Emotion recognition wearable device based on voiceprint learning and preparation method thereof

Abstract

The application provides a voiceprint learning-based emotion recognition wearable device and a preparation method thereof, and the preparation method comprises the following steps: placing each functional unit on the surface of a flexible piezoelectric sensor assembled by a piezoelectric film, and connecting each part with an electrode wire; spin coating an encapsulating material on the surface of the flexible piezoelectric sensor, and curing in an oven. The flexible piezoelectric sensor comprises a hydrogel layer, which can adhere to the skin to realize the wearable function of the device. The device collects the voice signal generated by the throat, and the voice signal is amplified, denoised, filtered and converted into an analog signal, and then transmitted to the mobile phone end by a Bluetooth module. The mobile phone end pre-processes the voice signal data through a band-pass filter, then performs fast Fourier transform on the acoustic frame to convert the time-domain waveform into a mel-frequency spectrum diagram, and sends the diagram into a convolutional neural network classifier, and then performs maximum pooling in the hidden state; finally, the linear layer and the full connection layer are used to realize emotion recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of patient emotion monitoring, and in particular to a wearable device for emotion recognition based on voiceprint learning and a preparation method thereof. Background Art

[0002] Emotions, as a key indicator of mental health, play a crucial role in daily work and life. For patients with mental illnesses such as anxiety, depression, and affective disorders, recording their emotional state can help monitor and treat mental illnesses, providing an important basis for the patient's recovery.

[0003] With the development of artificial intelligence, analyzing individual speech signals and extracting sound characteristics—voiceprints—to analyze and infer a speaker's emotional state has become an important research direction. By learning voiceprints and building a model that accurately describes and distinguishes the voice characteristics of different individuals, it is possible to identify and infer a speaker's current emotional state, such as joy, anger, or sadness.

[0004] Voiceprint learning requires collecting individual voice signals. Some common wearable devices used to collect human voice signals, such as smart headphones and glasses, are easily interfered with by environmental noise, which affects the quality of the voice signal and reduces the accuracy of emotion recognition. Summary of the Invention

[0005] The purpose of the present invention is to provide an emotion recognition wearable device based on voiceprint learning and a preparation method thereof, which can collect an individual's own voice signal to realize simple, convenient and accurate emotion recognition, so as to solve the problems encountered in the above-mentioned background technology.

[0006] To achieve the above object, the technical solution of the present invention is as follows:

[0007] A wearable device for emotion recognition based on voiceprint learning includes a flexible piezoelectric sensor and a signal processing module, a power module, and a Bluetooth module mounted on the outer surface of the flexible piezoelectric sensor. The power module is connected to the signal processing module and the Bluetooth module to provide power, and the flexible piezoelectric sensor is connected to the signal processing module. The signal processing module, power module, and Bluetooth module are placed on the upper surface of the flexible piezoelectric sensor, and wires are used to connect the modules to each other.

[0008] The power module connects to the Bluetooth module and sends data to the mobile device, which then recognizes and displays the user's emotions. The power module uses a rechargeable button battery or polymer lithium battery and is equipped with a Type-C charging port.

[0009] The flexible piezoelectric sensor collects the voice signal generated by throat vibration and transmits it to the signal processing module, which performs preliminary processing on the voice signal and transmits the voice data to the mobile phone through the Bluetooth module.

[0010] During manufacturing, the emotion recognition wearable device further includes a packaging layer, through which the signal processing module, the power module, and the Bluetooth module are integrated into a structure on the surface of the flexible piezoelectric sensor.

[0011] The flexible piezoelectric sensor includes two upper and lower electrode layers and a piezoelectric layer in the middle. Specifically, the flexible piezoelectric sensor includes a piezoelectric film, a single-sided conductive electrode sheet and a mechanical interlocking electrode. The piezoelectric film is located between the electrode sheet and the mechanical interlocking electrode, and the conductive side of the electrode sheet and the mechanical interlocking electrode is attached to the piezoelectric film.

[0012] Method for manufacturing mechanically interlocked electrodes: Fibers are prepared by spinning, using at least one of polyurethane, polyester, polypropylene, and polyamide as the fiber material. Furthermore, an electrode layer is formed on the upper surface of the fiber by evaporation, spin coating, or other methods. A hydrogel precursor is cast on the other side of the fiber and thermally cured to obtain a hydrogel electrode with a mechanically interlocking structure. The hydrogel and fiber form a mechanically interlocking region, and the hydrogel layer has a high degree of adhesion, allowing it to adhere to the skin, thereby achieving wearability of the device.

