Psychological modification sound generation ai system for pets, sound transmission system, and sound playback device
An AI system generates sound tools for psychological correction in pets, addressing the need for emotional well-being through wearable devices that efficiently transmit and play tailored sounds for stress relief.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- PET MUSE CO LTD
- Filing Date
- 2025-12-03
- Publication Date
- 2026-06-11
Smart Images
Figure KR2025020583_11062026_PF_FP_ABST
Abstract
Description
AI system for generating sounds for pet psychological correction, sound transmission system, and sound playback device
[0001] The present invention relates to an AI system for generating sound for psychological correction of pets. The present invention also relates to a system for transmitting sound using the results generated by such a system. The present invention also relates to a sound playback device that plays sound transmitted by such a transmission system, which is worn by a pet or placed around a pet.
[0002] In modern society, companion animals are recognized as part of the family, and interest in their health and welfare is increasing. In particular, psychological stability and stress relief are receiving attention as important factors directly linked to a pet's health.
[0003] Various studies are being conducted on the effects of music on the psychological stability of dogs. According to research, dogs experience a decrease in heart rate when listening to classical, reggae, and soft rock music, which can lead to stress relief. For example, a study by the University of Glasgow and the Society for the Prevention of Cruelty to Animals (SPCA) investigated the impact of various music genres on stress levels in pet dogs; the results showed that dogs tended to feel calmer and experience reduced stress when listening to specific types of music. Additionally, a study conducted by psychologist Deborah Wells at Queen's University revealed positive behavioral changes when playing classical music to shelter dogs, such as the dogs lying down and resting comfortably. On the other hand, heavy metal and grunge are genres of music that dogs do not prefer. These genres feature aggressive and fast sounds and tend to increase stress by inducing tension and anxiety in dogs.
[0004] Studies have also been conducted on the effects of music on cats, and according to these studies, cats may perceive auditory stimuli through autonomic nervous system activity even under general anesthesia. In particular, research suggests that using music in operating rooms can contribute to feline safety. This was confirmed through the effects of music and genres on the respiratory rate and pupil size of cats, suggesting that music can contribute to enhancing feline stability.
[0005] Meanwhile, pet collars have primarily been used to prevent pets from escaping during outings or to keep them tied up in a designated location. However, as the role of pets extends beyond that of mere companionship, the function of collars is also evolving. Recently, pet collars equipped with various functions beyond just decorative appearance are being developed, and along with technological advancements, there is a growing demand for innovative products that contribute to the welfare and care of pets.
[0006] The present invention aims to provide a device capable of generating sound through an artificial intelligence algorithm that can contribute to psychological corrective effects, such as stress relief and emotional stability for pets, and effectively transmitting and playing such sound. To this end, the present invention additionally aims to provide an AI system for generating sound for such purposes.
[0007] To achieve the above objective, the present invention comprises: an input module that receives learning data including a plurality of sound source data and one or more labels corresponding to the sound source data, wherein the psychological correction effect of a pet according to the playback result of each of the sound source data is labeled; and an AI module comprising a processor that performs supervised learning by a predetermined artificial intelligence algorithm based on the learning data received through the input module to generate a sound generation tool for generating a psychological correction sound for the pet corresponding to the psychological correction effect, and performs the artificial intelligence algorithm on the learning data received through the input module. The present invention provides an AI system for generating psychological correction sounds for pets, comprising: a memory that stores the sound generation tool generated by the AI module and provides the sound generation tool to a transmission system provided for transmitting the sound generation tool to a sound playback device; wherein each of the sound generation tools includes a function module and a parameter input value of the function module, the function module has program code for generating sound data corresponding to the psychological correction effect, and the parameter input value is a value generated in correspondence with the function module and set so that the function module generates the sound data corresponding to the psychological correction effect.
[0008] The above parameter input values may be configured to include a frequency set, which is a set of frequencies for sound, and a magnitude set, which is a set of magnitudes for sound.
