Research on Skin Listening System Based on MATLAB

At present, there are many deaf patients in the world, who are either born deaf or acquired deaf. At present, the deafness problem is mainly solved through traditional hearing aids and electronic cochleas, but both have obvious defects. The former is only used for patients with mild deafness, while the latter is too expensive and not suitable for promotion. In order to completely solve the problem of deafness, Professor Li Jianwen invented a transformer-type skin hearing device for the hearing problem of the deaf. Its essence is to use sound-to-electric conversion to allow the skin to feel the current signal, so that the deaf can get a certain hearing through the skin, which fundamentally solves the hearing problem of the deaf.
However, the transformer-type skin hearing device only allows users to feel electrical stimulation, but cannot distinguish speech. To solve this problem, this paper proposes a skin hearing device system based on MATLAB. The system can realize the functions of recording, playing, preprocessing, segmented filtering, feature extraction, etc. of speech signals, and can distinguish simple speech.

1 Skin hearing technology and skin hearing device
Skin hearing technology is a relatively cutting-edge technology, and its purpose is to use the skin to solve the hearing problem of deaf patients. The skin sound listener is an electronic tool that solves the hearing problems of deaf patients through skin sound listening technology. Its characteristics are similar to hearing aids and involve medicine, electronics, acoustics and physiology. Further development will rely on disciplines such as mathematics and computers. It belongs to a typical multidisciplinary high-tech field. It transmits sound signals through the skin and does not rely on any residual hearing of the person. It has obvious practical effects, is easy to wear, has almost no side effects, and can be used by anyone with hearing loss.
The first generation of skin sound listeners invented and developed by Professor Li Jianwen has achieved the effect of hearing. Next, the skin sound listener must meet the requirement of being able to distinguish speech. This requires converting the original analog circuit into a digital circuit, and then performing speech analysis on this basis. This paper designs a skin sound listener system based on MATLAB. The system uses MATLAB software as a platform to successively realize the functions of recording, playing, preprocessing, segmented filtering, feature extraction and final speech analysis of speech signals.

2 System Function Introduction
2.1 Main Functions
The role of the skin sound listener is to solve the hearing problems of deaf patients. Therefore, the energy of the skin sound listener to distinguish speech is a very important technical link. This paper designs a skin audio system that can basically distinguish simple speech. The system uses MATLAB software as a platform, and its main functions include recording speech, playing speech, preprocessing, segmented filtering, feature extraction and final speech recognition. The functional division of the platform is shown in Figure 1.

2.2 Core modules
According to the functions of the skin audio system in this paper, it is divided into three core modules, namely, the acquisition module, the processing module and the analysis module.
2.2.1 Acquisition module
The functions of the system acquisition module include: recording voice and playing voice.
1) Recording voice: Record the sound through the microphone and save it in .wav format for later use.
In MATLAB, use the function wavrecord to record voice. The result is a digitized voice data string, which can be directly used as a Matlab variable to participate in various operations. Then use the function wavwrite(y, filename) to write the data stored in the variable y to the WAVE file named filename.
2) Playing voice: Play the recorded voice WAVE file and output its waveform and spectrogram.
Here, use wavread and sound functions to read and play voice respectively. When obtaining the waveform and spectrogram, the length function and FFT transformation are used.
2.2.2 Processing module
The functions of this module include preprocessing, segmented filtering and feature extraction.
1) Preprocessing: Remove noise, pre-emphasize and detect endpoints of the previously recorded speech signal to provide a clearer and more effective signal for subsequent work.
2) Segment filtering: Segment filtering is performed on the preprocessed speech signal to separate speech signals of different frequencies so that the skin can better sense the speech signal.
3) Feature extraction: The characteristic parameters of speech can effectively reflect the characteristics of the speaker's speech that are different from other speakers, and it plays a vital role in the entire discrimination process. After speech filtering, this paper extracts a parameter that can characterize the speech segment from each frequency segment, and then combines all parameters to construct a feature vector, which is used as the characteristic parameter to characterize the speech signal, providing a more reliable reference data for the subsequent discrimination work.
2.2.3 Discrimination module
The functions of this module include: template library building, pattern matching and output results.
1) Template library building: The speaker repeats the speech unit many times, removes redundant information from the original speech sample, retains key data, obtains standard data representing the common characteristics of the recognition basic unit, and uses this to form a reference template. The reference templates of all the recognizable basic units are combined together to establish a reference template library.
2) Pattern matching: After the newly recorded speech is feature extracted, it is compared with the templates in the library according to a certain principle to find the pronunciation corresponding to the reference template with the highest similarity, which is the result of the analysis.
3) Output result: Output the previous analysis results so that users can better analyze speech through the skin audiometer. [page]

