|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
71.154.233.235
Hello all,
Recently, I have begun to digitize my music collection, and wonder...
Is there a difference between recording at the absolute maximum volume (right up to .5db of clipping on peak program level) and not worrying about it and recording 2-3 db below clipping.
Seems to me I have been told, back in the day, to record as hot as possible without clipping to get the most bit depth in the recording, BUT since I am getting to the point where "senior moments" are starting to occur, I thought I'd ask this august body for input.
I have been spending a significant amount of time tweaking and tweaking to get the volume at just under clipping. It takes some trial and error and analyzing the performance to get it right. I wonder if I am chasing a ghost.
Thanks,
Phil
Go, Muffin!!!!!!!
Follow Ups:
It might take a while to locate where in the track this takes place, but it's worth it. I would then adjust levels based on that point.
Thanks for the information, Todd. This is what I have been doing and not really worrying about the exact peak amount as long as it was not more than -3 or -6db (meter resolution precludes a more accurate measure without a magnifying glass!).
Phil and Muffin (RIP)
With analog, clipping is generally soft the first few dBs and gets progressively harsh as you get past the saturation point. In digital, there is no "gray area" saturation point -- only hard and nasty clipping after the "bit bucket" is full.
However, the fact is that some DACs process the signal in such a way that it is best that they not be fed a signal hotter than -3dB. At that point, it seems they will reach 0dB internally and a recorded signal above the -3dB point actually causes problems for the D-A converter. This applies to PCM. For DSD recording, I believe the max signal recorded should be no higher than -6dB (due to a similar reason).
There is a discussion at computeraudiophile.com (a few recording people contributed in the thread -- Barry Diament, Cookie Marenco and maybe a few others) re: the subject of recording levels in digital. Somewhere here at AA, I believe it was Christine Tham -- a few years ago -- brought up this subject because she was hearing what sounded like clipping from some of her discs, yet they were not "clipped" in the recording: they were (just) below 0dB.
I'm sure a search here and there will bring up relevant info.
... there seem to be exceptions. My only experience burning digital from analog sources is via a Tascam CD-RW700 recorder. While I always try to keep recordings out of the "red," there have been occasions where I slipped up. In some cases, REALLY slipped up. But after listening to the finished product, I have never detected anything harsh-sounding and have been wondering why. Just lucky?
It depends on what is clipped and how long. If a single drum hit is clipped for a few samples it will be inaudible. If a steady tone is clipped for multiple samples on multiple peaks of the waveform it will be audible if there is other music going on at the same time that is effectively "muted" while the clip is in effect. Generally it is possible to suspect something is wrong on a single clip in a classical orchestra climax. However, often this suspicion proves to be incorrect after inspecting the actual waveform. Sometimes several clean acoustic waveforms peak simultaneously and result in distortion in ones ear. (This can be experienced at live orchestra concert if one sits in front rows.)
There is software, such as iZotope RX, that can do a fairly good job of correcting clipped waveforms by "faking" the missing content. This will fool the ear if this is done sufficiently infrequently, so sometimes this works to fix a bad recording.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
There's a soft-clip circuit or limiter in the signal path. It would be set to only spring into action very close to 0dB.
0dB on the meter really isn't 0dB, but maybe a few dBs lower. I've seen peak levels levels on cassette tapes that I've recorded show up above and below my decks meter reading when played back on other tape decks.
What it has to say on the subject: "The red OVER indicators on the meters should never light. Unlike analog equipment, digital audio units produce extremely unpleasant sounds when distorted and there is no 'headroom' after the 0 mark."
After 10 years, still waiting for its CDs to emit any "extremely unpleasant sounds," so I guess I HAVE been lucky, given all the times the red OVER indicators have lit.
Thanks to you both.
If the cassette deck is playing a tape that is reading around 0 on the dec's VU meter then you will want the digital peak levels to be around -11 dBfs or even a bit lower if you are using metal tapes, as these can easily put out undistorted +10 analog signals.Soundforge has a mode for it's "channel meters" that provides a variety of VU scales that average over a short period. The best one to use is the Nordic PPM. If it's reading above 0 VU it gives a warm feeling of "loud enough" while still retaining substantial head room. Other editors may have similar capabilities.
You can make up a calibration tape of band limited pink noise, that you can get from Bob Katz' web site. Record this on the tape at 0 VU and then adjust your recording gain so that you get the same levels on a copy you make of the cassette tape as you get when you play the original file you downloaded. I believe this will work for all but the very hottest cassette tapes, but I haven't tried it because I'm not set up to record on my deck at present, so let me know if this doesn't pan out.
I list various ways of approaching this problem not because it is necessary to use all of them, but to give a feel for what might be available. There is even a better system that Bob Katz advocates and that is to ignore all the meters completely after calibrating one's speakers to a fixed setting. Then when one makes a recording (e.g. does a transfer) one simply adjusts the volume one hears in one's speakers until it sounds "right". One needs his test file and a Rat Shack SPL meter to set up the system plus repeatable settings on one's volume control.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
Edits: 03/26/12
... I've used the Tascam only to record from vinyl and off-air from a tuner.
If a DAC clips when playing a digital file that is unclipped it is, to put it simply, improperly designed. Either there was insufficient headroom in the analog circuitry (BAD) or insufficient headroom in the digital signal processing (WORSE). Designers do this because they are either (a) incompetent or (b) dishonest. One can get better measured signal to noise ratio by building in more gain to the upsampling routines inside the DAC (or DAC chip if this is done in the chip). Since many idiots buy equipment based on published S/N specifications rather than sound, there is economic pressure on designers to cheat.
It's pretty black and white if the DSP in a DAC clips while doing upsampling, as this creates distortion up and down the F scale due to aliasing, with horrible result. It's not so horrible (or necessarily obvious) when the clipping takes place in the analog circuitry as this creates harmonics up the F scale. If one suspects a DAC of clipping then one can put the digital file through a digital volume reduction. With most players this can be done without creating a new file, e.g. using a digital volume control built into the player.
It is helpful to be able to hear and recognize the difference between digital and analog clipping while making transfers. On one occasion I was transferring a "hot" metal cassette tape and heard clipping, so I reduced the input level to my converter. It turns out that the problem wasn't the converter clipping, it was the output stage in my tape deck that was clipping on the very hot tape. It was necessary to turn down the output volume control on the cassette deck to avoid this distortion and turn the gain back up in the ADC to get good digital levels. This kind of gain-staging problem is typical of consumer setups or improperly set up professional equipment.
To get the best possible results with one's equipment one needs two skills: an understanding of how the equipment and how to use it within its limitations and the ability to hear and reject various distortions created by misbehaving or abused equipment. Without both of these skills and incredible diligence motivated by love of music one will not get consistently good results as it only takes one slight screw-up to ruin an otherwise good recording. (At least when doing transfers one can go back and redo them, unlike live recording.)
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
The more I read Tony's insights, the more I appreciate them.
Having cranked through more that 250 vinyl-to-digital restorations, I have come to believe that the recording step should document the source as accurately as possible without manipulation or embellishment (i.e. as accurately as you normally play the material on your setup). You shouldn't take extraordinary steps because this may be the "last take." Recording is linear and real-time -- for a large collection, you only want to do it once, but if you do it right, once is enough. (Imagine the task in front of the team that has to digitize the 20,000 John Peel LP collection.)
The time to choose the optimal final sound level is after all the clean and restoration steps have been taken. You can always re-visit the decisions later, when the time investment is only changing to a new tool, or adding a new operation, or tweaking settings.
In the work that I do (mostly with cassette tapes) I keep three levels of each project:
1. The original digital transfer, as digitized with no editing, level changes, etc. (Sometimes there may be more than one transfer if there was some kind of a glitch when playing the tape, in which case I will keep both if there is the slightest possibility that the "inferior" one might not be so inferior in some dimension. These are generally done in high resolution, e.t. 88/24.
