Can anyone recommend a speech to text program

J

jrod81

Veteran
I have been asked to transcribe an 18 minute video on youtube... Rather than taking the time to pause the video ever so often and type it out I was going to see if anyone could offer a program that does this?
I know there is a program at nuance.com that does speech to text (its called dragon dictate) but i am not sure that it will work for websites or if it is something that i have to talk into a mic or something for it to work... that make sense at all?
thanks for the help
jared
 
jrod81 said:
I have been asked to transcribe an 18 minute video on youtube... Rather than taking the time to pause the video ever so often and type it out I was going to see if anyone could offer a program that does this?
I know there is a program at nuance.com that does speech to text (its called dragon dictate) but i am not sure that it will work for websites or if it is something that i have to talk into a mic or something for it to work... that make sense at all?
thanks for the help
jared
Click to expand...

I was under the impression that Google/youtube was working on a 'automatic captioning' capability. This has been a hotly debated subject around the Clark household, as the Wife is hard of hearing and almost no youtube clips are 'captioned'... especially the tutorials...

The Daughter has pointed to something a while back that was to do 'captioning' but it frequently botched things.

The Dragon software that I'm aware of requires training of the speaker (sometimes I think this is mostly training the speaker to say the same thing the same way always...), and so is not a particularly 'general' speech to text solution.

You could check out a local junior college and see if any people taking a court reporting curriculum are interested in a few hours of work... even if it is gratis... I mean internship...
 
As Jared wrote, Dragon by Nuance is software for your iPhone I heard of but I don't know if it's any good. The company actually asked me once to record native speakers for them here in Germany. I was on vacation at the time of the job though so I ended up not doing it.
I think speech recognition is not too advanced yet. But time moves fast. I'd be curious to try it.
 
there are plenty of transcription services out there. i don't know how much that costs, i'm usually just given login info from a client and will upload the files. so you could download the youtube video, convert it to an mp3 and then send it to a transcription service.

and adobe soundbooth will do a transcription too. here's a link to scott simmon's blogs where he checks it out
http://www.scottsimmons.tv/blog/2008/12/18/kicking-the-tires-on-adobe-cs4-speech-transcription/
 
its not going to work. sit down and start! It wont take longer than 1 hr of concentrated work, I do it all the time TRANSLATING my interviews from German-Italian-English in all directions. but I charge 250.-€ an hour :-)
 
i forgot i had soundbooth as part of cs5. i'm testing it out right now.

EDIT: transcribed a 7 minute audio clip, took about 3 minutes at medium resolution. terrible results.
 
Last edited:
Has anyone else found some software that might work?
I am helping a friend with some class interviews that he plans on recording by loaning some audio equipment. Some of these interviews may take up to 90 minutes. Part of the class credit is a full transcript of the interviews and he is a little leary of having to type them all out. He has a total of four to do.
 
I just don't think the technology is there yet.

Note that the worlds most sophisticated human text processing computer (hardware+software) is arguably the IBM "Watson" system that beat the human champions at Jeopardy. But note that the questions were fed to it in text form, they didn't even bother trying to get speech recognition to work. And remember IBM is the company that developed the "Dragon Naturally Speaking" product before it was sold off Nuance.

And there are dozens (scores?) of online transcription services available. That is what all the big-time productions use to create transcripts of interviews, etc. Many of them are located in other time zones which makes it convenient for customers to email their audio files at the end of the shooting day, and have the transcripts emailed back to them when they get to the office the next morning.
 
I have seen the Adobe speach recognition work pretty well, but even they don't think it is accurate enough to do subtitles. It's a little like OCR software, when it works it's amazing but if it's only say 95% accurate you spend more time fixing that it would have taken to just type it in the first place. They are all getting better but...

Dagon Naturally speaking is also software that runs on a computer. It's pretty good and once trained can be very good. If your friend is a terrible typist and doesn't want to hire it out he might try using the computer version trained to his voice. Listen to the interview and dictate into Dragon. That way you should get good accuracy and you won't have to type as much.
 
I own and use Dragon on my mac. It ships with a headset/mic and you read several paragraphs of text so it can recognize your voice. It require you to follow a strict protocol when dictating. You need to tell it to insert all punctuation marks, carriage returns, etc. I write a lot and hoped it would save me oodles of time. They are now running TV commercials about how easy it is to use. It is easy, and you can even tell it to run other apps and control your machine with you voice, but it is prone to error. I don't think I can recite a single sentence without there being something wrong with it. So, expect to do some serious editing. I found that in doing dictation, having to worry about each period and comma, that it inhibited my creativity so now I rarely use it! Hope that helps you understand one user's experience with one program. That said, it's the best I could find.
 
I tried Adobe speech recognition in CS4/5 and found it's typically more of a toy than a useful tool (to me). It'll take (much) more time to find, say, 20 random errors in a 500 word script, than just typing them out yourself while listening, pausing as often as necessary. Note this: statistically you'll have to listen to half of everything on average just to capture a single random error made in speech recognition. That said, it can still be very useful if you can't listen to something but need a grasp of what's being said. Or if you don't understand the language being used but need to take a written note of what's being said. Or if you just need a quick note for a short clip.

BTW, Windows Vista/7 has a robust built-in voice recognition feature. If you plan to buy Dragon, check out this feature first. It allows pretty much everyone who can speak clearly to command Windows solely via voice, including having Windows type an entire article for them. From what I read, some professional writers routinely use this feature: this way, they could literally lie in the bed in the dark, and let their laptop record/type all their words.
 
The "job" Adobe had in mind for the recognition is to enable filmmakers to locate video clips by searching for words. So it wasn't designed for transcription level accuracy. I don't know that that is even possible at an affordable level. I suspect that you would catch most errors just by reading the text and the plus side is that clicking on that part of the transcript will locate you to the corresponding video so you can check relatively easily.

But as pretty much everyone has said the fastest way to do it is to have a fast typist just do it. For me that would be hiring someone because I am neither fast nor a touch typist.
 
Noiz2 said:
But as pretty much everyone has said the fastest way to do it is to have a fast typist just do it. For me that would be hiring someone because I am neither fast nor a touch typist.
Click to expand...


If there's a court reporter program at a local college or school, put an ad up for a student in the program.

While I may be a touch typist, the ear, brain, finger connection is pretty bad for me...
 
j1clark@ucsd.edu said:
If there's a court reporter program at a local college or school, put an ad up for a student in the program.

While I may be a touch typist, the ear, brain, finger connection is pretty bad for me...
Click to expand...

The catch with that is you then need to get it translated. Court reporters don't type "english" but in a phonetic character set.
 
these speech to text utilities are like these nice kitchen fruit juice makers.
tired to spend 3 min to press orange ? use our fast kitchen robot, it will make a full glass in less than 30sec.....
except they forget to tell you will spend the next 15min to clean it.
 
You must log in or register to reply here.
Back
Top