Biometric Authentication - Research Paper Example

? Voice Whispered Biometric Authentication Smartphones contain a collection of personal data, and also provide access to even more data that is stored on the internet. Today most mobile devices provide password-based authentication, which is becoming more vulnerable. Recent advances in biometric authentication make it a viable alternative to password authentication, and voice biometric is one of them. However, voice biometric authentication still present weakness when the user presents changes on the voice. Whispering is an alternative speech production used by many to protect their private conversations. However, research has shown that the mismatched identification in a whispered speech is due to the phonemes hence providing an unvoiced consonant will prevent this mismatched in whispered identification. Key Words- Voice, Phoneme, Biometric Authentication I. RESEACH QUESTION/PROPOSAL A feature information of enrolled person’s biometric information in biometric authentication is normally enrolled as templates that are stored by secure databases and anti-tampered devices. However, the disadvantage is that biometric information is irrecoverable once compromised. Therefore, there is need to give such information special attention in order to ensure they are safely stored. Also due to increased internet penetration and access by many people has raised a need for the implementation of online biometric authentication i.e. a uniform, secure and reliable method of biometric authentication. Therefore, the research proposal in this case is to identify alternatives for the password-based authentication used in most smart devices today. This kind of authentication is vulnerable and can be compromised hence a proposal to focus and introduce better authentication ways i.e. biometric that uses one’s unique biological and psychological features. Hence the research question is; can Voice and Whispered Biometric Authentication provide a more secure way for ones data in smart device than the current password-based authentication? II. INTRODUCTION Biometric authentication is the automatic identification of living individuals by using their physiological and behavioral characteristics. When a password or PIN is forgotten or lost, it can be changed but a biometric identification cannot. Mobile devices are being stolen daily, which puts in danger important information that had been stored before. [1]. Therefore, a user will hope that his/her password is strong enough not to be decoded. Biometric authentication is giving a natural alternative to passwords. The wide variety of input sensors that mobile devices include nowadays like microphones, camera, PS, touch screens helps with the implementation of biometric authentication. Businesses are requiring an easy to use but secure authentication for their mobile devices in their corporations as well users around the world. Many of the biometric authentications are strong, but others have their limitations. [2] This paper will discuss and will show an improvement in the authentication of voice recognition. I. BACKGROUND Biometric can be classified in to two major categories based on their characteristics i.e. Behavioral Biometric and Physiological Biometric. Behavioral biometric is the study of the differences in how people do things while physiological biometric is the study of physical uniqueness of individual, which is unlikely to change very easily [3]. A. Importance of Voice Authentication Most of the biometric authentications works with sophisticated equipment that requires the physical presence of the person. For example, Retina biometric authentication needs a camera; which are contained in most smartphones. TABLE 1 PHYSIOLOGICAL BIOMETRIC AND BEHAVIOR BIOMETRIC Physiological Biometric Behavioral Biometric Voice Scan Iris Scan Finger Print DNA Matching Facial Scan Retina Scan Hand Scan Keystroke Scan Signature Scan Gait Recognition However, a Charge Coupled Device (CCD) which is complement with ANSI/IESNA RP-27.1-96 and IEC 60825-1 standards for radiation have to be included on the device for the retina biometric authentication to work [4]. On the other hand, voice biometric authentication would require a microphone, which it is already included on a phone. As well as the other biometric characteristics such as fingerprints, and retina; voice recognition have the same uniqueness on a person voice. However, voice biometric authentication has the advantage that individuals do not have to submit themselves to a retinal or fingerprint scans. The process that takes for voice recognition is actually really simple. First, a user has to record a sample of their voice, and that sample will be stored in the "authenticating system" and it will be known as a 'voice print'. After this, every time the user wants to access their device they will have to provide a sample of their voice, and this sample will be passed into the authenticating system which will determine if it match or it does not match. If the sample matches when it passed through the authenticating system then it will provide access to the data on their device. B. Voice Production Voice production is made when the air passes over the larynx, which will make the larynx vibrate and created acoustic wave. This acoustic wave produced sounds that are modified by tongue, and