Skip to main content

Metadata and MXF, part 1

This month, we will begin a two-part series on metadata and MXF. First, we will look at the importance of metadata to the professional media facility. Then, we will look at some of the specifics of metadata as applied to file interchange via MXF.

Most video professionals are now quite familiar with metadata. But, just to be sure everyone is on the same page, when I use the term metadata in this tutorial, I am referring to information (data) that typically refers to a program, commercial or other video/audio content. Examples of metadata include: the name of a program, the length of a segment or an ISCII code associated with a commercial.

Collectively, video, audio, subtitles, AFD and other things people consume as part of a program are known as essence. Just remember that essence refers to the stuff we are watching when we view a program. Importantly, essence includes data (subtitles, teletext, AFD and so on). If we want to talk about data that is part of the program, we talk about data essence. Data essence is not metadata; data essence is part of the program that is intended for “viewing” by the end consumer.

Technical vs. descriptive

Metadata tends to be treated as a single topic. But, in my mind, there are several different ways to classify metadata. One way to look at metadata is whether it is technical or descriptive. Technical metadata is metadata that is required to identify and play back essence properly. Examples include: a Unique Media Identifier (UMID), information about compression used and raster size. Descriptive metadata is typically the sort of information operators in professional media organizations care about — program title, show number, kill date, ISCII code and so on. Descriptive metadata might also include things like shot notes, scripts and music scores. Some metadata falls into a grey area. How would you classify GPS location, zoom and focus setting, crank rate, or robotic camera position? Is this technical metadata or descriptive metadata? Generally speaking, technical metadata is directly related to the essence or essence coding and does not include these data types.

Another way to classify metadata is to think about how frequently the metadata is expected to change. We would expect some metadata to remain constant for the duration of the content being viewed. For example, we would not expect the title to change halfway through a program. In the United States, we might expect that the technical metadata regarding frame rate would remain at 59.97Hz for the duration of the program. As you can see from these examples, the frequency of metadata change is an orthogonal classification (meaning it is completely unrelated) to the metadata type (technical or descriptive). Not surprisingly, metadata that remains constant is called static metadata.

Other metadata, such as timecode, would be expected to increase monotonically (increase by a count of one, only) on every frame. Timecode and other metadata may, therefore, be placed on a timeline that increases predictably, and should never decrease (run backwards) or unexpectedly jump ahead. This type of metadata might be called timeline-related metadata.

Finally, metadata associated with shot logging, describing actors in scenes or associating a music score to a section of a program may appear fixed for a period of time, may be duplicated in several locations during a program or may even overlap with other similar types of metadata. This metadata can be referred to as event metadata.

There is one other important family of metadata, and that is perpetual metadata. As you can guess, this is metadata that should never change. Systems rely on the fact that these metadata items are immutable. If something is immutable, it means that it cannot be changed after it is created. Examples of this include a unique ID and the location where a program was filmed. Remember, systems will count on the immutability of certain metadata. If you violate the rules, all bets are off.

Why does all this matter? Because, in the end, media professionals build and operate systems, and these systems need to be properly designed to handle all of these different types of metadata.

Metadata and essence

How do you tie timecode or program title to a specific copy of a movie? In the tape world, the question was irrelevant. The timecode was either stored in the VBI or in the ANC data, and the program title was printed on a label on the box or the tape itself.

But, things have changed in the file-based world. Associating metadata, especially technical metadata, to a particular program can be critical. Professional file formats must contain not only video and audio, but also the technical metadata necessary to play back the content. When it comes to descriptive metadata, the door was left open as to exactly what types of descriptive metadata were to be included in the file. This is both good and bad. One of the most critical linking mechanisms between metadata and essence is the UMID. The UMID is a unique identifier that is used to identify a particular piece of content.

UMIDs and the contract

It is important to understand that there is an implicit contract between systems that create content labeled with a UMID and systems that later manipulate the content or rely on the UMID to retrieve the content. Here is the contract: A system will never break the connection between a particular piece of content and the UMID that was originally associated with that content.

If you create a new piece of content, give it a new UMID. Simple, right? But, here is a twist: What if you create a bit-for-bit exact copy of the file? The rule is that you duplicate everything, including the UMID. But, now you have two copies of the content with the same UMID. Is this right? The answer is yes. And, if you ask a media asset management system to retrieve a piece of content using a UMID, you cannot say for certain which copy you will get. The particular instance of a copy of content is not uniquely identified by a UMID. For other reasons, a MAM system may uniquely identify each copy of the same content, but the UMID should not be used to do this.

In a file-based world, metadata becomes key. Understanding how unique identifiers work and treating them correctly is vitally important. Fortunately, manufacturers hide most of this from end users, so you may not even be aware that UMIDs exist.

Coming up

Next month, we will focus on why metadata is a critical part of file-based workflows. Also, we will explore issues with metadata rekeying in these workflows, and we will talk about how metadata contained in MXF files can reduce rekeying and errors. We will also discuss issues and solutions related to metadata interoperability in MXF files.

Brad Gilmer is executive director of the Video Services Forum, executive director of the Advanced Media Workflow Association, and is president of Gilmer & Associates.

Send questions and comments to: