Biological auditory inspired sound pulse coding and identification method and system

Abstract

The invention discloses a sound coding and recognition method and system for biological auditory inspiration. The method comprises the following steps: firstly, constructing a bionic auditory sensor chip constructed based on an organism auditory receptor; then, a bionic auditory coding model is constructed according to an electrophysiological response mode of the organism auditory receptor under the stimulation of sounds with different frequencies and intensities; and finally, coding the sound to be recognized into pulse information by using the coding model, processing the coded pulse signal through a pulse neural network, and finally realizing the recognition of the sound. According to the invention, the auditory feeling process of mammals and the information processing mode of the mammals are simulated, and a pulse coding and sound recognition algorithm with biointerpretability and high efficiency is constructed by fully utilizing the coding of sound information by organisms. Compared with an existing voice recognition method, the voice recognition accuracy is higher, and meanwhile the requirement for computing resources is low.

Description

Technical Field

[0001] The present invention belongs to the cross - field of bio - inspired computing and bionic sensors, and particularly relates to a method and system for sound pulse encoding and recognition inspired by biological audition. Background Art

[0002] Artificial neural networks (ANNs) have been widely used in sound recognition tasks. However, the energy requirement of artificial neural networks is huge. As the third - generation artificial neural network, spiking neural networks (SNNs) process information in the form of spikes. Information in the form of spikes is represented as discrete events at time points. SNNs transform continuous information into discrete spike events through spike encoding and perform calculations only when spikes occur, thus greatly reducing energy consumption. SNNs can not only significantly reduce the system power consumption but also be closer to the information - processing mode of the biological nervous system. How to reasonably encode sound information into spike signals is a prerequisite for SNNs to have good performance.

[0003] Currently, the main spike - encoding methods for sound information are rate encoding and time encoding. Rate encoding shows that the firing frequency of spikes is proportional to the intensity of the sound signal. Time encoding shows that the delay time of spike firing reflects the energy characteristics of the sound, and the stronger the energy, the earlier the spike - firing moment. However, these encoding methods have a large gap from real biological audition and have great deficiencies in biological interpretation, making it difficult to truly reflect the complex encoding mechanism in the biological auditory process. Therefore, how to design a reasonable spike - encoding method that is both low - power - consuming and has biological interpretability has become a key problem restricting the further development of SNNs in the field of sound recognition. In the biological auditory process, sound is conducted through the outer ear and middle ear to the inner ear. Especially in the organ of Corti in the cochlea, the organ of Corti is the primary auditory receptor, which is composed of hair cells, supporting cells, and spiral ganglion cells. Among them, hair cells are the main auditory receptors. Sound stimulation will cause the cilia on the hair cells to undergo mechanical displacement, thereby triggering the opening or closing of mechanically sensitive ion channels on the cell membrane, causing changes in the membrane potential and triggering the generation of action potentials, realizing the conversion of sound signals into nerve signals. This process not only involves the encoding of frequency and intensity but also integrates information in the time and space dimensions, thus achieving efficient and fine sound processing. So far, no complete system has been able to directly extract inspiration from biological auditory responses to construct a complete and reasonable spike - encoding method. Summary of the Invention

[0004] The purpose of the present invention is to propose a method and system for sound pulse encoding and recognition inspired by biological audition in view of the deficiencies of the prior art.

[0005] The purpose of the present invention is achieved by the following technical solutions: A method for sound pulse encoding and recognition inspired by biological audition includes the following steps: (1) A biological auditory receptor is used as a sensitive element, and a bionic auditory sensing chip is obtained by configuring a multi-electrode array and an optimized chamber structure, which is used to record electrophysiological responses under sound stimulation; (2) The bionic auditory sensing chip is used to detect the response patterns under pure tone stimuli of different frequencies and different intensities; (3) An optimization algorithm is used to adjust the parameters of the neuron model. Based on the neuron model, the output of the spike firing patterns generated at different frequencies and intensities is fitted with the response data recorded by the bionic auditory sensing chip to construct a bionic auditory coding algorithm. By preprocessing and feature extracting the sound signal, the obtained sound frequency and intensity information are used as the input of the bionic auditory coding algorithm, and the encoded spike output is obtained, realizing the encoding of continuous sound signals into discrete spike signals; (4) A sound recognition network based on a spiking neural network is constructed, and the spike signals obtained in step (3) are input into the network to realize the recognition of sounds.

[0006] Further, the biological auditory receptor is cells or tissues extracted from the biological auditory system, including any one of the organ of Corti, cochlea, inner ear organoids, hair cells or spiral ganglion cells. The biological auditory receptor is cultured on the surface of the multi-electrode array, and the biological auditory receptor is attached to the electrode surface.

[0007] Further, the bionic auditory sensing chip includes a biological auditory receptor, a multi-electrode array and an optimized-sized chamber, where the optimal size of the chamber is determined by acoustic simulation or experimental feedback to provide the liquid ion environment and ideal acoustic conditions required by the biological auditory receptor; the biological auditory receptor can generate extracellular action potentials under sound stimulation, and the corresponding potential signals are collected through the multi-electrode array to reflect the response pattern of the biological auditory receptor to acoustic stimuli.

[0008] Further, the response pattern is one or more of the firing rate, inter-spike interval and synchrony.

[0009] Further, the sound stimulation is a pure tone stimulation, and its frequencies include 100 Hz, 500 Hz, 1000 Hz, 2000 Hz, 3000 Hz and 4000 Hz, and the intensities include 60 dB, 70 dB, 80 dB and 90 dB.

[0010] Further, in step (2), pure tone stimuli of different frequencies with the same intensity are used to test the response of the bionic auditory sensing chip to the sound frequency, and pure tone stimuli of different intensities with the same frequency are used to test the response of the bionic auditory sensor chip to the sound intensity.

[0011] Further, in step (3), preprocessing is performed on the continuous sound signal, including voice activity detection, pre-emphasis, and constant-Q transform operations, to extract the frequency, intensity, and spectral features of the sound.

[0012] Further, the bionic auditory coding algorithm is based on the Izhikevich neuron model, and the parameters are adjusted by an annealing algorithm, a genetic algorithm, an ant colony algorithm, or a grid search optimization algorithm to minimize the root mean square error between the spike train pattern output by the model and the response data recorded by the bionic auditory sensing chip.

[0013] Further, the sound recognition network adopts a three-layer structure of an input layer, an intermediate layer, and an output layer based on a spiking neural network, where the spiking neuron is a LIF neuron, and the gradient replacement method is used to achieve backpropagation during the training process.

[0014] On the other hand, the present invention also provides a bionic auditory-inspired sound pulse coding and recognition system, which includes: A bionic auditory sensing chip construction module, which is used to use a bionic auditory receptor as a sensitive element, and is configured with a multi-electrode array and an optimized chamber to construct a bionic auditory sensing chip; A bionic auditory response module, which is used to use a variety of sound stimuli to test the response patterns of the bionic auditory sensing chip to pure tones of different frequencies and different intensities; A sound preprocessing and feature extraction module, which is used to preprocess and extract features from the sound signal to obtain the sound frequency and intensity information; A bionic auditory coding module, which is used to fit the spike train pattern output at different frequencies and intensities based on the neuron model with the response data recorded by the bionic auditory sensing chip to construct a bionic auditory coding algorithm; input the sound frequency and intensity information, and encode the continuous sound signal into a discrete pulse signal; A sound recognition module based on a spiking neural network, which is used to process the encoded pulse signal to achieve accurate sound recognition.

