Another common scenario occurs when ripping CDs. A technique of identifying the content involves gathering the characteristics of the CD and looking up those characteristics on the Internet. This technique usually consists of looking at the number of tracks on a CD and the length of each track. A unique signature is generated from these parameters and sent over the Internet to a central database to look up the CD information. This technique is 99.9 percent accurate, given that the content is in the database.

However, there are several drawbacks to this technique. First and most obvious is what happens when you are not connected to the Internet database. The content of the CD is ripped into some other format and potentially moved to a different structure other than artist-CD structure. The association of the CD to its content is lost, and this technique now falls apart. The end user is now required to associate the content manually.

There are lots of other scenarios in which the content association can and will get lost. So why is this important? A consumer electronic device known as a Digital Media Adapter was introduced into the market a few years ago. This device had high expectations and had a huge upside in the consumer electronics market; however, a major concern with its initial introduction into the marketplace was its high return rate. In some cases, the return rate was as high as 50 percent. This is in contrast to televisions, with a return rate of somewhere around 3 percent. A recent study cited the reason for the high return rate was not defects, but "ease of use": not meeting consumer expectations. Consumer ease of use is a big part of the consumer electronics experience. From the consumer’s perspective, a device has to be simple and just work. A 50 percent return rate is not a viable option to sustain a business in the competitive consumer electronics market. So what can be done to improve the user experience? That question is the focus of this paper.

Name that Tune
When I was growing up, my sister and I used to play a game called "Name that Tune." We would change radio stations and race to see who would be able to name the title and artist of the song. She would usually win. The reason this is relevant is because of the technique we used when associating content to the "information." In the game, there was nothing but the music source (radio) and a local database (in this case our memory) to associate the content to the artist and title. If we didn’t have it in our "local database," we would have to wait until the DJ "populated" the information by announcing the artist and or title. This was kind of like a deferred remote lookup. The winner of the game was usually the person who could associate the least amount of content (in this case the fastest) to the information in the local database. As time went on, the ability to assimilate the deferred lookup into the local database played a significant role in "winning." The same is true for audio content recognition. The source of the content should be encoding- and format-independent. The local database should be independent of content but easily associated using a standard method. The local database should have the capability to do a "deferred lookup" based only on content.

Content recognition
"Content recognition" is a vague term used throughout this paper, so perhaps it is time to make the phrase less ambiguous. Perhaps a better term is "content identification." In this context, the term "content recognition" means the "ability to uniquely identify the audio content based on the lowest and common form of audio content, the digital pulse code modulation format, and the ability to come up with a fairly unique identifying signature." The quality of content recognition is easily measured by the accuracy over a fairly large sample size, the size of content being compared, as well as the size of the compared-to content. These parameters, as well as the speed in which the identification, or recognition process, takes place will all play a role in the overall user experience.

Read the rest of this whitepaper: click download, below.

© 2005 Intel Corp.