I've just started evaluating the technology I'm going to need to help
me do the interviews for my doctoral dissertation. I expect to be
doing a dozen or more one-on-one interviews, some in person and some by
telephone. They will probably last in the neighborhood of 1-4 hours
each. I'm planning to record them in some form and transcribe them
into Word documents. I use a PowerBook G4 1.5 GHz, running OS 10.3.7,
with 1 GB of memory.

As I see it, I have two options:

--Get a digital voice recorder and speech recognition software, and
have my computer do the transcribing (realizing that, if this is even
do-able, I'll have to do some significant editing and cleanup of the
transcripts it produces); or

--Get a dictation machine with foot pedals and type in the suckers
myself.

Obviously, the first approach has the potential to save me a lot of
work--*IF* it works as planned. Has anybody out there done this? I'd
really like to hear about your experiences. In particular, can you
tell me:

--Did you use MacSpeech, Inc.'s iListen, IBM's Via Voice, or some other
speech recognition/transcription software? What did you think of it?

--How did it work for transcribing interviews with a variety of people?
(from what I've read, to get the best results, you're supposed to
"train" the software to work with your voice and speech patterns--is
this going to be a big problem when I'm recording a number of different
voices?)

--Did you try having the software transcribe recordings made over the
phone? How well did that work? (Olympus makes a "Telephone Recording
Device" to go with their Mac-compatible digital recorders, but it
doesn't appear to plug directly into the phone line--it records the
output from the telephone receiver. Has anybody worked with a setup
like that?)

--Do you have any experience with using speech recognition software
with recordings made on a lower-end, consumer-quality digital recorder
(the ones I'm seeing are in the $150-$200 range) vs. a higher-end,
professional-quality one (probably $400 and up)? Do I really need to
spend the money to get the recording quality I need for decent speech
recognition results?

--Any other advice you can share?

And, if I go the lower-tech, type-by-hand route, do you have any advice
or suggestions about that? In particular, any recommendations on
cassette vs. digital recorders, in terms of sound quality, ease of use,
and any other relevant features?

Thanks much!

--
MaryMc
marymc11@SPAM.BEGONE.mindspring.com
)
(
"Espresso est, ergo cogito." C[_]

In article <301220042249352215%marymc11@SPAMBEGONEmindspring. com>,
MaryMc <marymc11@SPAMBEGONEmindspring.com> wrote:

> --Any other advice you can share?

The present state of _all_ speech recognition software is rather
primitive, despite decades of effort creating such app's.

I have used all of the available Mac app's, iListen, MacSpeech,
ViaVoice, and others - - - most of them run about $100 in price.

The best available app' unfortunately only runs on a PC, it is called
"Dragon NaturallySpeaking" from ScanSoft Inc.

<http://www.ScanSoft.com/NaturallySpeaking/>


Unfortunately, the present owners/marketers of DNS Pro ver' 8.0 have
their website all screwed up so you will have a difficult time finding
that state-of-the-art version which is called:

"Dragon NaturallySpeaking Professional" - - - version 8.0

(DNS Pro) - - - USA English version.

It runs a modest $806.94 with shipping, tax, etc.

Cheer Up, the medical and legal versions of DNS Pro are much higher.

....and you really should buy the recommended digital recorder that they
sell to work with the software, which will run another $300.

They also sell wireless headsets, which are a nice convenience

With a couple of hours of training to your voice only, in a quiet room
with a quiet computer, you can expect approx' 99% dictation accuracy at
a dictation speed of 160 wpm.


I run DNS on an old HP laptop using the latest updated XP operating
system, with the so-called SP2 installed. (service pack 2)

Because of the RAM requirements of DNS, I had to jack up the RAM to
500-MBs, which I would consider as the bare minimum.


Once trained to _your_ voice, it is fully voice only operation. Even
your corrections can be dictated by voice, no keyboard interaction
required at all.

FYI, the correction process is what kills the usefulness of these
programs, it is very slow under the best of circumstances.