[0015] Advantages of the present invention: (1) The method of the present invention simulates the perception process of the biological auditory system and more reflects the biological response to sound; (2) At the same time, the present invention uses a bionic auditory receptor as a sensitive element for auditory perception to construct a bionic auditory sensing chip, so that the auditory response after the biological perception of sound stimuli can be directly obtained; (3) The bionic auditory coding algorithm constructed by the present invention encodes the sound information into pulse information by directly simulating the sound response pattern obtained from the bionic auditory sensing chip, which is both biologically reasonable and computationally feasible; (4) Compared with traditional voice recognition methods, the present invention encodes voice signals into discrete pulse signals and constructs a pulse neural network to perform voice recognition on the encoded pulse signals, which is more in line with the auditory perception process of organisms. (5) The present invention proposes an entire bionic auditory perception system from auditory sensation, conversion, to recognition, including a bionic auditory sensing chip, bionic auditory encoding, and a pulse neural network, and achieves better recognition results in voice recognition tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following-described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 It is a schematic diagram of the constructed bionic auditory sensing chip.

[0018] Figure 2 It is a schematic diagram of the electrical signals recorded by the bionic auditory sensing chip under different sound stimuli.

[0019] Figure 3 It is a schematic diagram comparing the output of the bionic auditory encoding algorithm with the response pattern of the biological auditory receptor.

[0020] Figure 4 It is a schematic diagram of the encoding results and recognition results of the bionic auditory encoding algorithm in different datasets. DETAILED DESCRIPTION OF THE INVENTION

[0021] The following further elaborates on the technical solutions of the present invention with reference to the drawings. It should be noted that the detailed description is only a detailed explanation of the present invention and should not be regarded as a limitation of the present invention.

[0022] The biological auditory sensitive element used in the bionic auditory sensing chip of the present invention is the organ of Corti, but it is not limited to the organ of Corti. It can also be other tissues or cells with auditory functions such as the cochlea, inner ear organoids, hair cells, and spiral ganglion cells.

[0023] The bionic auditory sensing chip used in the present invention for electrophysiological signal detection is a planar multi-electrode array chip, but it is not limited to a planar multi-electrode array chip. It can also be a patch clamp, a three-dimensional multi-electrode array chip, a flexible multi-electrode array chip, and other devices for detecting electrophysiology.

[0024] The optimized chamber used in the present invention is an acrylic chamber optimized by acoustic simulation. However, the optimization method is not limited to acoustic simulation and can also be optimization by experimental test feedback, multi-physics field coupling simulation, etc. The optimized parameters are not limited to dimensions and can also be parameters such as the shape of the chamber and the internal structure of the chamber. The material of the chamber is not limited to acrylic and can also be materials such as PDMS and PCL.

[0025] The sound stimulator used in the present invention is a loudspeaker. However, it is not limited to a loudspeaker and can also be devices and apparatuses that provide acoustic stimulation such as piezoelectric ceramics and MEMS acoustic transducers.

[0026] The bionic auditory coding algorithm constructed in the present invention is constructed by minimizing the root mean square error between the output optimized by the annealing algorithm and the electrophysiological signal (spike) response pattern induced by the sound recorded by the bionic auditory sensing chip. However, the optimization algorithm is not limited to the annealing algorithm and can also be optimization algorithms such as genetic algorithm, ant colony algorithm, and grid search. The optimization target, that is, the response pattern, is not limited to the firing rate of spikes and can also be response patterns such as interspike interval and synchrony.