2. A "studio master version" after all restoration work, editing, and mastering. This will be in the original hi-res sampling rate but at bit depth of 32 bits floating point. This allows for further processing with the minimum possible loss of resolution since there won't be any conversions to/from floating point to a integer file formats. (By working in 32 bit there is no danger of an intermediate result causing clipping in the work file, but to play it back or render as 24 bit or 16 bit integer it may be necessary to adjust levels.) This will typically be a single file divided into markers or regions. If there have been multiple steps of processing and some chance that there may be a desire to redo some aspect of the project, I may keep multiple working versions, e.g. an edited version but without any pre-mastering, and the final version that includes additional processing such as EQ. In addition, if there were many hand edits involved in the process of clean up (e.g. removing clicks and pops on a vinyl transfer) I may keep a additional versions. On most projects, however, I keep only the single studio master after I have completed the project. I also keep a text log file of all of the steps that I've done so that I can redo any of them if desired.
3. A distribution version, produced from the studio master version after reducing down to the final sample rate and bit depth. (e.g. 44/16). These files will be used to burn a master CD-R if the recording is going to be duplicated and as source for subsequent release as downloads. Each track will be a separate file and will be an integral multiple of 1/75 second frames (so the file can be burned correctly to CD without any problems in the gaps). In some cases there will be distribution versions at multiple sample rates, e.g. 88/24 and 44/16.
Each of these versions is stored on my audio workstation. In addition every day these files are synced to a file server on my network that keeps a second copy on a RAID array. Periodically, this file server is backed up to offsite storage. So as you can see, one hour of music may use a lot of storage space with at least three files (two hi-res) stored in three separate places. For those files that are actually released for download additional copies and additional backup copies occur on the server machine(s).
I'm not suggesting that everyone be so paranoid, it's just what has worked for me. It doesn't take much time to get into this workflow once the system is set up. It is a pain, however, to have to manage multiple computers. I have three Windows PCs that I have to manage, a Linux based NAS, a router and four hosting accounts on various server companies.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
I appreciate your replies.
I have been within -6db on peaks and am recording at 24/96 that I convert to flac.
So far, I haven't heard anything untoward, but the thought stuck me this afternoon as I was recording some classical which goes from a whisper to a roar. The quiet parts got me to thinking...
Anyway, thanks for your comments and your time.
Phil and Muffin (RIP)
-48dB peaks in 24 bit is the same resolution as CD at full scale. Only those who are interested in "the loudness war trials / competition" need worry about trying to make up for the 'small mans' syndrome...
Go for best sound quality and use your volume control to determine listening levels.
"-48dB peaks in 24 bit is the same resolution as CD at full scale."
This assumes that the -48 dB digitization was done nearly perfectly. One might achieve an approximation to this if one were to start with a proper digitization and reduce it arithmetically, since this would get a result that was correct to within 1 least significant bit. However, if one were to feed a low level analog signal to an ADC the result would not be what you said unless the analog noise in the converter was below -144 dB. This is far better from any ADC that I've heard of. ADCs tend to have more noise than DACs and the best current DAC chips are at about 135 dB and this is under laboratory conditions. A complete DAC running in a box connected to a system will not have this performance.
I've seen 16 bit digital recordings with peaks as low as -30 dBfs that ended up sounding acceptable (if not excellent) after they were boosted digitally to a reasonable level. This will depend on the converters used, i.e. for decent results they would have to have no gross non-linearity at low levels and no loud hum, birdies or other spurious signals, and have been properly dithered. (At this level turning up the analog volume on the playback of the unboosted file would be a very bad idea. If there were enough gain to get reasonably loud music than any glitch might blow up some gear.) Of course there will be a lot of noise, but if the original was noisy to start with this might not matter too much. My comments applied to some recordings that I was given that had been recorded or transferred digitally at very low levels. They were what they were. If I had been doing a digital transfer and gotten these results I would have redone them after thinking carefully about how and why I had made such a mistake.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
You are right! I have notice that some CD's don't -fizz out- or lose sonic connectivity (whatever that means..read between the lines if you will) at low levels but some, especially ill transferred opera and classical / symphonic works certainly do. I am sure you have seen the reconstructed waveform of low level,(recorded in 16 bit) convoluted waveforms. The old Cool Edit pro (now Adobe audition) shows the reconstructed approximation, unlike Sony Sound forge.(which does not)
Since the advent of 24 bit being made available, there is no reason at all not to be using it for the recording, editing or mastering stage. I tend to set the recorders to peak around a -10 on the loudest of sounds.
