June 30th, 2008 at 06:10am
I don’t know if any of you remember PodZinger as it came and went from the public spotlight very very quickly a few years ago, but basically it was a semi-promising new search engine which aimed to automatically index and transcribe podcasts. If it worked, it would have been a good way to find information in what is otherwise an awful medium to search.
But of course, it was doomed from the beginning as anybody who has ever used voice recognition software would have known. To get a computer to recognise a voice with any amount of accuracy, you have to train it. Usually this involves two steps, the initial preset “read these sentences” training exercises where the computer gets to hear you say things that is asks you to say, and then the ongoing “no, I said “cat” not “hat”, the hat is not in the hat, the cat is in the hat…no, the cat is not in the cat either” intermittent corrections which also help the computer to learn how to understand your voice.
The reason we need to teach the computer how to do recognise an individual’s voice is that everyone has a slightly different voice and a slightly different speech pattern…and if humans who are predisposed to understanding the speech of other humans have difficulty understanding people with accents, what hope does an untrained computer have?
As far as I can tell, PodZinger had no form of quality control…the robot listened to the audio, produced a mangled transcript of it, and nobody bothered to check the accuracy of it. A system where corrections could be submitted by listeners could have worked better, but I don’t think PodZinger were ever interested in having masses of voice samples floating around in their system, nor do I think that having masses of voice samples from different people would have helped with individual transcriptions.
So, why am I babbling about an ultimately failed search engine of little-to-know consequence? Because they’re still around under a different name (EveryZing) doing much the same thing, albeit with expansions in to the more profitable and sane market of search engine optimisation, and expecting people to pay them for it. Maybe the transcription software is better today than it was in 2007, but I wouldn’t be willing to pay them in order to test the theory.
I noticed that they are still around on the weekend when I was wondering if they are still around and was shocked to find that they are. They also have all of their old archives…and if you search for my name, you can be informed about my Chinese communist leanings by their transcription robot.
It has transcribed the Feedback segment from Samuel’s Persiflage #13, specifically the section from 47 minutes and 58 seconds. According to their robot, in that segment I said:
can go to get sort of course if you concentrate back podcasted Samuel Gordon Stewart — Communists the email address or you can leave comments in the China nights or irritants Samuels who supplies were sought to
Maybe the robot’s ears are blocked, because when I listen to that segment, I hear:
feedback to get through and of course if you’ve got some feedback, firstname.lastname@example.org is the email address or you could uh leave the comments in the show notes or go to the samuel’s persiflage website and
That said, I have been (jokingly) accused of encoding secret messages in Samuel’s Persiflage…maybe they were on to something…I would have to be among the last people on Earth that anybody would consider as a possible communist, so I would have to be the perfect vessel for hiding and broadcasting such messages.
It leaves me pondering the question: “Why am I giving the conspiracy nuts something to work with?”