[0027] The specific embodiments of the present invention are as follows: The bionic auditory sensor chip designed in the present invention is as Figure 1 shown, including a biological auditory receptor as a sensitive element, a multi-electrode array sensor for recording the electrophysiological response of the biological auditory receptor under sound stimulation, and an optimized chamber for providing the ionic environment and acoustic stimulation environment required by the biological auditory receptor; The bionic auditory sensing chip is composed of a biological auditory receptor, a multi-electrode array sensor, and a chamber. Among them, the chamber is connected to the multi-electrode array sensor through epoxy resin. The biological auditory receptor is cultured on the surface of the multi-electrode array, and the biological auditory receptor is attached to the electrode surface; The biological auditory receptor is used as an auditory sensitive element to sense sound stimulation and generate corresponding spike signals. In this embodiment, the organ of Corti tissue is selected as the auditory sensitive element. The organ of Corti tissue contains various auditory receptor cells, including hair cells and spiral ganglion cells, which can sense sound stimulation and generate responses; The multi-electrode array sensor is used to detect the electrophysiological signals of the biological auditory receptor induced by sound. It has 64 channels and is connected to a printed circuit board through gold wires. The detected signals are transmitted to a computer through a signal collector and an amplifier; The optimized chamber is used to provide the ionic environment and acoustic environment required by the biological auditory receptor. Through the optimization method, the bionic auditory sensing chip is in the sound detection range. The interpolation method is used to map the sound frequency response pattern obtained by the bionic auditory sensing chip to 0 - 4000 Hz, and the response pattern to the sound intensity is mapped to 0 - 100 to meet the construction requirements of the bionic auditory coding algorithm; when the sound frequency is in the range of 0 - 4000 Hz and the intensity is in the range of 60 - 90 dB, the energy attenuation is the smallest, ensuring that the bionic auditory sensing chip can record stable electrophysiological responses within this range of sound frequency and intensity stimuli. The specific simulated acoustic environment is as follows: A point source is set 5 cm away from the chamber to simulate the sound wave input of the speaker; air and a perfectly matched layer are set outside the chamber to prevent reflection; an ionic liquid layer with a thickness of 300 um is set at the bottom to simulate the ionic environment required by the auditory sensor, and a cube with a size of 1 mm × 1 mm × 0.2 mm in the middle represents the biological auditory sensor. The simulation process includes the generation of sound waves by the speaker, passing through the air, propagating to the solid chamber, and then propagating to the liquid. The acoustic wave transfer equations in air and liquid are as follows:

[0028] where p is the sound pressure, t is the time, c is the speed of sound in air in air and the speed of sound in liquid in liquid, is the Laplace operator, representing the characteristics of wave diffusion and propagation in space. The elastic wave equation for sound wave propagation in solids is:

[0029] where, is the fluid density, is the density of the solid, is the fluid sound speed, is the solid sound speed.

[0030] The biological auditory receptor is cultured on the surface of the multi - electrode array sensor, and the coupling between the tissue and the electrode is improved by modifying polylysine on the electrode surface.

[0031] Using the above bionic auditory sensing chip to detect the response patterns under different sound stimuli and constructing a bionic auditory coding algorithm based on the response patterns, the specific detection and construction process includes the following steps: (1) Detection of auditory response: 150 μL of artificial lymph fluid was added to the chamber before each test. Sound stimuli with different spectra and intensities were given through a speaker. The sound frequency range was 100, 500, 1000, 2000, 3000, 4000 Hz, and the sound intensity range was pure tone stimuli of 60, 70, 80, 90 dB. The stimulation time was 30 s, with an interval of 30 s, repeated 3 times, and the electrical signals of the bionic auditory sensing chip were synchronously collected. All experiments were carried out at room temperature of 25 degrees.

[0032] (2) Analysis of auditory response pattern: The signals recorded by the above bionic auditory sensing chip under different sound stimuli are as Figure 2 shown. First, the original signals were filtered. A Butterworth second-order high-pass filter was used to extract signals above 250 Hz. Further, spike signals were extracted from the filtered signals. Effective spike signals were extracted by using a threshold greater than 3 times the standard deviation, and the firing rate, interspike interval, and synchrony of the spike signals were analyzed as the response pattern under this sound stimulus.