"The old Cool Edit pro (now Adobe audition) shows the reconstructed approximation, unlike Sony Sound forge.(which does not)"
iZotope RX also allows such a plot. It is only an approximation as the peak factor depends on the reconstruction filter. With a perfect SINC filter there is no worst case bound for an infinitely long signal, but it can be over 10 dB for a CD track. (You wouldn't want to listen to such a recording even if it were reproduced "undistorted".)
"I tend to set the recorders to peak around a -10 on the loudest of sounds."
Sounds about right to me. Pink noise at -23 dB RMS (-20 according to the mathematically ignorant AES) peaks about at that level. This signal comes in around 0 VU according to the Nordic scale for digital audio, so there is comparable headroom here as with magnetic tape. The actual values vary a bit depending on the "crest" factor of the signal, which will depend on the musical instruments playing, the frequency response of the microphones and their location and the room acoustics.
I had one live recording where all but one of the peaks were around -15 dBfs that digitally clipped when a wind gust collapsed the performance tent. Sometimes there are worse problems than digital clipping. In this case the recording continued and a few minutes later the music resumed. :-)
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
Recording "hot" made sense on tape machines which had poor S/N ratio. It isn't necessary with digital where the S/N ratio is adequate. If you are getting peaks above -12 dB you are using enough. If you are concerned about quality, then record using 24 bits. If you recorded at 24 bits and feel the need to convert the final result to 16 bits then you may want to "normalize" the recording (e.g. complete album) to -0.5 dB before converting to 16 bits. In that case, I suggest keeping the original transfer as well as the 16 bit version. Also, it is worthwhile recording at higher sampling rates, depending on the equipment involved. The cost of the extra disk storage is minimal compared to the time spent doing transfers, even if you value your time at minimum wage. I suggest a minimum of 88/24 for digitizing LPs.
As with analog recorders it is necessary to become familiar with one's equipment and learn the effects of various use and abuse of the gear. There is no "right answer" rather it's what sounds best to you. Be aware, however, that if you have a large collection to digitize over the course of the project you will gain experience and may end up with a different perspective.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
For those who are trying to record @ 24-bit on Windows with Audacity, I was surprised to learn that the program cannot support it, because of a license issue that would not allow the project to stay open source. So even if you set Audacity to 24-bit input and your audio input device can support, Audacity will still pack 16-bit samples into 24-bit words.
Thanks for the example. This is just one example where audio software may "silently" trash the quality of one's recording. This problem is not unique to Audacity, there were similar problems with earlier versions of Soundforge.
There is no guarantee that any editing feature will not mangle the sound quality. Even more subtle than reducing 24 bits down to 16 are effects like converting to/from 24 bit integer and floating point. Here the conversion from floating point back to integer may add dither noise at an "inaudible" level of -135 dBfs. Unfortunately, under some conditions and on some systems this change of +1 or -1 in individual 24 bit samples may result in an audible difference. One has to trust one's tools to be an artisan, but the rule should be "trust but verify". Verification includes technical measurements (e.g. null tests, spectrum analysis, etc.) and careful listening tests, not casual 30 second A - B comparisons.
Other fun exercises one can play is to use the editing features to deliberately trash a recording in a subtle way and see how much trashing it takes before you hear it. Such things include failure to add dither, moderate amounts of clipping, suboptimal filtering for sample rate conversions, etc.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
nt
Post a Followup:
FAQ |
Post a Message! |
Forgot Password? |
|
||||||||||||||
|
This post is made possible by the generous support of people like you and our sponsors: