oggenc weirdness

Kerry Hoath kerry at gotss.net
Sun Feb 24 10:49:52 EST 2002

Sampling at 44100 is not a good idea at all, here is why.
Assume we have a 44100hz stream at 16 bits per sample so that is
88200 bytes per second. we want to compress that to 32 kilobits per second or
4000 bytes per second.
A quick devision shows a 22:1 compression is involved. At such
agressive compression settings lame downsamples by default and introduces noise
into the sample fromt he downsampling. This is agrivated by the mp3 encoding
process. If we start sampling at 22050 rather than 44100
then it is only a 11:1 compression ratio.
Also, mp3 is not free to use and has patent restrictions on its use.
ogg sounds better, is completely free and works well.
Below is the technical reason why you don't do 4-track tapes at 44100:

Take your average 4-track tape recorded at 15/16 inches per second.
This is half regular tape speed which is 1 and 7/8 inches per second.
On side 1 we have track 1 on the left channel and reversed track 4 on the right channel.
On side 2 we have track 2 on the left channel and track 3 reversed on the right channel.
the fastest way to sample these tapes is to do the following:
Sample side1 at 44100 16-bit stereo.
fiddle the wave file so that the header reads 22050 samples per second.
This halves the speed from normal tape speeds digitally and since no math
is involved we retain whatever quality we got into the sound card except see note below.
Split the wave file into left channel only track one and right channel only track 4.
Reverse the right channel. We have now got 2 tracks of aproximately 90
minutes in length but have only sampled for 45 minutes.
This means that if we sample both sides and slice them up we sample for 90 
minutes and fiddle for 10 minutes to get 6 hours of wave data on disk.
Not bad at all.
The problem is that even though the niquist frequency of 44100 samples per second is 22050hz
most low-end soundcards don't record high frequencies say above 16khz too well
and some tape recorders don't play them well either.
This means that some of the top end is lost by sampling at double speed and 
halving sample rate. This stuff is only spoken word so
the loss is not too great. Also lame puts a lowpass
polyphase filter on the wave data at about 7500hz anyway.
I could sample at 32000hz and half the rate to 16000 but I may loose more
high frequency component and would be incompatible with a lot of sound hardware that won't
play back 16khz sample rates.
Sampling on a 4-track machine would also preserve more quality but would take 4 times longer.
I just need to find a good card for the pc that is supported under Linux and that has good
quality analog inputs.
Sound blaster live cards tend to be a bit noisy in this regard. It's almost worth
dragging out the Gravis ultrasound max or the pas16 cards I have here.
The pas16 cards have the best line input i've heard on cheap cards but require
a dynamic microphone and do not provide phantom power so will
not run a condenser without a Microphone pre-amp.
Anyway I've ranted long enough :-)

Regards, Kerry.
On Sun, Feb 24, 2002 at 10:18:13AM -0500, Igor Gueths wrote:
> Hi Carry. I think that mp3s are a good way to go. Wouldn't you prefer to try to sample the audio at 44100 khz? The only problem with that is that you might get some tape hiss or noise depending on the playing device you are using and the age of the tape (s). Again, you don't have to re-record the samples, but that's just my opinion on quality. Hope this helps. 
> ----- Original Message ----- 
> From: Kerry Hoath <kerry at gotss.net>
> To: <speakup at braille.uwo.ca>
> Sent: Sunday, February 24, 2002 1:20 AM
> Subject: oggenc weirdness
> > Since we're all talking about anything and everything at the moment, here's
> > one for the list to think on:
> > I am currently sampling some 4-track material I own to keep it
> > safe in case the tapes age or snap.
> > I get the wav files as I want them; (I can summarize to the list if
> > anyone cares with the scripts I wrote yesterday to make it all work)
> > and these aren't Jim's scripts, I wrote these ones with comments to aid
> > in maintainance.
> > I have a .wav file 22050 samples per second 16-bit little endian mono.
> > I try to oggenc this file with the following command:
> > oggenc -b32 side01.wav
> > and get a message that "mode initialization failed"
> > This used to work fine in 1.0rc2 but appears to have broken in 1.0rc3
> > Anyone else running 1.0rc3 and seen this behaviour?
> > If I don't specify -b32 I get an encoded file which averages 51kb/s which is far too
> > much bandwidth to waste on this stuff. At the moment I am left to mp3 the files
> > with lame -h -b32 side01.wav side01.mp3
> > which does work but I would preferr to use ogg. Must I downgrade to 1.0rc2 again?
> > 
> > Any insites would be apreciated.
> > 
> > Regards, Kerry.
> > 
> > -- 
> > Kerry Hoath:  kerry at gotss.net kerry at gotss.eu.org or  kerry at gotss.spice.net.au
> > 
> > _______________________________________________
> > Speakup mailing list
> > Speakup at braille.uwo.ca
> > http://speech.braille.uwo.ca/mailman/listinfo/speakup
> _________________________________________________________
> Do You Yahoo!?
> Get your free @yahoo.com address at http://mail.yahoo.com
> _______________________________________________
> Speakup mailing list
> Speakup at braille.uwo.ca
> http://speech.braille.uwo.ca/mailman/listinfo/speakup

Kerry Hoath:  kerry at gotss.net kerry at gotss.eu.org or  kerry at gotss.spice.net.au

More information about the Speakup mailing list