What is meta? ×
Meta Stack Exchange is where users like you discuss bugs, features, and support issues that affect the software powering all 130 Stack Exchange communities.

How can we encourage more people to transcribe portions of the Stack Overflow Podcasts? Would a reCAPTCHA-like system be appropriate for this? Better yet, how can we encourage people to create a computer program that is capable of generating the transcriptions with no manual human effort? This seems like it could be a fun and challenging project.

share|improve this question
7  
Your latter suggestion sounds a little ambitious. Such a program would be worth hundreds of thousands of dollars. –  Ian Elliott Aug 14 '09 at 2:19
7  
Instead of speech-to-text, what about text-to-speech? Has anyone made a Markov chain to generate "new" podcasts using the existing transcriptions as a starting point? –  Kyle Cronin Aug 14 '09 at 2:25
    
which one of the sites that Meta supports is this question about exactly? –  EBGreen Aug 14 '09 at 3:00
3  
@EBGreen: This is about the Stack Overflow podcast. It is therefore an acceptable use of Meta Stack Overflow. –  Kyle Cronin Aug 14 '09 at 3:03
1  
All of them? It's the Stack Overflow podcast. –  Ian Elliott Aug 14 '09 at 3:03
    
StackOverflow because that was the only site when the podcast started, and is still the main site, as well as being the one that most of the others are based on –  a_m0d Aug 14 '09 at 7:27
1  
+1 For reCAPTCHA-like system for audio, would be interesting –  chakrit Aug 27 '09 at 0:05

6 Answers 6

We've tried the automated transcriptions before (machine voice detection) and they are horrible.

Require so much editing to be readable that you're better off transcribing manually.

share|improve this answer
7  
Just post your scripts. The ones where it makes the two of you sound all so ad-lib and as if you were just shooting the breeze and losing track and not listening and ticking off that bingo sheet to make sure all the listeners are hammered inside the first 6 minutes. –  random Aug 14 '09 at 7:49
    
Bother, you have to create a FogBUgz account if you want to keep track of edits and contributions? Why not OpenID option for the podcast wiki? –  random Aug 27 '09 at 4:33
2  
Here's another idea, post those podcasts on Youtube and let their magic autotranscript it for you. You'd have to have an image over it though. So something that won't scare the chillins. –  random Jan 11 '10 at 9:16
    
that's an interesting idea.. except.. "Please note that only 13 YouTube channels will feature automatic captions at this time" –  Jeff Atwood Jan 11 '10 at 9:43
    
Bother. How about speakertext.com/browse.php ? –  random Jan 11 '10 at 9:51
1  
Automatic captioning on youtube is now available for all videos. –  alex Mar 6 '10 at 12:19

Jeff/Joel,

Would you be interested in a community podcast transcription service that utilizes the power of your site's tribe/users/community. It would all be driven by a ridiculously simple-to-use UI and API. The goal would be for your users to stay on your site, but hit a little button/link on the side of the page that says something along the lines of "Contribute 10 seconds of your time to help transcribe Podcast #65" (or whatever you want it to say... it's an API, you have the power!!)


Feature Notes:

It would allow a user to transcribe a 1 to 3 second clip. In the end, it would auto-stitch each "transcriptlet" into the final transcript, then post it. The main goal is an unbeleiveabley simple to use interface that takes the user 15 seconds or less to transcribe a 1 to 3 second clip. If 1,800 (of your 1 mil users) transcribe 2 seconds then your 60 minute podcast is transcribed (potentially in an insanely short period of time).

There will of course be options: like must have 3 (or X) copies/matches of each "transcriptlet" ~roughly match before it is "auto-approved/stitched". And many more, I am not going to get into all of the details here...

Other features: Point-base sub-system, achievement sub-system, ...


Questions for Joel and Jeff:

Anyways, what I wanted to know is, would you guys (Jeff and Joel) be interested in a using a service like this. I need a good beta site to try this out on. You could empower your community and get your podcasts transcribed. And be part of building a (another) great service.

And of course, I want to know: Do you think other sites with podcasts and a community would pay for something like this. I am talking like $5, $10 or $15/hr for transcription (rather than the $200+/hr that some people charge). There will also be a free service (X minutes/month or whatever) too, but the premium service would cost a very small amount. Or possibly it will be ad supported. But I know I just listened to you guys talk about the dissapointment with Google AdSense... Anyways, it's starting out as free for sure (during beta). And most likely free for you guys always :) Because I want StackOverFlow to be transcribed (I will pay the $5 per podcast :)


What inspired this project:

  • The end of every single one of your podcasts (wiki comments)
  • The fact that I would like to search and replay segments of the podcast for someone, but can never remember which is which, due to the lack of podcast transcribing
  • I like your idea of the wiki, but that requires people to listen to the mp3 with a player, keep track of timing and stop/replay/pause the track, and update the wiki properly. It is a bit time consuming.
  • After listening to "Tribes" by Seth Godin, it really made me think that this would work no problem with a community like yours.
  • It would be a fun little app/service/site to write with my development team.
  • We would be solving a problem that exists for a lot of podcast/communities. (Hanselminutes, twit, ...) Anyone with a Tribe really.
  • And yes, money. Money is good side-effect of providing an elegant solution to a problem.


Current Project Status:

I had my first meeting with the development team, we have initial database design, very rough Balsamiq Mockups, FogBugz Wiki/Tasks created and a bunch of other planning done. They asked me:

"Well how do you know Jeff and Joel would even want to use something like this if we spent the 6 to 8 weeks (ok 4-6 months) to developed it?"

I said, I don't know, but I have very strong feelings that they would. So that is why I am asking. It would be GREAT if one of you or both of you could respond.

We can make this happen. I am ready to pull the trigger, we might do it anyways. But if I saw a reply from you guys that said "Heck yes we would use that". Then it would be a no-brainer for us to make it happen. If you said "You are a frikin' IDIOT"", then maybe I might back-off and think of another idea


Quick background:

I own/run a small development company in Grand Rapids, MI. We have run and successfully delivered many software products from the very big to the very small. We can rock it out.


Thanks!

Thanks for your time. I apologize for the very long message and the loose grammar. It happens when I get excited about something.

share|improve this answer
    
Cool idea - very similar to reCAPTCHA. –  Andrew Aug 26 '09 at 23:55
    
Yeah, very similar. Crowd-Sourcing is the way to go. –  mschmidt42 Aug 27 '09 at 11:04
    
5$/hr won't work as nicley as 0$/hr, as Jeff said [ codinghorror.com/blog/archives/001295.html ]. –  Steve Schnepp Nov 2 '09 at 7:58
    
> "I apologize for the very long message and the loose grammar" - You can edit it, you know? –  jmfsg Jan 11 '10 at 18:35

It seems to me that the earlier shows were fairly well-transcribed, but after Stack Overflow went public transcription rates went down. If that's an accurate assessment, it might have something to do with the fact that Jeff was allowing people who helped write down what was said in the podcasts to get into the SO beta.

Why not reward people who transcribe shows with reputation?

share|improve this answer
    
Yeah, good call. Or we could actually expose part of the API so StackOverflow could do some user account associatetion. And then pull all points/rep for a user from the tranascription service that are related to SO podcasts. So then SO developers could factor our point/rep system in with theirs. So maybe 500 points on our point system, might relate to 120 StackOverflow Rep/points or some other ratio. I dunno, just an idea. –  mschmidt42 Aug 27 '09 at 11:03
    
Forget free rep. Free tees are what's the swag haggle. –  random Jan 11 '10 at 9:38

Having taken a shot at transcribing, I would say that something that could aid my efforts would encourage me to do it more. What ends up happening is that I hear a segment I think is really good, and I want to transcribe it (because with other material I appreciate being able to locate it later, maybe quote it and send it to someone, etc). So I take 5 or 10 minutes to do the segment, and suddenly it just took 30 minutes! Why? It's tedious.

What would be nice is some kind of tool that is like the transcription machines from days gone by. I never used one but I know that the right features would make this faster. Some ideas:

  • Have the text entry and the controls in the same place
  • Make it controllable from the keyboard--mousing is time consuming
  • Easy way to back up some number of seconds
  • Set a bookmark during playing that I can return to
  • Set marks around a snippet so it can play repeatedly
  • Adjust the play speed; maybe pitch correct it.
share|improve this answer
    
My thoughts exactly. –  Ikke Mar 6 '10 at 9:45

I've tried some community transcriptions for another podcast, and some thoughts:

  1. It takes forever. My time estimate is 10-15 minutes for a 3 minute chunk.
  2. It's not that fun, so people tend to drift away.
  3. If you can't hear the whole sentence, it's very hard to transcribe it. Context is very important to make sense of garbled words or overtalking. This is a barrier to "chunking" it like ReCaptcha does.
  4. Attribution is another thing that could be difficult if you split up the audio. Although it might not be a problem with two speakers.
share|improve this answer

Revisit the idea of making teh Googles work in transcripting these with their rolling out feature of auto-captioning on Youtube.

You'd have to clean it up some later with all those dings and dongs that distract them so from forming proper words in the stream of things.

Bulk of the work would be done for you.

The hard part is actually listening to them talk for clarifications in the transcript.

share|improve this answer

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .