A method, device, equipment and readable storage medium for implementing CTC model decoding

Abstract

The present invention discloses a method, apparatus, device, and readable storage medium for decoding based on a CTC model, comprising: feeding a sound to be decoded into a decoding model; wherein the sound to be decoded includes blank characters and non-blank characters; and decoding the sound to be decoded by the decoding model; wherein the decoding model includes a DNN structure and an HMM structure; the DNN structure is used to obtain acoustic scores of blank characters and non-blank characters; and the HMM structure includes a structure that merges the hidden Markov models of blank characters and the non-blank characters to which they are connected, wherein each non-blank character has a sideband, the emission probability of the sideband self-jumping is the acoustic score of the blank character, and the emission probability of the sideband shifting is the acoustic score of the non-blank character. The present invention uses a new decoding model to delete blank signals that have no actual output meaning during the decoding process of the sound to be decoded, thereby solving the problem of excessive memory required for decoding.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to an implementation method, apparatus, device, and readable storage medium based on CTC model decoding. Background Art

[0002] The CTC algorithm is used to train deep neural network models, mainly to solve speech recognition, text recognition and other sequence tasks.

[0003] The posterior of the CTC model has obvious spike characteristics, that is, during the decoding process, many spike signals will appear. Spike signals are the embodiment of noise in decoding, and these spike signals include many blank signals with no actual output meaning. The existence of these blank signals increases the size of the decoding network and requires too much memory for decoding. Summary of the Invention

[0004] The embodiments of the present application solve the problem of excessive memory required for decoding in the prior art by providing an implementation method, apparatus, device, and readable storage medium based on CTC model decoding.

[0005] In a first aspect, an embodiment of the present application provides an implementation method for decoding based on a CTC model, including:

[0006] Sending the sound to be decoded into the decoding model; wherein the sound to be decoded includes blank characters and non-blank characters;

[0007] The decoding model decodes the sound to be decoded; wherein the decoding model includes a DNN structure and an HMM structure;

[0008] The DNN structure is used to obtain the acoustic scores of the blank characters and the non-blank characters;

[0009] The HMM structure includes a structure that merges the hidden Markov models of the blank character and the non-blank character connected to it, each of the non-blank characters has a sideband, the emission probability of the sideband self-jumping is the acoustic score of the blank character, and the emission probability of the sideband transfer is the acoustic score of the non-blank character.

[0010] In one embodiment, the blank character in the HMM structure is not a hidden Markov model.

[0011] In one embodiment, the HMM structure does not provide sidebands with blank characters as input for the non-blank characters in the sound to be decoded.

[0012] The present invention provides an implementation method for decoding based on a CTC model. By providing a decoding model, a sideband with blank characters as input is no longer provided to non-blank characters. Instead, the sideband of the non-blank characters is enabled to have the ability to self-jump and transfer. On the basis of ensuring that the output and input of the decoding of the sound to be decoded can be aligned, a large number of blank signals with no actual output meaning are deleted in the decoding process of the sound to be decoded, the memory required for decoding is reduced, and the problem of excessive memory required for decoding is solved.

[0013] In a second aspect, the present invention provides an implementation device based on CTC model decoding, comprising:

[0014] A device for implementing CTC model decoding, comprising:

[0015] A transmission module, configured to send the acquired sound to be decoded into a decoding model; wherein the sound to be decoded includes blank characters and non-blank characters;

[0016] A decoding module, configured to decode the to-be-decoded sound using the decoding model; wherein the decoding model includes a DNN structure and an HMM structure;

[0017] The DNN structure is used to obtain the acoustic scores of the blank characters and the non-blank characters;

[0018] The HMM structure includes a structure that merges the hidden Markov models of the blank character and the non-blank character connected to it, each of the non-blank characters has a sideband, the emission probability of the sideband self-jumping is the acoustic score of the blank character, and the emission probability of the sideband transfer is the acoustic score of the non-blank character.

[0019] In one embodiment, the blank character in the HMM structure is not a hidden Markov model.

[0020] In one embodiment, the HMM structure does not provide sidebands with blank characters as input for the non-blank characters in the sound to be decoded.

[0021] In a third aspect, the present invention provides a device comprising a memory and a processor;

[0022] The memory is used to store programs;

[0023] The processor is used to execute the program and implement any one of the above-described methods for decoding based on the CTC model when executing the program.

