[vox] Transcriber

Bill Kendrick vox@lists.lugod.org
Sun, 20 Jul 2003 08:27:10 -0700


On Sun, Jul 20, 2003 at 08:20:36AM -0700, Richard Burkhart wrote:
> How's the accuracy?  Have you tested it with faint audio / mixed voices
> (like a meeting situation) ... or voices that use slang and/or have
> strong accents?  (going from the perspective of someone whh spent a good
> 2 hrs one day translating sometimes-garbled voicemail messages off of a
> micro-tape)

Ah - Sorry for any confusion.  Transcriber is not an ASR (automatic speech recognition) tool.
It's simply a nice GUI that shows you your sound file's waveform, and lets you break it up to
segments, which get assigned speakers and then the actual text.

You have to do it all by yourself, though.  (Ears + brain + fingers)


It REALLY does look like it will make the job a few hundred times easier for me, though!


I DID play around with some ASR stuff last night.  Specifically, Sphinx2.  It looks promising,
but isn't exactly the easiest thing to figure out.


(Also, I played with Speak Freely a little... Melissa and I chatted.  She was downstairs on her laptop
with the built-in mic next to the keyboard - going over wireless.  I was upstairs on my desktop with a
big mic - going over 100baseT.  Worked great...  Need to figure out full-duplex, though :^P
It could also use a GUI.  I might look into GnomeMeeting<sp?> or some other GUI 'phone' app.)

-bill!

-- 
bill@newbreedsoftware.com                           Got kids?  Get Tux Paint! 
http://newbreedsoftware.com/bill/       http://newbreedsoftware.com/tuxpaint/