-
Transcribing interviews and such
Hi gang…
I just thought I’d pass along a resource that I stumbled on that I’m pretty darn impressed with.
We just finished a week-long corporate shoot in Las Vegas, where we shot tons of interviews with our client’s customers. We do interview-type shoots for this client all the time, and when the shoot wraps the first order of biz is always to give them written transcripts of what we shot.
In the past we have done this with out bare hands like animals, lots of listening and typing. Painfully monotonous, the only upside is that we bill a lot for it with no heavy lifting.
Well this go-’round, I decided I was just going to farm out the transcribing. A little Googling led me to a company that I was going to try… they promised real live human transcribers, a turnaround time of 12 hours, and a cost of a dollar a minute… which was a huge bargain I thought.
But before I pulled trigger on that, I ran across another company, Temi, that does automated speech recognition and text transcription online. You can try one for free, so I gave it a shot.
I wasn’t very hopeful because some of our interviewees spoke at around 900 miles per hour, some had accents, and their dialog was full of industry jargon, acronyms and such. We had tried speech-recognition-to-text voodoo several times before, with poor results.
I was very surprised though that the Temi transcripts are DARN NEAR FLAWLESS, even industry terms that aren’t in the usual vernacular, acronyms, and even company names that are not exactly common. Most of these interviews were about 15 minutes long, and I’d say on average there were anywhere from a half dozen to a dozen places where I had to tweak the transcript. Otherwise they were perfect.
AND… it happens almost instantly (takes about 2-3 minutes to process an audio file upload), and is 10¢ a minute. TEN CENTS!!! The average interview cost me a buck and a half, for what previously was about two hours of typing.
Works like a charm… and I heartily recommend…
T2
__________________________________
Todd Terry
Creative Director
Fantastic Plastic Entertainment, Inc.
fantasticplastic.com
![](\"https://i1.creativecow.net/u/77996/smalllogo.gif\")
-
13 Replies
![](\"https://images.creativecow.net/77996/smalllogo.gif\")
-
I have been using Trint for some time and really like it. I imagine that most of these services use IBM’s watson so will have similar outcomes. TRINT has lots of features I like and the fact that you can get your client involved is great.
Video Production – Videographer
Melbourne, Australia
http://www.mcmm.com.au -
Yep, there’s no reason to manually transcribe any more, audio to text technology is pretty darn good. I also used to transcribe myself but now I use Ebby.co (pretty similar to Temi in features and pricing)
-
Thanx, Todd!! The referral is MUCH appreciated. Ben using the $1 a minute services so this “sounds” like a great deal.
-
REV.com is also one of the best….
Rich Rubasch
Tilt Media Inc.
Video Production, Post, Studio Sound Stage
Founder/President/Editor/Designer/Animator
https://www.tiltmedia.com -
Hi Rich
I like the look of REV, TRINT is a software based translation and while you get your translation back quickly (30 minutes for a 10 min read) it’s about 90% accurate. So you then have to spend 30 minutes or so correcting the translation.
Can you share a translation using the API with a client? This is so they can mark up the text with highlights for selecting good takes etc.
Video Production – Videographer
Melbourne, Australia
http://www.mcmm.com.au -
I can’t speak to the other flavors, but with Temi you can share with clients and they can mark up the text, etc… you can keep it all “live” on the cloud or download various file types… docx, PDF, etc.
My interviews were about 15-20 minutes long each. I’d say on average the processing took about 2-3 minutes if I did them one at a time. Once I uploaded a batch of about 10 of them all at once, the transcribing took about 5 minutes for those.
The transcriptions were very VERY accurate… I’d say on average I had to make between six and a dozen corrections on each one. Even this one guy who had a pretty heavy (actually extremely heavy) German accent and in several instances said the names of fairly obscure companies, and the translation was still correct.
Just an example… let’s say I had an instance where the interviewee said the company name “MagiCorp” several times. Temi would spell it “Magicorp,” (only getting the capitalization wrong) which is pretty amazing considering most of the company names were either acronyms or just plain made-up words.
T2
__________________________________
Todd Terry
Creative Director
Fantastic Plastic Entertainment, Inc.
fantasticplastic.com
-
Sounds great. Can you also get them formatted as .srt for captioning purposes?
I’m still using Youtube to auto-caption/transcribe, then I have to manually add capitalization and punctuation corrections and fix the occasional zen koan that pops up when the audio isn’t clean or the speaker mumbles or has an accent. But it’s free.
-
Yes Mark…
You can either do your edits and such live via a dashboard, or export files.
You can export text as .docx, .pdf, or .txt files.
Caption files can be exported as .srt or .vtt files.
It’s not free like YouTube, but at 10¢ a minute it’s about as close as you’ll get. The first one is free so give it a try.
T2
__________________________________
Todd Terry
Creative Director
Fantastic Plastic Entertainment, Inc.
fantasticplastic.com
-
Rev does all the flavors, can edit online and download in many formats. They do Captioning, subtitles and can even put in all the ums and ahs etc or leave them out. The only thing they don’t do yet is stripe the transcription with timecode from an MP4 with window burn…they can add a timestamp but it is like control track from the old tape days.
So when we get the transcript back we add timecode stamps every few sentences from our sequences.
Rich Rubasch
Tilt Media Inc.
Video Production, Post, Studio Sound Stage
Founder/President/Editor/Designer/Animator
https://www.tiltmedia.com -
Temi puts a timecode stamp every paragraph or so… or every couple of sentences…whenever there seems to be a good clean break in the audio.
It does start from 00:00:00 though, so if you want it synced with the actual media’s timecode it will require some math, IF you are uploading a clip that has timecode starting at something other than zero. I don’t, the way I’m using it, so that’s a non-issue for me.
It will also differentiate and name different speakers (I haven’t used or needed that feature yet). You can turn off or on the speaker names as well as the timecodes if you want before exporting.
I don’t think it has a way to turn off the umms and ahhs… those were all accurately in my transcripts, but fortunately I wanted them there.
T2
__________________________________
Todd Terry
Creative Director
Fantastic Plastic Entertainment, Inc.
fantasticplastic.com
Reply to this Discussion! Login or Sign Up