Hello,
I am currently using an EMU 1616M to rip vinyl and using Sony Sound Forge 10 to record with. I am wondering what the best settings are to use for exporting the .WAV files. I have the option to record in 24bit or 32 bit at 192 Khz and save the
.WAVs in using IEEE Float or PCM. I am wondering if it would be best to record in 32bit and bounce down to 24 when I export to .WAV, or just record in 24 if there won't be any difference. No matter what I will need a 24 bit file in the end to
convert to .flac as it is the highest bit rate the format supports.
Furthermore, I am confused as to whether to use IEEE Float or PCM. I understand the IEEE float was developed for the broadcast industry while PCM is used in the red book CD standard. I'm guessing when I convert to .flac it will end up as PCM
anyway and am not sure if I would get any benefits from utilizing IEEE Float. If anybody can shed some light on this I would be most appreciative and believe this sort of information should be included in a wiki somewhere as I have been unable to
find any useful information as to the pros can cons of IEEE Float and PCM .WAV files with regards to audiophile needs.
You do know that 24-bit with vinyl as a final product is complete overkill, right? I doubt you will be able to present a single example properly mastered to make use of the more significant bits demonstrating 16-bit to be inadequate through a properly controlled double-blind test. I don't even think you'll need to dither; as a medium, vinyl simply doesn't provide enough SNR.
Gosh, greynol! The OP didn't even mention audibility or sound quality. If I didn't know any better, you were putting words in his mouth.
Given that: this question is still reasonable to ask, strictly from a "correctness" point of view (in the same sense that we might care about bit-perfect output even if we freely admit that it does not make an audible difference).
I am wondering if it would be best to record in 32bit and bounce down to 24 when I export to .WAV, or just record in 24 if there won't be any difference. No matter what I will need a 24 bit file in the end to convert to .flac as it is the highest bit rate the format supports.
Your 1616M won't have a resolution remotely close to 24 bits (to say nothing of the preamp); the noise involved is probably going to make meaningless any sort of quality improvement effected by a higher bit depth on any DSP operations you might do inside of SoundForge.
Floating-point WAVs can be nice for a ton of different reasons, but those reasons don't apply for this sort of thing. (They can be useful for encoding >0dbFS samples, but a lot of editors will just clip that.)
Furthermore, I am confused as to whether to use IEEE Float or PCM. I understand the IEEE float was developed for the broadcast industry while PCM is used in the red book CD standard. I'm guessing when I convert to .flac it will end up as PCM anyway and am not sure if I would get any benefits from utilizing IEEE Float. If anybody can shed some light on this I would be most appreciative and believe this sort of information should be included in a wiki somewhere as I have been unable to find any useful information as to the pros can cons of IEEE Float and PCM .WAV files with regards to audiophile needs.
I'm gonna have to nitpick you on terminology here. Virtually every possible way you could save the audio will always be PCM; saving as floating-point is still PCM; lossless/FLAC is still PCM etc. You only wouldn't use PCM if you saved as DSDIFF or with lossy codecs like MP3 or any number of other totally obscure formats you will never use.
Now, if you replace "PCM" with "fixed point" then your question makes sense.... The question becomes, should you use a fixed point (integer) format, ie 24-bit or 32-bit signed, or should you use 32- or 64-bit floating point? Basically, while there are very strong theoretical reasons why you'd want to save intermediate files in floating point, for what you are dealing with, they are pretty much guaranteed to be utterly insignificant, both from mathematical and audible points of view. The best reason why one would really need to save floating point WAVs is if either they are dealing with data that intrinsically cannot be limited to <0dbFS, or if they are dealing with purely synthetic signals, generated mathematically rather than acquired via ADC. Neither situation applies to you.
Where did you hear that the broadcast industry cares about floating point formats? That's somewhat surprising to me, but I am not knowledgeable about their practices.
Gosh, greynol! The OP didn't even mention audibility or sound quality. If I didn't know any better, you were putting words in his mouth.
The comment, "No matter what I will need a 24 bit file in the end to convert to .flac as it is the highest bit rate the format supports," raised a flag. If the OP already knows the information, then I don't see any harm as opposed to the potential risk in validating a possible and all-to-common misconception that throwing more of those pesky 1s and 0s at an analog source will better preserve the quality of the audio.