[0024] In a fourth aspect, the present invention provides a readable storage medium, which stores a program, and when the program is executed by a processor, the processor implements a method for implementing CTC model decoding as described in any one of the above. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a schematic diagram of an implementation method based on CTC model decoding provided by the present invention;

[0026] Figure 2 Schematic diagram of the HMM structure in the decoding model provided by the present invention;

[0027] Figure 3 This is a schematic diagram of an implementation device based on CTC model decoding provided by the present invention;

[0028] Figure 4 This is a schematic diagram of a device provided by the present invention. DETAILED DESCRIPTION

[0029] To better understand the above technical solutions, exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. Instead, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0030] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0031] See Figure 1 、 Figure 2 As shown, Figure 1 This is a schematic diagram of an implementation method based on CTC model decoding provided by the present invention. Figure 2 It is a schematic diagram of the HMM structure in the decoding model provided by the present invention.

[0032] The present invention provides a voice wake-up method based on keyword recognition technology, comprising the following steps:

[0033] Step S1: sending the acquired sound to be decoded into a decoding model; wherein the sound to be decoded includes blank characters and non-blank characters.

[0034] Specifically, the sound to be decoded is the sound input into the decoding model for decoding. The composition of the sound to be decoded is analyzed based on the decoding result. The sound to be decoded can be regarded as consisting of blank characters and non-blank characters.

[0035] More specifically, taking the CTC model decoding of the sound to be decoded as an example, the output decoding result contains many blank characters that have no actual output meaning. The presence of these blank characters increases the size of the decoding network and requires too much memory for decoding.

[0036] It is understandable that the memory required for decoding can be reduced by reducing these blank characters.

[0037] Specifically, the role of these blank characters in the original CTC model is to provide a sideband with blank characters as input for all non-blank characters that cannot self-jump, so that the output decoding result can be aligned with the input sound to be decoded; in a decoding structure provided by the present invention, the blank characters are merged with the non-blank characters connected to them, and each non-blank character has a self-jumping sideband. The emission probability of the self-jumping of the sideband is the acoustic score of the blank character, and the emission probability of the transfer is the acoustic score of the blank character, so that the output decoding result can be aligned with the input sound to be decoded.

[0038] More specifically, the decoding model includes a DNN structure and an HMM structure, and the acoustic scores of blank characters and non-blank characters are obtained by the DNN structure.

[0039] Step S2: The decoding model decodes the sound to be decoded; wherein the decoding model includes a DNN structure and an HMM structure.

[0040] The DNN structure is used to obtain the acoustic scores of blank characters and non-blank characters.

[0041] The HMM structure includes a structure that combines the hidden Markov models of blank characters and the non-blank characters connected to them. Each non-blank character has a sideband, the emission probability of the sideband self-jumping is the acoustic score of the blank character, and the emission probability of the sideband transfer is the acoustic score of the non-blank character.

[0042] In this embodiment, a decoding model is provided in which sidebands with blank characters as input are no longer provided to non-blank characters. Instead, the sidebands of non-blank characters are enabled to self-jump and transfer. This ensures that the output and input of the sound decoding to be decoded are aligned, thereby achieving the deletion of a large number of blank signals that have no actual output significance in the decoding process of the sound to be decoded, reducing the memory required for decoding, and solving the problem of excessive memory required for decoding.

[0043] See Figure 3 As shown, Figure 3 This is a schematic diagram of an implementation device based on CTC model decoding provided by the present invention.

[0044] The present invention provides a voice wake-up device based on keyword recognition technology, comprising:

[0045] The transmission module 1 is used to send the acquired sound to be decoded into the decoding model; wherein the sound to be decoded includes blank characters and non-blank characters.

[0046] The decoding module 2 is used to decode the sound to be decoded using a decoding model; wherein the decoding model includes a DNN structure and an HMM structure.

[0047] The DNN structure is used to obtain the acoustic scores of blank characters and non-blank characters.

[0048] The HMM structure includes a structure that combines the hidden Markov models of blank characters and the non-blank characters connected to them. Each non-blank character has a sideband, the emission probability of the sideband self-jumping is the acoustic score of the blank character, and the emission probability of the sideband transfer is the acoustic score of the non-blank character.

[0049] In this embodiment, the blank characters in the HMM structure are not hidden Markov models.

[0050] In this embodiment, the HMM structure does not provide sidebands with blank characters as input to non-blank characters in the sound to be decoded.

