Voice wake-up method, device, electronic device, storage medium and product

Abstract

The present invention provides a voice wake-up method, device, electronic device, storage medium and product. The method comprises: when a target voice is received, an audio segment in the target voice is input into a preset acoustic model at a preset time interval to obtain the path identification information and path confidence corresponding to each audio segment; based on the path identification information and path confidence corresponding to each audio segment, the target text path corresponding to the target voice is determined; the target text path is queried in a dictionary map to obtain the text information corresponding to the target text path; when a preset wake-up word exists in the text information, a wake-up operation is performed. In an embodiment of the present invention, by inputting each audio segment in the target voice into the acoustic model at a preset time interval for decoding operation, it is ensured that the audio segment corresponding to each text path obtained is a continuous sentence, thereby avoiding triggering a false wake-up operation or failing to perform the wake-up operation, and improving the efficiency of voice wake-up.

Description

Technical Field

[0001] The present invention relates to the field of audio processing technology, and in particular to a voice wake-up method, device, electronic equipment, storage medium and product. Background Art

[0002] Artificial Intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. With the continuous development of artificial intelligence technology, the application of artificial intelligence wake-up is becoming more and more extensive, and more and more smart devices, such as smart speakers and smart TVs, are equipped with voice wake-up functions.

[0003] The existing voice wake-up method is that the smart device continuously receives the voice sent by the user, performs the same decoding operation on the above voice, and divides the voice into an audio segment at each preset time interval, and then decodes each audio segment as a unit to obtain the text information corresponding to each audio segment, and then detects whether there is a wake-up word in the text information corresponding to each audio segment.

[0004] However, the above-mentioned method of forcibly segmenting the input voice may cause the audio segments received by the decoding module in the device to be not continuous sentences, thereby causing the device to wake up incorrectly or the device to fail to perform the wake-up operation, which reduces the accuracy of the smart device performing the wake-up operation. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide a voice wake-up method, device, electronic device, storage medium and product to solve the technical problem of low efficiency in performing wake-up operations. The specific technical solution is as follows:

[0006] In a first aspect of an embodiment of the present invention, a voice wake-up method is first provided, including:

[0007] When a target speech is received, an audio segment in the target speech is input into a preset acoustic model at a preset interval to obtain path identification information and path confidence corresponding to each audio segment; the path identification information is used to characterize a text path of the corresponding audio segment in a preset dictionary graph, and the path confidence is a probability value of the text path of the corresponding audio segment, and the target speech includes at least one audio segment;

[0008] Determining a target text path corresponding to the target speech based on the path identification information and the path confidence corresponding to each audio segment;

[0009] Query the target text path in the dictionary graph to obtain text information corresponding to the target text path;

[0010] In the case where the text message contains a preset wake-up word, a wake-up operation is performed.

[0011] In a second aspect of the embodiment of the present invention, a voice wake-up device is also provided, including:

[0012] A processing module, configured to input an audio segment in the target speech into a preset acoustic model at a preset interval when receiving the target speech, and obtain path identification information and path confidence corresponding to each audio segment; the path identification information is used to characterize a text path of the corresponding audio segment in a preset dictionary graph, and the path confidence is a probability value of the text path of the corresponding audio segment, wherein the target speech includes at least one audio segment;

[0013] A determination module, configured to determine a target text path corresponding to the target speech based on the path identification information and the path confidence corresponding to each audio segment;

[0014] A query module, used for querying the target text path in the dictionary graph to obtain text information corresponding to the target text path;

[0015] The execution module is used to execute a wake-up operation when the text information contains a preset wake-up word.

[0016] In a third aspect of the embodiments of the present invention, an electronic device is further provided, wherein a computer program is stored in the electronic device, and when the computer program is run on a computer, the computer executes the voice wake-up method described in any of the above embodiments.

[0017] In a fourth aspect of the embodiments of the present invention, a computer-readable storage medium is further provided, wherein instructions are stored in the computer-readable storage medium, and when the computer-readable storage medium is run on a computer, the computer executes the voice wake-up method described in any of the above embodiments.

