Re: Digital Voice Recorder - Voice to Print feature

[David Neeley <dbneeley -at- gmail -dot- com>]

"Training" a voice recognition engine is necessary for accuracy, since
there are so many different pronunciations that must be interpreted by
the engine. However, you will get some output without specific
training. In the case of a clear speaker of standard "non accented"
English it can be surprisingly good under ideal conditions.

In addition, the task is made infinitely harder when you have multiple
voice levels by using too few microphones. A wide-area conference mic
means that voice levels vary all across the board--as you have
observed if you have recorded such a meeting with a single mic.

Thus, to have a prayer of getting results that can be remotely
adequate, I'd couple your recorder with a small audio mixer and
multiple microphones. Then you have the problem of fidelity on the
telephone participants, since most telephones are relatively
low-fidelity devices from a recording perspective.

You will also be rather appalled at the tasks your brain handles with
ease in a multi-person conference that may well baffle the conversion
engine--such as people who speak simultaneously with others.

The conversion engine will also not discriminate in text output
between different speakers, unless perhaps there is one that can use
multiple channels with highly directional individual microphones.

In short, I seriously doubt you will have much success with that
approach without extensive "help" by a human in making sense of the
results.

In the end, if you must have an accurate transcript, it will probably
be by the time-honored method of listen-stop-type-start-listen that
seems so tedious. It may prove less tedious than correcting and
manipulating the conversion engine output (which would probably entail
a great deal of the above manual method simply to make the
corrections, only now you must go to the spot in the recording where
each problem exists that you can't figure out without comparing the
recording.

The other benefit of doing it manually is that you can do a reasonable
amount of editing in the process, eliminating the inevitable off-topic
comments--unless, of course, it is a sworn transcript that must be
word-for-word faithful to the content. I assume it is not that, since
court reporters long since have learned to deal with this kind of
thing.

David




On 9/9/05, Lucero, Peggy <plucero -at- atsva -dot- com> wrote:
> 
> All,
> I bought an IC Recorder (by Sony) that I'm still making it through the
> paperwork/manual that came with it.
> I am wondering if any participants on this list knows  if the Naturally
> Speaking Dragon Systems software by ScanSoft, Inc. (and some other
> software that came with it) will actually work for recording a meeting
> with multiple voices (9 in the room and 3 on the phone).
> I recall hearing a while back that you have to teach the software the
> individual voices before  it works....

^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Now Shipping -- WebWorks ePublisher Pro for Word! Easily create online
Help. And online anything else. Redesigned interface with a new
project-based workflow. Try it today! http://www.webworks.com/techwr-l

Doc-To-Help 2005 converts RoboHelp files with one click. Author with Word or any HTML editor. Visit our site to see a conversion demo movie and learn more. http://www.componentone.com/TECHWRL/DocToHelp2005

---
You are currently subscribed to techwr-l as:
archiver -at- techwr-l -dot- com
To unsubscribe send a blank email to leave-techwr-l-obscured -at- lists -dot- techwr-l -dot- com
Send administrative questions to lisa -at- techwr-l -dot- com -dot-  Visit
http://www.techwr-l.com/techwhirl/ for more resources and info.