Speech-To-Text Podcast Indexing

Support and Suggestions Volunteers
#1

I can’t remember how many times we’ve all wanted to find a particular bit of GWC at one time or another, but with over 200 hours of podcasts (and growing), it’s darn near impossible. I’ve heard of a number of services that claim do combine imperfect speech-to-text indexing with timecode to give some semblance of search functionality, but I can’t seem to find any we can participate in.

The most popular Google result seems to be Podzinger, which apparently has become Everyzing.com:

http://www.everyzing.com/

But I can’t see any way to sign up – and they seem to be all about video now.

Obviously the long-term solution to this is that when advertising money allows we’ll just pay to have the podcasts transcribed (which will also allow the deaf to enjoy the 'cast as well), but we’re a good way from that.

Any ideas?

#2

I know that MuggleCast, which is a Harry Potter podcast, has listeners transcribe their podcasts. I can check with their transcript section editors to see how they do it.

We may need to have GWCers volunteer to transcribe sections of each podcast, like a 15-20 minute section, and then have an editor combine them into one transcript.

I would be willing to do some transcribing, if others showed some interest as well.

#3

Oof…a formidable task, but one that may have to be kept as a “last resort”. I started the process when going over archives for Chuck’s “Best Of” compilation, and soon realized that a lot of the really engaging/heated/smackdown conversations, when the Triumvirate is really fired up and speaking simultaneously, proved tricky to parse with any accuracy, and that’s manually. (Could it be that AI can eventually surpass my own human ears? Gasp!)

#4

As you can see, it’s a formidable task. I’ve been leery of asking anyone to actually transcribe 'em because that strikes me as about as close to pure hell as it can get. It looks like Everyzing would work great for this, but all the links to their signup pages are broken. I get the feeling they’re just not taking new 'casts on.

Anyone know someone over there?

#5

Chuck check out http://rbytes.net/software/wave-to-text-review/
it coverts .wav files to to text. we could probably get someone to convert the podcast to wav then to text. A ton easier then transcribing by hand. (or keyboard)

#6

Might take a majority percentage of the dictation effort out of the mix…leaving maybe only one or two passes for correcting overlaps, mispronunciations, or other quirks that may occur (I’m foreseeing a clunk when scratchy voicemail calls are taken, things like that). Anyone ever used this shareware, or any tool of its kin? I’ve only used voice-to-text dictation tools for word processors, etc., and they were dodgy at best.

#7

I’ve done some transcribing for battlestarwiki’s podcast section. It’s doable, especially if people take a fifteen minute segment at a time.

#8

I converted the most recent GWC to .wav ( it took less the four minutes with switch) I will report my test of the software results when I have them.
(sometimes work gets in the way of this important stuff. :slight_smile: )

#9

You are the man, thank you THX. Way to step up! I am very interested to see the results. No matter how good I am sure it will take manual intervention to make it formatted, indicate who is speaking etc…but simply having a level of text that could be searchable would be awesome.

#10

it sort of tried to work a little bit. ok maybe not. but, Hope is not lost. this bit of software has the option for making custom library files for translation. I will give that shot over the weekend and let you know what I find out.

#11

I really don’t think an automated solution will solve things. Why? Character names are nearly impossible for EveryZing to understand. Half the time EveryZing cannot understand a bloody thing I say when using slightly technical jargon. Machine transcription is likely quite unfriendly to sci-fi discussion. Attempting to create neologisms definitely gives such software fits too.

There is a house that does transcription services in LA but they’re not cheap.

#12

Speech Recognition software has a hard time with one person, it’d be a neat trick to get it to do three (plus Sean’s various voices, the occasional guest, etc.) I think this is going to end up a transcription project. It’d probably be easiest to set up a wiki (easiest in the long run, that is.)

#13

A formidable task, not for the faint of heart, but I think worthy of the effort!
It certainly gives one the opportunity to brush up on their grammer and typing skills.

#14

…Or in my case , to develop grammer and typing skills!

#15

I have a couple of weeks of vacation at the end of the year, so I would be willing to give a podcast a try and see how long it takes to do one, or how much I can accomplish for a 30 minutes of podcast, or whatever. Anyway, if I do it should I start with a recent podcast, or one of the first? I had a portable hard drive failure , so I don’t have any before 80 or so. If you want me to give it a try let me know.

#16

NO. sorry. its just garble. really not a good piece of software for this. I will keep looking around and let you know what I find. ( i don’t admit defeat easily.)

#17

I’m kinda slow at getting to this, but I’m good for a quarter of a podcast every other week. It does not appear from the earlier responses that transcription software is going to work (or at least not easily), so I will offer my skilz as a former word processor.

Any other takers? With some practice and decent typing speed, a half hour of air time shouldn’t take more than an hour to an hour and a half to transcribe.

#18

I’m a decent typer, and it’s taken me about twice as long to do the RDM podcasts when I was helping BSGwiki. Multiple-voice podcasts (e.g., when he was with Eick) took longer.

But yeah, no reason we can’t start chipping away here. Pick one and get started!

#19

Should we make a list of who is doing what?

#20

I don’t imagine there’ll be stampede (but prove me wrong.) Just post which one you’re taking. PM me when done, and I’ll figure out where it should live.