[0018] In a fifth aspect of the implementation of the present invention, a computer program product comprising instructions is also provided, which, when executed on a computer, enables the computer to execute the voice wake-up method described in any of the above embodiments.

[0019] In an embodiment of the present invention, when a target voice is received, an audio segment in the target voice is input into a preset acoustic model at a preset time interval to obtain path identification information and path confidence corresponding to each audio segment, wherein the path identification information is used to characterize the text path of the corresponding audio segment in a preset dictionary map. By inputting each audio segment in the target voice into the acoustic model at a preset time interval for decoding, it is ensured that the audio segment corresponding to each obtained text path is a continuous sentence, thereby avoiding triggering a false wake-up operation or failing to perform a wake-up operation, and improving the efficiency of voice wake-up. Further, according to the path identification information and path confidence corresponding to each audio segment, the target text path is determined, the target text path is queried in the dictionary map, and the text information corresponding to the target text path is obtained. When the text information contains a preset wake-up word, the wake-up operation is performed. Since the above-mentioned target text path is a text path with a higher path confidence determined based on the path confidence corresponding to each audio segment, the accuracy of performing the wake-up operation can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0021] Figure 1 Schematic diagram of the flow of the voice wake-up method in an embodiment of the present invention;

[0022] Figure 2 This is a schematic diagram of the structure of a voice wake-up device in an embodiment of the present invention;

[0023] Figure 3 The figure is a schematic diagram of the structure of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present invention will be described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0025] To facilitate understanding of the overall technical solution, the relevant technical contents of the present invention are described below.

[0026] In the related art, the voice wake-up device includes an acoustic model and a decoder. The user's input voice is continuously input into the acoustic model, and the input voice is processed using the acoustic model to obtain the path identification information corresponding to the input voice, and the input voice is segmented. Subsequently, the above path identification information is input into the decoder for decoding, and the decoder is used to query the path identification information in a preset dictionary map to obtain the text information corresponding to each voice segment, and then the wake-up operation is performed when the text information contains a wake-up word.

[0027] However, in the above process, the acoustic model is only used to process the input speech once, and the speech is forcibly segmented, which may result in the text information corresponding to the segmented audio segments may not be continuous sentences, thereby causing the device to wake up by mistake or the device failing to perform the wake-up operation, which reduces the efficiency of the smart device in performing the wake-up operation.

[0028] In order to solve the above-mentioned technical problems that may exist, an embodiment of the present invention provides a voice wake-up method, which divides the input voice into at least two audio segments, and inputs each audio segment into the acoustic model in the input order at a preset time interval, and obtains the path identification information corresponding to each audio segment, so as to ensure that the text information corresponding to each audio segment is a continuous sentence, avoid triggering false wake-up operations or failing to perform wake-up operations, and improve the efficiency of voice wake-up.

[0029] See also Figure 1 , Figure 1 Flow chart of the voice wake-up method in the embodiment of the present invention. The voice wake-up method provided by the embodiment of the present invention includes:

[0030] S101, when a target speech is received, an audio segment of the target speech is input into a preset acoustic model at every preset time interval to obtain path identification information and path confidence corresponding to each audio segment.

[0031] The voice wake-up method provided in the embodiment of the present invention can be applied to a voice wake-up device, and the target voice can be a voice input by a user, or a voice downloaded from the outside by the voice wake-up device.

[0032] In this step, when the target speech is received, an audio segment of the target speech is input into the preset acoustic model at a preset interval to obtain the path identification information and path confidence corresponding to each audio segment. The path identification information corresponds to the audio segment one-to-one, and the path confidence corresponds to the audio segment one-to-one. Optionally, each audio segment of the target speech can be input into the acoustic model in the order in which the target speech is input.

[0033] The acoustic model may be a Markov model, a neural network-based model or other types of models, which are not specifically limited here.

[0034] It should be understood that the above path identification information includes multiple conversion IDs, and the above conversion ID represents the node of each phoneme in the audio segment in the preset dictionary graph; the above path identification information is used to represent the text path of the corresponding audio segment in the preset dictionary graph.

