Podcast Search Mission: Mostly Possible
If a tree falls in a forest and no one is there to hear it, does it make a sound?
Updated to the digital era, the Zen koan could be: If a word is uttered in a podcast, can it be found by a search engine?
The modern quandary, at least, now has an answer.
Two online services are using voice recognition technology to translate speech in podcasts -- as well as the latest twist on the form, video blogs -- into text that can be searched for specific words, names and phrases.
Neither of the services -- PodZinger and Blinkx -- is perfect by any means, mainly because voice recognition is still an inexact art, at best. This leads to bothersome gaps and mistakes in searches that are sometimes inadvertently amusing.
But without these free services, searches can be done only on the brief text descriptions that accompany podcasts or vlogs on the Web.
“It’s like trying to judge a book from its cover,” said Alex Laats, who oversees PodZinger for parent BBN Technologies.
Like text search engines such as those operated by Google and Yahoo, PodZinger (www.podzinger.com) and Blinkx (www.blinkx.com) continuously examine sites on the Web in a process known as spidering. They look for audio and video files encoded with RSS, which stands for either rich site summary or really simple syndication, depending on which reference work you’re using. RSS allows users to “subscribe” to podcasts and blogs and thus automatically receive future installments.
When PodZinger or Blinkx finds a podcast or vlog, it begins translating the audio track into text, an operation that is far from instantaneous. Laats said an hourlong podcast can take about 45 minutes to complete the voice recognition procedure.
PodZinger and Blinkx each claim to have translated about 100,000 podcasts and vlogs -- on professional sites, such as National Public Radio, and personal pages -- into searchable text. The user can click on a result to be taken directly to the audio or video stream.
Judging by test searches, PodZinger and Blinkx have each found items the other has not.
But the basic difference between the two is that while PodZinger focuses almost exclusively on podcasts and vlogs, Blinkx also processes videos posted by select broadcast sources, including CNN, Fox News, Bloomberg and certain local TV stations.
Blinkx evidently favors these broadcast clips when generating search results. For example, a search for “Medicare” on PodZinger produced 51 hits, all from podcasts and vlogs. The same search on Blinkx returned 50 hits, with all but eight from broadcast videos.
Because you can use PodZinger and Blinkx free, it costs you nothing but time to search with both services to get the most complete results.
You’ll come up with finds you probably couldn’t have imagined. A PodZinger search on opera composer “Puccini” found a podcast in which the theremin -- an electronic musical instrument used to make weird sound effects in 1950s science-fiction films -- played selections from “Madama Butterfly.” Picture a performance of the opera as it would be done in “Forbidden Planet” and you get the idea.
On Blinkx, the same search found an informative half-hour TV show, produced at UC San Diego, about the opera “La Boheme.” But as often happened with searches on Blinkx, many of the hits were out of date, leading to sites where the video had been discontinued.
Indeed, of the first 10 hits on a Blinkx “Teri Hatcher” search, only six were still functioning.
PodZinger was generally easier to use because it showed exactly where a mention was found within a podcast or vlog.
For example, when I searched for college basketball star Adam Morrison of Gonzaga, a hit showed the name -- and a bit of the text before and after it -- at exactly 26 minutes, 13 seconds into the podcast.
With a click, PodZinger took me to that point in the audio stream. An on-screen controller then allowed me to move back and forward in the show, 20 seconds at a time. Blinkx was not nearly as exact. The link would go to the beginning of a show or video segment.
Neither service offers full text translation of shows. Blinkx founder Suranga Chandratillake said that could lead to legal complications. “While it’s OK to create an index to the content,” he said, “what you are not supposed to do is present it in its entirety.”
A full transcript might not be very useful, anyway, because of the inaccuracies of voice recognition.
A PodZinger search on the wine “Pinot Noir” found the podcast series “Grape Radio.” On the audio stream, the announcer talked about an area “too far north to grow vines, especially varieties like Riesling and Pinot Noir.” PodZinger translated it as “two foreign office two Croat specialty varieties like Riesling and Pinot Noir.”
Well, it did find Pinot Noir, and that’s all that was needed to track down the podcast clip.
This also means it probably misses some podcasts because of mistranslations. And sometimes you have to search with phonetic spellings (“blue ray” for the video format “Blu-ray”) for more complete results.
Still, the fact that the audio tracks of podcasts and videos are now searchable is mighty impressive. I look forward to someone figuring out how to get the Web to make the sound of one hand clapping.
David Colker can be reached at technopolis@latimes.com. Previous columns can be found at latimes.com/technopolis.