lips. This sounds is called phonated or better known as voiced sounds. Also, other sounds can be produced by other parts of the vocal tract. For instance, when a person whispers the sounds that are created by air goes over the arytenoids cartilage at the back of the throat. The sounds that are produced will have great effect depending on the shape of the tract. [5]. So far we have seen that voice/speech is a physiological biometric characteristic, but it also contains a behavioral biometric component. This behavioral component includes voice accent, pronunciation, and how fast the words are spoken. Furthermore, physiological and behavioral biometric will make voice patterns that are unique making them hard or even impossible to imitate their voice. II. VOICE AUTHENTICATION Knowing now how voice is produced and how we transfer sound wave produced by the vocal tract into computer data. Now we have to look how this data does is used in authentication. The user will be asked to give a sample of his/her voice; in this case a sample of their whisper, and this will be compare with the whisper print. There are several ways for the authentication system to recognize and register the voice of the users. The first method is presented when the user says the command word once. In this case the system will try to record as many low tones as it can in order to have a template of that word. Once the user says that word, the system will compare the registered template with the word being said. A second method for recognizing the command is seen when the user says the command word more than once. The repeating of the command will register the different tones and ways the user pronounces that specific command. This method is more effective than the previous one since it allows the system to have a broader range of recognition, especially when the user pronounces the word in a different way than the other one previously recorded. A way to decrease this vulnerability is by making use of more sophisticated systems that record longer voiceprints by speaking or repeating sentences. This recording is known as a general voiceprint. Adding to this process, the user must say a random word to authenticate the sentence previously said. The only disadvantage of this method is that the general voiceprint template will encounter difficulty-making use of Template Matching. [5] In this case, the general template cannot be compared with a random phrase that have the same reliability and identical registration in the voiceprint. The reason for this is due to the complex mathematical processes involved in this comparison. The general voiceprint will then be less reliable than the specific voiceprint method. A. Problems with Voice Biometric and Solution Even though voice biometric authentication is simple and secure, many times the system will reject the match. Comparing the sound wave coming from a person that lose their voice would be different from the voice sample sound wave stored in the database device. The loss of voice could occur from many reasons. There is no precise solution for this, since we cannot substitute the voice. However, every human being has the ability to whisper. Whisper is consider as an alternative vocal effort style from neutral speech which can be employed between speakers when the speaker may consider the information to be personal. [6]. Most of the people may use whispering as an alternative way of speaking to protect a private conversation. Also an individual that speaks with low vocal capability, such as those who smoke, whispering is their main way of communication [7]. Whispering is not affected by any sickness or any other vocal problems. This is why whispering biometric is a good alternative to voice biometric authentication. However, some studies show that whispering speech has the most deficiency for speech processing system. [8] This mainly because in neutral speech, the vocal fold makes a periodic vibration that produces phonemes, which this control the air flow into the pharynx. On the other hand when someone whispers, the pharynx is adjusted in way so the vocal fold do not vibrate, and this makes the air stream continuous without any cycle. [7] To address unvoiced consonant there will be need to extract some features using exponential and linear frequency, and also the speakers will need to have an effective capturing of the voice. III. DIFFERENCE BETWEEN NEUTRAL AND WHISPERED SPEECH There are some few differences in the speech production between neutral and whispered speech. They can be classified into four main differences. Figure 1: Voice wave form shows a higher frequency called upper The waveform shows that in whispered speech there is no harmonic structure or periodic excitation than in the neutral speech. The position where the lower frequency is located in the whispered speech waveform, are shifted to a higher frequency, than the neutral speech waveform. The time duration in the whisper speech is longer than the neutral speech. Also, the spectral slope of whispered is flatter. There is a higher energy in neutral speech, and in whispered it is so much lower [7]. Figure 2: Voice wave form shows a higher frequency called upper Figure 3: This waveform Image can show a lower frequency and longer Figure 4: Spectrogram waveform shows the most energy on whispered IV. PHONEMES CONSONANTS