[0013] The signal processing module includes a preamplifier circuit, a 50Hz notch circuit, a bandpass filter circuit, and an A / D conversion circuit; the preamplifier circuit is used to amplify the voice signal; the 50Hz notch circuit is used to filter out power frequency interference; the bandpass filter circuit is used to limit the frequency range of the voice signal; the A / D conversion circuit is used to perform analog-to-digital conversion on the voice signal; the power module is used to supply energy to all circuits in the signal processing module; and the Bluetooth module is used to send the collected voice data to the mobile phone.

[0014] A method for preparing an emotion recognition wearable device based on voiceprint learning comprises the following steps:

[0015] A piezoelectric film is prepared by using a piezoelectric material, and is assembled with an electrode sheet and a mechanically interlocked electrode into a flexible piezoelectric sensor, wherein the flexible piezoelectric sensor is used to convert the vibration of the throat when the user speaks into an electrical signal;

[0016] Placing various functional units on the surface of the flexible piezoelectric sensor, including: a signal processing module, a power module, and a Bluetooth module; arranging electrode wires on the flexible piezoelectric sensor according to corresponding circuits to connect the various functional units;

[0017] Insulating and biocompatible organic materials are spin-coated on the surface of the flexible piezoelectric sensor to encapsulate various functional units and electrode wires.

[0018] Furthermore, the piezoelectric material uses at least one of polyvinylidene fluoride, fluorinated polymer, polylactic acid, polyacrylonitrile, and cellulose, adds a certain concentration of filler, dissolves it in a solvent according to a certain proportion, and obtains a suspension with uniformly dispersed particles through magnetic stirring, and the piezoelectric film is obtained by electrospinning or spin coating.

[0019] Among them, the manufacturing method of piezoelectric film is: the piezoelectric material is prepared into a solution of a certain concentration, and a uniform piezoelectric film with a thickness of about 100 microns is prepared by spin coating, spraying, stretching, calendering, spinning and polarization.

[0020] Furthermore, to enhance the piezoelectric performance of the piezoelectric sensor, a simple and effective method is to add a certain concentration of additives or fillers to the piezoelectric material. The fillers are used to enhance the piezoelectric conversion efficiency of the flexible piezoelectric sensor and are selected from one or more of metal nanoparticles, metal oxides, piezoelectric ceramics, and nano-zinc oxide particles.

[0021] The piezoelectric sensor is assembled by sandwiching the prepared piezoelectric film between an electrode sheet and a mechanical interlocking electrode. The electrode sheet can be made of conductive tape or conductive copper. The electrode sheet and the mechanical interlocking electrode are single-sided conductive, meaning the side in contact with the piezoelectric film is conductive and the other side is non-conductive.

[0022] An emotion recognition wearable device based on voiceprint learning. The algorithm for implementing emotion recognition includes the following steps:

[0023] Multiple speech signals were recorded using the flexible piezoelectric sensor. Each signal was manually classified and labeled with four emotion labels: anger, calm, happiness, and sadness. During training, the training set to test set ratio was set to 8:2.

[0024] Before training the model, speech signals expressing four different emotions were preprocessed using a bandpass filter to remove extremely low-frequency or high-frequency signals. The filtered signals were then segmented into small frames using a Hanning window. The frames were then subjected to a fast Fourier transform and an 80-band Mel filter to convert the time-domain waveform into a Mel-spectrogram.

[0025] The size of the Hanning window is 1024 milliseconds and the hop length is 320 milliseconds.

[0026] The 80-channel Mel-spectrogram is fed into a convolutional neural network (CNN) classifier and then passed through a max pooling layer in the hidden state to reduce computational cost and variance. The classifier consists of 5 layers of convolutional neural networks (CNNs), with the convolution kernel size of each layer being [1, 1, 3, 3, 5].

[0027] Finally, the linear layer and the fully connected layer are used to project emotions as the final classification result to achieve emotion recognition.