I have the option to record in 24bit or 32 bit at 192 Khz and save the .WAVs in using IEEE Float or PCM. I am wondering if it would be best to record in 32bit and bounce down to 24 when I export to .WAV, or just record in 24 if there won't be any difference.
There are no 32-bit or floating-point analog-to-digital converters or digital-to-analog converters. Your interface is 24-bits. (Most "consumer" soundcards are 16-bits.) If you "record to" higher resolution than you can capture,
you are just wasting disk space.Most audio editors work in floating-point (32 or even 64 bits) internally, no matter what you load-in. The sample rate (kHz) stays the same. As Axon suggested, you
may have reasons for saving an intermediate file in floating-point. But, for your original recording and your "final release", there is no benefit.
If you listen to a vinyl record between tracks, and to a CD (16-bits) between tracks, (or very-quiet passages) it's pretty clear that 16-bits has
far more dynamic range than analog vinyl... In other words,
16-bits is plenty! It turns-out that 16/44 is better than human hearing...
Don't believe the audiophile hype! Most of these guys making claims about "high-resolution audio" have never done proper
blind ABX tests, and they are fooling themselves. And/or they make excuses about why blind testing doesn't work... Maybe there are a few "golden ear" audiophiles that can hear a difference between CD and a higher-resolution recording, but I remain skeptical. And, some of these guys actually
prefer the "warm-crackly sound" of vinyl! Anyone who "likes noise" has different goals than me and his/her opinions & advice are not of much use to me.
convert to .flac as it is the highest bit rate the format supports.
FLAC is lossless, period. When you decode/playback a FLAC, you
always get-back the exact-original PCM data. When you encode FLAC, you don't "choose" a bitrate or quality setting. The different settings determine how much "work" goes into squeezing the file. A FLAC made from a 24/192 file will have a much higher bitrate (and larger file size) than a 16/44 file (just like the uncompressed WAVs).
P.S.
Speaking of noise..... I don't use Sound Forge, and I'm not sure what tools it has for removing "snap", crackle", and "pop". For me, noise reduction is the most important part of digitizing vinyl. (Far more important than bit-depth & sample rate! )
If Sound Forge doesn't do an adequate job (this is special type of noise reduction) here (http://www.delback.co.uk/lp-cdr.htm#clean_pops) are some software suggestions (and a ton of other helpful-related information). I use Wave Repair[/u] (http://www.delback.co.uk/wavrep/)[/url] ($30 USD). It does an amazing job with most clicks & pops in the manual mode, and it only "touches" the few-milliseconds of audio where you mark a defect. The downside is that it usually takes me a full weekend to clean-up a vinyl transfer. (I always buy the CD if it's available!)
Audacity (http://audacity.sourceforge.net/) is a free program analogous to SF/CoolEdit(Audition) that does click removal and NR out of the box. I hope SoundForge does a better job, since Sony charges for it, but Audacity will give you a free point of comparison. I've heard good things about the ClickRepair Programs (http://www.clickrepair.net/software_download/), which have a three week trial period, but no direct experience. Goldwave has a trial version as well IIRC.
As everyone else has pointed out, 24 bit recordings (of an analog medium as inherently limited as a vinyl LP) are a massive waste of hard drive space with no value other than a psychological one. Perhaps if you had some good studio master tapes (with dolby studio or whatever) then it might be worth considering.
Even assuming never played, perfectly stored, new-in-sleeve 180/200 gram virgin vinyl records perfectly cut directly from the master tapes at half speed , 24 bit is a massive waste of space (and time, the larger the file the longer it takes to open, save, manipulate, etc.). The SNR of records in real life is crap. To be precise (http://www.angelfire.com/super2/animorphs/Tintin.html), warmed over, day-old refried dog crap of the most offensively odoriferous variety.
I'm going to assume that your target records have been previously played (and hence have notably decreased HF content), are imperfect to begin with (record masters being what they were), and are being captured from a good quality record player that is not calibrated daily.
Provided the EMU 1616 ADCs perform comparably well at all available resolutions and word lengths, 16 bit is what you want. I've seen a few computer audio solutions that perform rather differently (at least judged by the fallible RMAA) depending on setting, but it seems nearly impossible that an EMU semipro rig would have such issues.