[0033] (3) Construction of bionic auditory coding algorithm: The firing rate of the spike signals under the above different sound stimuli serves as the biological basis for constructing the bionic auditory coding algorithm. The Izhikevich model was used as the skeleton model of the bionic auditory coding algorithm, and its formula is as follows:

[0034]

[0035] where I represents the sound stimulus, and a, b, c, d represent the parameters of the model. When the membrane voltage exceeds 30, a spike signal is obtained. By counting the pulse signals, the response pattern of the bionic auditory coding algorithm is Model_response. The response pattern of the spike of the above-obtained bionic auditory sensing chip under different sound stimuli is OC_response. The model parameters a, b, c, d were optimized by an optimization algorithm to minimize the root mean square error between Model_response and OC_response. Among them, the response pattern can be one or more of the firing rate, interspike interval, and synchrony, and the bionic auditory coding algorithm was obtained. Among them, the value of I is mapped to 0 - 100. The optimized bionic auditory coding algorithm and the response of the bionic auditory sensing chip are as Figure 3 shown.

[0036] The above bionic auditory coding algorithm was used to encode the sound signal, making the continuous sound signal become a discrete pulse signal. A sound recognition algorithm based on a pulsed neural network was constructed to decode the pulse signal to achieve efficient sound recognition. The specific construction process includes the following steps: (1)Selection of test data: In this implementation, two different data sets are selected for the voice recognition task. The TIDIGITS data set is a publicly available data set of the Linguistic Data Consortium and is one of the most commonly used speech data sets for benchmarking speech recognition algorithms. This data set contains oral digit utterances of 111 male and 114 female speakers. 4,950 isolated digit spoken sentences are selected from it for pulse coding and voice recognition. The data contains 11 isolated digit spoken words (i.e., "zero" to "nine" and "oh"). These isolated digit spoken words are randomly split, with 2,464 used for training and 2,486 used for testing.

[0037] The sound scene data set of the Real World Computing Partnership (RWCP) is recorded in a real acoustic environment. For a fair comparison with the TIDIGITS data set, the same 10 sound event categories in the data set are used: "bells5", "bottle1", "buzzer", "cymbals", "horn", "kara", "metal15", "phone4", "ring", "whistle1". 40 sound segments are randomly selected from each category in the present invention, with 20 used for training and the remaining 20 used for testing. Therefore, there are 200 training samples and 200 testing samples each.

[0038] (2)Preprocessing and feature extraction of sound data: First, downsample the sound data of different types to 8 kHz. Secondly, perform voice activity detection in the sound stream, and calculate the maximum energy and spectral centroid of each frame as thresholds. Using these thresholds, frames are regarded as voice activity or silence, and signal segments containing sound are extracted. The constant-Q transform (CQT) is used to extract the spectral features of the preprocessed sound signal, and its formula is:

[0039]

[0040] where represents the calculated energy information of the sound, which is input into the bionic auditory coding algorithm and satisfies .

[0041] (3) Pulse Coding and Recognition of Sound: The sound information after the above preprocessing and feature extraction is input into the bionic auditory coding algorithm to obtain the corresponding pulse coding results. A sound recognition algorithm based on a spiking neural network is constructed to decode the pulse information to obtain the sound recognition results. The spiking neural network in this implementation has three layers, including an input layer, an intermediate layer, and an output layer. The structure of the recognition model for the TIDIGITS dataset is 6889-1024-11, and the structure of the recognition model for the RWCP data is 6889-1024-10. The spiking neurons in the SNN use LIF neurons, and its formula is:

[0042] where . Since the pulse is the Heaviside step function in the form of 0 or 1, during the training process, the gradient substitution method is used for training, that is, the step function is used in the forward propagation, and a substitution function with a shape similar to the step function and continuous differentiability is constructed in the backward propagation, and its derivative is used for backward differentiation. The definitions of the substitution function and its derivative are as follows:

[0043]

[0044] where 𝛼 is the width of the surrogate gradient, set to 2. During the training process of the spiking neural network, MSELoss is used as the loss function, the AdamW optimizer is used for training, the weight decay regularization coefficient is 0.02, the initial learning rate is 0.01, and ReduceLROnPlateau is used to dynamically adjust the learning rate, epoch = 200, batchsize = 64. The results of its sound coding and recognition are as Figure 4 and Table 1 shows.

