Field of invention:
This invention is proposed to achieve the protection of speech data for voice based social media platforms using machine learning. Social media platform and many applications use voice and speech data. Thus, it deals with the data collection and transmission process and to achieve seamless transmission of the speech data, it requires larger amount of data, and this data requirement increases along with the numerous vulnerable threads. Thus, the thread control technique have been implemented in the speech data to achieve the privacy of the speaker on the social media platforms.
Background of the invention:
Machine learning, which is the subset of artificial intelligence empowers the computer to learn by itself and based on the analysis it proves the possible response. Numerous machine learning algorithms have been developed to perform the prediction or data classification task to ensure the required outcome without any human intervention. Thus, machine learning makes the computer to act as human brains and its decision making capacity and predictions are much more efficient than the humans* predictions. Applications of the machine learning includes detection of the stock price index, prediction of the risks in hospital readmission.
Massive data collection is the demand of the machine learning to produce appropriate response over comparison of the collected relevant data by the machines. This higher requirement of data includes hospital information, speech recordings, medical history, private details of a person, and many more sensitive information's.
Most common attacks performed by hackers revolves around the social media platform where the data breach is a most common factor since its much based on speech data input. Recordings or voice data which are collected through the loudspeakers, watches, intelligent digital assistances, televisions, smart cars are transmitted over public networks and then the data is stored and processed on cloud based infrastructures. Hence to protect these privacy concerns while achieving the required outcome is implemented by machine learning using various techniques.
The surge of machine learning across billion of device have changed the way of learning, work, and function. The privacy obligation must be taken into serious discussion to enhance the speech data privacy protection. According to the Sifan Ni, a researcher at Donghua university in China has put forward the challenging task of the protection of the speech data by emphasizing the balance between the privacy and data utility. While preserving the analytical utility of the data, speech content, the speakers voice, and data setdescription are focused to achieve the data privacy and quality of the signal.
Voice based social platform allows users to enjoy the platform with supportive languages such as Tamil, Telegu, Kannada, Bengali, Gujarati, Marathi. Allowing users to communicate with each other by using the voice data, and attractive posts to help better reverberate with the followers. And the application allows user to record the voice data over 60 seconds only and after the

recording is completed user can add music, backgrounds, and captions to ensure better attractive posts. These voice input platforms help the users to reward the quality content, aiming to create a transparent social media platform and it has opened doors for numerous user friendly applications.
Voice assistance such as Amazon Alexa, Siri, Cortana, Google Assistance, and other social media platforms which are given as of Facebook, Twitter, Instagram have become increasingly popular, and millions of peoples are using these platforms on day to day life. Voice assistance will greatly make simpler the way in which the PC's, tablets and smart phones are being used. But there are numerous vulnerabilities that are caused because of the hacker's involvement in theses software and the data stored on the system are stolen over a phone channel via voice assistance. One of the researchers Zheng Xian He carried out the study of data breaching of sensitive data. As voice based social media platform has been the most promising mode of communication and entertainment to the users. Though we have numerous advantages of the speech data processing for voice based social media platform, still it's being a challenging task due to the numerous risks which arises due to the data theft by the hackers.
Objective of the invention:
1. In this proposed model the security and privacy of the speech data are protected by the development of the speech data protection model for voice based social media platforms through the implementation of cryptography in machine learning.
2. This proposed model includes blending of both the hardware and software components to achieve the secured transmission of the collected speech records.
3. In this proposed model the data collected and recognized through the speech recognizing technique and later the obtained data are stored in the cloud using the source of internet.
4. In this proposed model the data is broken into bits and are converted to a digital format helping in the analyses and with the help of machine learning the system determines the appropriate response comment through the mode of speech by considering the historical data that had been already stored for cloud processing.
5. In this proposed model, the data control bits for Hadamard gate are obtained with the hyperchaotic system and the wavelet transformation is carried out through bit-flip operation.
6. In this proposed model the security of the signal is strongly achieved with the help of the secret nonorthogonal angle and hyperchaotic quantum system.
7. In this proposed model the input data are encoded with the help of the encryption cryptography and the encrypted data is stored in the encryption domain to achieve the avoidance of the possibilities of data breach.
8. Thus, in this proposed model with the implementation of sophisticated cryptographic algorithm along with encrypting standard using machine learning, the speech data especially for voice based social media platforms can be protected through this speech data protection method.

Summary of the invention:
Rapid development of the machine learning implementation in all fields to achieve automation has numerous advantages along with the numerous hazards and result in the mishandling of the acquired data. However, with the advancements in the machine learning technologies various algorithm are being proposed to provide security to the speech data. Cryptography in machine learning is being the most essential encryption model of the data transmission for secure communication. The security of the acquired speech data is carried out with the help of the encryption and decryption process with the secret key for actual data retrieval. The input speech data is converted to cipher data by the cryptographic encryption process and the same data gets reversed at the decryption end to provide actual input and thus ensures the security over network transmission. With this cryptography encryption techniques various differential and statistical data theft can be controlled with the higher key security, wider keyspace.
Detailed description of the invention:
Social media platforms such as news telecasting, e-commerce and few confidential areas which also includes the voice-conference, VoIP, speech encryption techniques have already been emphasized to secure the speech data and in much more fields it's still being a challenging task to avoid the hacker's attention with the encryption models. Encrypting the signal at the receiver end and sending the encrypted data prevents the unauthorized access to the signal. There are several techniques implemented to encrypt the data which are given as of the Fast Fourier transform, discrete cosine transforms, analogue algorithm. Based on the efficiency of the encryption model, the techniques are chosen to secure the speech data and achieve privacy over communication.
Initially the speech data received as an input to this model are captured through the speech capturing and analog to digital conversion process along with the frontend processing such as acoustic feature extraction, subtraction of additive noise, speech enhancement, deconvolution of sophisticate noise. The data theft starts at the very initial stage speech capturing stage. With the advancement in technologies various technique have been proposed to achieve the security of the data. Over the limited bandwidth channel, the analog algorithm for encryption is used for the better performance.
Cryptography algorithms are the most suitable form of data encryption model to achieve the protection method for voice based social media platform with the help of the scrambling techniques. The algorithm is based on the chaotic maps, blind source separation, blowfish, fuzzy commitment schemes and doffing maps are the most important techniques implemented to achieve the secured data transmission. All the cryptographic algorithms deal with the process of conversion of the actual acquired data to an unreadable or unclear image and the data are then moved through the unprotected network and later at the receiver end the original data is recreated based on the decryption.
Over several algorithm, the process which encrypt, and decrypt signal determines the strength of the signal. The speech encryption is sampled at the process end and the data is applied to the Fast Fourier transformation. At the output region the process of encryption is reversed to obtain the desired output is proven by proper selection of the scrambling matrix and limited bandwidth with

key flexibility. In chaotic maps the random production of the cipher block is the added advantage, where the input signal is segmented at the input region and later combined to new shape of fixed size. While in this method secret key plays the vital role and the strength of the signal is determined with the selection of the secret key which are used at both the encryption and decryption end of the protection method.
In the blowfish algorithm, the encryption of the data happens via 16-round Feistel network and each of the rounds includes the precomputed key for encryption. Fuzzy commitment combines error correction code and harsh function to perform safer authentication and the process is completed carried out with the enrolment and verification procedures. LSPR algorithm is the algorithm used for VoIP signals to efficiently encrypt and decrypt the streamlined speech data and it is carried out by converting the input to plain text with the help of the analog transform. The expression is provided as of q = pL + kL, The cipher text is sent over the network without any transmission error gives the added advantage to the LSPR algorithm. Through cryptography algorithm we can obtain the solution towards the speech data theft especially on social media platforms to avoid financial as well as personal attack to any individuals.
In Quantum cryptography, improving the keyspace and security hyperchaotic system is implemented along with the Hadamard gate and controlled-NOT gates. The initial process includes the speech samples in standard and Hadamard basics. Converting the sample speech to a nonorthogonal quantum series improves the security against the vulnerabilities. Through unitary rotation of secret polarization angle is chosen for the encoding of the speech signal. The speech signal after encoded in nonorthogonal quantum states is given as of a{0,l}. The transmitter selects the encoding method by choosing the polarization angle 6[ and m* representing quantum state is decided based on the rotational operator Rit
Corresponding to the quantum state the superposition states are generated. The secret angle in the opposite direction, is rotated to the ith quantum angle to retrieve the original data form. For the quantum information processing, in this method we have implemented the quantum gates and controlled-NOT gates for the better performance. As classical network is less efficient than the suitable quantum gate to perform much faster in producing the quantum. C-NOT gate is implemented to make the target signal same and the control bit along with target bits of this gate performs the encryption process. Hadamard gates derives the relation between the ground state and the excited state, and the operation of this gate is derived as of
Based on the keyspace and security of the hyperchaotic system, its widely used in the data encryption process and it generated through the 4-D hyperchaotic system.
Encryption process of the model:
The speech data protection model initial stage is the encryption process, and the systematic representation of the encryption process is illustrated as of,

The generated sequences are provided as of {*t}{yi}{zi} and {wi} for the initial set of sample speech data parameter of the hyperchaotic system. With the runge-kutta method the four different hyperchaotic sequences are proposed. And this signal gets converted into integer sequence provided as of
x\ = | fix {Xi - fix (xi)) * 1014|mod 2n, (2)
Similarly for all other integers the same integration is applied and the generation of the key stream klt k2 is carried out with the help of the Hadamard and C-NOT operations.
For C-NOT operation and Hadamard operation control bits are proposed as an input and the flip flop operation is carried out with the help of controlled NOT gate and the expression is provided as of
ck = / when z[ = 0 (3)
cki = x when z{ = 1 (4)
Where X is given as the bit flip operator on the quantum state. And the Hadamard gate operation is provided as the Wfc2and the expression is derived as
Hk2 = l,whenw[ = 0 (5)
Hk2:=Hrwhenw{ = 1 (6)
Hadamard gate for the n qubit operation is provided as of
wn = v= ZuC-D'y (?)
The final stage of the proposed model deals with the decryption of the encoded message with the help of the security key. This process is just the reverse process of the encryption, and the unitary operators are of rotation operator, H-gate, C-NOT gate and the shared key parameters are used for decrypting the original message from the quired data.
The secured data is converted from the encrypted signals to the random like noise signal with low correlation analogue technique to provide the secured speech data on any social media platform. With this proposed model, the speakers or the users will be happy to use any social media platform with the confidence that their data are secured over any unsafe networks and even the effective result of the speech data protection model along with the wider keyspace and higher key security can be achieved with the help of the privacy preserving machine learning