[0028] Compared to existing technologies, the present invention offers the following advantages: This wearable emotion recognition device uses a flexible piezoelectric sensor to collect throat voice signals, transmits the data to a mobile phone through a series of circuits, and then uses a trained voiceprint learning method to perform emotion recognition. Therefore, compared to existing technologies, this device can collect an individual's own voice signals to achieve simple, convenient, and accurate emotion recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The disclosure of the present invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. In the accompanying drawings, the same reference numerals are used to refer to the same components. Among them:

[0030] Figure 1 This is a system block diagram of a wearable device for emotion recognition based on voiceprint learning;

[0031] Figure 2 Schematic diagram of the working process of wearable device for emotion recognition based on voiceprint learning;

[0032] Figure 3 This is a schematic diagram of the structure of a wearable device for emotion recognition based on voiceprint learning;

[0033] Figure 4 Schematic diagram of the integrated packaging of a wearable device for emotion recognition based on voiceprint learning;

[0034] Figure 5 This is a schematic structural diagram of a flexible piezoelectric sensor according to an embodiment of the present invention;

[0035] Figure 6 Schematic diagram of the machine learning model for wearable devices that recognize emotions based on voiceprint learning;

[0036] Figure 7 This is a Fourier infrared spectrum of the piezoelectric film introduced in the embodiment of the present invention;

[0037] Figure 8 This is an example of a speech signal image collected by the flexible piezoelectric sensor introduced in an embodiment of the present invention;

[0038] Figure 9 This is an example of a Mel spectrum graph collected by the flexible piezoelectric sensor introduced in an embodiment of the present invention.

[0039] Explanations in the figure: 1. A wearable device for emotion recognition based on voiceprint learning; 2. Flexible piezoelectric sensor; 21-piezoelectric film; 22-electrode sheet; 23-mechanical interlocking electrode; 231-electrode layer; 232-mechanical interlocking area; 233-hydrogel layer; 3. Signal processing module; 4. Power module; 5. Bluetooth module; 6. Type-C charging port; 7. Packaging layer. DETAILED DESCRIPTION

[0040] In order to make the technical means, creative features, objectives and effects of the present invention easier to understand, the present invention will now be further described in detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the relevant components of the present invention.

[0041] According to the technical solution of the present invention, without changing the essential spirit of the present invention, a person skilled in the art may propose a variety of interchangeable structural modes and implementation modes. Therefore, the following specific embodiments and drawings are merely illustrative of the technical solution of the present invention and should not be regarded as the entire invention or as a limitation or restriction of the technical solution of the present invention.

[0042] The technical solution of the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0043] Example 1, please refer to Figure 1 and Figure 2 A wearable device for emotion recognition based on voiceprint learning includes a flexible piezoelectric sensor 2 and a signal processing module 3, a power module 4, and a Bluetooth module 5, which are arranged on the outer surface of the flexible piezoelectric sensor 2. The power module 4 is connected to the signal processing module 3 and the Bluetooth module 5 to supply power to them. The flexible piezoelectric sensor 2 is connected to the signal processing module 3. The signal processing module 3, power module 4, and Bluetooth module 5 are placed on the upper surface of the flexible piezoelectric sensor 2 and connected to each other with wires.

[0044] The power module 4 connects to the Bluetooth module 5 and sends data to the mobile device, which then recognizes and displays the user's emotions. The power module uses a rechargeable button battery or polymer lithium battery and is equipped with a Type-C charging port 6.

[0045] The flexible piezoelectric sensor collects the voice signal generated by throat vibration and transmits it to the signal processing module, which performs preliminary processing on the voice signal and transmits the voice data to the mobile phone through the Bluetooth module.

[0046] Example 2, please refer to Figures 3 to 5This embodiment introduces a method for preparing a wearable device for emotion recognition based on voiceprint learning, which mainly includes a flexible piezoelectric sensor, a functional unit, and a packaging layer arranged in sequence. The steps of the preparation method are as follows:

[0047] Step 1: Prepare piezoelectric film 21 by electrospinning. Piezoelectric film 21 can be a piezoelectric fiber film. A certain amount of nano-zinc oxide particles is placed in a mixed solvent of acetone and dimethylformamide. Ultrasonication is performed for two hours to uniformly disperse the nano-zinc oxide particles in the mixed solvent. Polyvinylidene fluoride powder is then added and magnetic stirring is performed at 50°C for 12 hours to obtain a spinning solution.