[0045] Table 1 Comparison of recognition accuracies of the bionic auditory coding algorithm in SNNs and traditional non-SNN models with the same structure

[0046] Note: 1. The model compares SNNs with the same structure with convolutional neural networks (CNNs) and recurrent neural networks (RNNs); 2. The number of parameters represents the number of neurons required by the neural network.

[0047] The features of the present invention are as follows: (1) The present invention constructs a bionic auditory sensing chip with a biological auditory receptor as a sensitive element for auditory perception, which can directly obtain the auditory response after the biological senses the sound stimulus, that is, the present invention can directly obtain the electrophysiological signals of the biological receptor after being stimulated by different sound frequencies and different sound intensities, which more reflects the biological response to sound; (2) The bionic auditory coding algorithm constructed by the present invention is highly consistent with the response pattern of the biological auditory receptor to sound, has biological rationality, and based on a computable neuron model, the bionic auditory coding algorithm has computational feasibility and can meet large-scale computational requirements; (3) Compared with the traditional sound recognition algorithm, the present invention encodes the sound signal into discrete pulse signals and constructs a pulse neural network to recognize the encoded pulse signals, which is more in line with the biological auditory perception process. And on the basis of the same model structure, it has fewer parameters, higher accuracy, requires less computing resources, and at the same time, the processing process in the form of pulses more meets the low-power requirements.

[0048] On the other hand, corresponding to the foregoing embodiment of a method for sound pulse coding and recognition based on biological auditory response, the present invention also provides an embodiment of a system for sound pulse coding and recognition based on biological auditory response. The system includes: a bionic auditory response module, a sound preprocessing and feature extraction module, a bionic auditory coding module, and a sound recognition module based on a pulse neural network. For the specific implementation process of each module, please refer to the specific implementation steps of the foregoing embodiment of a method for pulse-driven sound coding and recognition inspired by biological auditory response.

[0049] A bionic auditory sensing chip construction module, configured to use a biological auditory receptor as a sensitive element, and configure a multi-electrode array and an optimized chamber to construct a bionic auditory sensing chip; A bionic auditory response module, configured to obtain the response pattern of the biological auditory receptor to pure tones of different frequencies and different intensities, and the response pattern of the organism is used as the construction basis of the bionic auditory coding module; A sound preprocessing and feature extraction module, configured to perform preliminary processing on continuous sound, including operations such as extraction of the effective segment of the sound and pre-emphasis, and further extract the sound frequency and intensity features; A bionic auditory coding module, configured to use the foregoing bionic auditory sensing chip to fit the response pattern of the sound signal, and based on the neuron model, output the pulse firing pattern generated at different frequencies and intensities to fit the response data recorded by the bionic auditory sensing chip, and construct a bionic auditory coding algorithm; encode the sound features extracted by the sound preprocessing and feature extraction module into pulse signals; A sound recognition module based on a pulse neural network, configured to perform sound recognition on the sound signal encoded in the form of pulses.

[0050] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A biological hearing-inspired sound pulse coding and recognition method, characterized in that: The following steps are involved: (1) Using biological auditory receptors as sensitive elements, and configuring a multi-electrode array and an optimized chamber to construct a bionic auditory sensor chip, which is used to record electrophysiological responses under sound stimulation; (2) Using the bionic auditory sensor chip to detect response patterns under pure tone stimulation of different frequencies and intensities; (3) using an optimization algorithm to adjust the parameters of the neuron model, fitting the pulse emission pattern output generated at different frequencies and intensities with the response data recorded by the bionic auditory sensor chip based on the neuron model, constructing a bionic auditory coding algorithm, and using the sound frequency and intensity information obtained by preprocessing and feature extraction of the sound signal as the input of the bionic auditory coding algorithm to obtain the encoded pulse output, thereby encoding the continuous sound signal into a discrete pulse signal; (4) Construct a sound recognition network based on a pulse neural network, and input the pulse signal obtained in step (3) into the network to realize sound recognition.