[0035] It should be understood that the dictionary graph can be a finite state transducer graph, and the above FST graph is a directed acyclic graph. The dictionary graph includes multiple nodes and text corresponding to each node. An optional implementation method is to determine the nodes corresponding to multiple conversion IDs in the dictionary graph, use these nodes to form a text path, and determine the text information corresponding to the text path based on the text corresponding to each node, and determine the text information as the text information corresponding to the audio end.

[0036] It should be understood that the above path confidence is the sum of the selected probability values ​​corresponding to each conversion ID in the path identification information, and the above path confidence is the probability value of the text path of the corresponding audio segment.

[0037] To understand this, see the following example:

[0038] Assuming that the path identification information corresponding to an audio segment includes 4 conversion IDs, and these 4 conversion IDs represent nodes 1, 2, 3, and 4 in the dictionary graph respectively, the text path represented by the path identification information can be determined to be "node 1 to node 2 to node 3 to node 4". If the selected probability value corresponding to node 1 in the dictionary graph is 0.5, the selected probability value corresponding to node 2 is 0.3, the selected probability value corresponding to node 2 is 0.2, and the selected probability value corresponding to node 2 is 0.4, then the path confidence can be determined to be the sum of 0.5, 0.3, 0.2, and 0.4, that is, 1.4.

[0039] S102: Determine a target text path corresponding to the target speech based on the path identification information and the path confidence corresponding to each audio segment.

[0040] In this step, after obtaining the path identification information and path confidence corresponding to each audio segment, the optimal text path can be selected from the text paths represented by multiple path identification information according to the path identification information and path confidence corresponding to each audio segment, and the optimal text path can be determined as the target text path.

[0041] For a specific technical solution on how to determine the target text path corresponding to the target speech, please refer to the subsequent embodiments.

[0042] S103, querying the target text path in the dictionary graph to obtain text information corresponding to the target text path.

[0043] As described above, a text path is a path on a dictionary graph consisting of multiple conversion IDs. In this step, after determining the target text path, the conversion ID included in the target text path is queried in the dictionary graph to obtain the node corresponding to the target text path, and the characters represented by each node are combined into text information, and the text information is determined as the text information corresponding to the target text path.

[0044] S104: When the text message contains a preset wake-up word, perform a wake-up operation.

[0045] In this step, a wake-up word is preset. When the text message includes the preset wake-up word, the wake-up operation is performed.

[0046] For example, the wake-up word is "Xiaozhi". If the text information corresponding to the target text path is "Xiaozhi, help me turn on the TV", since the wake-up word "Xiaozhi" exists in the text information, the voice wake-up device performs a wake-up operation, and the above wake-up operation can be a power-on operation or other operations.

[0047] It should be understood that if the text message does not contain a preset wake-up word, the wake-up operation will not be performed.

[0048] In an embodiment of the present invention, when a target voice is received, an audio segment in the target voice is input into a preset acoustic model at a preset time interval to obtain path identification information and path confidence corresponding to each audio segment, wherein the path identification information is used to characterize the text path of the corresponding audio segment in a preset dictionary map. By inputting each audio segment in the target voice into the acoustic model at a preset time interval for decoding, it is ensured that the audio segment corresponding to each text path obtained is a continuous sentence, thereby avoiding triggering a false wake-up operation or failing to perform a wake-up operation, and improving the efficiency of voice wake-up. . Further, according to the path identification information and path confidence corresponding to each audio segment, the target text path is determined, the target text path is queried in the dictionary map, the text information corresponding to the target text path is obtained, and the wake-up operation is performed when the text information contains a preset wake-up word. Since the above-mentioned target text path is a text path with a higher path confidence determined based on the path confidence corresponding to each audio segment, the accuracy of performing the wake-up operation can be improved.

[0049] The following specifically describes how to determine the target text path corresponding to the target voice:

[0050] Optionally, determining the target text path corresponding to the target speech based on the path identification information and the path confidence corresponding to each audio segment includes:

[0051] In the case where there is only one maximum path confidence, the text path of the audio segment corresponding to the maximum path confidence is determined as the target text path.

[0052] As described above, the path confidence is the sum of the selected probability values ​​of each node in the corresponding text path. In this embodiment, after obtaining the path confidence corresponding to each audio segment, the maximum path confidence is determined. It should be understood that the higher the value corresponding to the path confidence, the more the text path corresponding to the path confidence matches the target speech, and the more it can reflect the text information corresponding to the target speech.

[0053] Furthermore, if there is only one maximum path confidence, the text path corresponding to the maximum path confidence is determined as the target text path.

[0054] To understand this, see the following example:

[0055] The target speech is divided into 4 audio segments, the path confidence corresponding to the first audio segment is 4, and the corresponding text path is "node 1 to node 2 to node 3 to node 4"; the path confidence corresponding to the second audio segment is 2, and the corresponding text path is "node 1 to node 3 to node 5 to node 6"; the path confidence corresponding to the third audio segment is 1.8, and the corresponding text path is "node 1 to node 4 to node 5 to node 7"; the path confidence corresponding to the fourth audio segment is 2.6, and the corresponding text path is "node 1 to node 5 to node 6 to node 8", then the maximum path confidence is determined to be 4, and the text path corresponding to the first audio segment is determined to be "node 1 to node 2 to node 3 to node 4" as the target text path.

[0056] In this embodiment, when there is only one maximum path confidence, the text path of the audio segment corresponding to the maximum path confidence is determined as the target text path, and the text paths other than the target text path are pruned, and then the text information that best matches the target voice is obtained through the target text path, thereby improving the accuracy of executing the wake-up operation.

[0057] In other embodiments, a pruning operation may be performed based on the path confidence corresponding to each text path and the number of text nodes included. Exemplarily, a pruning operation is performed on a text path whose path confidence is less than a first preset threshold and whose number of text nodes is greater than a second preset threshold.

[0058] Optionally, determining the target text path corresponding to the target speech based on the path identification information and the path confidence corresponding to each audio segment includes:

[0059] In the case where there are at least two maximum path confidences, an audio segment corresponding to each maximum path confidence is determined as a candidate audio segment;

[0060] Get the number of text nodes included in the text path corresponding to each candidate audio segment;

[0061] The text path with the least number of text nodes is determined as the target text path.

[0062] In this embodiment, if there are multiple maximum path confidences, the audio segment corresponding to each path confidence is determined as a candidate audio segment. Further, the number of text nodes included in the text path corresponding to each candidate audio segment is obtained, and the text path with the least number of text nodes is determined as the target text path.

[0063] It should be understood that if the path confidences corresponding to two text paths are the same, but one text path contains more text nodes than the other text path, it means that the average selection probability value corresponding to each node in the text path is low, and the text information represented by the text path is not the text information that best matches the target speech. For easier understanding, please refer to the following example:

[0064] There are two maximum path confidences for the target speech, where the first text path corresponding to one maximum path confidence is "node 1 to node 2 to node 3 to node 4", and the second text path corresponding to the other maximum path confidence is "node 1 to node 3 to node 4 to node 5 to node 7", the number of nodes included in the first text path is 4, and the number of nodes included in the second text path is 5, then the first text path is determined as the target text path. In this embodiment, when there are multiple maximum path confidences, the text path including the smallest number of text nodes is determined as the target text path, and the text paths other than the target text path are pruned, and then the text information that best matches the target speech is obtained through the target text path, thereby improving the accuracy of executing the wake-up operation.

[0065] Optionally, determining the target text path corresponding to the target speech based on the path identification information and the path confidence corresponding to each audio segment includes:

[0066] Performing path confidence compensation on each target audio segment in the audio segment set to obtain a compensated path confidence of each target audio segment;

[0067] The text path of the audio segment corresponding to the maximum path confidence is determined as the target text path.