[0048] The spinning solution was placed in a 10 ml syringe, and a 21G needle was connected to the syringe needle. The syringe was then mounted on a syringe pump to provide a constant flow rate of 0.5 ml per hour. Electrospinning was performed at a voltage of 15 kV, with a fixed distance of 15 cm between the spinning head and the collector. A rotating drum collected the piezoelectric film 21 at a speed of 100 revolutions per minute. The Fourier transform infrared spectrum of the piezoelectric film 21 is shown in FIG. Figure 7 As shown, it contains the crystalline phase of polyvinylidene fluoride.

[0049] Mechanically interlocked electrodes 23 are prepared by spinning fibers using at least one of polyurethane, polyester, polypropylene, and polyamide. For example, a certain amount of polyurethane is dissolved in a mixed solvent of acetone and dimethylformamide and stirred at room temperature for 24 hours. The mixed solution is then spun into polyurethane fibers at an electrospinning voltage of 20 kV.

[0050] A layer of silver electrode is prepared on the surface of the polyurethane fiber by evaporation. Then a hydrogel precursor is cast on the other side of the fiber and cured at 50°C for 2 hours to finally obtain a mechanically interlocked electrode. The hydrogel precursor consists of a certain concentration of acrylamide, sodium alginate, calcium carbonate nanopowder and tetramethylethylenediamine. An electrode layer 231 is prepared on the upper surface of the fiber by evaporation, spin coating and other methods. A hydrogel precursor is cast on the other side of the fiber and thermally cured to obtain a hydrogel electrode with a mechanically interlocking structure. The hydrogel and the fiber form a mechanically interlocking region 232, and the hydrogel layer 233 has a high degree of adhesion and can be attached to the skin, thereby achieving the wearability of the device.

[0051] Step 2: Prepare flexible piezoelectric sensors, Figure 5 Schematic diagram of the structure of the flexible piezoelectric sensor 2. The prepared piezoelectric fiber film is assembled into a sandwich structure with a single-sided conductive sheet and mechanical interlocking electrodes. The conductive sheet 22 can be a copper sheet with its conductive side attached to the piezoelectric fiber film, i.e., the piezoelectric film 21.

[0052] Step 3: If Figure 4As shown, the functional units are adhered to the preset positions on the surface of the flexible piezoelectric sensor 2, and the various units are connected according to the corresponding circuit using conductive silver paste.

[0053] The functional unit includes a signal processing module, a power module, and a Bluetooth module. The signal processing module 3 includes a preamplifier circuit, a 50Hz notch circuit, a bandpass filter circuit, and an A / D converter circuit. The signal processing module 3 further processes the voice signals collected by the flexible piezoelectric sensor and transmits these signals to the mobile phone via the Bluetooth module 5. Voiceprint learning is performed on the voice data, ultimately enabling emotion recognition.

[0054] The entire device is powered by the power module 4 and is equipped with the Type-C charging port 6 charging interface.

[0055] It should be noted that the upper and lower electrodes of the flexible piezoelectric sensor 2 need to be connected to the signal processing module so that the voice signal collected by the sensor enters the signal processing module.

[0056] Step 4: Coat the surface of the device prepared above with the encapsulation layer 7 to protect each unit and circuit.

[0057] One specific method involves mixing polydimethylsiloxane (PDMS) as the encapsulation material with a crosslinker in a 10:1 ratio. Stirring the mixture with a glass rod for 5 minutes ensures a uniform mix. A centrifuge is then used to remove air bubbles from the solution for subsequent use. The prepared PDMS is then spin-coated onto the device surface and cured in a 60°C oven for 5 hours.

[0058] In this embodiment, a flexible piezoelectric sensor is prepared, and various functional units are integrated on the surface of the piezoelectric sensor, and further packaged to obtain the wearable device 1. Figure 2 The device shown can convert vocal cord vibration into a voltage signal, amplify, denoise, filter and convert the signal into analog-to-digital through the signal processing module, and then transmit it to the mobile phone through the Bluetooth module for emotion recognition.