3 System Design
In the main window of MATLAB, select the New menu item in the File menu, and then select the GUI command, and the design template of the graphical user interface will be displayed. The system in this paper mainly includes three functions: voice acquisition, voice processing and voice analysis.
3.1 Acquisition module
The system will record through the computer microphone. In MATLAB, y=wavrecord(n, fs, ch, dtype) is used to complete the recording, where n is the number of sampling points. It determines the recording length; fs is the sampling frequency, the default value is 11 025 Hz, and you can also choose the appropriate sampling rate according to your requirements; ch is the number of channels, the default value is 1, indicating mono, if it is specified as 2, the sampling is two-channel stereo data; dtype is the storage format of the sampled data, which is specified by a string, which can be 'double', 'single', 'int16', 'int8'. When specifying the storage format, it also specifies the quantization accuracy of each sample value. Int8 corresponds to 8-bit precision sampling, and the others are 16-bit sampling accuracy.
After the recording is completed, the system will save the recorded voice signal as a WAVE file. Here, the function wavwrite is used to complete it.
Finally, the previous voice WAVE file will be played, and its waveform and spectrogram will be generated to provide a basis for subsequent processing.
3.2 Processing module
The system will preprocess the previously recorded voice signal, including: voice signal denoising, pre-emphasis and endpoint detection. Noise removal is to remove useless information (noise) from the voice signal as much as possible, while retaining the useful information (voice). The purpose of pre-emphasis is to enhance the high-frequency part and make the signal spectrum flat, so as to facilitate spectrum analysis or vocal tract parameter analysis. The endpoint detection of voice is to determine the starting and ending points of the voice, exclude the silent segment from the voice signal, and provide an operation object for subsequent processing.
Then, the signal is filtered. The filter is a two-port network that is selective to the frequency of the input signal. It allows certain frequencies (usually a certain frequency range) to pass through, while the amplitudes of other frequencies are attenuated, thereby separating voice signals of different frequencies, so that the skin can better sense the voice signal. Wavelet packet transform divides the frequency band into multiple levels, and can adaptively select the corresponding frequency band according to the characteristics of the analyzed signal to match it with the signal spectrum, thereby improving the time-frequency resolution. Using this feature of wavelet packets, this paper divides the speech signal into 16 frequency bands with filters, each frequency band is a speech channel, and each channel is connected to an electrode sheet to contact the skin respectively. The division of the speech frequency range is shown in Table 1.
Finally, click the "Feature Extraction" button to extract the characteristic parameters of the speech signal. Calculate the signal energy value of each frequency band above, and then combine the 16 energy values ​​as component vectors to form a feature vector, which is used as the characteristic parameter of the speech signal. The energy value reflects the speech energy carried by the speech signal in each frequency band, which can characterize the speech segment, so it can be used as a feature parameter. In addition, the characteristic parameters of each speech signal will be output in the bottom window.
3.3 Analysis module
First, establish a database. The database stores the characteristic parameters of the speech signal, and uses SOL Server2003 to build the database.
Then, perform pattern matching. After the database is built, the feature parameters of the new speech signal are extracted and matched with the data in the database. If a template with the closest similarity is found, the template is the speech to be analyzed, and the speech signal is output in the window below; otherwise, the feature parameters of the new signal are stored in the database to enrich the data in the database, and "There is no such speech in the database!" is output.
Finally, the analysis result is output.

4 Feature Analysis
The purpose of this system is to enable deaf patients to distinguish simple speech through the skin audiometer. It is accomplished in two aspects: 1) After the system is filtered, the speech signal is transmitted to the skin through a group of (16 in this paper) electrodes. Since the energy of the speech signal in each frequency band is different, the electrical stimulation to the skin will also be different. The user will feel different frequencies and degrees of stimulation, and will be able to distinguish different speech to a certain extent. 2) After the signal is filtered, feature extraction is performed, and then the speech is analyzed by software according to a certain algorithm, and finally the analysis result is output. This paper uses
the high-resolution characteristics of the wavelet packet transform to first decompose the speech signal, then filter it through a filter, extract feature parameters on this basis, and then use the feature parameters to perform speech analysis.
This is to analyze speech from the two aspects of skin touch and visual effects, which is bound to enhance the ability of the skin audiometer to analyze speech.

5 Conclusion
This paper first introduces the skin listening technology and the skin listening device; then, it explains the main functions and module division of the system, as well as the usage methods and core technologies of each module; finally, it summarizes the characteristics of this system, that is, to distinguish speech from the two aspects of skin touch and visual effects, which will improve the ability of the skin listening device to distinguish speech to a certain extent.
Of course, this system has some imperfections, such as whether the characteristic parameters can be more detailed and more distinctive? Can it distinguish more complex speech? These are problems that need to be solved urgently and are also the focus of future research work.