[0068] The number of text nodes included in the text path corresponding to the above-mentioned target audio segment is greater than or less than a preset node number threshold, and the target audio segment is an audio segment in an audio segment set. Optionally, the above-mentioned preset node number threshold can be the maximum number of nodes included in the text paths corresponding to multiple audio segments, or it can be a pre-set node number threshold.

[0069] In this embodiment, path confidence compensation is performed on each target audio segment to obtain the compensated path confidence of each target audio segment. For a specific technical solution for compensating the path confidence of the target audio segment, please refer to the subsequent embodiments.

[0070] After performing path confidence compensation on the target audio segment, the text path corresponding to the maximum path confidence is determined as the target text path.

[0071] In an optional embodiment, if there are multiple maximum path confidences after path confidence compensation is performed on the target audio segment, the text path corresponding to the maximum path confidence, including the text path with the smallest number of nodes, can be determined as the target text path.

[0072] Optionally, performing path confidence compensation on each target audio segment in the audio segment set to obtain the compensated path confidence of each target audio segment includes:

[0073] For any first target audio segment, adding the path confidence of the first target audio segment to the first confidence to obtain the path confidence of the first target audio segment after compensation;

[0074] For any second target audio segment, the second confidence is subtracted from the path confidence of the second target audio segment to obtain the compensated path confidence of the second target audio segment.

[0075] The audio segment set includes a first target audio segment and a second target audio segment, the number of text nodes included in the first target audio segment is less than a preset node number threshold, and the number of text nodes included in the second target audio segment is greater than the preset node number threshold.

[0076] Among them, the first confidence is associated with a first difference, which is the difference between the preset node number threshold and the number of text nodes included in the first target audio segment; the second confidence is associated with a second difference, which is the difference between the number of text nodes included in the first target audio segment and the preset node number threshold.

[0077] To understand this, see the following example:

[0078] The audio segment set includes 4 audio segments, among which the path confidence corresponding to the first audio segment is 4, and the corresponding number of nodes is 4; the path confidence corresponding to the second audio segment is 2, and the corresponding number of nodes is 3; the path confidence corresponding to the third audio segment is 5, and the corresponding number of nodes is 7; the path confidence corresponding to the fourth audio segment is 1, and the corresponding number of nodes is 2.

[0079] The preset node number threshold is set to 5, and the compensation path confidence corresponding to one node is set to 0.5. Then the first audio segment, the second audio segment and the fourth audio segment are the first target audio segments, and the third audio segment is the second target audio segment.

[0080] The path confidence corresponding to the first compensated audio segment is 4.5, the path confidence corresponding to the second compensated audio segment is 3, the path confidence corresponding to the third compensated audio segment is 4, and the path confidence corresponding to the fourth compensated audio segment is 2.5.

[0081] Since the path confidence corresponding to the first audio segment is the maximum path confidence, the text path corresponding to the first audio segment is determined as the target text path.

[0082] In this embodiment, path confidence compensation is performed on the target audio segment whose number of nodes does not reach the preset node number threshold, and the path confidences corresponding to the compensated audio segments are compared, and the text path corresponding to the maximum path confidence is determined as the target text path, and the text paths other than the target text path are pruned, and then the text information that best matches the target voice is obtained through the target text path, thereby improving the accuracy of executing the wake-up operation.

[0083] Optionally, the step of inputting at least two audio segments corresponding to the target speech into a preset acoustic model in a preset order comprises:

[0084] Using a preset sliding window to segment the target speech to obtain at least two audio segments;

[0085] At each preset time interval, one of the audio segments is input into a preset acoustic model in a preset order.

[0086] In this embodiment, after receiving the target speech, the target speech is segmented using a preset sliding window to divide the target speech into at least two audio segments, wherein each audio segment has the same duration. It should be understood that one audio segment can be regarded as one frame of speech signal, that is, when the target speech is received, the target speech is divided into at least two frames of speech signal.

[0087] At each preset time interval, an audio segment is input into a preset acoustic model in a preset order. Optionally, the time length corresponding to each audio segment can be set to 10 milliseconds. In an optional application scenario, the voice wake-up device inputs the target voice into the acoustic model at every 10 millisecond interval. Optionally, the above preset order is the input order of the target voice.