ANALYSIS As it was shown previously there is a significantly difference between neutral and whispered speech when the acoustic characteristic of vowels and voiced consonant are spoken. However, there is a relatively similarity on the spectral properties of unvoiced consonant between neutral and whispered speech. In [7], it was shown an analysis on phoneme using only the read portion of the UT-Vocal-Effort II corpus. These consist of the separation of unvoiced consonant from all other phonemes in both neutral and whispered speech. For better understanding phonemes are also called voiced consonants, vowels, non-unvoiced consonant, diphthongs, etc. When a person whispered the voiced and unvoiced consonants are similar because of the absence of periodic excitation. The unvoiced consonant ‘to’ and the voiced consonant ‘do’ are some examples that are not separate. Fig.5 shows the waveform of 'to' first and 'do' second. Now, to know which frame belongs to unvoiced consonant there must be three measurements for each frame i: 100-400Hz, 4000-8000Hz, and 100-8000Hz from the total energy of the frequency bands. In (1) is used to calculate the relative symmetric entropy to compared the spectral structure of the nears frames in the high frequency Figure 5: Imaged compared first a Unvoiced and second a Non-unvoiced Where Which mean higher frequency over everything in the frequency, and this is due because most of the spectral energy is concentrated in the higher frequency domain. On the other hand, the energy of most the vowels are concentrated in the lower frequencies. Furthermore, a threshold can be set for both Rn and En to separate out unvoiced consonant frames [7]. In this analysis it was shown the effectiveness of the method to detected unvoiced consonants. Therefore; by using this procedure, unvoiced consonants in neutral/whispered speech can be obtained. V. SUMMARY AND CONCLUSION The importance of a secure authentication on mobile devices has become more and more important as mobile devices evolve to become one of the main computing platforms. Therefore, voice biometric authentication has been called to be as secure as possible to protect mobile devices from unknown individuals that try to hacked data on the devices. As it was previously mentioned, voice biometric authentication was not able to recognize the voice of the user when someone loses their voice, or because of sickness their voice changed. Whispered biometric is alternative speech production that is employed by individual to protect their private conversation, and even though an individual get sick whispering would not change. However, by analyzing neutral and whispered speech it has been shown that voiced consonants makes a difference between both of them, but if unvoiced consonants are used then there would be a similarity between neutral and whispered speech. Therefore these solution of this research paper is to provide easiest, private and user friendly methods of authentication. This is important in this era of increased use of smart phones and devices to store some sensitive information and also access this information over the cloud. Biometric technology is currently the safest and most accurate authentication technology since it provides a way of identifying living being using their unique identifier. The technology is currently used in most harbors in developed countries and at airports to control access to certain areas and facilities. References [1] S. Trewin, C. Swart, L. Koved, J. Martino, K. Singh, and S. Ben- David, “Biometric authentication on a mobile device: a study of user effort, error and task disruption,” in Proceedings of the 28th Annual Computer Security Applications Conference, ser. ACSAC ’12. New York, NY, USA: ACM, 2012, pp. 159–168. [Online]. Available: http://doi.acm.org/10.1145/2420950.2420976 [2] H. Sieger and S. Muller, “Gender differences in the perception of security of mobile phones,” in Proceedings of the 14th international conference on Human-computer interaction with mobile devices and services companion, ser. MobileHCI ’12. New York, NY, USA: ACM, 2012, pp. 107–112. [Online]. Available: http://doi.acm.org/10.1145/2371664.2371685 [3] M. Sujithra and G. Padmavathi, “Next generation biometric security system: an approach for mobile device security,” in Proceedings of the Second International Conference on Computational Science, Engineering and Information Technology, ser. CCSEIT ’12. New York, NY, USA: ACM, 2012, pp. 377–381. [Online]. Available: http://doi.acm.org/10.1145/2393216.2393280 [4] M. Dunker, “Don’t blink: iris recognition for biometric identification,” November 20 2003. [5] D. Currie, “Shedding some light on voice authentication,” 2003. [6] A. R. Vijaya Santhi Nellore, Polasi Phani Kumar, “Whispered speech identification using lfcc feature extraction algorithm,” D. of Electronics and V. Communication, Velagapudi Ramakrishna Eng. College, Eds., vol. 2, December 10 2012, pp. 124–129. [7] X. Fan and J. Hansen, “Speaker identification within whispered speech audio streams,” Audio, Speech, and Language Processing, IEEE Transactions on, vol. 19, no. 5, pp. 1408–1421, 2011. [8] C. Zhang and J. Hansen, “Advancements in whisper-island detection using the linear predictive residual,” in Acoustics Speech and Signal Processing (ICASSP), 2010 IEEE International Conference on, 2010, pp. 5170–5173. Read More