If you're enamored of hand-waving and demonstrably useless audiophilic woo-woo (http://www.synergisticresearch.com/galileo-system/galileo-system-speaker-cable/), record at 16bit/96kHz, process the recording for noise and then resample down to 16/44.1. That way people with $100k tube amps (http://www.soundscapehifi.com/kondo.htm) won't point and laugh (especially when the roles should be reversed (http://www.parts-express.com/pe/showdetl.cfm?Partnumber=310-300)).
If for some inexplicable reason you do decide to capture at 24 bit, dither down to 16 bit for storage & playback anyway. Your dog, your hard drive, your portable player, and the universe itself (http://en.wikipedia.org/wiki/Heat_death) will thank you.
For editing I´d leave it at 32 bit floating point, that´s what I do. The only consideration is that it takes slightly longer than editing in 16 Bit - as one poster above stated most programs work with 32 bit floating point internally anyway (Sound Forge: 64). I´d even leave it at 24 bit after you´ve finished your de-noising, de-crackling, equalization etc. If you have the space why not leave it at that? Your PC won´t get slower, most hardware can play it. Folks around here will tell you that it´s overkill, which is true for vinyl - but have you considered that you might have the desire to re-edit your recording, for example when you encounter a problem you didn´t see before? For such cases it´s best to leave it at such bit depths.
And if you rip with 192 you might want to downsample to 48 kHz afterwards. Vinyl rarely has frequencies beyond 16 kHz. Most of the things you can see with a spectrum analyzer are distortions.
convert to .flac as it is the highest bit rate the format supports.
FLAC is lossless, period. When you decode/playback a FLAC, you always get-back the exact-original PCM data. When you encode FLAC, you don't "choose" a bitrate or quality setting.
From the context it's clear he meant the highest bits/sample that FLAC supports. Or bit rate of the uncompressed file.
Thankyou all for the information. I quickly realized after posting the 32 bit conundrum considering as you said there are no 32 bit dacs. I also appreciate the information regarding integer vs floating point as I was merely regurgitating program parameters. As far as the broadcast industry is concerned I can't remember where I cobbled up that information. I had orignially assumed that a higher sampling rate would result in a more accurate copy of an analog signal given the increase in samples per second, yet it seems as you say 44.1khz is more than enough to replicate everything in the 20hz-20khz bandwidth according to Nyquist. This then makes me wonder why we even have 24/192 dacs in the first place except to sell stuff or make use of some psychoacoustical effect of higher frequencies or if there is a secret audiophile community of dogs somewhere................. ...................................... While certainly noise will always be an issue with vinyl to an extent, I will not take sides in the vinyl vs cd debate as I have heard examples playing to the strengths and weaknesses of both formats and putting all the technical aspects aside preference for one or the other is merely subjective in the end.
Soundforge is a terrific program with very powerful dsp features, although I fear none of them are very helpful in dealing with pops and whatnot, however its editing interface is fantastic. I am somewhat curious how well something like Izotope RX2 works as it is a professional grade VSTi
Thanks
The argument that 16 bits is overkill for vinyl is that the SNR of vinyl playback is less than what 16 bits can provide (96 dB or so). That sounds completely reasonable. I have a question though, about the nature of noise in an audio recording. Can we hear into the noise? In other words, can we hear things below the noise floor? If so, would we need to use some figure in addition to SNR to describe the amount of audible information that could potentially be captured in the playback of a vinyl recording?
We can certainly hear single tones well below the rms noise, but people don't generally listen to single tones. Speech, music, etc. are not heard nearly so well below noise level.
I don't know if anyone has mentioned this, but while 16 bits is surely more than adequate for recording vinyl, the same may not be true for 44.1 kHz. Clicks and pops can have frequency content well above 22 kHz, and including the higher frequency content can be useful for their removal. This is not to say that one needs anywhere near 192 kHz, however.
For post-processing sake, wouldn't it be better to have 192k samples/s than 44.1k samples/s and bit-resolution as many bits above 16 than it's possible (meaning use of ADC max bit-resolution))?
Juha
The argument that 16 bits is overkill for vinyl is that the SNR of vinyl playback is less than what 16 bits can provide (96 dB or so). That sounds completely reasonable. I have a question though, about the nature of noise in an audio recording. Can we hear into the noise? In other words, can we hear things below the noise floor? If so, would we need to use some figure in addition to SNR to describe the amount of audible information that could potentially be captured in the playback of a vinyl recording?
