Session 3220

Production of Digital Internet Video Material for Streaming Applications

Z. Chambers, M. B. Taylor, J. Iannelli and A. J. Baker University of Tennessee Knoxville, TN 37996-2030

Abstract

The rapid growth of Internet-based teaching curricula has prompted a new direction for distance education - the streaming of live video lectures to remote student sites for on-demand education. This live material is exceptional while the post-processed static files are better than nearly all currently produced streaming video formats. The necessary compression software and computer hardware is readily available and surprisingly inexpensive. The development of quality video material, however, is a time-consuming process which requires both technical savvy and an artistic touch. This paper therefore provides a detailed recipe for creating digital video material for streaming applications.

Introduction

The emergence of the Internet, and in particular high performance communications, has allowed the traditional classroom-based educational environment to transcend to a multimedia, computer- driven venue, admitting global students whose only participation requirement is a modern computer with an Internet connection. The deployment of synchronized audio and video to support a developed static curriculum, i.e., HTML and PDF documents, fosters a sense of “human-ness” to the remote user while allowing an entire curriculum to be “taught” at any time. The advent of streaming compression technology has removed the restrictive file-size limitations of previous video compression-decompression algorithms (codecs). A one hour video lecture which is nearly 10 Gigabytes in uncompressed digital video form can be compressed to under 6 Megabytes and streamed to a remote user for real-time reception through a 28.8Kbps modem.

Basic recipe directions are as follows. The video is first recorded with a Canon Optura digital camcorder. Next, it is captured via an Osprey 100 video capture card to an 18 GB Quantum hard drive, connected to a Pentium II 333 PC, and hand-edited in Adobe Premiere. The audio track is exported, noise-reduced, and amplified for optimum clarity with CoolEdit. The audio track is then recombined with the video and compressed using the Indeo 5.06 codec. Having completed the first round of compression, the file is then archived onto a compact disc for future use. The file is then further compressed, using RealMedia’s propriety compression technology, for uploading to a Linux powered video server. Students are able to access the videos using nothing more than a standard web browser and the free RealPlayer plugin. Page 4.427.1 Discussion and Results

Step 1: Recording the Video Lecture To record the video lecture, three key elements are required: a high quality camcorder, a lavaliere microphone, and a well-lit whiteboard. The choice of camcorder is primarily dictated by budget - a $500 8mm camcorder will generate acceptable entry-level video while a $1500 digital camcorder will provide exceptional video quality. An important theorem is that final video quality is directly proportional to initial video quality, hence obtaining the highest quality camcorder is of utmost importance.

The Canon Optura digital camcorder was selected based on its excellent features, competitive price (WB Hunt, $1500), and pure digital video output. After experimenting with the built-in wide-area microphone, a lavaliere microphone (Radio Shack, $39.99) was incorporated to better focus the audio on the instructor and to minimize room noise. The final element, a well-lit whiteboard, required the construction of an incandescent light rack (Home Depot, $100) over the whiteboard to eliminate shadows caused by the regular fluorescent room lighting.

The actual recording of the lecture should take full advantage of the optical zoom features of the camera. The digitally enhanced zoom feature should not be used as it introduces video “noise” which is unacceptable for clean compression. The purpose of the camera is to relay the classroom experience to the remote user, not to focus on the instructor’s head. Equations and key notes on the board take precedence while the instructor should be in focus only for monologues greater than a minute. Zooming must be slow and fluid so as not to disorient the remote participant. To insure that written text is fully legible, the character height must be approximately 1/4 the height of the viewfinder. Finally, 15 seconds of silence should be recorded at the beginning of each lecture. This allows the room noise to be later filtered out, allows the instructor to prepare for the “take,” and permits the camera to get up to full speed internally.

Step 2: Capturing the Video Lecture to the Hard Drive Capturing the video to the hard drive has four requirements: a video capture card, a sound card, an audio/visual (AV) rated SCSI hard drive, and a CPU powerful enough to convert the torrent of digital video data into a Windows AVI format and write it to disk. We first experimented with a Radius MotoDV capture card (Radius, $420) because it allowed for the transfer of pristine digital video data via its IEEE 1394 “Firewire” interface. Unfortunately, the 740x480 pixel window size of the captured video, while phenomenally clear, required nearly an hour per minute of video for the first round of compression. The Osprey 100 capture card (RealNetworks, $200) was tested next. Although the digital-to-analog-to-digital conversion process introduced noise into the video signal, the 160x120 pixel window size required only three minutes per minute of video for the initial compression. The Osprey card was therefore selected for this project.