I average about 2 corrections per minute. (by voice, of course)

That means if I dictate 1600 words by talking for 10 minutes, then I
could have 16 or more significant mistakes. Correcting 16 mistakes
could easily take another 10 minutes, lowering my "effective" overall
dictation speed from 160 wpm down to 80 wpm.

Correcting minor punctuation could drop that 80 wpm even further.

that that is is that that is not is not

....to its punctuation-corrected version:

That that is, is; that that is not, is not.



Okay, now that I have shocked you, let's get down to reality. :)

If you want to experiment only, you could buy their substantially
lower-cost "Preferred" version, which runs $200. You might even be
able to run that on a junk PC with only 256 MBs of RAM.

I was not too happy with that "Preferred" version, too many missing
features for my taste.

Here I run Macs almost exclusively, only use the PC for that one app'.

Sorry for this relatively bad news about the present state of
speech-recognition software.

The app's are still very user-voice-dependent, so your plan to take
dictation from several different speakers would likely result in an
intolerable amount of corrections needing to be done.

Mark-

In article <301220042249352215%marymc11@SPAMBEGONEmindspring. com>,
MaryMc <marymc11@SPAMBEGONEmindspring.com> wrote:

> And, if I go the lower-tech, type-by-hand route, do you have any advice
> or suggestions about that? In particular, any recommendations on
> cassette vs. digital recorders, in terms of sound quality, ease of use,
> and any other relevant features?

Yep, go the digital recorder route, if for no other reason than they
can pick up lower volume speech than cassette recorders can.

I have an older Sony ICD-MS1 model, which completely disappears in a
man's small shirt pocket.

I have successfully used it to record my own voice for literally hours,
on-location, talking at 160 wpm.

After the on-site dictation session was over, I later removed the
"memory-sticks" from the recorder, plugged the "sticks" into a PC.

The hours of dictation get automatically downloaded into the PC, where
DNS-Pro automagically converts them into text.

If I then want to correct mistakes, I can do so at my leisure.

DNS Pro keeps the audio from the recording in the computer, so it can
always "automatically" refer back to the original audio when making my
corrections.

All this with the "Pro" version of DNS. The lower cost "Preferred"
version of DNS will not handle it. Specifically, if you forgot what
you had dictated, you would have to laboriously manually play back the
recorder version, until you found the section of the recording that you
were interested in "correcting" in the text version.

The "Pro" version finds the of the exact section of the recording you
want, instantly and automatically, so correction is quick and easy.

The sound quality of using the built-in microphone of today's digital
recorders is great. I did not notice any significant inprovement by
using a seperate microphone.

Other's milages may vary, they might prefer a seperate microphone to
use with their digital recorder.

Mark-

On 12/31/04 7:07 PM, in article 311220041709244320%NoSpamDammit@invalid.com,
"Mark Conrad" <NoSpamDammit@invalid.com> wrote:

> In article <301220042249352215%marymc11@SPAMBEGONEmindspring. com>,
> MaryMc <marymc11@SPAMBEGONEmindspring.com> wrote:
>
>> And, if I go the lower-tech, type-by-hand route, do you have any advice
>> or suggestions about that? In particular, any recommendations on
>> cassette vs. digital recorders, in terms of sound quality, ease of use,
>> and any other relevant features?
>
> Yep, go the digital recorder route, if for no other reason than they
> can pick up lower volume speech than cassette recorders can.
>
> I have an older Sony ICD-MS1 model, which completely disappears in a
> man's small shirt pocket.
>
> I have successfully used it to record my own voice for literally hours,
> on-location, talking at 160 wpm.
>
> After the on-site dictation session was over, I later removed the
> "memory-sticks" from the recorder, plugged the "sticks" into a PC.
>
> The hours of dictation get automatically downloaded into the PC, where
> DNS-Pro automagically converts them into text.
>
> If I then want to correct mistakes, I can do so at my leisure.
>
> DNS Pro keeps the audio from the recording in the computer, so it can
> always "automatically" refer back to the original audio when making my
> corrections.
>
> All this with the "Pro" version of DNS. The lower cost "Preferred"
> version of DNS will not handle it. Specifically, if you forgot what
> you had dictated, you would have to laboriously manually play back the
> recorder version, until you found the section of the recording that you
> were interested in "correcting" in the text version.
>
> The "Pro" version finds the of the exact section of the recording you
> want, instantly and automatically, so correction is quick and easy.
>
> The sound quality of using the built-in microphone of today's digital
> recorders is great. I did not notice any significant inprovement by
> using a seperate microphone.
>
> Other's milages may vary, they might prefer a seperate microphone to
> use with their digital recorder.
>
> Mark-

