I use this method with Dragon NaturallySpeaking (DNS) Preferred and DNS
Professional 4&5 and it works well. I use DNS daily for nearly all my work
(RSI) and have to write up lots of meeting notes. In stable acoustic
environments I average recognition accuracy better than 98% -- and no need
to correct typos!
You need a PC with Pentium II or better; 500 Mhz or better and 256 Mb of RAM
or better. You also need a good soundcard. SoundBlaster cards usually work
well. Most failures in recognition result from a combination of poor
hardware (inc soundcard) and poor training of the software
BUY GOOD QUALITY HEADSET MICROPHONE -- Andrea QW1000 is the best I have
found so far over five years' experience with DNS. Don't rely on the
microphones bundled with the software. You get what you pay for.
Then...
1. Train software in basic recognition of your voice (1 hour max)
2. Prepare word list (in plain text file) which includes all the 'terms of
art' or jargon you expect to encounter in your interviews and any unusual
personal names.
3. Import word list and train software in recognition of special words to
improve accuracy. (1 hour max)
4. Identify up to ten interview (transcript) documents. Dictate them (see 6
below) and correct misrecognitions. Provide additional training DNS to
handle the syntax of interview transcripts. (30 mins max)
All the above covered in the DNS manual. Additional info on
http://www.dragonsystems.com
5.Set up your transcription machine with the headphone of your headset
microphone plugged into the taperecorder and your microphone plugged into
the soundcard
6. Start the transcription machine. Listen to a unit of text (sentence or
paragraph) Dicate same into DNS software.
7. Repeat....
Happy to offer advice on DNS or text to speech to fellow list members.
David Cairns
Dr David Cairns, Head, Shadow Information Unit, QAA
Southgate House, Southgate Street, Gloucester, GL1 1UB
Tel: 01452 55 7106 (direct line) Fax: 01452 55 7011
Mobile: 07774 816875
PA Mrs Tracey Hendry 01452 55 7108 e-mail [log in to unmask]
-----Original Message-----
From: [log in to unmask] [mailto:[log in to unmask]]
Sent: 26 November 2001 13:26
To: [log in to unmask]
Subject: Re: use of voice to text conversion
Dear Marten,
My experience of using voice recognition mirrors that of Heather.
I also used Via Voice and found it often interpreted what i said
very inaccurately indeed. I found keying the text directly from
an audio machine with footpedal was by far the quickest way to do
it.
Hope this helps.
Elaine Welsh
Department of Sociology and Social Policy
Oxford Brookes University
Oxford
OX3 0BP
|