• be available for the full duration of the e-learning program,
• cover all of the topics in the e-learning program, and
• complete all the assignments for the true student.
• physical state (e.g., head cold, tiredness),
• mental/emotional state (happiness, nervousness, depression, etc.), and
• other long-term changes due to aging and physiological condition.
1. The system is installed and commissioned by appropriate technical staff, which is easily accomplished by staff with general software skills.
2. Annually, the authentication system is updated (by the technical staff), in consultation with administrative staff, to update the list of students (list of current students or those whose applications have been accepted).
3. Upon registration, each student on the ME program is required to deposit a "voiceprint," consisting of a standard phrase (see Section 7.2), in the student database. The voiceprint logging is facilitated by automated response software, as described in Section 5.1, and records are stored for use in conjunction with the student authentication system. Note that each student's registration is not deemed to be completed until their voiceprint has been successfully logged, sending a clear message to students that plagiarism is taken seriously and counteracting measures are in place. Note that the rationale behind the logging of the voiceprint is clearly explained to students.
4. Preprocessing is now carried out (offline), via the execution of a macro utility, on the student voiceprints to minimize the wait time during the use of the live authentication system. The preprocessing procedure is articulated in Section 6.
The system is now ready for use.
5. In the ME program, each module is coordinated by a "module coordinator" (MCO) who has overall module responsibility in the case of multiple instructors on the module. Each MCO makes an assessment of the risk of plagiarism for that module, based on:
• the nature of the material,
• the degree to which group assignments are employed, and
• the degree to which assignments are individualized.
Based on this assessment, the MCO establishes a sample size appropriate for oral testing. In any event, a minimum of 1/10th of the student enrollment for each module is recommended for oral testing.
6. Oral testing, with the support of the student authentication system, is carried out in the last quarter of the semester (each teaching semester is 12 weeks long).
7. For each oral test, the call is initiated by the MCO at a time agreed with the student. Following the initial contact, the student is switched to the voice authentication system, which requests a voiceprint (using the same standard phrase as in Step 3). The voiceprint is then preprocessed and submitted to the classifier (see Section 8), which returns a level of confidence (LoC) that the student being examined is the same student which registered for the program. The computation due to preprocessing and classification may take up to 15-20 seconds. Following the authentication result, the MCO should proceed as follows:
• the difficulty in getting information on the exact Ericsson protocol used on the digital lines,
• the long lead time and comparatively high installation and rental costs on a dedicated (true) analogue line, though this would provide better quality, and
• the flexibility of using an analogue line with a range of PC telephone interface cards.
• the student's name,
• a sequence of numbers, and
• phonetically-balanced phrases.
• the name of the user,
• the call number for that user (note that users can call multiple times),
• the type of phone channel being used,
• the nature of the speech segment, i.e., name, number, or phrase, and
• the repetition index for that utterance,
• easy to measure,
• stable over time,
• high interspeaker variation,
• low intraspeaker variation, and
• robust against noise and distortion.
7.1.1 LP Cepstral Coefficients Linear predictive coding (LPC) is also based on the source-filter model, with the filter constrained to be an all-pole filter. The analysis performs a linear prediction so that the next sample is predicted by using a weighted sum of past samples:
where is the predictor order and are the filter coefficients, which can be calculated using correlation analysis. In practice, raw LPC coefficients are rarely used as features due to the high correlation between adjacent coefficients. Instead, complex cepstrum coefficients are often used, which can be computed easily from the LP coefficients using
7.1.2 Mel-Frequency Cepstral Coefficients Another popular technique for extracting useful features from speech is to use a filterbank-based cepstral representation. A bank of 15 to 20 channels, or bandpass filters, whose bandwidth and spacing increase with frequency, is generally used, motivated by studies of the human ear. The filterbank represents power logarithmically, which is of phonetic significance—the lower formants are emphasized more. The distribution in each of the channels tends to be Gaussian [ 9]. The locations of the center frequencies of the filters are given by:
The mel-frequency cepstral coefficients (MFCCs) are obtained using the following procedure:
1. the FFT is applied to the signal, or a windowed version of it,
2. spectral power values are then mapped onto the mel scale using (4),
3. the logarithm is taken of the spectral powers, and
• B. Hayes is with Intel (Ireland) Ltd., Leixlip, Co. Kildare, Ireland. E-mail: firstname.lastname@example.org.
• J.V. Ringwood is with the Department of Electronic Engineering, NUI Maynooth, Maynooth, Co. Kildare, Ireland. E-mail: email@example.com.
Manuscript received 11 Nov. 2008; revised 30 Dec. 2008; accepted 5 Jan. 2009; published online 8 Jan. 2009.
For information on obtaining reprints of this article, please send e-mail to: firstname.lastname@example.org, and reference IEEECS Log Number TLT-2008-11-0099.
Digital Object Identifier no. 10.1109/TLT.2009.2.
Barry Hayes received the electrical and electronic engineering degree from University College Cork in 2005 and the MS degree in electronic engineering from the National University of Ireland (NUI), Maynooth, in 2008. He is currently employed in industry as a semiconductor process engineer at Intel Ireland's headquarters near Dublin. He is a member of the Institute of Engineers of Ireland.
John V. Ringwood received the electrical engineering diploma from the Dublin Institute of Technology and the PhD degree in control systems from Strathclyde University, Scotland, in 1981 and 1985, respectively. He was with the School of Electronic Engineering at Dublin City University from 1985 to 2000 and, during that time, held visiting positions at Massey University and the University of Auckland, New Zealand. He is currently a professor of electronic engineering with the National University of Ireland (NUI), Maynooth, and is the associate dean for engineering with the Faculty of Science and Engineering. He was the head of the Electronic Engineering Department at NUI Maynooth from 2000 until 2005, developing the department from a greenfield site. He has acted as a consultant to a number of companies in the power, servomechanism, and process industries. His interests cover a number of areas, including time series modelling, control of wave energy systems, control of plasma processes, biomedical engineering, and e-learning. He is a chartered engineer, a senior member of the IEEE, and a fellow of the Institution of Engineers of Ireland.