2. The method for bio-auditory inspired sound pulse coding and recognition according to claim 1, characterized in that: The biological auditory receptors are cells or tissues extracted from the biological auditory system, including any one of the organ of Corti, the cochlea, the inner ear organoids, the hair cells or the spiral ganglion cells. The biological auditory receptors are cultured on the surface of a multi-electrode array, and the biological auditory receptors are bonded to the electrode surface.

3. The biological hearing-inspired sound pulse coding and recognition method according to claim 1, characterized in that: The bionic auditory sensor chip includes a biological auditory receptor, a multi-electrode array and an optimized-size chamber, wherein the optimal size of the chamber is determined through acoustic simulation or experimental feedback to provide a liquid ion environment and ideal acoustic conditions required by the biological auditory receptor; the biological auditory receptor can generate an extracellular action potential under sound stimulation, and the corresponding potential signal is collected through the multi-electrode array to reflect the response mode of the biological auditory receptor to the acoustic stimulation.

4. The biological hearing-inspired sound pulse coding and recognition method according to claim 1, characterized in that: The response mode is one or more of firing rate, firing interval and synchronization.

5. The biological hearing-inspired sound pulse coding and recognition method according to claim 1, characterized in that: The sound stimulation is a pure tone stimulation, and its frequency includes 100 Hz, 500 Hz, 1000 Hz, 2000 Hz, 3000 Hz and 4000 Hz, and its intensity includes 60 dB, 70 dB, 80 dB and 90 dB.

6. The biological hearing-inspired sound pulse coding and recognition method according to claim 1, characterized in that: In step (2), pure tones of different frequencies with the same intensity are used to stimulate the response of the bionic auditory sensor chip to the sound frequency, and pure tones of different intensities with the same frequency are used to stimulate the response of the bionic auditory sensor chip to the sound intensity.

7. The biological hearing-inspired sound pulse coding and recognition method according to claim 1, characterized in that: In step (3), the continuous sound signal is preprocessed, including voice activity detection, pre-emphasis and constant Q transform operations, to extract the frequency, intensity and spectral characteristics of the sound.

8. The biological hearing-inspired sound pulse coding and recognition method according to claim 1, characterized in that: The bionic auditory coding algorithm uses the Izhikevich neuron model as a framework, and adjusts parameters through an annealing algorithm, a genetic algorithm, an ant colony algorithm or a grid search optimization algorithm to minimize the root mean square error between the pulse emission pattern output by the model and the response data recorded by the bionic auditory sensor chip.

9. The biological hearing-inspired sound pulse coding and recognition method according to claim 1, characterized in that: The sound recognition network adopts a three-layer structure of input layer, middle layer and output layer based on pulse neural network, wherein the pulse neurons are LIF neurons, and the gradient substitution method is used to realize back propagation during the training process.

10. A biological hearing-inspired sound pulse coding and recognition system for implementing the method of claim 1, characterized in that: The system includes: A bionic auditory sensor chip construction module is used to use biological auditory receptors as sensitive elements and configure a multi-electrode array and an optimized cavity to construct a bionic auditory sensor chip; Bionic auditory response module, used to test the response mode of the bionic auditory sensor chip to pure tones of different frequencies and intensities using a variety of sound stimuli; The sound preprocessing and feature extraction module is used to preprocess and extract features of sound signals to obtain sound frequency and intensity information; A bionic auditory coding module is used to fit the pulse emission pattern output generated at different frequencies and intensities with the response data recorded by the bionic auditory sensor chip based on a neuron model to construct a bionic auditory coding algorithm; input sound frequency and intensity information to encode continuous sound signals into discrete pulse signals; The sound recognition module based on the pulse neural network is used to process the encoded pulse signal to achieve accurate sound recognition.

Images