[0088] like Figure 2 As shown, the embodiment of the present invention further provides a voice awakening device 200, and the voice awakening device 200 includes:

[0089] The processing module 201 is used for inputting an audio segment of the target speech into a preset acoustic model at a preset interval when the target speech is received, so as to obtain path identification information and path confidence corresponding to each audio segment;

[0090] A determination module 202, configured to determine a target text path corresponding to the target speech based on the path identification information and the path confidence corresponding to each audio segment;

[0091] A query module 203, configured to query the target text path in the dictionary graph to obtain text information corresponding to the target text path;

[0092] The execution module 204 is used to execute a wake-up operation when the text message contains a preset wake-up word.

[0093] Optionally, the determining module 202 is specifically configured to:

[0094] In the case where there is only one maximum path confidence, the text path of the audio segment corresponding to the maximum path confidence is determined as the target text path.

[0095] Optionally, the determining module 202 is further specifically configured to:

[0096] In the case where there are at least two maximum path confidences, an audio segment corresponding to each maximum path confidence is determined as a candidate audio segment;

[0097] Get the number of text nodes included in the text path corresponding to each candidate audio segment;

[0098] The text path with the least number of text nodes is determined as the target text path.

[0099] Optionally, the determining module 202 is further specifically configured to:

[0100] Performing path confidence compensation on each target audio segment in the audio segment set to obtain the compensated path confidence of each target audio segment; the number of text nodes included in the text path corresponding to the target audio segment is greater than or less than a preset node number threshold;

[0101] The text path of the audio segment corresponding to the maximum path confidence is determined as the target text path.

[0102] Optionally, the determining module 202 is further specifically configured to:

[0103] For any first target audio segment, the path confidence of the first target audio segment is added to the first confidence to obtain the path confidence of the first target audio segment after compensation; the first confidence is associated with a first difference, and the first difference is the difference between the preset node number threshold and the number of text nodes included in the first target audio segment;

[0104] For any second target audio segment, the second confidence is subtracted from the path confidence of the second target audio segment to obtain the compensated path confidence of the second target audio segment; the second confidence is associated with a second difference, which is the difference between the number of text nodes included in the first target audio segment and the preset node number threshold.

[0105] Optionally, the processing module 201 is specifically configured to:

[0106] Using a preset sliding window to segment the target speech to obtain at least two audio segments;

[0107] At each preset time interval, one of the audio segments is input into a preset acoustic model in a preset order.

[0108] In an embodiment of the present invention, when a target voice is received, an audio segment in the target voice is input into a preset acoustic model at a preset time interval to obtain path identification information and path confidence corresponding to each audio segment, wherein the path identification information is used to characterize the text path of the corresponding audio segment in a preset dictionary map. By inputting each audio segment in the target voice into the acoustic model at a preset time interval for decoding, it is ensured that the audio segment corresponding to each obtained text path is a continuous sentence, thereby avoiding triggering a false wake-up operation or failing to perform a wake-up operation, and improving the efficiency of voice wake-up. Further, according to the path identification information and path confidence corresponding to each audio segment, the target text path is determined, the target text path is queried in the dictionary map, and the text information corresponding to the target text path is obtained. When the text information contains a preset wake-up word, the wake-up operation is performed. Since the above-mentioned target text path is a text path with a higher path confidence determined based on the path confidence corresponding to each audio segment, the accuracy of performing the wake-up operation can be improved.

[0109] The embodiment of the present invention further provides an electronic device, such as Figure 3 As shown, it includes a processor 301 , a communication interface 302 , a memory 303 and a communication bus 304 , wherein the processor 301 , the communication interface 302 , and the memory 303 communicate with each other via the communication bus 304 .

[0110] Wherein, the memory 303 is used to store computer programs;

[0111] The processor 301 is configured to execute a program stored in the memory 303. When the computer program is executed by the processor 301, when a target speech is received, an audio segment of the target speech is input into a preset acoustic model at every preset time interval to obtain path identification information and path confidence corresponding to each audio segment;

[0112] Determining a target text path corresponding to the target speech based on the path identification information and the path confidence corresponding to each audio segment;

[0113] Query the target text path in the dictionary graph to obtain text information corresponding to the target text path;

[0114] In the case where the text message contains a preset wake-up word, a wake-up operation is performed.

