The Library is interested in investigating computer-file based storage and migration as a paradigm for the preservation of video and audio. (See Attachment 4 for a discussion of preservation paradigms.) A goal for preservation reformatting is to eliminate or minimize the loss of information when copying an original. Thus an ideal approach for computer-file preservation is the storage and migration of uncompressed files. For comparison, the Library's approach for the digitization of maps and photographs is to create and store uncompressed high-resolution master files on the order of 200 MB for maps and 60 MB for color photographs. (The Library is considering the use of non-proprietary lossless compression algorithms like LZW but has not implemented this.) Of course, compressed and/or lower-resolution versions of the maps and photographs are also produced for convenient access.
The Library's understanding of the field indicates that it is practical to create, store, and migrate uncompressed audio files, even if the master is the large file that would result from 96 MHz sampling and 24-bit word length. The practicalities for video, however, are less clear.
5.2 Uncompressed or Minimally Compressed Video
Uncompressed digital video means the "linear" bitstream referred to in (1) ITU-R Rec. BT.601-4 (often called "CCIR 601"), sampled as 4:2:2 and recorded on, say, a D-1 videotape, or (2) the bitstream sampled at "4fSC" (four times frequency subcarrier) and recorded on, say, D-2 videotape. The former is a digital component signal with a data rate of 270 megabits/second, while the latter is a digital composite signal with a data rate of 143 megabits/second. An example of minimally compressed digital video is the signal recorded by Sony's proprietary Digital Betacam format, which includes modest intra-field compression in the bitstream as recorded on the tape.
It is worth noting that most recordings in the Library's archive of historical video are in analog composite form, although many commentators have suggested that they be transformed to component when they are made digital. The Library is consulting with a number of experts in the field to explore this question and the practicality of computer-file storage of the bitstreams in the formats listed above. The sizes of the resulting files are in and of themselves a barrier to computer storage, at least in the 1999-2000 environment. A thirty-minute program in uncompressed digital component video has a theoretical extent of about 60 gigabytes (roughly 270 Mb/s times 1,800 seconds). Manufacturers have produced devices to support the transition to digital broadcasting that cater to component video but devices that cater to digital composite material do not appear to be available.
The question of which format to select for preservation reformatting is also challenging. Recorders for uncompressed 4:2:2 signals are very expensive and budgetary concerns may argue for minimally compressed recordings (e.g., Digital Betacam) instead of uncompressed (e.g., D-1). Digital composite recordings may have a shorter "life in the marketplace" (and smaller installed base of players) than digital component recordings, which strengthens the case for converting historical composite analog recordings to component when they are digitized, even though that may have a slight adverse effect on picture quality. The case for converting composite to component when digitizing is further strengthened by the fact that compressed formats like MPEG--needed when producing derivative versions for research access--must receive component signals as an input.
5.3 Delivery to Researchers
The challenges to the effective delivery of audio and video (even compressed video) over computer networks are well known. At the same time, the computer and online industry has a very high interest in overcoming these challenges and many initiatives are under way to find better ways to compress and stream audio and video at a variety of data rates. The Library plans to take advantage of these initiatives and wishes to use the Digital Audio and Video Repository Prototyping Project to continue its investigation of this matter, understanding that new approaches and options will become available every year.
It is worth noting one approach for the delivery of moving-image material that could be employed for this project: "decode" digital video at or near the storage site and then to transmit an ordinary video signal to the researcher, like that carried by cable system or broadcaster. This approach would be workable here because there will be only a small number of workstations and each workstation could be equipped with both a video and a computer monitor. Today, the approach would employ analog video signals. Tomorrow, the transition to digital transmissions could be made, in step with the industry's own transition.
5.3.1 Researcher Workstations
The Library foresees that cost factors will lead to the offering of varying levels of quality to of researchers at different workstations. The constraints will be most sharply felt in the case of video materials because of their high data rates. However, the shortage of soundproof listening rooms in which loudspeakers can be erected will also constrain the design of researcher workstations for audio. Thus, content requiring (1) high bandwidth on the network and high-end hardware and software or (2) loudspeakers may only be offered in a few locations in the Library. Other workstations will be provided with lower-resolution versions of content and use such devices as headphones.
In the future (post-Prototyping Project), the Library foresees offering researchers a small number of high-performance workstations and a larger number of moderate-performance workstations. During the Prototyping Project, the Library anticipates that from two to four high-performance workstations and an indefinite number of moderate-performance workstations will be established.
5.3.2 Researcher Interface
What might the interface on a high-performance workstation look like? For video, will there be two monitors: a video monitor for the program and a computer monitor for the bibliographic (cataloging) information and a "remote playback controller?" Will a thumbnail of some kind be displayed with the catalog record and, if so, what will an audio or video thumbnail look like?
What playback controls will be desired and which of these will be in practical reach? Preliminary discussions at the Library have highlighted interest in some of the following capabilities:
Indiana University has developed a fully realized research-access system, including input, storage, and research access features, for students of music.
5.4 Video Conventional Media Output Workstation
The Motion Picture, Broadcasting, and Recorded Sound (M/B/RS) Division laboratory regularly responds to requests for copies of audio and video recordings from authorized purchasers. Typically, these are producers who wish to incorporate historical recordings in radio or television programs, audio compact disks, and the like. The copies provided to these purchasers must be made from the highest quality (preservation quality) masters available to the Library. The Library anticipates that the delivery of audio will increasingly be accomplished by delivering a file, say a WAVE-format preservation file, on a CD-ROM or other media. Since this can be easily accomplished, the issue to be addressed in the Prototyping Project concerns the delivery of video.
Video purchasers require the delivery of content in a conventional format like a videotape, of a type and at a quality that permits the insertion of segments into the purchaser's newly created program. If the preservation master files are in the repository, then the Library requires a means to write this data back to conventional media. In effect, the device or devices needed represent ingestion in reverse: instead of playing a videotape to put the program into the computer, the computer writes/records the program back out to a videotape.
5.5 Production of Derivative Audio and Video Versions
Derivative or "browse-quality" versions are files at lower resolution and requiring less bandwidth than "master" or preservation-quality versions (see also Attachment 1). Increased ease of accessibility has made derivative versions a part of most digital information access systems, including the American Memory WWW collections of manuscripts, printed matter, maps, photographs, audio, and video produced. American Memory collections are produced by the Library's National Digital Library Program (NDLP). In American Memory, for example, a book-page master image is typically a 300 or 400 dpi bitonal TIFF image compressed by the lossless ITU Group IV algorithm. The derivative book-page file for presentation in web browsers is typically a grayscale GIF image, scaled to about 450x700 pixels for easy, one-direction-scrollable screen display.
Derivative or lower-resolution audio and video files may include some of the following types:
In the long term, there are many advantages to producing derivative versions on the fly: the need for storage and maintenance is reduced and only items in demand need receive this treatment. But as a practical matter, it is not always possible to create derivatives in real time. The CPU power of a server might be strained or the system might otherwise become clogged. For a variety of reasons, the Library's experience with derivative-making thus far has not included production on the fly, although the institution has high interest in the topic.
Thus far, the Library's derivative versions have been produced in advance of need. Although specialist contractors have produced virtually all of the master files, the NDLP has taken a varying approach to producing derivatives. In some cases, the most efficient and cost-effective approach was to have the specialist contractor deliver both the master and derivative versions. In other cases, it was efficient for the Library to produce the derivatives itself. Typically, this was done by running a batch process on the master files after they had been inspected, approved, and loaded into the server. For example, the software Image Alchemy can quickly produce a large number of GIF page images within a UNIX directory.
What will be the best way to produce derivatives from audio and video master files? In the modular design proposed for the Prototyping Project repository, this question may be rephrased in this way: