Application of embedded voice and digital forensics system in financial cost management

This article first describes the embedded voice processing process and uses the two-door restriction method for endpoint detection. Next, the wavelet + MFCC algorithm is proposed. The Fourier transform will replace it with ripple. The disadvantage of the ripple is that the size of the Fourier transform window cannot move with frequency. On this basis, the overall goal of the system is proposed, the overall structure is constructed, and the functions of the system are divided. Android mobile phone data collection technology has been extensively researched, including logical search, temporary access to permissions, JTAG search and ISP search. The design and implementation of the logical search engine based on root authority, the physical search engine based on Qualcomm processor, and the data analysis system have been completed, and the evidence collection of e-commerce applications has been completed. Based on the digital forensics system, this article studies and designs an Android device digital forensics system suitable for China's national conditions and laws from the three perspectives of data collection, analysis and storage. At the same time, brainstorming methods must be used between corporate financial management and cost management personnel. Finally, through research, scientifically construct the framework of the financial cost management system, distinguish the various requirements of the construction system, classify the importance, urgency, and feasibility of the requirements, establish the basic requirements and extension requirements, set the demand function to complete the schedule and Perform as planned. Through the research of embedded voice and digital forensics system, this paper applies it to the research of financial cost management, thereby promoting the development of enterprise financial management.


Introduction
With the advancement of science and technology, embedded speech recognition technology has become more and more mature, and it is the basis for the full realization of "artificial intelligence" [1].At present, the embedded speech recognition system based on the microprocessor platform can train from the speech library, create a training model, and complete the recognition by combining the human voice and pattern, rather than a specific person for speech recognition.In this paper, DTW algorithm combined with polynomial kernel function is used to effectively process the time-varying features of embedded speech, and a new PDTW-SVM algorithm is obtained [2].It has local interpolation capabilities and global generalization capabilities, as well as the time-varying characteristics of speech signals.Finally, the PDTW-SVM algorithm is transplanted to the LD3320 platform, and an echo noise reduction circuit conforming to CX20921 is added according to the actual application scenario, which improves the anti-noise ability and stability of the system [3].The rapid development of mobile Internet technology has completely changed the way people live and work.The Internet has accumulated a large amount of personal information, and it has obviously become an important information transmission medium.However, new types of criminal activities targeting mobile devices have emerged in an endless stream, including SMS and email fraud, e-commerce fraud, intellectual property infringement, the spread of harmful information, and computer viruses [4].The number of criminal activities is endless, mobile digital forensics systems continue to exist, and the development of the Internet has attracted much attention.Among them, the mobile device market share of the Android operating system accounts for approximately 86%.

Related work
The literature introduces a trace analysis method, and uses this method to analyze the traces left by the client program when accessing Baidu Netdisk cloud storage service through the browser [9].This article introduces the format and content of each file in detail to lay the foundation for the design of the cloud storage client forensic system in the next step.The literature introduces the reverse analysis method [10][11].The online disk client program uses the reverse analysis mechanism to decrypt account information files, uses the cloud to store the account information files of the client program, converts the user account, and obtains the password.The literature introduces the cloud storage client forensics system to realize the client forensics of Baidu cloud storage service [12].First, determine the required functions of the forensic system, and carry out the overall design of the forensic system, such as case management, data extraction, data extraction and analysis.Then each function is designed separately, the realization method of each function is introduced, and the execution result of each function is introduced.The literature introduces the Android system framework, Android Bootloader startup principle, Root principle, debugging bridge, database, NET Framework and other key skills and basic knowledge required in the development process of the system, which has obtained a good theoretical foundation for the development of the system [13].The literature introduces the detailed research of Android mobile phone related data collection technology, Android data collection investigation and solutions, logical search, routing search, JTAG search, ISP acquisition and other logical identification for obtaining root permissions [14][15].

Embedded platform
In order to be able to perform voice recognition under any circumstances, not to be affected by the surrounding environment or poor network, and to occupy as little storage space as possible, this system uses an offline voice recognition system to identify unidentified people.At the same time, considering the price, space storage and production requirements, the embedded platform that can be selected must meet the following conditions: a It has an integrated chip that can handle audio signal processing and high-speed D/A conversion.
b The system has powerful computing capabilities and can perform complex transformation processes such as Fourier signal transformation.
c Has a rich library of speech recognition and keyword recognition.d Provide a human-computer interaction interface, which can dynamically add recognition commands to users to evaluate the rationality of the recognition results, such as the recognition effect of broadcast recognition results, voice, and text display.
This article chooses the LD3320 development kit as the platform of the language recognition system.This is a chip that realizes non-specific speech recognition function.Because of the built-in digital/analog conversion module, it can be converted without adding A/D and D/A interfaces.At the same time, the chip has internal registers, so there is no need to connect flash and RAM to store data.In addition, users can add a keyword list of up to 50 words according to their actual needs, and the chip can provide dynamic delete and write functions that take up very little memory.The main control of the MCU is the STC10L08XE chip, all software programs are written into the internal flash of the MCU, and the language recognition process of the LD3320 is controlled by the read/write register of the LD3320.
Figure 1 shows the architecture of the LD3320-based embedded platform.It is usually composed of multiple main modules such as voice collection, language recognition, main control module, and so on.Collecting the human voice is usually realized by using a voice collection module.After a series of preprocessing and A/D conversion, the analog audio signal becomes a digital signal, which is inserted into the speech recognition processing chip to form its collection and imaging.The main control of the MCU is used to control the voice recognition chip, and it uses I2C to communicate and program with it.The command recognition library is used to temporarily store voice commands and feed back the recognition results.Because the user must program the ROM registry code every time the instruction is changed, the operation is inconvenient.In this article, we will add an EEPROM module, which can also directly write the contents of the registry for easy association.

Speech recognition
Digital signals are usually expressed in binary.For example, the length of the measurement word is p, which means there are p binary numbers, and the authentication value is 2p, which means that the amplitude is divided into 2p areas.Example: "101" means that the amplitude value is divided into 5 areas, and this value uses the same amount of space, so there is a certain error in the quantization process.The error function is: The value range of e(n) in the formula is: (2) Then the measured signal-to-noise ratio is: Equation ( 3) becomes:  = 6.02 − 7.2 (4) After the audio signal is sampled and counted, it is put into the internal storage area, and the data is enthusiastically stored in a "queue".The characteristic parameters of the speech signal are processed and retrieved and then stored in the database to realize the dynamic balance of the input and output signals.
Studies have shown that above 800 Hz, the average power spectrum of the sound is reduced by 6 dB/octave reduction, or 6 dB/oct or 20 dB/dec, so the signal must be pre-weighted.The high frequency part of the signal is processed first in the time domain, so that the signal changes slowly and the overall trend is relatively stable.The same signal-to-noise ratio calculation method can be used to solve the cross-band problem, which is very useful for analysis.For example, as in equation ( 5), the first-order digital filter is H(z): (5) In this equation, μ is the first-order delay factor, which is a value close to 1.As shown in equation ( 6), the weighted signal is y(n): ) As shown in (7) and ( 8), the commonly used window functions are rectangular window and Hamming window: Different analysis fields, such as time domain and frequency domain, require different window functions to capture different appearance parameters.In the time domain, the frame signal is "split" between frames.The corresponding window function should have a smooth slope, which makes the transition from frame to frame smooth and eliminates the sudden change of "split".
Rectangular window: The frequency response is: −(−1)/2 (10) f01 is the bandwidth, which is the frequency when the first zero value is taken.f01 is as shown in formula (11): The minimum gap between two different frequencies is the frequency resolution, expressed by △f.The relationship between △f, window length and sampling time Ts is: A larger value of N is equivalent to a signal passing through a narrow low-pass filter, with suppression of highfrequency phase, small short-circuit energy changes, and unrecognizable voice signal behavior.If N is too low, high-frequency components will be amplified, and short-term energy changes will cause insufficient correction of the energy function, which will suddenly lead to "faults." Before compounding, sample counting and initial signal weighting are very important processing methods.When extracting the feature function, each frame is extracted through the "queue" method, and the extracted parameters are combined to extract the feature function of the entire speech.

