[MV] Re: A couple of transcription comments.

[T. Patrick Henebry]

>Actually, accuracy is relatively good. I see no difference in 
>recognition from a CD where I have recorded speech as AIFF files 
>input via the audio jack in my pismo vs. using the supported Olympic 
>dictation recorder via USB. Once again (and I speak as someone who 
>has actually tried multiple ways of inputing information into 
>iListen), the program really doesn't care where the sound comes 
>from, as long as it has been trained to recognize it.

Interesting, At what frequency was the sampling done on those AIFF 
files? The Olympus DS-2000 and DS-330 sample at 12 KHz (I thought it 
was 11 KHz but I double checked.) when set to their highest quality. 
The quality of the microphone in the recording device will also make 
a difference.
>
>The caveat really is three-fold:
>
>1). Use a digital recording device.
>
>2). Read the training stories into the recording device and play it 
>back to create a profile for that device.
>
>3). You MUST speak punctuation.

All true whether you're using the device with either a Mac or a PC.
>
>Now, that having been said - recognition is "relatively" good. While 
>you may get recognition as good as the supported device, the user 
>must still decide if 80-90 percent recognition is good enough to 
>warrant using the device vs. just re-speaking or typing the 
>information.

Also true.

>In my opinion, the only advantage of using a "supported device" is 
>that you can transfer the sound files much more quickly - in other 
>words, not in real time. I see no advantage whatsoever in accuracy 
>of a "supported device" over other digital recording methods.

That time savings is often a key factor for those most likely to use 
the combination of speech recognition and a digital recorder. Not all 
digital recording devices will produce acceptable results. Besides 
the time savings using a supported device eliminates the possibility 
of poor performance due to an incompatible device (digital or 
otherwise). This is no small consideration for users who are not 
speech recognition experts. Trust me on this one. Though it *is* 
possible to get acceptable results using an unsupported device I 
still feel it's a BAD idea in general.
-- 
----------
T. Patrick Henebry