39.2 Producing Digital Talks for the Web

The basic design process of the web-based DAGS'95 proceedings is very similar to the one for the DAGS'92 CD-ROM proceedings, although some details had to be refined due to improved video processing technology, limitations in network bandwidth and previous experiences with DAGS'92 and DAGS'93. Not much could be done to improve the initial information gathering and documentation steps, as this is still a mostly manual process, although we noticed that most of the speakers were now using presentation software to prepare their slides or transparencies, which made it easier for us to digitize their material. We just had to make sure to collect the material not only on paper or foils, but also in digital format.

http://town.hall.org/university/network/estrin

Due to the limited network bandwidth, special care had to be taken not to annoy modem users by forcing them to download huge sound files. When looking at other digitized talks on the web, as, e.g., the talk by Deborah Estrin, we noticed that the sound track was offered as one large sound file per slide. We chose a different approach. Because of the size of the sound files (an uncompressed 30 to 45 minutes sound file digitized in low quality 8kHz "AU"-format can easily take up to 40 MBytes) it was obviously impractical to store entire files. Rather we cut the files into 200-300 KByte pieces. Restricting the average file size of the audio clips to 200 kilobytes implies that transmission via a 14400 baud modem should generally take less than two minutes. This means, that frequently the sound track belonging to one slide had to be broken into multiple sound clips. Given the common transfer rate achievable for modems we nevertheless expect modem users to employ the "listen to audio" feature very selectively[31]. Avoiding chopped sentences required intensive human postprocessing, defining all sound-clip-boundaries manually. The target duration of the available audio clips was set at 40 seconds. As there were different factors to consider in the audio editing process (see point 4 below), the audio clip size varies considerably.

Figure IV.23 The creation process for producing digital talks for DAGS 95

The numbering in the following sequential description of the creation process corresponds to the graphical representation of the process in figure IV.23.

Collect material:

Each talk was filmed with HI-8 video. Based on previous experience we made a movie not of the speaker, but of the slides or transparencies. The audio input was taken from a microphone that speakers were requested to carry.

Speakers had previously been asked to give us a slide set after the talk, preferably both in paper and electronic format. Slide originals were collected from the speakers after the talk. Last minute corrections or annotations by the speakers could also be included this way. The slide formats included hand-written slides, color transparencies and files produced by presentation programs.

Digitize material:

Transparencies were scanned and screen-shots from the speakers software-based presentations were converted to images.

Basic editing and indexing:

The videotape was used to capture the audio, i.e., the speakers voice, and to provide synchronization between the speaker's voice and the slides. Slides were ordered based on the video of the talk. The audio track of the video was digitized using SoundEdit(TM) and Adobe Premiere(TM). The samples were manually cleaned from noise and superfluous "uhm"s and "ahh"s using the same applications. Unused slides were omitted while some others were concatenated into one image.

Semantic editing and indexing:

The sound track for each slide was cut into 30 to 50 second sound clips. The audio edit guidelines were (sorted by decreasing priority) (1) to keep slides and audio-clips in synch, (2) to let speakers complete a sentence, and (3) to have sound clips of at most approximately 300 KBytes. Sound samples that were shorter than 20 seconds were either added to the previous or next sample.

Create HTML and auxiliary files:

Once the slides were available as sorted GIF files, we used a HyperCard stack (named "AWCF") to automatically generate the HTML, GIF control panel description, and imagemap files based on the listing of the slide titles per talk. This step is described in detail in the next section.

Create images for control panels:

In the next step the GIF control panel image files had to be generated based on the description file generated in the previous step. We used DeBabelizer(TM), a commercial Macintosh graphics program that allows to create and edit collections of graphics files in batch mode.

Integration and testing:

In the last step we linked together all the files, placed them on the server and tested the validity of the links.