Key technologies of Digital Forensics System
With the progress and continuous development of computer and network technology, digital forensics has become a new technology, which comes from computer forensics technology.With the development of digital technology and the actual needs of forensics, the fields of forensics related to digital technology are expanding.At the same time, we need to maintain a tight chain of information integrity and data protection.The working group on digital forensics defines digital forensics as the use of scientific and proven tools to modify the process of digital crime, or to predict and eliminate existing harmful fraud.In the process of storing, collecting, verifying, identifying, analyzing, interpreting, archiving and presenting digital evidence of digital devices and other resources, one of the concepts related to digital forensics is digital evidence, which is the progress and enhancement of the connection with traditional evidence.It must be reliable and accurate, and complement with national laws and regulations.Generally speaking, digital evidence can prove all kinds of digital data of the situation, including text, image and others in the operation and activities of digital equipment.
Traditional digital forensics technology is facing many challenges, which makes the forensic of cloud server face a series of challenges to be solved.In today's real forensic work, cloud server-side forensic relies on helping cloud service providers easily obtain client evidence.We can capture digital evidence in client by using traditional digital forensics technology for analysis, and we can also get many important evidences from app.According to the forensic research of cloud storage client and most files, it stores some valuable information, such as the time of access and operation time and category, and the ability to retrieve user cloud storage account and password.This information helps forensic investigators determine whether the evidence related to the event is available on the cloud storage server, And allows investigators to decide whether to collect evidence on the cloud server.
Meanwhile, cloud storage client has the ability to obtain the user password account information, which will help forensic investigators to conduct further forensic investigation.For example, if required by law, the user's account password is used to access the cloud store to retrieve user data stored in the cloud.The user can access the cloud storage service directly using client programs, which have the function of automatically logging in the account.Forensic investigators can not simply analyze the data on the file to obtain encrypted information, which hinders the evidence collection.
The client program analyzes the decryption algorithm and decryption key of encrypted information in reverse interpretation technology, so as to quickly decrypt the encrypted information.See Figure 2 for details: If SVM uses a Gaussian kernel function, a causal penalty of C = 16 is determined for the maximum Gaussian kernel activity, and the Gaussian kernel radius is 0.2.Table 1 compares the optimized DTW-SVM algorithm with the traditional SVM algorithm, and the experimental results are shown in Table 1.It can be seen from Figure 3 that the recognition accuracy of the SVM algorithm gradually increases with the increase in the number of training samples, and the recognition accuracy gap between the DTW-SVM algorithm and the DTW-SVM algorithm is gradually narrowing.When the number of samples is 50, the recognition accuracy is close to the DTW-SVM algorithm.In terms of identification time, the larger the number of samples, the longer the response time of SVM and DTW-SVM algorithm identification.See Figure 4 for details: The DTW-SVM algorithm needs to calculate the minimum weight distance, so the training and testing time is longer than the SVM algorithm.The detection accuracy of the DTW-SVM algorithm is at least 1.1% higher than that of the SVM algorithm, and the maximum time difference of the average test time is 0.45 seconds, which can be ignored.In general, DTW-SVM has a better overall detection effect than SVM.See Figure 5 for details:

System requirement analysis
If we want to continue the overall analysis and design of the system, we need to have a more comprehensive understanding of the characteristics of the system.Through the research, we have developed a set of management system which fully understands the user's software requirements and is accepted by users.In the user needs analysis, we investigate and analyze the actual situation of software use and users.After getting the survey results, we design and describe the software requirements in detail, focusing on the system design and function.Analysis of the detailed description of the applicable modules in the article, the specific objectives and tasks need to be achieved are as follows: The beginning of the paper describes the database data processing, focusing on understanding the user needs and analyzing the security problems in the process of data processing, so as to give the design corresponding design.Given the corresponding design of the scheme, investigate and create a management system to meet the needs of users.In the design of the corresponding database management system, the most important is certainly the type of data structure.Users can integrate good data structure to get better data query and search services.When considering the ability to analyze and process data, the ability to recover data should also be checked to ensure that data security reaches 100%, so that users can easily manage their data.
Functional requirements analysis is based on the most effective use of management information system necessary analysis operation, you can understand the specific functions of the system, and according to the requirements of users for software function test.Only a more comprehensive understanding of the characteristics of the system, users can use the final software development to meet their needs.In the analysis of system functional requirements, public interviews are always used to understand the user requirements.Then, using the relevant analysis tools, the whole business process described by the user in the interview is captured and decoded, and combined with a programming language, the above-mentioned special process is modified into software that can be implemented on the computer, and converted into a more readable system, so that the user can meet the actual needs of the system.

System architecture design
Because users only need to open the web to operate the software, it is less difficult to use, which improves the cross platform effect of the system and the transplantation effect.But this method often has high requirements for the server, because it not only involves data storage, but also involves the logical operation of the middle layer, such as the need to interact with data through the Internet, the execution speed of the whole software will slow down.Therefore, the three-level system architecture used in this paper will make the system layered when executing the system layout method to achieve load balancing.

Requirements of database design
First, analyze the data storage capacity.In the process of data storage operation, the system connects to the backend database at this time and saves the data from the front-end later.For data storage, it also includes processing logic, which is an integrated module to facilitate data filtering.The efficiency of data storage is effective.Assuming that the data in the previous table is the same or related, it can be repaired.In other words, if the front-end and database linked data, it is necessary to maintain the mapping relationship between them, which is possible, and gradually determines the amount of data storage.
Second, safety requirements, how safe the system is, even if it can run different modules during production and is still relevant, software developers need to pay attention to software security in the design process.When setting security, we need to implement two methods: authentication and digital authentication.As we all know, in order to improve the system security and make the system more secure on the basis of ensuring the efficiency of system operation, we need to meet many design requirements and avoid being attacked by viruses and hackers, so as to make the system security meet the standard.

Database conceptual design
The key point of this scheme is to establish a data model.The representation of the data model is an ER graph.The shape of the entity is rectangle and diamond, which represents the corresponding relationship between two entities.The ellipse represents the corresponding attributes of the entity.Quality is usually ignored when drawing E-R diagram, because the allocated internal capacity is very large.

Database logic design
The logical structure of the database is to transform the basic E-R diagram in the conceptual structure stage into a logical structure that matches the data model supported by the selected DBMS product.The following describes the database table layout, table relationship and table structure in this system. (

1) System user table
This table is the basic information of storage system users.Details are shown in Table 2. (2) Project list This table is used to store the uploaded project file list and is the main basis for making material plans and equipment leasing plans.Therefore, according to the E-R diagram, the specific structure of the data table is shown in Table 3.

Design of system function module 4.4.1 Design of project list management module
The list management module of China railway engineering construction project is mainly to create the list of materials required for the project.From the industrial value, the types and quantity of materials needed for construction projects are huge for construction enterprises.Before the cost control of construction works, detailed engineering lists need to be created for specific projects, and all costs incurred are managed by using the project list method.However, if the project list has been created, the list of items cannot be accessed in a simple way of information input because of its size and structure.Therefore, in the project list management design, the project order table upload method will be used for the entry of the item list.Users can upload all the item list information to an excel file at a time.

Design of material cost management module
The specific function and function of this module is to manage many materials needed for the project.The material cost management process requires the operator to make the material plan in advance and then submit it to the competent leading office.The project creates a master plan and creates a specific plan for each monthly material; Instead, if the review fails, it needs to be updated or corrected.

Design of financial business management module
It is the main financial module used to repay the debt of each project.Based on the huge debt and machine debt, the main obligation of the company's financial management is to pay the company's financial bills according to the company's requirements.Among them, the repayment process is carried out according to the submission process, similar to the loan repayment approval process executed by the application process.

Development strategy of financial cost management
Cost management information system can only be established after the public fully understand, from a visible point of view, to formulate and summarize the medical reform policies of different countries, in order to determine the direction of medical reform.From our own point of view, we have made clear the strategic goal and social orientation of development.The problems that need to be solved in the world need to follow the trend, such as introducing new government accounting system, proposing cost control, etc.Because there is no unified standard for the construction of cost management system, hospital managers and cost managers can draw on the hospital benchmark and consult external experts at the same time.Finally, through scientific research and construction of cost control system, identify various system construction requirements, sort the importance and feasibility requirements, and complete the schedule and corresponding plans.

Conclusion
In this paper, we will study an embedded micro processor compatible audio recognition system, optimize the audio image capture and standardization algorithm, and optimize the embedded platform based on the characteristics of the actual hardware structure of the embedded platform.We have applied the transmission algorithm to the financial cost management information system, actively implemented it, and accelerated the informatization and digitization construction, To maximize the cost-effectiveness.In the digital era, we should grasp and make use of this opportunity to improve the management level of hospitals, promote healthy and sustainable development, and provide many high-quality medical services for the society.

Funding
This research has been financed by The Anhui Provincial Quality Engineering Project for the institution of higher learning in 2020 "The Virtual Simulation Experiment Project for Procurement Service" (2020xfxm22) and The Teaching and Scientific Research Project in 2019 of The Anhui Vocational College of Auditing "The Team of Cost Accounting Professional Teaching" (2019sjjxtd008).

Fig. 1
Fig. 1 Embedded speech recognition system architecture diagram

Fig. 3
Fig. 3 Comparison chart of DTW-SVM algorithm and traditional SVM algorithm

Fig. 4
Fig. 4 Comparison chart of DTW-SVM algorithm and traditional SVM algorithm (training time)

Fig. 5
Fig. 5 Comparison chart of DTW-SVM algorithm and traditional SVM algorithm (evaluation test time)

Table 1
Comparison of DTW-SVM algorithm and traditional SVM algorithm

Table 2
System user table

Table 3
Bill of quantities