[0059] Example 3: This example introduces another method for preparing an emotion recognition wearable device based on voiceprint learning, which also includes a flexible piezoelectric sensor, a functional unit, and a packaging layer arranged in sequence. The specific steps are as follows:

[0060] Step 1: Ultrasonic dispersion of the filler in the solvent. Add the piezoelectric material and stir until uniform. The filler can be one or more of metal nanoparticles, metal oxides, and piezoelectric ceramics. The piezoelectric material can be at least one of polyvinylidene fluoride, a fluorinated polymer, polylactic acid, polyacrylonitrile, and cellulose.

[0061] The mixed solution was spin-coated on a glass substrate and heat-treated in a drying oven at 60° C. for 2 hours to obtain a piezoelectric film.

[0062] It should be noted that the piezoelectric film needs to be polarized at high voltage to have piezoelectric properties.

[0063] The second step is to deposit electrodes on both sides of the polarized piezoelectric film. Electrode materials can be made of conductive metals such as gold, silver, and copper. Alternatively, a conductive polymer can be sprayed on both sides of the piezoelectric film to create a flexible piezoelectric sensor. A polyacrylamide hydrogel precursor is then cast onto one side of the flexible piezoelectric sensor and cured at 50°C for two hours to create the polyacrylamide hydrogel, allowing the entire device to adhere to the skin surface.

[0064] Step 3: Add a 10-micron-thick flexible substrate to the flexible piezoelectric sensor 2 to house the various functional units. The substrate material can be silicone, polyethylene terephthalate, or polyurethane. A conductive circuit is pre-printed on the flexible substrate using methods such as inkjet printing or thermal evaporation, and the functional units are then connected to the circuit at predetermined locations.

[0065] Step 4: Encapsulate the surface of the device to fix the functional units into one and improve the mechanical strength. On the other hand, it prevents the device from direct contact with the outside world and protects the functional units.

[0066] Example 4: This example introduces an emotion recognition wearable device based on voiceprint learning, which mainly includes a flexible piezoelectric sensor, a functional unit and an emotion recognition algorithm.

[0067] The flexible piezoelectric sensor is obtained by the preparation method described in Example 1 or Example 2. The flexible piezoelectric sensor can convert the mechanical vibration of the vocal cords through the skin into a voltage signal. When a person speaks, the voltage signal is the voice signal.

[0068] The functional units include a signal processing module, a power module, and a Bluetooth module, which are integrated into a flexible piezoelectric sensor surface through an encapsulation layer. Each functional unit is connected by electrode wires.

[0069] The flexible piezoelectric sensor collects individual voice signals in real time and transmits the signals to the signal processing module, amplifies, denoises, filters and performs analog-to-digital conversion on the signals, and then transmits the data to the mobile phone through the Bluetooth module.

[0070] The principle diagram of the machine learning model for emotion recognition wearable devices is as follows Figure 6 As shown, the emotion recognition algorithm mainly includes the following steps:

[0071] The above equipment is used to collect speech signals of different emotions in advance, such as Figure 8The speech signals shown in the figure were manually classified and labeled with four emotion labels: angry, calm, happy, and sad. The ratio of the training set to the test set was set to 8:2. The speech signal data expressing the four different emotions was preprocessed through a bandpass filter and split into small frames using a Hanning window. The sound frames were then subjected to fast Fourier transform and 80-band Mel filtering to convert the time domain waveform into a Mel spectrogram, as shown in the figure. Figure 9 The 80-channel Mel-spectrogram is fed into a convolutional neural network (CNN) classifier for emotion recognition. The classifier consists of a five-layer CNN, with kernel sizes of [1, 1, 3, 3, 5] in each layer. A max-pooling layer is then used to reduce the number of features in the hidden state. Finally, linear and fully connected layers are used to project emotion into the final classification result.

[0072] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. 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 wearable device for emotion recognition based on voiceprint learning, characterized by: The flexible piezoelectric sensor (2) comprises a signal processing module (3), a power module (4), and a Bluetooth module (5) arranged on the outer surface of the flexible piezoelectric sensor (2), wherein the power module (4) is connected to the signal processing module (3) and the Bluetooth module (5) to supply power to them; the flexible piezoelectric sensor (2) is connected to the signal processing module (3), and the power module (4) is connected to the Bluetooth module (5), and data is sent to a mobile terminal, and the user's emotions are recognized and displayed through the mobile terminal; The flexible piezoelectric sensor (2) comprises a piezoelectric film (21), a single-sided conductive electrode sheet (22), and a mechanical interlocking electrode (23); the piezoelectric film (21) is located between the electrode sheet (22) and the mechanical interlocking electrode (23), and the conductive side of the electrode sheet (22) is in contact with the piezoelectric film (21); The mechanical interlocking electrode (23) includes an electrode layer (231), a mechanical interlocking region (232) and a hydrogel layer (233), wherein the electrode layer (231) is attached to the piezoelectric film (21), and the hydrogel layer (233) has mechanical softness and high skin adhesion, and the mechanical interlocking region (232) is distributed throughout the hydrogel layer (233).

2. The wearable device for emotion recognition based on voiceprint learning according to claim 1, characterized in that: It also includes a packaging layer (7), through which the signal processing module (3), the power module (4), and the Bluetooth module (5) are integrated into a structure on the surface of the flexible piezoelectric sensor (2).

3. The wearable device for emotion recognition based on voiceprint learning according to claim 1, characterized in that: The signal processing module includes a preamplifier circuit, a 50Hz notch circuit, a bandpass filter circuit, and an A / D conversion circuit; the preamplifier circuit is used to amplify the voice signal; the 50Hz notch circuit is used to filter out power frequency interference; the bandpass filter circuit is used to limit the frequency range of the voice signal; the A / D conversion circuit is used to perform analog-to-digital conversion on the voice signal; the power module is used to supply energy to all circuits in the signal processing module; and the Bluetooth module is used to send the collected voice data to the mobile phone.

4. The method for preparing a wearable device for emotion recognition based on voiceprint learning according to any one of claims 1 to 3, characterized in that: The following steps are involved: A piezoelectric film is prepared by using a piezoelectric material and assembled with an electrode sheet to form a flexible piezoelectric sensor, wherein the flexible piezoelectric sensor is used to convert the vibration of the throat when the user speaks into an electrical signal; Placing various functional units on the surface of the flexible piezoelectric sensor, including: a signal processing module, a power module, and a Bluetooth module; arranging electrode wires on the flexible piezoelectric sensor according to corresponding circuits to connect the various functional units; Insulating and biocompatible organic materials are spin-coated on the surface of the flexible piezoelectric sensor to encapsulate various functional units and electrode wires.

5. The method for preparing a wearable device for emotion recognition based on voiceprint learning according to claim 4, characterized in that: The piezoelectric material is made of at least one of polyvinylidene fluoride, fluorinated polymer, polylactic acid, polyacrylonitrile, and cellulose, a certain concentration of filler is added, dissolved in a solvent according to a certain proportion, and a suspension with uniformly dispersed particles is obtained by magnetic stirring. The piezoelectric film is obtained by electrospinning or spin coating.

6. The method for preparing a wearable device for emotion recognition based on voiceprint learning according to claim 5, characterized in that: The filler is one or more of metal nanoparticles, metal oxides, piezoelectric ceramics, and nano zinc oxide particles; the filler is used to improve the piezoelectric conversion efficiency of the flexible piezoelectric sensor.

7. The method for preparing a wearable device for emotion recognition based on voiceprint learning according to claim 4, characterized in that: The algorithm for implementing emotion recognition includes the following steps: Multiple speech signals were recorded using the flexible piezoelectric sensor, and each signal was manually classified and labeled with four emotion labels: angry, calm, happy, and sad; The speech signal data is preprocessed through a bandpass filter to remove extremely low or high frequency signals, and then divided into small frames through a Hanning window. The sound frames are then subjected to fast Fourier transform and 80-band Mel filtering to convert the time domain waveform into a Mel spectrogram. The 80-channel Mel-spectrogram is fed into the convolutional neural network classifier and then passed through a max pooling layer in the hidden state to reduce computational cost and variance. Finally, the linear layer and the fully connected layer are used to project emotions as the final classification result to achieve emotion recognition.

8. The method for preparing a wearable device for emotion recognition based on voiceprint learning according to claim 7, characterized in that: The size of the Hanning window is 1024 milliseconds and the hop length is 320 milliseconds; the classifier consists of a 5-layer convolutional neural network, and the convolution kernel size of each layer is [1, 1, 3, 3, 5].

Images