The sound card is responsible for the capture of the audio component of the video track. The SoundBlaster AWE64 ProGold card ($149) has performed excellently - do not settle for anything that can not handle at least 16 bit sound at 44KHz. Page 4.427.2 An AV rated hard drive is one which is guaranteed to handle the rigors of Audio-Visual recording. Specifically, these hard drives to do not pause to perform thermal recalibrations. Two Atlas Quantum III 18 GB hard drives (NetExpress, $1200 each) were selected to handle the projected 10 GB per hour lectures. Over the last three months, these drives have been reliably filled and erased on a near daily basis. Because digital video (DV) is not amenable to direct editing, it must be converted to either a Windows AVI or a Macintosh MOV file as it is captured. A Pentium 333 PC with 384 MB RAM (Gateway, $3500) easily handled this chore.

A critical obstacle in the acquisition of the video data was the 2 GB filesize limit inherent to the AVI filetype. Thus, only about 13 minutes of video could be recorded before reaching the filesize limit. This in turn required careful hand editing in Adobe Premiere 5.1 to splice the files together. A 15-second overlay between segments insures a good splice point.

Step 3: Editing and Archiving the Video Data As stated previously, Adobe Premiere 5.1 (UT Computer Store, $383) was used to splice the video segments together1. Having reassembled the video, the audio track was exported as an AVI file for post-processing with CoolEdit 96 (CoolEdit, $50). The 15-second room noise sample was used in an FFT algorithm to eliminate those offending frequencies from the lecture. In the author’s opinion, this is the most valuable of all post-processing steps. The cleaned audio was then visually inspected for aberrations - noise spikes due to dropping a marker, clearing one’s throat, closing a door, etc. These aberrations were isolated and their volume reduced to match the average lecture volume. The entire track was then amplified to 95% of the soundcard’s internal cutoff. The post-processed audio track was then recombined with the video track in Premiere and the entire lecture was compressed using the Indeo 5.06 codec.

Multiple layers of compression are traditionally bad practice. The losses incurred by compressing a video once are frequently revealed when it is compressed again. However, the first round of compression was necessary due to the 2 GB filesize limit. The Indeo 5.06 codec with 100% quality compressed the 10 GB file to approximately 500 MB with minimal incurred losses. The advantage was twofold - the source file was small enough to handle with Windows and it could be burnt onto a compact disc for archival purposes. Our CD burner, a Yamaha 4260 CDRW (NetExpress, $420), has proven quite reliable for this procedure.

Step 4: Final Compression The final step is to compress the 500 MB file using RealMedia’s propriety compression technology2. Here another theorem is appropriate: increasing the quality of the video increases the bandwidth requirement, i.e. the type of Internet connection, which decreases the potential viewing audience size. Thus, encoding a video for reception through a 112K dual ISDN connection may look great but will be unreceivable by all remote viewers with a 56K dial-up connection. It was therefore advantageous to encode for several bandwidths - 28.8K modem, 56K modem, and 112K dual ISDN - and allow the remote viewer to choose their desired connection*. We have found that setting the optimized frame rate to maximum sharpness and the audio to mono yields superb results. Page 4.427.3 *Note that the recently released Real G2 server automatically adjusts the signal bandwidth to accommodate the user’s Internet connection. Step 5: Upload to Server It is suggested that a separate machine be acquired to serve the RealMedia files. If live broadcast is desired, a dedicated server is required. We used a 333Mhz PII computer with a 9 GB SCSI hard drive, 128MB of memory, and a 10 MB/sec Ethernet connection running Debian Linux for this purpose. This hardware was more than adequate with the exception of the small hard drive size. An additional 18 GB SCSI external drive was purchased but, for small audiences, EIDE performance is adequate and yields a more cost-effective storage solution.

Samba file serving software3 was installed on the RealServer machine and configured to make the directory containing the RealMedia files appear as a networked drive on the capture/compress machine. Transferring a RealMedia file from the capture machine to the server machine was a simple drag and drop mouse operation. Once the RealMedia were moved to the server, remote students could select a lecture for viewing by clicking a password-protected web page link to launch the free RealPlayer utility.

RealServer runs on nearly all operating systems. We chose Linux4 because it is very stable, free, and runs on inexpensive hardware. Once installed, RealServer ran continuously for the entire semester, except for one down weekend due to a power failure.

The RealServer installation manual was clear on how to start RealServer interactively. However, it is preferable to run RealServer as a daemon, started automatically at boot. A short shell script was written to start and stop RealServer, combining information from Real's documentation and Debian's. The script was placed in the /etc/init.d directory and update-rc.d was run with a sequence number of 40. update-rc.d automatically updated all the rcx.d scripts, so that RealServer is automatically started and stopped in the proper sequence at boot and shutdown.

Summary and Conclusions

Internet streaming video marks a turning point for modern education endeavors. The use of video lectures to support static material makes “on-demand education” a reality to be capitalized upon by academia and industry alike. The key component of on-demand education is the uninterrupted reception of clear audio and video with low bandwidth requirements. Meticulous deployment of high quality AV equipment, desktop computers, and contemporary editing and compression software allows even the novice user the ability to create TV quality video for a modest time and financial investment. This paper demonstrated and described tools that can be effectively combined to yield superior on-demand video for Internet delivery.

Acknowledgments

This research and development could not have been performed without the generous support of the Tennessee Governor’s School for Manufacturing and the Condra Collaboratory / CFDLab. We are grateful to Pat Watson, of UTK Telephone and Networking Services, who got us started with RealMedia and provided us with exceptional network support. We are grateful to Tom Murrel of Tennessee High School, Bristol TN, and Dr. Terry Walker of Louisiana State Page 4.427.4 University, for providing valuable feedback on the video from a remote user’s perspective. Finally, Mr. Chambers and Mr. Taylor would like to thank Dr. Iannelli and Dr. Baker for their dedication to education and their perpetual desire to make all levels of learning accessible in this revolutionary Internet environment.

Bibliography 1. Adobe Premiere User Manual 2. RealPublisher and RealServer User Manuals: http://www.real.com 3. Samba Website: http://www.samba.org 4. Debian Linux Website: http://www.debian.org

Example Distance Education Websites 1. Finite Elements for the Engineering Sciences http://cfdlab.engr.utk.edu/551w 2. Tennessee Governor’s School for Manufacturing http://www.engr.utk.edu/~gschool 3. Theoretical Aerodynamics http://cfdlab.engr.utk.edu/AE422w

Z. CHAMBERS Zachariah Chambers is a Ph.D. student in the Engineering Science CFD curriculum. He holds B.S. and M.Sc. degrees in Mechanical Engineering from Rose-Hulman Institute of Technology, Terre Haute, Indiana. He has been actively involved with Internet based distance education at UTK and is exclusively responsible for the video content and HTML evolution of the above distance education websites.

M.B. TAYLOR Mike Taylor is a Ph.D. student in the Engineering Science CFD curriculum. He holds a Bachelor of Mechanical Engineering from the Georgia Institute of Technology, and an M.Sc. degree from the University of Tennessee Space Institute. He was a commissioned officer in the United States Air Force for twelve years, with the last assignment as an Assistant Professor of Aeronautics at the Air Force Academy. He has configured and maintained the RealServer machine and provided excellent computer support to this video processing project.

J. IANNELLI J. Iannelli, Ph.D., is Associate Professor of Mechanical and Aerospace Engineering and Director of the Tennessee Governor’s School for Manufacturing. He joined the faculty in 1991 and has also been involved in the development of computer laboratories for engineering mechanics. His research area is in Characteristics Finite Element Methods for CFD and has published several articles in CFD. Mr. Chambers and he have pioneered at UTK the use of Internet-based distance education programs and developed the first Internet Governor’s School for Manufacturing in the United States.

A.J. BAKER A.J. Baker, Ph.D., PE, is Professor, Engineering Science, and Director of the CFD Laboratory at the University of Tennessee/Knoxville. He joined the faculty in 1975, following a research stint in aerospace industry, with the specific goal to develop the graduate curriculum in computational fluid dynamics and heat transfer. He has authored more than 240 technical papers on the subject, including 60 archival journal articles, and has published two textbooks, each with international editions and one Japanese translation. The CFD program matriculates an average of one each Ph.D. and M.Sc. degree per year, and currently has six graduate students at various stages of degree completion. Page 4.427.5