After a bit of searching, I failed to find any specific answers, so I hope I'm not repeating a commonly-asked question. I'm planning to make digital recordings from vinyl by using the line-in jack on my PC with a Sound Blaster Audigy 2 sound card. I was planning to use Audacity to do the recording, and I was wondering if there is any real purpose to using 24 bit and/or a high sample rate like 88,200 or 96,000 Hz when originally recording. The final destination will be typical 16 bit 44,100 Hz, but if I decide to try do do click/pop or noise removal, is there any benefit from doing such work at a higher quality and then reducing the sample rate and bit depth later, rather than doing all the manipulation at an initial setting of 16/44? The only topic on this subject that I found was this one:
http://www.hydrogenaudio.org/forums/index....showtopic=61758 (http://www.hydrogenaudio.org/forums/index.php?showtopic=61758)
Also, Audacity allows you to record with 32 bit float as well, but I'm quite sure that my sound card doesn't support it. Is there any way to verify that the sound card actually does output 24 bit when you're recording at that depth? If there is, I haven't found any way to configure it, except in the media player supplied by Creative, which allows you to select the bit depth for recording (but presumably only for it, not for other programs like Audacity). Any assistance would be greatly appreciated.
I'd probably just record at whatever the highest your sound card supports (probably 96/24) and then convert to 44.1 at the last step using good software like sox or foobar. Its unlikely to make a practical difference but it eliminates the possibility that your hardware does a bad resample somewhere and theres no real cost to it.
If I select to record at 96/24 in a program like Audacity, is there any way to tell if the D-A conversion is actually 96/24 instead of an up-conversion from 44/16? I assume that if I selected to record in an unsupported format like 96/32, it would just be converting either 16 or 24 to 32 and no actual extra precision would be gained. Am I supposed to check anything in the sound card settings, or should the D-A conversion be automatic as long as I select supported rates? Obviously I haven't done this before, so I don't know how it works.
If I select to record at 96/24 in a program like Audacity, is there any way to tell if the D-A conversion is actually 96/24 instead of an up-conversion from 44/16?
You'll have to check the documentation that came with your card or google to see if anyone knows the answer.
Audigy 2 does support 24/96 but, it's same result if you record using 16/96 (measured using RMAA) -> no true 24-bit recording possible.
To test if the resulting samplerate is 96 just record something and then check with some spectrum analysis software (as for an example, iZotope RX http://www.izotope.com/) (http://www.izotope.com/)) if the frequency range goes beyond 22.05 kHz.
Juha
Audigy 2 does support 24/96 but, it's same result if you record using 16/96 (measured using RMAA) -> no true 24-bit recording possible.
Hmm, I thought that was a limitation of the Audigy 1 that was fixed in the Audigy 2, but maybe that was just for playback and not recording. I guess I might as well just stick with 16, then. Thanks for the information.
Hi,
I recorded a lot LP's and I have found that doing so at 24bit 88.2KHz or higher is best.As clickrepair and other software works better and play back sounds better.But to get true 24bit you need to have a audio interface that does 24bit like e-mu and it best to bypass the windows kmixer with AISO or WDM/KS for the best sound.
Per our Terms of Service (http://www.hydrogenaudio.org/forums/index.php?showtopic=3974) to which you agreed upon registering, I'd like to see some samples and ABX logs indicating that you can tell the difference, else you are in violation of #8.
The final destination will be typical 16 bit 44,100 Hz, but if I decide to try do do click/pop or noise removal, is there any benefit from doing such work at a higher quality and then reducing the sample rate and bit depth later, rather than doing all the manipulation at an initial setting of 16/44?
The advantage of processing at 24 bit is that if you do a heck of a lot of DSP, then you are less likely to accumulate sufficient rounding errors that the quantisation noise becomes audible above the intrinsic noise of the source material. That said, the level of vinyl surface noise is so high that you'd have to do dozens of passes before this theoretical danger actually becomes real.
The disadvantage of processing at 24 bit is that there are some useful vinyl restoration tools out there that only operate at 16 bit. Working at 24 bit means you deny yourself access to those tools.
Also, Audacity allows you to record with 32 bit float as well, but I'm quite sure that my sound card doesn't support it. Is there any way to verify that the sound card actually does output 24 bit when you're recording at that depth?
That's not an issue when you're dealing with vinyl sources. Even if you decide to process at 24 bit, it is safe to record at 16 bit and convert to 24 bit afterwards (either as a separate step or by the recording software doing it on-the-fly).
Thanks for all the input so far. I do have a few more questions, though.
That's not an issue when you're dealing with vinyl sources. Even if you decide to process at 24 bit, it is safe to record at 16 bit and convert to 24 bit afterwards (either as a separate step or by the recording software doing it on-the-fly).
Since no extra precision is gained if you increase the bit depth afterwards, is this just a case of using a higher-quality "container" for the editing process so that each edit produces minimal rounding errors, thus when you convert back to 16 at the end, the final result has fewer rounding errors than if you did multiple edits on the original 16-bit recording?
Since recording at 24/96 seems to be out of the question for my hardware, is there really any point in recording at 16/96 or 16/88, or is there no real benefit? I would assume if there is any, it would again just be to reduce potential errors from multiple passes of processing, but since error would also be introduced by resampling back to 16/44 in the final step, is it better to just record at 16/44 in the first place?
Finally, is there a "best" strategy to deal with the recording volume? Is it better to ensure there is no clipping by setting the recording volume at ~50% and then using a program like SoX to raise the volume later while guarding against clipping, or should the recording volume be set higher (like say, to make the loudest track on the record be as loud as possible without clipping)?
Since no extra precision is gained if you increase the bit depth afterwards, is this just a case of using a higher-quality "container" for the editing process so that each edit produces minimal rounding errors, thus when you convert back to 16 at the end, the final result has fewer rounding errors than if you did multiple edits on the original 16-bit recording?
Since you computer doesn't support 16 bit float point, pretty much any nontrivial processing is going to be done at 32 bit precision regardless of what you specify. The settings you're referring to are just for recording, not processing.
Since no extra precision is gained if you increase the bit depth afterwards, is this just a case of using a higher-quality "container" for the editing process so that each edit produces minimal rounding errors, thus when you convert back to 16 at the end, the final result has fewer rounding errors than if you did multiple edits on the original 16-bit recording?
Pretty much, yes.
Finally, is there a "best" strategy to deal with the recording volume? Is it better to ensure there is no clipping by setting the recording volume at ~50% and then using a program like SoX to raise the volume later while guarding against clipping, or should the recording volume be set higher (like say, to make the loudest track on the record be as loud as possible without clipping)?
You might want to experiment a bit and see if your soundcard gives you any extra headroom when recording to floating point. At any rate I try to stay within 2-3 dB of the 16 bit max and normalize at the end using that function in Audacity. For the most part I don't have to mess with the record level once it's set.
Other things to do before going after the scratches (I think these are addressed in the Audacity documentation) are 1) normalize the zero average (remove any DC bias), and 2) apply a high pass filter at 20 hz (higher if you know there aren't any deep bass fundamentals)
Unless you know that you're going to have a problem with aliasing or plan on doing something that may require resampling (eg. pitch/time changes), I think you're probably best off recording at the sample rate you plan on using for playback.
Since recording at 24/96 seems to be out of the question for my hardware, is there really any point in recording at 16/96 or 16/88, or is there no real benefit? I would assume if there is any, it would again just be to reduce potential errors from multiple passes of processing, but since error would also be introduced by resampling back to 16/44 in the final step, is it better to just record at 16/44 in the first place?
With (24/)96kHz you'll get quite straight frequency response in frequency range of 20Hz-20kHz.
Couple reviews:
Audigy 2 - http://ixbtlabs.com/articles2/creativeaudigy2/ (http://ixbtlabs.com/articles2/creativeaudigy2/)
Audigy 2 Zs - http://ixbtlabs.com/articles2/creative-audigy2-zs/ (http://ixbtlabs.com/articles2/creative-audigy2-zs/)
Juha
I only scanned those links, but didn't see anything that suggested that frequency response wouldn't be flat from 20-20k when sampling at something less than 96kHz.
I think you're probably best off recording at the sample rate you plan on using for playback.
But, remember, if it's 44.1kHz then, maybe it's better to record using 96kHz and then after post processings, apply SRC down to 44.1kHz using good SRC software (as like the voxengo r8brain free) because of, 44.1kHz isn't native samplerate for Audigy 2 series card (it's SRC isn't very good).
I only scanned those links, but didn't see anything that suggested that frequency response wouldn't be flat from 20-20k when sampling at something less than 96kHz.
Cosh ... they have taken all graph off from review ...
Here those are within: http://ixbtlabs.com/articles2/creative-audigy2-platinum-ex/ (http://ixbtlabs.com/articles2/creative-audigy2-platinum-ex/)
(Use Internet Explorer)
Juha
Ok so they still didn't get it right with the Audigy 2. Thanks for pointing this out, Juha.
Shouldn't 48kHz still be ok, though?
Finally, is there a "best" strategy to deal with the recording volume? Is it better to ensure there is no clipping by setting the recording volume at ~50% and then using a program like SoX to raise the volume later while guarding against clipping, or should the recording volume be set higher (like say, to make the loudest track on the record be as loud as possible without clipping)?
Recording at a lower level means you sacrifice some resolution (signal-noise ratio). If you're recording at 16 bit precision and your recording level peaks at 50%, you're only using 15 bits and the S/N ratio will be 6dB worse.
HOWEVER... this is only a problem if the source material you're recording has a low enough noise level that it requires the full 16 bits of resolution. Vinyl is most definitely NOT in this category - even the best audiophile LPs pressed on virgin heavyweight vinyl will achieve a S/N ratio of no better than 70dB (on a good day with a following wind). Provided your soundcard has a noise floor below about -80dB you are safe to record at around 50% and then normalise later.
Provided your soundcard has a noise floor below about -80dB you are safe to record at around 50% and then normalise later.
That depends a lot on just what "50%" means. IF it means that the maximum sample recorded is half of the maximum 16 bit integer (16384) then it should be ok, the case you outlined. If it means the VU meter peaks at half scale, that's losing about 30 dB of the dynamic range. If it means the slider on the record level control is at 50% then it depends on the input signal and the sound card... could be way too high.
Per our Terms of Service (http://www.hydrogenaudio.org/forums/index.php?showtopic=3974) to which you agreed upon registering, I'd like to see some samples and ABX logs indicating that you can tell the difference, else you are in violation of #8.
Some declickers seems to have fixed algorithms which don't adapt to sample rate. So the same audio at a different sample rate can give a different result (e.g. different clicks fixed / missed).
My best guess is that they work better (or at least, as designed) at 44.1kHz.
I can imagine declicking agorithms that would work better with higher sample rates, but I don't know if these are used in practice. Probably not. You can't rely on clicks stretching up into the ultrasonic range, so any declicker that needs this range to work at its best probably isn't very good.
Cheers,
David.
It was actually primarily directed at the last part:
it best to bypass the windows kmixer with AISO or WDM/KS for the best sound.
Has anyone done any testing to determine the minimum number of bits required for a digital recording to sound the same as a vinyl recording? We've been doing the calculations based on the SNR of the medium (e.g. 12 bits for a ~72 dB SNR), but has anyone determined a threshold for transparency via listening tests? Does anyone think they can hear stuff below the noise floor?
Has anyone done any testing to determine the minimum number of bits required for a digital recording to sound the same as a vinyl recording? We've been doing the calculations based on the SNR of the medium (e.g. 12 bits for a ~72 dB SNR), but has anyone determined a threshold for transparency via listening tests?
The dynamic range determines the number of bits needed to represent a signal. If the noise floor and full scale value are known, the number of bits for PCM encoding is trivially computed barring unusual circumstances (sparse input signal, etc). Theres no obvious way to do a listening test here. You'd only be testing the transparency of the original vinyl (which is probably not transparent) or your own ability to hear noise over music (which is really just a test of how much damage you've done to your own hearing).
We've been doing the calculations based on the SNR of the medium (e.g. 12 bits for a ~72 dB SNR)
I have the test record by Floyd Toole, with excellent Dynamics, and other studio test records..
The typical level of the noise floor running the test record is about -40db with a well running thorens td 125 and a denon dl 160 on an SME III arm. Where do you get the -72 db from?
There is the noise level as recorded and the noise level after final processing, and what one is measuring as the noise level. Between`tracks, the Average RMS level runs about -56 to -54dBfs on my recordings. Obviously the numbers could be different with a different cartridge and/or preamp.
I run cartridge to phono preamp to soundcard inputs. A few orchestral recordings have had peaks within about -0.2dBfs, although many have had nothing above -12dBfs, sometimes lower. Unmodulated track portions don't vary because of these differences.
I normalize most recordings, so the final figures can be different, but most of my processing is done before normalization. After rumble filtering, declicking, and noise reduction, that between tracks number has often dropped below -70dBfs.
Peak readings will be significantly higher than -70dBfs (and start out around -40dBfs before any processing) but I've investigated selections a number of times and find this is nothing like the peak-RMS of a sine wave. Those higher level peaks are only infrequent spikes making up a tiny percentage of the total unmodulated recording.
Has anyone done any testing to determine the minimum number of bits required for a digital recording to sound the same as a vinyl recording?
I just did a very casual test using small section of a typical vinyl rip that I did some time back. The original rip was made at 16 bits with an M-Audio AP2496 soundcard. (The turntable/arm/cartridge/preamp was Linn LP12/Ittok/Karma/Naim NAC42.5). The LP was a fairly typical album from my collection - not a test record or "audiophile" pressing, but a normal album in good condition. I deliberately chose a section that was fairly quiet (final part of a tune with delicate piano and tinking cymbals).
I prepared the reduced bit depth samples by cutting the amplitude by 6dB and then raising it back by 6dB. Each such cut/boost cycle gives a 1 bit reduction (as near as makes no odds). I did all this at 16 bit without dither to provoke approximately the same sort of quantisation noise you'd get with native recording at those bit depths.
It was trivially easy to hear the difference between 16 bit and 11 bit - ABX log follows:
foo_abx v1.2 report
foobar2000 v0.8.3
2010/12/15 09:23:00
File A: file://C:\clive\audio\The Guest Stars\Out At Night\x11.wav
File B: file://C:\clive\audio\The Guest Stars\Out At Night\x16.wav
09:23:00 : Test started.
09:23:33 : 01/01 50.0%
09:23:37 : 02/02 25.0%
09:23:40 : 03/03 12.5%
09:23:48 : 04/04 6.3%
09:23:55 : 05/05 3.1%
09:23:59 : 06/06 1.6%
09:24:03 : 07/07 0.8%
09:24:06 : 08/08 0.4%
09:24:12 : 09/09 0.2%
09:24:18 : 10/10 0.1%
09:24:20 : Test finished.
----------
Total: 10/10 (0.1%)
I could not tell the difference in an ABX between the 16 bit and 12 bit versions. Note that I was doing this test on a laptop with a not especially sophisticated audio path, using fairly run of the mill headphones (Sennheiser 435). Therefore I think it is conceivable that it might be possible to distinguish the 12 bit version with better playback equipment and/or other music samples.
So: how does this answer the question of the minimum number of bits needed to transparently record an LP? Well, my argument is that if there is no difference between 12 and 16 bits, adding still more isn't going to help. My conclusion is that 13 bits is probably enough to faithfully record typical vinyl sources, although you'd then have to be very careful about recording levels to ensure you actually made use of all 13 bits.
We've been doing the calculations based on the SNR of the medium (e.g. 12 bits for a ~72 dB SNR)
I have the test record by Floyd Toole, with excellent Dynamics, and other studio test records..
The typical level of the noise floor running the test record is about -40db with a well running thorens td 125 and a denon dl 160 on an SME III arm. Where do you get the -72 db from?
Good question. The -72 dB does
not reflect data. I used it as a hypothetical number for a dynamic range calculation, which is why I was careful to write "(e.g. 12 bits for a ~72 dB SNR)," instead of "(
i.e. 12 bits for a ~72 dB SNR). Sorry for not stating that more explicitly.
It was trivially easy to hear the difference between 16 bit and 11 bit - ABX log follows:
I could not tell the difference in an ABX between the 16 bit and 12 bit versions.
Interesting. Thanks for doing the experiment.
To test if the resulting samplerate is 96 just record something and then check with some spectrum analysis software (as for an example, iZotope RX http://www.izotope.com/) (http://www.izotope.com/)) if the frequency range goes beyond 22.05 kHz.
You're confusing low-pass filtering with sampling rate - the two things are not related.
To test if the resulting samplerate is 96 just record something and then check with some spectrum analysis software (as for an example, iZotope RX http://www.izotope.com/) (http://www.izotope.com/)) if the frequency range goes beyond 22.05 kHz.
You're confusing low-pass filtering with sampling rate - the two things are not related.
I think the point was that if the samplerate was 96 kHz then there
may (and probably will) be frequencies above 22.05, but if the samplerate was 44.1 then there
can not be frequencies above 22.05.
You're confusing low-pass filtering with sampling rate - the two things are not related.
I guess we need not concern ourselves with aliasing.
I recommend you study up on sampling theory before resurrecting dead discussions in a clueless attempt to strike down legitimate information.
There's no such thing as a dead discussion - this is a forum. If you disagree with what I said then I welcome a refutation, as long as it's civil.
To test if the resulting samplerate is 96 just record something and then check with some spectrum analysis software (as for an example, iZotope RX http://www.izotope.com/) (http://www.izotope.com/)) if the frequency range goes beyond 22.05 kHz.
You're confusing low-pass filtering with sampling rate - the two things are not related.
No he isn't.
There's no such thing as a dead discussion - this is a forum. If you disagree with what I said then I welcome a refutation, as long as it's civil.
pdq refuted you nice and succinctly. Of course, even entry level signal processing / EE training would disabuse you of the tangent you're trying to drag this thread off into.