 
|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
|
|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
128.48.10.72
Hi:I sent a note to Alesis, whose ML-9600 I use while doing location recordings. They failed to properly address the following question, however:
Are any audio artifacts introduced as a result of originally recording a solo piano at 24/96, editing at that resolution and then dithering down to produce a Red Book 16 bit/44.1 kHz sampled cdr?
What happens to pitch when such a down conversion takes place?
Is recording at 24/96 only for the purpose of eventually producing a standard "cd" sampled recording a waste of higher resolution, or should I just stick with recording solo piano at 16 bit/44.1 sampling?
Again, what advantage would I have in recording at higher resolution only to yield a Red Book product, other than preserving the master for some future applicaiton?
Richard Links
Berkeley, CA
Follow Ups:
Dithered or not, an asynchronous sample-rate "downconversion" from 24/96 to 16/44 will be lossy compared to just sampling at 16/44 in the first place. The 16/44 data is calculated off of an interpolated 24/96 signal will not be as true to the original signal as a direct A/D to 16/44.
An asychronous conversion (by which you appear to mean something that isn't a small integer ratio, which is not what I would mean by asynchronous) does not have to be any more lossy than a ratio 2:1 conversion, etc.However, it is possible for the 16/44 version to be as good or better, since you are capturing originally with less noise, and as a result have less residual noise to add to the requantization and dithering to 16 bits. You also get a teeny-tiny (3dB) bit of oversampling advantage when you do this.
because rational fraction sample rate conversion is likely used and it is a completely synchronous process – the only “lost” information is the unavoidable bandwidth limiting lowpass filtering required to meet the Nyquist limit of the 44.1K output streamlogically you:
"upsample" and zero stuff 96K 147x
filter and decimate by 1/320 to 44.1K
but,
since you don't have to multiply out the "0" entries in the "upsampled" data in the filter and you only need to evaluate the filter for each output sample the actual fir bandlimiting/image reject filtering you need anyway doesn't cost much more except for fancy indexing
The "320/147" ratio was brought up by the late Julian Dunn of Prism Sound, which would indeed make this a "synchronous" conversion. The problem is I've yet to see the a 16/44 to 24/96 conversion actually implemented this way.....I don't know where the term "zero stuffing" came from.... The upsampled signal is not "zero stuffed", but calculated using a mathematical process called "convolution," where a set of coefficients representing the "Fourier transform" (within a truncated time period or "window") of the brickwall filter function is first multiplied with the a set of corresponding values from the raw signal at a specific time, and then added together, to provide the "interpolated" sample value.
Consider.If I do f&y (& meaning convolution), f a filter, y the signal in the time domain...
If I then calculate F (fourier transform of the filter) and Y (fourier transform of the signal), then F*Y is the exact dual of f&y.
(* being point by point multiply, complex)If I then do IFFT(F*Y), I get exactly f&y. You need some guard band to prevent wraparound, but that's easy to calculate.
Perhaps you're mistaking this for something else?
but software to do synchronous conversion certainly existshttp://www.soundslogical.com/support/resample/documentation/english/documentparts/resamplehelp-30.html
cruise the site, matlab files, textbook references
http://www.analog.com/UploadedFiles/Application_Notes/154260303EE183Rev5.pdf
just from the first few hits from
google: 320 147 sample rate conversion software
The polyphase filter implementations can be thought of as the “fancy indexing” I referred to previously
I have designed/implemented fir filters using polyphase concepts to deskew/time align data from a multiplexed multichannel adc system, so I am familiar with those big word too
There are two issues to deal with, bit depth and sample rate. Since the A/D converter probably runs at its native (highest) bit depth even when recording at the 16-bit level, there really should be no difference whether downconversion is done in real time or is performed after the recording is complete. In either case, the unit would perform the same operation on the data.Sampling rate is different. I agree, recording at 44.1kHz is different than recording at 96kHz and then downsampling. If my understanding is correct, the A/D converter runs exactly at the sampling rate you select. The difference in quality depends on how good the dithering algorithm is. I usually record at 48kHz and then downsample to 44.1Khz in the computer using Cool Edit Pro at its highest quality setting. Initially I was concerned the sound might suffer, but my worries proved groundless as the software does an excellent job indeed. The only "lossiness" I hear is from the difference in sampling rate. I don't know how the Alesis performs in that regard, but I should hope it's also very good. So, it could be an issue... or not.
Richard... it's probably a good idea to perform a simple test and compare the results between "native" 16/44.1 and downconverted/downsampled 16/44.1 for your own peace of mind. You could simply use the analog output from your CD player and record the same track using both methods.
-Anthony
I second Bersani's comments. Cutting and pasting is just a matter of moving stuff around, but adding EQ, reverb, changing level, compressing, etc., all alter the signal, and it's best to do that at the highest resolution.On the other hand, if you're not doing those kind of things (or not much of it), then you might as well just record at 16/44.
I don't have any experience with the Alesis recorder, but it's not fundamentally different than other combination A/D converter & HD recorders I'm more familiar with. Hope this helps...(1) There should be no significant audio artifacts (i.e. "ill effects") from downsampling. Granted, the final sound will be different to some degree, but that's due to the lower resolution and the noise-shaping method that is used in the downsampling process.
(2) There is no change in pitch.
(3) There are a few reasons to record at 24/96. One is obviously having the means to listen to a 24/96 recording, either through DVD-A or a computer-based system. Another would the issue of editing: if you're going to alter the data through EQ or other digital filters, that should be done at the highest resolution. ("Cutting and splicing" types of edits on the other hand, do not alter the values of the sample points, so that can be done at any resolution.) A third, but unlikely reason would be if you intend to use a specific dithering algorithm in the downsampling process (Apogee's UV22 is an example of a commercial dithering algorithm). Finally, the reason you stated: to preserve the master for some future use.
If you have no intention of listening to the 24/96 master or using digital filters, you may as well just set your recorder to 16/44.1. You'll obviously be able to store a lot more data on the disk. Unless I'm mistaken, the unit's A/D converter will still operate at its highest bit-depth and then downconvert the data to 16-bit while applying its own dithering algorithm in real time. Therefore, the result is the same as you would get if you downsampled to 16/44.1 after the recording session.
-Anthony
Hi, Anthony!Much thanks for your very helpful analysis and suggestions!
In fact, I did edit and re-eq/normalize and added limiting to the recording in the 24/96 domain and then burned a Red Book copy for the client.
My understanding was that at certain high resolution (in this case, 24/96), there would be some sort of an advantage regarding possible headroom. I failed to also mention that the two-channel recording involved a pair of Milab microphones set in figure-of-eight pattern going through an Aphex 1100 with the mic lim circuit activated. Even so, it was necessary to tweak the eq a bit and then normalize before yielding a standard Red Book burn.
Thanks again!
Richard Links
Marantzguy: "Even so, it was necessary to tweak the eq a bit and then normalize before yielding a standard Red Book burn."Inmate: It doesn't matter what microphones you use. There is always some tonal adjustment to be made. Either the direct sound is off a little due to mic characteristics, or the reverberant sound adds too much unwanted color, or the mic doesn't have a good off-axis characteristic. There's always some reason to tweak it. Just don't let your speakers or headphones color your judgement. ;)
"Normalizing" is one of the most useful tools in the digital arsenal. Because of it, I can comfortably record at a reasonable level, and not worry about a heart attack if a big sfz comes along. Although it does raise the noise floor, it's not an issue for the vast majority of listeners.
As you rightly pointed out recording in 24bit gives you more headroom and you are less likely to trip your limiter, which is always a good thing.
It also gives you a couple of bits at the bottom where digital noise 'collects' and leaves you a clean 16bit 'window' for your signal...
As for sampling frequency I would use 88.2kHz since higher sample rates produce audibly better treble, even after conversion, and simply halving it requires substantially less number crunching when you eventually dither to 16/44.1.This is just my opinion based on a little practical experience, so please don't nail me technicalities...
Richard,
One of the nice things about 24-bit recording is the freedom to be more conservative with recording levels while still maintaing better resolution than 16-bit's full potential (thereby providing greater headroom). I suppose when you combine that with normalization at the 24-bit level before downconverting to 16-bit, there would indeed an advantage... and since you ARE using digital EQ, sticking with 24-bit is a good idea.Most of my recording is still done at 16-bit, and I run my levels hot enough that they reach 0dBFS for a few inaudible samples here and there. Obviously, normalizing is not an issue in my case, and with only 16 bits to work with, I don't usually mess around with EQ. Since it's a hobby for me, I can live with less than perfect spectral balance.
-Anthony
This post is made possible by the generous support of people like you and our sponsors:
Top of Page ]
[ Contact Us ]
[ Support/Wish List ]
[ Copyright © 2025, Audio Asylum, all rights reserved ]
