Feb 23, 2008

Searching Podcasts

In stark contrast to text documents, which are easily searched — Google just looks for words in documents, after all — podcasts are difficult to search because they must be transcribed by man or machine first and then searched. The science of transcription is young and, based on our experience with companies claiming to search audio and video, partially an act of smoke and mirrors. Most of the search tools we've reviewed look only in metadata — the text associated with a podcast — rather than searching the content of programs. In many cases, people who are as inexact or error-prone as machines are involved at some point, and almost always there is a very spotty result. As computational power increases, this will improve. These are the major players:

- PodScope: This site searches both audio and video on the Web, mostly podcasts. Our results showed very poor results, where, for example, a search for Microsoft Vista (without quotes) yielded only three results while a search for "Vista" returned dozens of results, many referring to Microsoft. Some unique neologisms from podcasts, such as "Ninternship" from the popular Ask A Ninja podcast, were not found at all. http://www.podscope.com/

- PodZinger: A much more complete index, PodZinger is offered by BBN Technologies, a commercial provider with a long history. It purports to search almost 300,000 podcasts at this writing and returns much more complete results for common words, like Microsoft Vista and Vista — more than 1,000 each — while failing to find unique words, such as Ninternship. http://www.podzinger.com/

- Pod Razor: The service is obviously using metadata describing shows rather than transcribing the shows themselves before searching. http://www.podrazor.com/

- AOL Search Podcast Beta: This has all the hallmarks of beta software, but the results are impressive, if sparse, because they return results from within the body of shows. Keep your eye on this one. http://podcast.search.aol.com/

- Digital Podcast: An interesting effort in podcast search, we found the results poor, but the excerpts of programs displayed were more comprehensive than many other search providers delivered, even if there were lots of extraneous characters in the results indicating imperfect machine translation. http://www.digitalpodcast.com/

No comments: