fall within the scope of the claims of the present invention and their equivalents, they still fall within the protection scope of the present invention.

Claims

1. A method for recognizing multi-intention speech signals, characterized in that: The steps include: S1: The device terminal receives a voice signal in real time, and segments the voice data in the voice signal in real time through a local voice recognition model to obtain several groups of audio segment data in sequence; S2: the device terminal obtains, through the local speech recognition model, several groups of speech command recognition results corresponding to several groups of audio segment data; S3: The device terminal executes control actions in sequence based on several groups of voice command recognition results.

2. The multi-intention speech signal recognition method according to claim 1, characterized in that: An audio comparison database is provided in the local speech recognition model, and a group of wake-up word comparison data and several groups of command word comparison data are pre-recorded in the audio data comparison database.

3. The multi-intention speech signal recognition method according to claim 2, characterized in that: In S1, the device terminal receives the voice signal in real time, further comprising the following steps: S11: The device terminal uses the local speech recognition model to accumulate the wake-up word audio data in the speech signal at fixed capture time intervals starting from the starting point of the speech signal, and compares the wake-up word audio data with the wake-up word comparison data in the audio comparison database in real time until the wake-up word audio data matches the wake-up word comparison data, then the device terminal enters the voice command recognition state.

4. The multi-intention speech signal recognition method according to claim 3, characterized in that: In S1, the device terminal obtains the audio segment data, further comprising the following steps: S12: the device terminal uses the local speech recognition model to remove the wake-up word audio data from the speech signal as a starting point, accumulates and obtains first audio data in the speech signal at fixed capture time intervals, and compares the first audio data with several groups of command word comparison data in the audio comparison database in real time until a group of command word comparison data matches the first audio data, and then marks the first audio data as first audio segment data; S13: If there is still remaining voice data after removing the wake-up word audio data and the first audio data from the voice signal, the device terminal uses the local voice recognition model to remove the wake-up word audio data and the first audio data from the voice signal as a starting point, and accumulates and obtains the second audio data in the remaining voice signal at fixed capture time intervals, and compares the second audio data with several groups of command word comparison data in the audio comparison database in real time until there is a group of command word comparison data that matches the second audio data, and then marks the second audio data as second audio segment data; S14: Repeat step S13 until the voice signal transmission is completed.

5. The multi-intention speech signal recognition method according to claim 4, characterized in that: Limited to a single group of the voice signals, the device terminal can obtain at most four groups of the audio segment data based on the local voice recognition model. After the device terminal obtains four groups of the audio segment data based on the local voice recognition model, it automatically exits the current voice command recognition state.

6. The multi-intention speech signal recognition method according to claim 4, characterized in that: In steps S11-S14, the device terminal uses the local speech recognition model to compare the wake-up word audio data with the wake-up word comparison data, and in the process of comparing the audio data with the command word comparison data, when at least part of the wake-up word audio data has an overall time domain and frequency domain information similarity with the wake-up word comparison data that is higher than 80%, and at least part of the audio data has an overall time domain and frequency domain information similarity with the command word comparison data that is higher than 80%, the local speech recognition model determines that the wake-up word audio data contains the wake-up word comparison data, and the audio data contains the command word comparison data.

7. The multi-intention speech signal recognition method according to claim 4, characterized in that: In S1, the device terminal obtains the audio segment data, further comprising the following steps: S15: After the device terminal obtains any group of the audio segment data through the local speech recognition model, it detects whether there is a new voice signal input within a unit time through the human voice recognition algorithm of the local speech recognition model. If there is a new input voice signal, the recognition process of the next audio data is executed. If there is no new input voice signal, the device terminal automatically exits the current voice command recognition state.

8. The multi-intention speech signal recognition method according to claim 4, characterized in that: A voice command mapping table is also provided in the local voice recognition model, and the voice command recognition results corresponding to several groups of command word comparison data are pre-recorded in the voice command mapping table.

9. The multi-intention speech signal recognition method according to claim 8, characterized in that: In S2, the device terminal obtains the voice command recognition result, further comprising the following steps: S21: The device terminal obtains several groups of command word comparison data corresponding to several groups of audio segment data, and obtains the voice command recognition results corresponding to the several groups of command word comparison data by querying the voice command mapping table in the local voice recognition model.

10. A multi-intention speech signal recognition device, characterized in that: Execute a multi-intent speech signal recognition method as described in any one of claims 1-9, which is used to realize multi-intent recognition of speech information and multi-intent speech control functions of a device terminal through a local speech recognition model.

Images