[0009] The above sound source data may be generated by using a plurality of sound sources artificially created by applying one or more of frequency modulation (FM), amplitude modulation (AM), and frequency modulation synthesis (FM Systnesis) to a predetermined number of sound sources, or by performing a predetermined operation on digital data generated from the predetermined number of sound sources and / or the artificially created sound sources in one or more ways of pulse code modulation (PCM) or delta modulation (DM).
[0010] According to another aspect of the present invention, a transmission system for transmitting a sound generating tool generated by an AI system of the above configuration to a sound playback device is provided, comprising: an input unit that receives a selection command for a psychological correction effect requested by a user; and a transmission unit that transmits the sound generating tool corresponding to the selected psychological correction effect to a sound playback device provided for the pet as the selection command is input to the input unit.
[0011] Additionally, a pet psychological correction sound playback device is provided, characterized by comprising: a receiving unit for receiving a sound generating tool transmitted by the transmission system; a storage unit for storing the received sound generating tool; an execution unit for generating sound data by applying a parameter input value included in the sound generating tool to a function module included in the sound generating tool and executing the function module; and a speaker for playing the sound data generated by the execution unit.
[0012] According to another aspect of the present invention, a transmission system for transmitting a sound generation tool generated by the AI system as described above to a sound playback device is provided, comprising: an input unit that receives a selection command for a psychological correction effect requested by a user; and a transmission unit that transmits function modules within the sound generation tool to a sound playback device provided for the pet in advance and stores them within the sound playback device, and transmits a parameter input value within the sound generation tool corresponding to the selected psychological correction effect to the sound playback device as the selection command is input to the input unit.
[0013] Additionally, a pet psychological correction sound playback device is provided, characterized by comprising: a receiving unit that receives a function module and a parameter input value transmitted by the transmission system; a storage unit that stores the received function module; an execution unit that stores the parameter input value in the storage unit upon receiving the parameter input value in the receiving unit, applies the parameter input value to a corresponding function module to execute the function module and generate sound data; and a speaker that plays the sound data generated by the execution unit.
[0014] According to the present invention, sound that can contribute to psychological corrective effects, such as stress relief and emotional stability for pets, can be generated through an artificial intelligence algorithm. In addition, the sound generated in this way can be effectively transmitted and played back.
[0015] FIG. 1 is a schematic diagram illustrating an AI system for generating psychological correction sounds for pets according to the present invention.
[0016] FIG. 2 is a schematic diagram illustrating the configuration of a transmission system according to the present invention.
[0017] FIG. 3 is a schematic diagram illustrating the configuration of an acoustic playback device according to the present invention.
[0018] The present invention will be described in more detail below with reference to the drawings.
[0019] In the following description, configurations described as "~part," "~module," etc., may each be implemented by a single computer, server, or device, or may be configured to be provided as separate functional modules within a single computer or server. Here, a functional module may, for example, be a chip manufactured in the form of an ASIC (Application Specific Integrated Circuit) independently to perform the corresponding function; if the corresponding function is to store data so that it can be written and retrieved, it may be a general volatile or non-volatile memory for data storage; if the corresponding function is to read from another memory or process data transmitted from another device according to a specific algorithm, it may be a microprocessor; and if the function is to receive data from another device and transmit it to yet another device, it may be a data communication interface. Furthermore, configurations described as "~system," "~device," etc., may be implemented as a system equipped with a server and auxiliary peripherals, or may be configured in the form of a small terminal. In the description of the present invention, detailed illustration and description of general configurations that are not directly related to the gist of the invention or are typically self-evident are omitted to avoid obscuring the gist of the invention.
[0020] FIG. 1 is a schematic diagram illustrating an AI system (10) for generating psychological correction sounds for pets according to the present invention. The AI system (10) of the present invention is configured to include an input module (12), an AI module (14), and a memory (16).
[0021] The input module (12) receives training data used for training the AI module (14). The input module (12) includes a communication interface that is directly connected to another device or connected via a network to enable communication with another device, for example, and receives training data transmitted from another device.
[0022] Training data consists of audio data and label data.
[0023] Sound source data is digitized general sound, consisting of, for example, data encoded through PCM (Pulse Code Modulation). The sounds used to generate this sound source data include multiple diverse sounds, such as natural sounds like thunder, wind, and waves; mechanical sounds like horns; biological sounds like human and animal voices; and created sounds like music.
[0024] However, due to limitations in the number and type of the sounds obtained in this way, they are insufficient to constitute a vast amount of sound sources with various frequencies or sizes required for the learning of the AI module (14). Therefore, the present invention uses a method of artificially generating a number of new sound sources for the learning of the AI module (14). Specifically, multiple sound sources can be generated by frequency modulating (FM) and / or amplitude modulating (AM) each sound obtained as described above. Additionally, multiple additional sound sources can be generated by selecting a predetermined number of sounds obtained as described above and performing frequency modulation synthesis (FM Systnesis). In the generation of sound sources, only one of frequency modulation, amplitude modulation, and frequency modulation synthesis may be applied, or two or more may be applied simultaneously. Furthermore, multiple additional sound sources can be generated by modulating or synthesizing the sound sources generated in this manner again in the same way. In this way, as many as 10,000 sound sources of various frequencies and sizes can be generated as needed for the learning of the AI module (14).
[0025] Furthermore, additional audio data can be generated by performing specific operations on the audio sources obtained through the above method. Specifically, audio data is data obtained by converting the aforementioned analog audio sources into digital data through Pulse Code Modulation (PCM) or Delta Modulation (DM); additional training data can be generated by applying specific operations to each of these training data. Here, specific operations refer to specific function operations, such as logarithmic scaling on a single piece of digital data, or operations that synthesize two or more pieces of digital data using a specific algorithm.
[0026] Label data refers to data in which the playback results of multiple audio data prepared in this manner are each labeled. The labels added through labeling indicate information regarding the psychological correction effects on pets. For example, a label table is constructed that classifies the effects on pets into various categories, such as "label 1: calming, label 2: cheerful, label 3: comfortable, label 4: patience, label 5: sleep." This label table can be subdivided by type of pet, such as "dog, cat," or by body size, such as "large dog, medium dog, small dog" for dogs, or by breed, such as "Maltese, Poodle, Spitz." Accordingly, the label table is constructed as a set of labels for multiple required psychological correction effects.
[0027] One label data corresponds to the audio data, and one label data includes one or more labels. That is, if two or more effects are confirmed for any single audio data, two or more labels may be assigned. For example, if the effects are actually tested by playing audio 1 to various types of pets, and the test results confirm 'sedative effect on large dogs' of label 1-5 and 'sleep effect on poodles' of label 5-12, then label data including two labels, label 1-5 and label 5-12, is assigned to audio 1.
[0028] In this way, a pair of audio data and its matched label data constitutes a single training dataset. By assigning one or more labels to all audio data in this manner, the entire training dataset is constructed.
[0029] The AI module (14) learns based on the training data received through the input module (12). The AI module (14) includes a microprocessor that performs a process of learning by an AI algorithm using the training data. Here, the AI algorithm may be various existing algorithms such as CNN, RNN, and Neural Network, or a combination of two or more of these algorithms. The learning of the present invention corresponds to supervised learning using label data containing a psychological correction effect, which is the actual playback result for each audio source data.
[0030] The AI module (14) generates a sound generation tool as a result of learning. The present invention does not provide a method of generating the sound itself that provides a psychological correction effect as a result of learning, but rather a method in which the AI module (14) generates a sound generation tool as a result of learning. This sound generation tool is a tool having the function of generating a sound that provides a psychological correction effect, and is composed of program code that implements an algorithm to perform this function.
[0031] These sound generation tools are generated in correspondence with each psychological correction effect. Here, a psychological correction effect refers to any effect that a pet owner may request, such as, for example, "sleep effect for poodles." This effect can be configured to be identical to, for example, the composition of the label. That is, an effect table identical to the label table as exemplified above is prepared, and for each effect included in this effect table, one sound generation tool is generated as a learning result of the AI module (14). For example, sound generation tool 5-12 is provided for the case where the psychological correction effect is "sleep effect for poodles" of label 5-12. It may also be configured so that two or more sound generation tools are provided for a single psychological correction effect (e.g., a single label). In this case, the two or more sound generation tools are named, for example, sound generation tool 5-12-a and sound generation tool 5-12-b.
[0032] The sound generation tool consists of a function module and parameter input values. The function module has the function of generating sound data corresponding to a psychological correction effect. This function module is the main part of the sound generation tool and consists of program code. For example, in the above example, the function module 5-12 included in the sound generation tool 5-12 consists of program code that generates digital data of sound that brings about a sleep effect on a poodle. This function module is configured to require parameter input values. These input values are generated in correspondence with the function module and are values set so that the function module generates sound data that brings about a corresponding psychological correction effect. The parameter input values are also generated according to the learning results of the AI module (14).
[0033] The function of a function module can be conceptually expressed as shown in the following formula.
[0034] R = F 5-12 (fs, as)
[0035] Here, F is a function module, and the subscript 5-12 indicates a function for generating acoustic data that produces a psychocorrecting effect 5-12 corresponding to label 5-12. R is acoustic data that generates a sound exhibiting a psychocorrecting effect of label 5-12 as the result of executing the function module.
[0036] fs and as are parameters, F 5-12 These are parameters required for the function module to generate acoustic data exhibiting the psychocorrecting effect of label 5-12. These parameters (fs, as) are composed of, for example, a frequency set (fs: frequency set), which is a set of numerical values representing the frequency of the sound, and an amplitude set (as: amplitude set), which is a set of numerical values representing the magnitude of the sound. Examples of input values for each parameter can be expressed mathematically as follows.
[0037] fs = [ f1, f7, f35, .... ]
[0038] as = [ a10, a24, a43, .... ]
[0039] Each value of the fs parameter is a frequency, and the above formula means that the fs parameter requires frequencies f1, f7, f35, ... as input values. Each value of the as parameter is a magnitude, and the above formula means that the as parameter requires magnitudes a10, a24, a43, ... as input values. That is, F 5-12 The function requires frequency values such as f1, f7, and f35 as input values as fs, and magnitude values such as a10, a24, and a43 as input values as as. In short, F 5-12 The function is a function that generates acoustic data capable of obtaining a psychocorrecting effect 5-12, and requires frequency values included in fs and magnitude values included in as as parameter inputs.
[0040] For a single function module, at least one set of input values is generated. Here, the set of input values refers to a set that includes the input values of the fs parameters and the input values of the as parameters. It is desirable to configure the system so that two or more sets of input values are generated for a single function module.
[0041] Meanwhile, as an embodiment of the present invention, a value included in fs, e.g., f17, is expressed as a frequency value in Hz units such as 1700, and a value included in as, e.g., a24, is expressed as a magnitude value in dB units such as 24. Accordingly, in this case, the input value set consists of fs, which includes one or more frequency values in Hz units, and as, which includes one or more magnitude values in dB units.
[0042] However, as another example of the present invention, the input value of the fs parameter in the present invention may be composed of one or more acoustic data. In this case, the acoustic data is data having frequencies f1, f7, f35, .... For example, one acoustic data may be configured to have only the frequency f1, and acoustic data having frequencies f1, f7, f35, etc., may be provided in this manner. As another example, one acoustic data may be provided that includes all frequencies f1, f7, f35, etc. In such an example, the input value of fs is not a frequency value, but the acoustic data itself having the frequency included in fs. Similarly, the input value of as is not a magnitude value, but the acoustic data itself having the magnitude included in as. At this time, in order to minimize the size, it is preferable that the acoustic data used as the parameter input value be composed of data fragments of the smallest possible capacity within the limits of including the required frequency and the required magnitude.
[0043] In this way, the process of generating function modules and parameter pressure values is performed for each of the psychological correction effects, that is, for each label, thereby providing sound generation tools corresponding to each psychological correction effect. In short, the AI module (14) generates multiple sound generation tools corresponding to the psychological correction effects of each label, wherein each sound generation tool includes one or more function modules and also includes one or more sets of parameter input values corresponding to each function module.
[0044] As described above, the AI module (14) generates a sound generation tool by supervised learning using each sound source data and label data resulting from the psychological correction effect thereon. In that the sound generation tool includes a function module, the present invention differs from general supervised learning, which generates data of the same or similar type as the data input for learning. Below, an example of a method in which the AI module (14) generates such a sound generation tool through an AI algorithm is described.
[0045] The AI module (14) groups the audio data corresponding to each label for all labels and vectorizes the audio data within each group. Accordingly, a group containing multiple audio data vectors is established for each label. Then, the AI module (14) uses the vectors within each group to extract a group vector representing the characteristics of that group. By repeating this process for all groups, multiple group vectors are obtained.
[0046] The AI module (14) generates an arbitrary function. Various types of existing functions and combinations thereof may be used to generate this arbitrary function. Any known function may be employed as an existing function. For example, various functions such as polynomial functions, exponential functions, logarithmic functions, trigonometric functions, Bessel functions, zeta functions, Z-transform functions, convolution, Lagrangian, etc. may be employed. As an example, the AI module (14) may generate the following function arbitrarily by combining existing functions. The following function is provided as an example for convenience of explanation and does not imply that such example functions are appropriate implementation examples of the algorithm of the function module of the present invention.
[0047] f(a, b) = 3cos(a) + 2log(b)
[0048] Here, a and b, which are parameters of the function, are parameters input to an arbitrary function, and these are values corresponding to the aforementioned fs and as, respectively. The AI module (14) generates input values for these parameters together when generating an arbitrary function. At this time, the AI module (14) generates multiple sets of input values so that multiple sounds can be generated through multiple input values for a single function. The AI module (14) inputs the arbitrarily generated parameters into the arbitrarily generated function to generate the result.
[0049] In short, the AI module (14) generates an arbitrary function and a set of multiple input values to be input into the arbitrary function, and generates multiple operation results by sequentially inputting parameters into the arbitrary function.
[0050] Then, the AI module (14) vectorizes the result of the operation and determines the similarity between the vectorized result of the operation and each group vector. Typical similarity measurement algorithms, such as Cosine Similarity and Euclidean Distance, may be used to determine the similarity. The AI module (14) determines which group vector has the highest similarity to the vector of the result of the operation and is above a certain level, and a label corresponding to the group vector thus determined is assigned to an arbitrary function and a parameter input thereto. The arbitrary function and the parameter input determined in this way are adopted as the sound generation tool of the present invention, and the arbitrary function determined thereto becomes the function module of the present invention, and the determined parameter is adopted as the parameter input value of the present invention. Meanwhile, any function or combination of any function and its parameter with a similarity level below a certain level is not adopted as the sound generation tool, even if it is most similar to any group vector. Accordingly, when determining the sound generation tool, not only relative similarity but also absolute similarity is reflected.
[0051] The AI module (14) repeats this process infinitely. That is, by repeating the process of generating an arbitrary function and its input parameters, vectorizing the result of the operation, and determining whether to adopt it based on similarity verification, an acoustic generation tool is provided for all labels.
[0052] A plurality of sound generation tools generated by the AI module (14) are stored in the memory (16). The sound generation tools stored in the memory (16) are utilized as described below.
[0053] Below, a system and device utilizing the AI system (10) of the present invention as described above will be explained.
[0054] FIG. 2 is a schematic diagram illustrating the configuration of an artificial intelligence sound transmission system (20) for generating psychological correction sounds for pets according to the present invention, and illustrates a transmission system (20) that transmits the results generated by the AI system (10) shown in FIG. 1.
[0055] As described above, the AI module (14) of the AI system (10) of the present invention does not generate the 'sound' itself that brings about a psychological correction effect, but rather generates a 'sound generation tool' that performs the function of generating such sound. The transmission system (20) of the present invention transmits the sound generation tool itself that is thus generated. In the present invention, the transmission system (20) may be configured as a separate system from the AI system (10) to receive the sound generation tool, which is the result, from the memory (16) of the AI system (10), or the transmission system (20) and the AI system (10) may be configured as a single integrated system.
[0056] The transmission system (20) is configured to include a storage unit (22), an input unit (24), and a transmission unit (26).
[0057] The storage unit (22) receives and stores a plurality of sound generation tools that are generated using an AI module (14) learned by the AI system (10) and stored in memory (16). If the transmission system (20) is configured by a separate terminal or server, such a storage unit (22) may be provided separately within the transmission system (20). The storage unit (22) may be composed of a simple data storage chip such as a memory IC.
[0058] The input unit (24) receives a selection command for a psychological correction effect requested by the user from the user terminal (30). To receive the psychological correction effect, the user terminal (30) is provided with a selection table having the same configuration as the label as described above, or such a selection table is provided to the user terminal (30) from the transmission system (20) and displayed in a GUI manner on the display of the user terminal (30). The user can select any one from this selection table, for example, the psychological correction effect of label 5-12, "sleep effect for poodles." The input unit (24) is configured to include a communication interface that performs communication with the user terminal (30) via a communication network to receive such selection.
[0059] When a selection command is input to the input unit (14), the transmission unit (26) extracts a sound generation tool corresponding to the input psychological correction effect from the storage unit (22) and transmits it to the sound playback device (40) provided for the pet. As described above, since the sound generation tool includes a function module and a parameter input value of the function module, the transmission unit (26) transmits both the function module and the parameter input value to the sound playback device (30). The storage unit (22) is not an essential component of the sound playback device (20), and, for example, the transmission unit (26) may be configured to directly receive the sound generation tool from the memory (16) of the AI system (10) and transmit it to the sound playback device (40). The transmission unit (25) includes a communication interface that performs communication with the sound playback device (40) through a communication network.
[0060] FIG. 3 is a schematic diagram illustrating the configuration of a pet psychological correction sound playback device (40) according to the present invention. The sound playback device (40) of the present invention is a device that plays sound to ultimately provide psychological correction sound to a pet, and may be manufactured in the form of a pendant or accessory worn by the pet, or may be manufactured in the form of a small device installed close to the pet in an environment where the pet resides.
[0061] The sound playback device (40) is configured to include a receiver (42), a storage unit (44), an execution unit (46), and a speaker (48). The receiver (42) receives a sound generating tool transmitted by the transmission system (20). The storage unit (44) stores the sound generating tool received by the receiver (42). The execution unit (46) generates sound data by applying parameter input values included in the sound generating tool to a function module included in the sound generating tool and executing the function module. The speaker (48) plays the sound data generated by the execution unit (46). Accordingly, a sound with a psychological correction effect selected by the user is played, and the pet that listens to it obtains effects such as mental and physical calming. The receiving unit (42) is configured to include a communication interface that performs communication with the transmission system (30) through a communication network, the storage unit (44) is configured to include a memory IC, and the execution unit (46) includes a memory in which program code is recorded to generate sound data by applying parameter input values to a function module stored in the storage unit (44), and a microprocessor that generates sound data by executing the program code of the memory.
[0062] According to the present invention, sounds with a psychological correction effect for pets can be played to pets according to the user's choice. In this process, the playback device that plays the sounds to the pet does not receive the entire sound data, but only the sound generation tool for generating the sound data. Therefore, compared to a method of directly transmitting sound data, the amount of data communication is significantly reduced, enabling rapid data transmission and playback. Furthermore, since it is not a method of pre-transmitting, for example, various types of sounds to the sound playback device, the burden on the storage capacity of the sound playback device is greatly reduced.
[0063] Furthermore, even if a new sound with a psychological correction effect is created, or a new sound that achieves the same psychological correction effect more effectively is created, these new sounds can be played simply by transmitting, for example, additional parameter input values, without the need for processes such as downloading the sounds in advance. Therefore, there is an advantage in that it is convenient to use as the process of data communication for updates is unnecessary every time.
[0064] The present invention can be modified in various ways as follows.
[0065] There are two ways in which the transmission unit (26) transmits the sound generating tool to the sound playback device (40).
[0066] The first method is a method in which, as a selection command is input into the input unit (24), the corresponding sound generating tool, i.e., the function module and its input value, are both extracted from the storage unit (22) and transmitted.
[0067] The second method is a method in which function modules are pre-transmitted to the sound playback device (40) in advance, and when a selection command is input to the input unit (24), the corresponding parameter input value, specifically the parameter input value corresponding to the function module of the sound generation tool corresponding to the input psychological correction effect, is transmitted to the sound playback device (30). In this case, the transmission unit (26) pre-transmits all function modules within the sound generation tool. When a selection command is input, only the corresponding parameter input value is transmitted to the sound playback device (40).
[0068] In the first method, the sound playback device (40) does not require a separate storage unit (44) or requires a storage unit (44) of very small capacity, so the configuration of the sound playback device has the advantage of being very simple.
[0069] In the second method, a storage unit (44) with a capacity to store all function modules in advance is required in the sound playback device (30), but since the transmission system (20) transmits only the parameter input value after the user's selection command input, the capacity to be transmitted is significantly reduced. Generally, since the function module is fixed unless separate update learning and update generation are performed after the learning of the AI module (14) and the generation of the sound generation tool are completed, the burden of data communication volume is further greatly reduced when the second method is adopted.
[0070] As a variation of another aspect, in FIG. 2, the user terminal (30) and the transmission system (20) are shown as separate configurations, but for example, the user terminal (30) itself may be configured as a device including the transmission system (20). In this case, for example, the user terminal (30), which is a smartphone, can access the memory (16) of the AI system (10) through a mobile web network to download and store a sound generation tool.
[0071] As another variation of the aspect, although the sound playback device (40) and the user terminal (30) are illustrated in FIG. 2 as separate configurations, the sound playback device (40) may be configured to receive a selection of psychological correction effects from the user by providing a separate input unit (not shown) that can receive the user's selection of psychological correction effects, thereby allowing the user to input a command for selecting psychological correction effects without the need for a separate user terminal (30), or in parallel with the user terminal (30).
[0072] Meanwhile, in the description of the present invention above, it was described that the label data used to train the AI module (14) is assigned based on the actual test results for each psychological correction effect. However, training the AI module (14) requires a vast number of sound source data, and, for example, assigning one or more labels to each of hundreds of thousands of sound source data requires such a massive amount of work that it is difficult to perform manually.
[0073] In consideration of this, the present invention proposes a method for performing the process of labeling audio data itself using an artificial intelligence algorithm. Specifically, for instance, an AI system with a function to verify a pet's response to sound is constructed by introducing a Neural Network algorithm. Then, audio data is played to the pet, and a video of the pet's behavioral changes is recorded. The video obtained in this way and the corresponding audio data are input into the AI system to label the psychological correction effect obtained by the audio data. By introducing this method, label data for a vast number of audio data can be obtained.
[0074] Meanwhile, the output produced by the AI module (14) in the present invention is not acoustic data that brings about a psychological correction effect, but includes a function module for producing such acoustic data. Generally, since AI algorithms do not generate a function as their output, separate measures are required to make the AI algorithm generate a function as its output. In the above-described embodiment of the present invention, as an example of a specific method for this, a method of generating an arbitrary function by combining existing functions and deriving a result by inputting an arbitrary parameter input value is exemplified. As an alternative method, for example, if the AI module (14) is a module that employs a neural network algorithm, the AI module (14) can be configured to output a function module and its parameter input value as its output by applying a Symbolic Regression algorithm that approximates a certain output with a mathematical formula, Neural Network Compression that compresses the complex structure of the neural network output into a simple module or derives a model that shows similar performance to the output data by mathematically interpreting it, or Piecewise Linear Approximation that linearizes the activation function of the neural network piece by piece to create a mathematical formula for each piece.
[0075] Although the invention has been described with reference to an embodiment illustrated in the drawings, this is merely illustrative, and those skilled in the art will understand that various modifications and equivalent alternative embodiments are possible therefrom. Accordingly, the true technical scope of protection of the invention should be determined by the technical spirit of the appended claims.
Claims
1. An input module that receives learning data comprising multiple sound source data and one or more labels corresponding to the sound source data, wherein the psychological correction effect of a pet according to the playback result of each of the said sound source data is labeled. An AI module comprising a processor that performs supervised learning based on a predetermined artificial intelligence algorithm on the training data received through the input module, generates a sound generation tool for generating a psychological correction sound for the pet in correspondence with the psychological correction effect, and performs the artificial intelligence algorithm on the training data received through the input module; and A memory that provides the sound generation tool to a transmission system configured to store the sound generation tool generated by the AI module and transmit the sound generation tool to a sound playback device; Includes, Each of the above-mentioned sound generation tools includes a function module and a parameter input value of the function module, and The above function module is equipped with program code that generates acoustic data corresponding to the above psychological correction effect, and An AI system for generating psychological correction sounds for pets, characterized in that the above parameter input value is a value generated in correspondence with the above function module and configured so that the above function module generates the above sound data corresponding to the above psychological correction effect.
2. In Paragraph 1, An AI system for generating psychological correction sounds for pets, characterized in that the above parameter input values include a frequency set, which is a set of frequencies for the sound, and a magnitude set, which is a set of magnitudes for the sound.
3. In Paragraph 1, The AI system for generating psychological correction sounds for pets is characterized by the fact that the above-mentioned sound source data is generated by using a plurality of sound sources artificially created by applying one or more of frequency modulation (FM), amplitude modulation (AM), and frequency modulation synthesis (FM Systnesis) to a predetermined number of sound sources, or by performing a predetermined operation on digital data generated from the predetermined number of sound sources and / or the artificially created sound sources in one or more ways of pulse code modulation (PCM) or delta modulation (DM).
4. A transmission system for transmitting a sound generation tool generated by the AI system of claim 1 to a sound playback device, An input unit that receives a selection command for the psychological correction effect requested by the user; and A transmission unit that transmits the sound generating tool corresponding to the selected psychological correction effect to a sound playback device provided for the pet, as the selection command is input into the input unit; A transmission system characterized by including 5. A transmission system for transmitting a sound generation tool generated by the AI system of claim 1 to a sound playback device, An input unit that receives a selection command for the psychological correction effect requested by the user; and A transmission unit that transmits function modules within the sound generation tool in advance to a sound playback device provided for the pet and stores them within the sound playback device, and transmits parameter input values within the sound generation tool corresponding to the selected psychological correction effect to the sound playback device as the selection command is input into the input unit; A transmission system characterized by including 6. A receiver that receives a sound generating tool transmitted by the transmission system of claim 4; A storage unit for storing the received sound generating tool; An execution unit that generates sound data by applying parameter input values included in the sound generation tool to a function module included in the sound generation tool and executing the function module; and A speaker that plays sound data generated by the above execution unit; A pet psychological correction sound playback device characterized by including 7. A receiving unit that receives function module and parameter input values transmitted by the transmission system of claim 5; A storage unit that stores the received function module; An execution unit that stores the parameter input value in the storage unit upon receiving the parameter input value in the receiving unit, and applies the parameter input value to the corresponding function module to execute the function module and generate sound data; and A speaker that plays sound data generated by the above execution unit; A pet psychological correction sound playback device characterized by including