The Vice Chancellor's Office at Vanderbilt University Medical Center is
going to iPod-based digital voice recording. Use the third party microphone,
synch to a PC using Dragon, and you wind up with 90-95% accuracy on first
pass. Not bad for a totally hands-off solution!
--
Brian Ehni

In article <BDFB7E55.19835%behni@comcast.net>, Brian Paul Ehni
<behni@comcast.net> wrote:

> On 12/31/04 7:07 PM, in article 311220041709244320%NoSpamDammit@invalid.com,
> "Mark Conrad" <NoSpamDammit@invalid.com> wrote:
> >
> > The sound quality of using the built-in microphone of today's digital
> > recorders is great. I did not notice any significant inprovement by
> > using a seperate microphone. Other's milages may vary, they might
> > prefer a seperate microphone to use with their digital recorder.
> >
> > Mark-
>
> The Vice Chancellor's Office at Vanderbilt University Medical Center is
> going to iPod-based digital voice recording. Use the third party microphone,
> synch to a PC using Dragon, and you wind up with 90-95% accuracy on first
> pass. Not bad for a totally hands-off solution!

Interesting, I did not know that Apple's iPod could deliver its voice
files to a PC for processing via Dragon.

How is the syncing to a PC done. via a USB cable connection?

If the iPod recording is of a high enough quality, I would imagine that
a Dragon user could even achieve the advertised 99% accuracy on their
own "trained" voice usage.

The regular "Dragon approved" digital recorders with their built-in
microphones have accuracy as good as speaking directly into the PC
laptop via Dragon's USB headset which comes included with Dragon.

Other users claim that a good external microphone is a necessity,
however I had no problem maintaining 99% average accuracy using the
recorder's built-in microphone.

The 99% accuracy figure mainly applies to a well trained language file
being provided by the user of the program.

Even then, if I dictate complex language, instead of everyday simple
language, the accuracy goes down, of course.

These speech recognition programs have a long way to go before they
will become as good at recognizing speech as we humans can.

It was not too many years ago when Dragon NaturallySpeaking Pro, then
called "Dragon Dictate", could only handle 40 wpm, and we had to pause
between words.

Now at least it recognizes continuous speech, clipping along at
160-200wpm with acceptable accuracy.

I wish they would port it over to the Mac, but no chance of that
happening, I guess.

Mark-

THANK YOU to everyone who has responded with their advice and input. I
can't tell you how helpful your suggestions have been. You've probably
saved me a whole lot of time, money and frustration.

I posted this query in a few different places--here, plus a newsgroup
for medical transcriptionists, and a bulletin board for people doing
research similar to mine. Here's a summary of the main points I
gathered from all the responses:

Using speech recognition software to transcribe interviews with
multiple interviewees sounds great, but it probably won't work in the
real world. There were widely differing opinions on the speech
recognition software in general--some people think it's all rudimentary
and flawed at best, some say it can be surprisingly accurate and
useful--but several people pointed out that a lot depends on training
it to work with your own voice and speech patterns. It's just not
likely to perform well with recordings of many different speakers--the
amount of time you'd spend correcting and cleaning up the output will
probably negate the time you'd save on the initial data entry. Some
people suggested playing back the recordings of the interviews and
repeating them in chunks in my own voice into the computer's
microphone, but that doesn't really sound like a viable option to
me--it seems too time-consuming to be much of a time-saver, for one
thing.