[0051] See Figure 4 As shown, Figure 4 This is a schematic diagram of a device provided by the present invention.

[0052] The present invention provides a device comprising:

[0053] The memory is used to store programs;

[0054] The processor is used to execute the program and implement the above-mentioned CTC model-based decoding implementation method when executing the program.

[0055] In a device provided by the invention, the processor is used to provide computing and control capabilities to support the operation of the entire computer device.

[0056] The memory may include non-volatile storage media and internal memory.

[0057] The non-volatile storage medium can store an operating system and a program. The program includes program instructions, which, when executed, can cause the processor to execute any of the above-mentioned methods for implementing CTC model-based decoding.

[0058] The internal memory provides an environment for running the program of the method for detecting appearance defects of traditional Chinese medicine plugs as a non-volatile storage medium. When the program is executed by the processor, the processor can execute any of the above-mentioned implementation methods based on CTC model decoding.

[0059] The network interface is used for network communication, such as sending assigned tasks.

[0060] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the device to which the solution of the present application is applied. The specific device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0061] It should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0062] Those skilled in the art will understand that the structure within the above-mentioned computer device is merely a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those described above, or combine certain components, or have a different arrangement of components.

[0063] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to ROM / RAM, magnetic disks, or optical disks) containing computer-usable program code.

[0064] The present invention is described with reference to flowcharts and / or block diagrams of methods, apparatuses, and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0065] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0066] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 Any modifications, equivalent substitutions and improvements of the present invention should be included in the protection scope of the present invention.

[0067] In this embodiment, the image acquisition block or multiple blocks specify the steps of the function.

[0068] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program, and the computer program includes program instructions. The processor executes the program instructions to implement any one of the CTC model decoding-based implementation methods provided in the embodiments of the present application.

[0069] The computer-readable storage medium may be an internal storage unit of the computer device described in the aforementioned embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a SmartMedia Card (SMC), a Secure Digital (SD) card, a flash memory card, etc., equipped on the computer device.

[0070] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. All works shall be made within the spirit and principle of the present invention.

[0071] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for implementing decoding based on a CTC model, characterized in that: include: The acquired sound to be decoded is sent to the decoding model; wherein the sound to be decoded includes blank characters and non-blank characters; The decoding model decodes the sound to be decoded; wherein the decoding model includes a DNN structure and an HMM structure; The DNN structure is used to obtain the acoustic scores of the blank characters and the non-blank characters; The HMM structure includes a structure that merges the hidden Markov models of the blank character and the non-blank character connected to it, each of the non-blank characters has a sideband, the emission probability of the sideband self-jumping is the acoustic score of the blank character, and the emission probability of the sideband transfer is the acoustic score of the non-blank character.

2. The method for implementing CTC model decoding according to claim 1, wherein: The blank character in the HMM structure is not a hidden Markov model.

3. The method for implementing CTC model decoding according to claim 1, wherein: The HMM structure does not provide a sideband having a blank character as input to the non-blank character in the sound to be decoded.

4. A device for implementing CTC model decoding, characterized in that: include: A transmission module, configured to send the acquired sound to be decoded into a decoding model; wherein the sound to be decoded includes blank characters and non-blank characters; A decoding module, configured to decode the to-be-decoded sound using the decoding model; wherein the decoding model includes a DNN structure and an HMM structure; The DNN structure is used to obtain the acoustic scores of the blank characters and the non-blank characters; The HMM structure includes a structure that merges the hidden Markov models of the blank character and the non-blank character connected to it, each of the non-blank characters has a sideband, the emission probability of the sideband self-jumping is the acoustic score of the blank character, and the emission probability of the sideband transfer is the acoustic score of the non-blank character.

5. The device for implementing CTC model decoding according to claim 4, wherein: The blank character in the HMM structure is not a hidden Markov model.

6. The device for implementing CTC model decoding according to claim 4, wherein: The HMM structure does not provide a sideband having a blank character as input to the non-blank character in the sound to be decoded.

7. A device for implementing CTC model decoding, characterized in that: The device includes a memory and a processor; The memory is used to store programs; The processor is used to execute the program and implement the implementation method based on CTC model decoding according to any one of claims 1 to 3 when executing the program.

8. A readable storage medium, characterized in that: The readable storage medium stores a program, and when the program is executed by the processor, the processor implements a method for decoding based on a CTC model according to any one of claims 1 to 3.

Images