The 16-bit digital quantisation noise is
way below the vinyl noise - and it's possible to hear "into" either (but not at the same time, since vinyl noise completely hides digital noise, which is already inaudible at reasonable volume settings anyway).
Plenty of threads on this. Some even in the FAQ
Cheers,
David.
The argument that 16 bits is overkill for vinyl is that the SNR of vinyl playback is less than what 16 bits can provide (96 dB or so). That sounds completely reasonable. I have a question though, about the nature of noise in an audio recording. Can we hear into the noise? In other words, can we hear things below the noise floor? If so, would we need to use some figure in addition to SNR to describe the amount of audible information that could potentially be captured in the playback of a vinyl recording?
Things you could hear in/below the vinyl noise floor would be recorded with 16 bits as well. Guess you mean 'dynamic range' which 16 bit offers plenty (above 140 dB with dithering iirc).
IIRC, I read long ago on some site that using high samplerate (192kHz) in recording improves the AD conversion quality at higher frequencies (inside Nyqvist range) because of there are more samples to present the analog data as digital then (i.e. how many samples represents lets say 12kHz tone if I have understood it correctly).
It wasn't this site - http://electronics.howstuffworks.com/analog-digital3.htm (http://electronics.howstuffworks.com/analog-digital3.htm) but, maybe this site supports what I tried to remember.
Juha
It's very simple, really.
If the analog signal contains NO content above Nyquist, then it can be reproduced EXACTLY by sampling at 2x and then converting back to analog. End of story!
It's very simple, really.
If the analog signal contains NO content above Nyquist, then it can be reproduced EXACTLY by sampling at 2x and then converting back to analog. End of story!
In reality however it's not that simple. You'd need perfect brickwall filters ... so in reality you can expect to see the effects of a low pass filter down to something like 0.9*nyquist.
Doesen't the analog (vinyl) signal always contain content above Nyqvist (when recorded > 44.1kHz)?
Juha
Doesen't the analog (vinyl) signal always contain content above Nyqvist (when recorded > 44.1kHz)?
Juha
"content"? No. Not always. Not even usually. Very occasionally.
Noise and distortion: yes.
Anyway, how well do your ears work above 22kHz?!
Cheers,
David.
"content"? No. Not always. Not even usually. Very occasionally.
Noise and distortion: yes.
Anyway, how well do your ears work above 22kHz?!
Hmm... least 2nd - nth harmonies of tone can be seen above 22kHz in spectrum analysis (aren't those part of the base tone least when tone is from acoustic instrument)? In my understanding, if harmonies are removed from tone the original 'sound' changes a bit (another question is can this difference be heard if the removed harmony is outside human hearing range)?
My ears works well even @ 192kHz but I can't hear/feel those high frequencies if levels are that low (I'm over 50 so maybe I can hear somewhere upto around 18kHz nowadays).
Juha
Doesen't the analog (vinyl) signal always contain content above Nyqvist (when recorded > 44.1kHz)?
'Nyquist frequency' refers to the frequency that is half the sampling rate, i.e. if the sampling rate is 1000 kHz the nyquist frequency is 500 kHz, so this question doesn't make much sense.
A vinyl record can contain frequencies of 96 kHz, probably even higher.
Hmm... least 2nd - nth harmonies of tone can be seen above 22kHz in spectrum analysis (aren't those part of the base tone least when tone is from acoustic instrument)?
[...]
My ears works well even @ 192kHz but I can't hear/feel those high frequencies if levels are that low (I'm over 50 so maybe I can hear somewhere upto around 18kHz nowadays).
Guess you're talking about fundamental frequencies and harmonics here.
see frequency instruments chart (http://www.independentrecording.net/irn/resources/freqchart/main_display.htm)
Which instrument produces fundamentals above 10 kHz?
I don't understand how your 'ear works well @ 192 kHz' but cannot hear above 18 kHz.
Guess you're talking about fundamental frequencies and harmonics here.
Yes harmonics.
Juha
As everyone else has pointed out, 24 bit recordings (of an analog medium as inherently limited as a vinyl LP) are a massive waste of hard drive space with no value other than a psychological one.
24-bit files use 50% more space than 16 bit. With storage at less than 10 cents per gigabyte, that certainly is not a massive concern.
I don't think we'll get anywhere trying to determine the inherent resolution of an analog medium. What we do know is that usable resolution of a decent ADC is greater than 16 bits and no more than 24 bits. To
ensure no loss of data, 24-bit is a reasonable choice.
For post-processing sake, wouldn't it be better to have 192k samples/s than 44.1k samples/s and bit-resolution as many bits above 16 than it's possible (meaning use of ADC max bit-resolution))?
Some ADCs have better S/N performance at lower sample rates. Switch to 192 and you get more bandwidth but reduced S/N. It is not safe to assume that highest sample rate gets you the best performance.
This then makes me wonder why we even have 24/192 dacs in the first place except to sell stuff or make use of some psychoacoustical effect of higher frequencies or if there is a secret audiophile community of dogs somewhere.
In some situations, it is possible to hear subtle difference between 48 kHz and 192 kHz recordings. As far as I understand this is because of differences in the anti-aliasing filters, not due ultrasonic hearing or anything exotic like that. The differences, when they exist, are probably too subtle to accurately compare quality but listeners and sales people will uniformly assume that 192 is better.
IIRC, I read long ago on some site that using high samplerate (192kHz) in recording improves the AD conversion quality at higher frequencies (inside Nyqvist range) because of there are more samples to present the analog data as digital then (i.e. how many samples represents lets say 12kHz tone if I have understood it correctly).
It wasn't this site - http://electronics.howstuffworks.com/analog-digital3.htm (http://electronics.howstuffworks.com/analog-digital3.htm) but, maybe this site supports what I tried to remember.
Juha
That's audiophile crap. 24/44.1 is more than enough to make needledrops.
It's very simple, really.
If the analog signal contains NO content above Nyquist, then it can be reproduced EXACTLY by sampling at 2x and then converting back to analog. End of story!
It's a common nonsense. In short, no, unless you have an infinite stream of samples.
Longer answer (https://extrememoderate.wordpress.com/2011/05/25/dont-abuse-nyquist-shannon-theorem-like-this/)
This has been an interesting thread which has degenerated into the debate between "audiophile crap" vs. 44.1/16.....I wonder, is there a definitive be all and end all empirical study (or set of studies) that one can read to prove EITHER side? Just a simple google search the other way and one can also find other zealots pitted against the "HA crap." If there are no such studies for either side, then it is crap+crap (or maybe crap^2).
It's a common nonsense. In short, no, unless you have an infinite stream of samples.
Longer answer (https://extrememoderate.wordpress.com/2011/05/25/dont-abuse-nyquist-shannon-theorem-like-this/)
I already explained that things look different in practice. If something is nonsense then the linked 'longer answer'.
@ron: http://www.aes.org/journal/online/JAES_V55/9/ (http://www.aes.org/journal/online/JAES_V55/9/)
A carefully controlled double-blind test with many experienced listeners showed no ability to hear any differences between formats. The results were that 60 listeners over 554 trials couldn’t hear any differences between CD, SACD, and 96/24.That's just one but an in depth search should yield a couple of such tests (the results usually are that no one could hear any differences).
I already explained that things look different in practice. If something is nonsense then the linked 'longer answer'.
No, you stated that things are different. But anyway, I started writing the blog post before you did and when it was ready and I realized that there was some talk about it, I stopped for a while to think whether posting it here still makes sense and decided that it makes more sense than not doing so.
Now if you think what I wrote is nonsense, provide some arguments because such statements are useless w/out a stated basis.
@xnor
thanks for the links.....will read
I do find it very interesting the zeal with which both sides of the argument state their case. Seems like there are numerous labels remastering to 96/24, perhaps the better master is all that is really causing the effect?
Now if you think what I wrote is nonsense, provide some arguments because such statements are useless w/out a stated basis.
"decoder doesn’t use Shannon-Nyquist algorithm to recreate it"
This is nonsense because Shannon deals with recreating
analog signals; an audio decoder has nothing to do with this.
"music rarely consists on infinite repetitions of the same signal, and that’s what Shannon-Nyquist theorem requires to work"
Shannon doesn't require any repetitions and an infinite signal is only needed for
exact reconstruction. It still works with finite signals but the result will be an approximation of the original which can be good enough for us not to be able to detect a difference.
The 3999.5 Hz wave cannot be reconstructed perfectly because you have to give the bandlimiting low pass filter some room (like I've explained in my initial post).