[0115] When the computer program is executed by the processor 301, the computer program is further configured to, when there is only one maximum path confidence, determine the text path corresponding to the maximum path confidence as the target text path.

[0116] When the computer program is executed by the processor 301, the computer program is further configured to, when there are at least two maximum path confidences, determine the audio segment corresponding to each maximum path confidence as a candidate audio segment;

[0117] Get the number of text nodes included in the text path corresponding to each candidate audio segment;

[0118] The text path with the least number of text nodes is determined as the target text path.

[0119] When the computer program is executed by the processor 301, it is further used to perform path confidence compensation on each target audio segment in the audio segment set to obtain the compensated path confidence of each target audio segment;

[0120] The text path of the audio segment corresponding to the maximum path confidence is determined as the target text path.

[0121] When the computer program is executed by the processor 301, the computer program is further configured to, for any first target audio segment, add the path confidence of the first target audio segment to the first confidence to obtain the path confidence of the first target audio segment after compensation;

[0122] For any second target audio segment, the second confidence is subtracted from the path confidence of the second target audio segment to obtain the compensated path confidence of the second target audio segment.

[0123] When the computer program is executed by the processor 301, it is further used to segment the target speech using a preset sliding window to obtain at least two audio segments;

[0124] At each preset time interval, one of the audio segments is input into a preset acoustic model in a preset order.

[0125] In an embodiment of the present invention, when a target voice is received, an audio segment in the target voice is input into a preset acoustic model at a preset time interval to obtain path identification information and path confidence corresponding to each audio segment, wherein the path identification information is used to characterize the text path of the corresponding audio segment in a preset dictionary map. By inputting each audio segment in the target voice into the acoustic model at a preset time interval for decoding, it is ensured that the audio segment corresponding to each obtained text path is a continuous sentence, thereby avoiding triggering a false wake-up operation or failing to perform a wake-up operation, and improving the efficiency of voice wake-up. Further, according to the path identification information and path confidence corresponding to each audio segment, the target text path is determined, the target text path is queried in the dictionary map, and the text information corresponding to the target text path is obtained. When the text information contains a preset wake-up word, the wake-up operation is performed. Since the above-mentioned target text path is a text path with a higher path confidence determined based on the path confidence corresponding to each audio segment, the accuracy of performing the wake-up operation can be improved.

[0126] The communication bus mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0127] The communication interface is used for communication between the above terminal and other devices.

[0128] The memory may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0129] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0130] In another embodiment of the present invention, a computer-readable storage medium is provided, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer executes the voice wake-up method described in any of the above embodiments.

[0131] In another embodiment of the present invention, a computer program product including instructions is provided. When the computer program product is run on a computer, the computer executes the voice wake-up method described in any one of the above embodiments.

[0132] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Solid State Disk (SSD)), etc.

[0133] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0134] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0135] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A voice wake-up method, characterized in that: The method comprises: When a target speech is received, an audio segment in the target speech is input into a preset acoustic model at a preset interval to obtain path identification information and path confidence corresponding to each audio segment; the path identification information is used to characterize a text path of the corresponding audio segment in a preset dictionary graph, and the path confidence is a probability value of the text path of the corresponding audio segment, and the target speech includes at least one audio segment; Determining a target text path corresponding to the target speech based on the path identification information and the path confidence corresponding to each audio segment; Query the target text path in the dictionary graph to obtain text information corresponding to the target text path; If the text message contains a preset wake-up word, performing a wake-up operation; The determining the target text path corresponding to the target speech based on the path identification information and the path confidence corresponding to each audio segment comprises: Performing path confidence compensation on each target audio segment in the audio segment set to obtain the compensated path confidence of each target audio segment; the number of text nodes included in the text path corresponding to the target audio segment is greater than or less than a preset node number threshold; Determine the text path of the audio segment corresponding to the maximum path confidence as the target text path; The audio segment set includes a first target audio segment and a second target audio segment, the number of text nodes included in the first target audio segment is less than a preset node number threshold, and the number of text nodes included in the second target audio segment is greater than the preset node number threshold; The performing path confidence compensation on each target audio segment in the audio segment set to obtain the compensated path confidence of each target audio segment comprises: For any first target audio segment, the path confidence of the first target audio segment is added to the first confidence to obtain the path confidence of the first target audio segment after compensation; the first confidence is associated with a first difference, and the first difference is the difference between the preset node number threshold and the number of text nodes included in the first target audio segment; For any second target audio segment, the second confidence is subtracted from the path confidence of the second target audio segment to obtain the compensated path confidence of the second target audio segment; the second confidence is associated with a second difference, which is the difference between the number of text nodes included in the second target audio segment and the preset node number threshold.

2. The method according to claim 1, characterized in that The determining the target text path corresponding to the target speech based on the path identification information and the path confidence corresponding to each audio segment comprises: In the case where there is only one maximum path confidence, the text path of the audio segment corresponding to the maximum path confidence is determined as the target text path.

3. The method according to claim 1, characterized in that: The determining the target text path corresponding to the target speech based on the path identification information and the path confidence corresponding to each audio segment comprises: In the case where there are at least two maximum path confidences, an audio segment corresponding to each maximum path confidence is determined as a candidate audio segment; Get the number of text nodes included in the text path corresponding to each candidate audio segment; The text path with the least number of text nodes is determined as the target text path.

4. The method according to claim 1, characterized in that The step of inputting an audio segment of the target speech into a preset acoustic model at a preset time interval includes: Using a preset sliding window to segment the target speech to obtain at least two audio segments; At each preset time interval, one of the audio segments is input into a preset acoustic model in a preset order.

5. A voice wake-up device, characterized in that: The device comprises: A processing module, configured to input an audio segment in the target speech into a preset acoustic model at a preset interval when receiving the target speech, and obtain path identification information and path confidence corresponding to each audio segment; the path identification information is used to characterize a text path of the corresponding audio segment in a preset dictionary graph, and the path confidence is a probability value of the text path of the corresponding audio segment, wherein the target speech includes at least one audio segment; A determination module, configured to determine a target text path corresponding to the target speech based on the path identification information and the path confidence corresponding to each audio segment; A query module, used for querying the target text path in the dictionary graph to obtain text information corresponding to the target text path; An execution module, configured to execute a wake-up operation when a preset wake-up word exists in the text information; The determining module is further specifically used for: Performing path confidence compensation on each target audio segment in the audio segment set to obtain the compensated path confidence of each target audio segment; the number of text nodes included in the text path corresponding to the target audio segment is greater than or less than a preset node number threshold; Determine the text path of the audio segment corresponding to the maximum path confidence as the target text path; The audio segment set includes a first target audio segment and a second target audio segment, the number of text nodes included in the first target audio segment is less than a preset node number threshold, and the number of text nodes included in the second target audio segment is greater than the preset node number threshold; The determining module is further specifically used for: For any first target audio segment, the path confidence of the first target audio segment is added to the first confidence to obtain the path confidence of the first target audio segment after compensation; the first confidence is associated with a first difference, and the first difference is the difference between the preset node number threshold and the number of text nodes included in the first target audio segment; For any second target audio segment, the second confidence is subtracted from the path confidence of the second target audio segment to obtain the compensated path confidence of the second target audio segment; the second confidence is associated with a second difference, which is the difference between the number of text nodes included in the second target audio segment and the preset node number threshold.

6. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor is used to implement the voice wake-up method according to any one of claims 1 to 4 when executing the program stored in the memory.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the voice wake-up method according to any one of claims 1 to 4 is implemented.

8. A computer program product, characterized in that The computer program product is stored in a non-volatile storage medium, and the computer program product is executed by at least one processor to implement the voice wake-up method according to any one of claims 1 to 4.

Images