Several people said they prefer digital recorders over cassettes
because you get better quality recordings. One person said that
digital recorders can pick up lower-volume speech that cassette
recorders can't. A few people said that even lower-end digital
recorders produce good results on the highest-quality recording mode
(several people recommended recording only in that mode). But a major
limitation on the cheaper models is storage space--you may not be able
to fit all of a long interview in the high-quality mode in the storage
capacity they offer. One person recommended a higher-end model for
this reason, preferably with removable storage media. Several people
mentioned Olympus as a good brand (and I can attest that they have good
tech support--I called them the other day with some questions about
their products, and the person I talked with was extremely helpful and
knowledgeable). One person did point out that it's easier to lose (or
accidentally delete) a digital recording file than it is to erase a
whole interview on tape.

Several people recommended investing in a really good microphone--not
the one that came with the recorder. One person suggested getting a
two-channel recorder and separate microphones for you and your
interviewee, so you can separate voices when both of you speak at once
(she said she learned this from a journalist who does interviews for a
living). If you're using just one mic, another person recommended
using a tabletop model, since a lapel mic on your interviewee may not
pick up your voice as well. He also pointed out that you and your
interviewee aren't tethered to a tabletop mic, and can just walk away
from it if necessary without being unhooked. He said that his tabletop
mic--a Sony ECM-R100 with "boundary effect" and a "noise reduction"
setting--worked very well. Still another person said he thinks the
built-in microphones in current-model digital recorders are very good
and an add-on mic isn't necessary.

Several people emphasized the importance of taping the interviews in a
quiet setting, as free from background noise as possible (several
strongly recommended against coffee houses, bars, restaurants and other
public locations, and one warned about noise from wind, air
conditioning, etc. that you won't hear, but the microphone will).
Recordings with less junk in the background are easier to transcribe,
and if you're paying someone to do it for you, you'll end up paying
them less.

For recording a telephone conversation, one person recommended the
TMP-636S from Dynametric.com. It plugs directly into your phone line
and sends output to the microphone jack of a computer or a tape
recorder. He said it works on Windows or Mac (although the company's
website only mentions PC). NextWaveSolutions.com also makes a
Telephone Record Coupler (#TRX2035B) that does the same thing.

Lots of people recommended a playback device with foot pedal controls
(the professional transcriptionists think you'd be crazy not to use
one). You can get these to use with both cassette and digital
recordings. Many people suggested looking for a used one on eBay, and
selling it there when you're done.

Opinions among the researchers about paying a transcriptionist were
mixed. Some said that doing it yourself lets you immerse yourself in
the data in a way that's invaluable to the analysis. Others were glad
they'd found the money to invest a transcriptionist, and didn't think
they missed anything of value in letting go of that task. Some said
that what they got from the transcriptionist was really a first draft,
and they spent a fair amount of time correcting and polishing it, so
they got plenty familiar with the data in that process.

A couple of the transcriptionists recommended word expansion software,
like SpellCatcher. With this program, besides the spell checking
functions, you can create shortcuts that let you type in a few letters
or a short abbreviation, and it automatically expands this into a
longer word or phrase. I can imagine this being very handy for some of
the cumbersome and repetitive terminology that a lot of us use in our
writing. It's available for Mac and Windows from
http://www.rainmakerinc.com/. They have a 15-day free trial download
on the website, and the full version is $40. I'm going to try it.

And finally, one person pointed out that at least one qualitative data
analysis program--HyperResearch for the Mac--lets you code audio
recordings directly, without having to transcribe the entire interview.
You'd only have to type the utterances you were quoting directly in
your paper.


I hope all this is as helpful to others here as it has been to me!

--
MaryMc
marymc11@SPAM.BEGONE.mindspring.com
)
(
"Espresso est, ergo cogito." C[_]