Premier Pro’s new Voice Recognition- Killer!

[Chris Borjis]

Hopefully in the pursuit of a bell & whistle like that they
are not sacrificing stability. That was and has been a major thorn in
premiere’s side since its beginning.

[Bob Cole]

I’d be very surprised if this actually worked very well. I’ve used a supposedly great VR product, Dragon Naturally Speaking, to transcribe interviews, and it would only work if I revoiced each line into a mic, because it needed to “learn” a particular voice. That said, Naturally Speaking (ver. 9) did pretty well, but not perfectly, especially for unexpected words. Amazingly, context and grammar seemed to give the software cues which improved the accuracy. Fun to watch it change its mind about the third word in a sentence when it heard the last phrase.

As for percentage of accuracy, it’s not only a matter of percentage. Usability also depends on two more subtle factors: what KIND of errors, and HOW you’re using the transcripts.

What kind of errors:
Human errors are more tolerable to other humans (like myself and probably you) than software-generated errors. It’s an odd thing, but it’s true. Even “professional” transcripts are rarely more than about 98% accurate, but that 2% doesn’t slow me down, because a human made the mistake, it’s understandable, and you can easily “get it.” On the other hand, software errors are so far out in left field that they can really make a transcript hard to read.

How you’re using the transcripts:
If you’re using it to read an immense quantity of material, less than 95-98% accuracy is going to be very tedious (especially because of the TYPE of errors – see above). But if you only use it to find the occasional word with a computerized search, it will be useful at much lower percentages of accuracy.

Would someone enlighten me as to how this VR feature compares with Avid’s similar efforts at text-based editing?

Bob C

[Walter Biscardi]

[Bob Cole] “I’d be very surprised if this actually worked very well. “

After the utter fiasco of Encore with BluRay authoring on the Mac in CS3, I’d be surprised if half of what Adobe claims actually works well in CS4.

Walter Biscardi, Jr.
Biscardi Creative Media
HD and SD Production for Broadcast and Independent Productions.

STOP STARING AND START GRADING WITH APPLE COLOR Apple Color Training DVD available now!
Read my Blog!

[David Frisk]

EDIT: Nevermind, I think we’ve argued (or “debated”) enough :)~

[David Frisk]

Would someone enlighten me as to how this VR feature compares with Avid’s similar efforts at text-based editing?

With Avid, you have to supply the script first, and then it will match your clips to the script, but it doesn’t generate a script for you. I don’t use it at all, but from what I’ve heard, Avid’s scriptsync works extremely well.

[Mark Raudonis]

I’d just like to add that anyone contemplating this product (Auto voice transcription) should do their own test.

I did. And it just ain’t good enough.

mark

[Bob Cole]

Thank you. That sounds really useful.

If the new PP (or better yet, the next version of FCP) could integrate a human-generated transcript a la Avid scriptsync, and do it well, I would tend to find that more useful than a semi-accurate VR product.

otoh, VR that actually worked on the file level (that is, you don’t have to play the audio into a mic) would be a plus even at low accuracy — just for those occasional searches for a random word or phrase that can otherwise drive you crazy.

Bob C

[Gary Adcock]

[David Roth Weiss] ” using pristine audio from one of my own one on one interviews, with zero background noise and an interviewee with a good voice and no accent. Let me just say that the results were not what I’d hoped for, and not what I’d been led to expect. “

david,
let me say that was not my result. My test was done via live capture then processed with myself and one other being the talent -those that know me understand that I can talk very fast – and the software performed better than I had expected.

Both subjects were on hardwired LAV’s and recorded onto separate tracks in my test, I agree that when both subjects are recorded together as stereo instead of using dual mono the software transcription does suffer,

gary adcock
Studio37
HD & Film Consultation
Post and Production Workflows
Inside look at the IoHD

[Mark Maness]

Let me jump into this…

I think what people are missing right now, is the fact that the software is only as good as the person or talent can speak. Meaning, if your talent speaks clearly, the software will work pretty accurately. If your talent has alot of sloppy speech or slurred words, of course, the software is going to choke.

Different areas of the country will have different speech patterns. Gary is the Chicago area and folks around there speak very clearly. Whereas, I’m in the Memphis, TN area, and people here speak slow and slurred – southern – with a bunch of local terms that only we would understand. This makes ANY speech recognition software choke or at best, struggle.

Think about it… BUT, this is a real start.

_______________________________

Wayne Carey
Schazam Productions
https://web.mac.com/schazamproductions
schazamproductions@mac.com

[Walter Biscardi]

[Wayne Carey] “Gary is the Chicago area and folks around there speak very clearly. Whereas, I’m in the Memphis, TN area, and people here speak slow and slurred – southern – with a bunch of local terms that only we would understand”

Now Waaaynnee whut y’all messin up thisum f’rum thred fr? Yew syin’ folks tawk funny r smtin? ha ha haaaaaaa!

Walter Biscardi, Jr.
Biscardi Creative Media
HD and SD Production for Broadcast and Independent Productions.

STOP STARING AND START GRADING WITH APPLE COLOR Apple Color Training DVD available now!
Read my Blog!