On Sunday, 12 Oct 2003, at 3:07:17 PM, Art Shifrin
<[log in to unmask]> wrote:

I'd like to chime in here -

My experience has been that majority of problems on tapes of this
nature are either excessive room/background noise from speakers being
too far off-mic, excessive tape noise from low original recording
levels or or multiple tape transfers, or other background noise (air
conditioners / electronic equipment / refrigerators / wind noise /
etc...).  In this case, Alyssa's recommendation to digitize the source
material "flat" is a good one.  If you have good outboard analog gear
(EQs/compressors/etc...) and know how to use them, then great, but
otherwise I would tend to let today's powerful software plug-ins do the
lions share of the work removing noise.  Also, digitizing at 24 bits is
a good idea if your hardware/software can handle it (as most newer
soundcards/software can).  The more data your plug-ins have to work
with, the more likely you are to have good results - 24 bits of
recording gives your system a lot more to work with when using
broadband noise reduction or various kinds of EQ (hum removal,
emphasizing speech-related frequencies and attenuating noise-related
frequencies, etc...).

When your problem is coughing, dogs barking, or the like, things get a
bit trickier...  Sure, if the major problem on your tape is multiple
transient noises that are 20dB higher than the speech you are trying to
hear, then using an outboard analog limiter/compressor on the way in to
the computer to suppress the levels of the transients will help you get
better levels on the speech you are hoping to hear and restore.  Either
way, the problem one encounters once the audio is digitized is that the
frequencies of many of these random sounds are in the same range as the
speech you are trying to hear, making total removal of the offending
sound difficult at best.  I have had the best results using EQ on
noises of this nature, usually removing low frequencies (under 200-250)
and high frequencies (above 8k-10k), but the bulk of a dog bark or
police siren will still be there over your speech.

Remember, Sam's original post from Duke University said that they had
outsourced the transfer of their reels and were receiving flat
transfers on cdr and  reel.  I believe he said that they were thinking
of using Wavelab as their main audio software, and he was asking for
recommendations of how to make spoken-word recordings clearer so the
speakers could be identified and the recordings transcribed, with audio
restoration somewhere down the line if they could find "easy" software
to do that.  I seem to recall that he said their budget was modest and
he was thinking that Wavelab (at $449) might be "more sophisticated"
than they were able to handle.  Personally, I wish that more libraries
would consider "upgrading" their budget just a hair and demo a few of
the leading noise reduction system out there - Waves Restoration Bundle
(software, $900), T. C. Electronics Restoration Suite (software $999,
requires TCE Powercore Firewire interface), or Cedar systems (very
expensive - available as stand-alone hardware or as software) - to name
a few...  A lot of Art's advice below is good if one has the budget for
multichannel analog->digital converters, outboard gear, and the
like...but sadly I think the majority of institutions with audio
collections aren't yet thinking of audio preservation that seriously.

Unfortunately, I have to agree with Alyssa's post from last week
pointing out that a lot of folks in libraries and archives don't seem
to have a the experience, budget, or motivation to take a
"professional" approach to audio archiving and preservation.  I have
worked as an advisor on more than one recent project where the job of
specifying an archival audio system and software was done by college
interns with virtually NO audio experience, or by an administrator who
may have spent a week or two asking for advice here and there from the
"audio geeks" they knew, all of which has led to an interesting
assortment of mismatched audiophile equipment and amateur
software/hardware (think $500 digital cables or $2000 digital routers
with $99 soundcards and $150 software).  There is a lot more awareness
in libraries and archives about printed material and photographs -
hence the budgets and technology are better.  What many places consider
"archiving" of audio would be the moral equivalent of preserving a
photographic or manuscript collection by having an intern snap a
picture of the original with a disposable point-and-shoot, using a
low-res scanner to import the image, and then using an inferior
knock-off of Photoshop to do the restoration...

With the right advice, budget, and training, good audio preservation
doesn't have to be hugely expensive, but as a professional archivist
and restoration engineer, I would like to think that there is more
skill (and investment in training/materials/equipment) involved than I
have seen in most places I have advised.  When people are asking if a
$449 software package and minimal to no audio training is "too much",
then you begin to see how far down on the totem pole audio preservation
is in many people's minds...

dave nolan


> You'd be setting the maximum digitization (recording)
> level for the loudest ambient noise (coughing, knocking, clock
> chiming, dog
> barking, or the loudest uttered phoneme).
> In poorly made original tapes (any analog format), the conversations
> (or their most useful frequencies)
> might be 20 or more db below that.  Assuming that you'd be starting @
> 16 /
> 44.1, after processing, you'd be working with a bit depth of just a few
> bits.  Even if the A/D were initialized at 24 / 96, then significant
> losses
> would still occur but be less sonically costly.  When subsequently
> readjusting the panoply of characteristics for maximum intelligibility,
> DIGITAL NOISE AND DISTORTION would be imposed upon the results.
> Unless a tape is self evidently, mortally deteriorating, it is often
> desirable to replay it more than once.  Otherwise, if it does not
> contain
> unambiguous level documentation, (what % of tapes have tones &
> annotated db
> = nWb/m?)  how will you achieve best S/N & least distortion?
> IF the source tape is monaural, then it's SIMPLE to achieve
> "preservation"
> and "preprocessed" digitizations:
> split the output: run one into "left" & the other into "right" of the
> A/D.
> IF the source tape is stereo, bin-naural, whatever, then it's equally
> to achieve "preservation" and "preprocessed" digitizations: split the
> two
> pairs.  Run one pair 'straight' and the other pair "preprocessed" into
> 4
> A/Ds.


> From: "Alyssa Ryvers" <[log in to unmask]>
> To: <[log in to unmask]>
> Sent: Sunday, October 12, 2003 12:23 PM
> Subject: Re: [ARSCLIST] poor speech intelligibility
>> I would do the initial transfer as "flat" as possible; that way you
>> will have a preservation copy.
>> You can always subsequently send the audio out of the computer to any
>> outboard gear you want to try, if you so choose. Besides, you can make
>> as many attempts at *tweaking* that way as you want, without playing
>> the original more than once.
>> Alyssa.
>> _____________
>> Alyssa Ryvers