At 03:48 PM 8/27/2003 -0600, Jon Noring wrote:
>[Cross-posted from 78-L to the ARSC mailing list for their learned
>Some of the discussion in this thread (on Pitching 78s) talks about
>pitching during the digital transfer of source material such as 78rpm
>records. My present view (subject to being changed with a persuasive
>argument) is that pitching should not be a worry during the transfer
>itself (just get it close.) Once digitized, during the restoration
>process the recording can be pitched digitally using sound processing
>software (effectively by resampling).
>Some may argue that digitally pitching a digital audio file will
>introduce errors (from what is effectively resampling) that we need to
>worry about and thus pitching should be done during transfer, but am I
>right that the "bit-error" introduced in doing this "resampling" is so
>minimal as to not even worry about?
>(Also, in reality, is 44.1K or 48K sampling more than adequate for
>all 78 digital restoration? Or should one double sample, e.g. at 96K,
>for digital restoration purposes, and only at the end of restoration
>to resample downward to 44.1K for "Red Book"?)
In my opinion, pitching should be done at recording if possible to avoid
the time and inaccuracy of resampling. The time is, of course, a function
both of the speed of your system and the quality of the sampling algorithm.
Using a program such as CoolEdit, which allows you to adjust quality, you
can hear and (more significantly in some ways) see in the spectrum the
effects. Oversimplifying, resampling 48 ksps to 44.1 ksps introduces
components at the difference 'frequency' of 3.9 ksps and multiples. The
higher the quality of interpolation (or extrapolation in the other
direction), the lower the amplitudes of those components.
NOTE: It's been forty years since I worked with z transforms. If someone is
up on the matter today, please provide the math to replace my memories. <G>
Note, too, that the error introduced here is not LSB or fractional bits as
in most transforms but actual introduction of spurious audio.
>On a related topic is equalization. Here, when I talked with George
>Morrow earlier this year (he passed away in late May), he commented
>that it is much better to set the proper analog equalization during
>transfer than to do it digitally (but he was of the opposite opinion
>with regards to speed/pitch, as I note above.) I'd like comment from
>the experts here on 78-L and ARSC on this topic.
Unless the starting signal-to-noise ratio is extraordinary - say, 90 db -
the losses due to modest equalization are at the fractional-bit level. In
the case of severe equalization, I'd suggest that if the maximum effect (db
difference between max and min equalization) plus SNR of the source is
substantially less than the 96 db of redbook audio, digital EQ is fine.
Thus, if you have 4 db/octave over the usual range (20 - 20K Hz) with a
good open-reel tape as input, you would approximate as 4db x 10 + 60 db =
100 db and digital EQ will cost SNR. Of course, with very low-noise input
and ADC, capturing and processing at 24 or more bits, the loss would be
buried when the signal is finally reduced to 16 bits.
[log in to unmask]