But it's questionable how this file was created and resampled in the first place..
The interpretation of the waveform doesn't make any sense to me, nor does the conclusion for reasons I just mentioned.
I'm sorry if all of this sounds harsh.. and also for off-topic.
edit:
I do find it very interesting the zeal with which both sides of the argument state their case. Seems like there are numerous labels remastering to 96/24, perhaps the better master is all that is really causing the effect?
Discussions get heated when people can't control their emotions (I'm not referring to anyone in this thread).
You're right, the master dictates everything and a bad master will always sound worse pretty much regardless of the format. (I have seen that it's not uncommon in the industry to use different masters for CDs compared to 'high-quality' downloads or SACDs for example. I was pretty shocked when I first saw this: the waveform of a 16/44.1 download and the 24/96 version of the same song looked very different. I don't know what they did exactly but at least the dynamic range and compression was different.)
The 3999.5 Hz wave cannot be reconstructed perfectly because you have to give the bandlimiting low pass filter some room (like I've explained in my initial post).
Bingo!
"Now before I get to details, a remark: I never studied signalling theory"
Clearly.
Anybody who is seriously claiming that 16 bits is unequivocally sufficient for recording vinyl is nuts. By all means, record at 24 bits if you can: the benefits will at best be barely significant in a rare percentage of cases, but unless you are really crunched on space, the cost is basically zero.
I just measured the minimum spot noise margin of a raw 24-bit recording of a needledrop (vinyl background noise to the 16-bit TPDF floor). It's 13db. And I wasn't even trying very hard to find a quiet record, either.
Certainly recording at 24 bits is not a sufficiently big deal so as to care about rerecording something that was done at 16 bits, and certainly 16 bits is much more than adequate for the final product, and it may be true that I just have an unusually quiet setup (I do try). But "vinyl is far noisier than 16 bits" doth not a 13db noise margin support.
Man so many people came out of the wood work over this, so many different opinions. I think I am going to have to just experiment and try everything
It's very simple, really.
If the analog signal contains NO content above Nyquist, then it can be reproduced EXACTLY by sampling at 2x and then converting back to analog. End of story!
That isn't what the theorem says. You have to sample at greater than 2x. At exactly 2x you will always sample at the same phase, but no information about what that phase is, so no information about the amplitude. You might sample at the zero crossings and the signal will not appear to be there at all.
"music rarely consists on infinite repetitions of the same signal, and that’s what Shannon-Nyquist theorem requires to work"
Shannon doesn't require any repetitions and an infinite signal is only needed for exact reconstruction. It still works with finite signals but the result will be an approximation of the original which can be good enough for us not to be able to detect a difference.
Yep. That's the point of my post, reconstruction has to be imperfect. I think I made it clear I don't know how does it sound in practice.
The 3999.5 Hz wave cannot be reconstructed perfectly because you have to give the bandlimiting low pass filter some room (like I've explained in my initial post).
Are you trying to tell that if the stream was infinite it wouldn't work too? I'm not 100% positive about it, but I think it would. And with long enough signals it could be decoded with arbitrary precision, but still it wouldn't be because decoders don't attempt to do it.
But it's questionable how this file was created and resampled in the first place..
Sure. You're free to repeat the experiment.
You're free to repeat the experiment.
Contrary to what you state, the file is not sampled at 8000 Hz; it's sampled at 44100. Presumably what you have done is taken a 8k file and resampled it, badly. A resampler with better stop-band rejection would have left nothing audible in the output; a resampler with a higher cut-off frequency would have rendered the sine wave correctly. As has been pointed out already, filters with such narrow stop bands as would be needed in this example are difficult to achieve in practice; however, they are unnecessary: no-one is claiming that 44.1k is good (in practice) for any signal <22.05k (or that 8k is good for anything <4k). 8k came about from telephony where the target upper frequency was around 3.3k, far below the 3.9995k you are trying to squeeze out of it here.
Contrary to what you state, the file is not sampled at 8000 Hz; it's sampled at 44100. Presumably what you have done is taken a 8k file and resampled it, badly. A resampler with better stop-band rejection would have left nothing audible in the output; a resampler with a higher cut-off frequency would have rendered the sine wave correctly. As has been pointed out already, filters with such narrow stop bands as would be needed in this example are difficult to achieve in practice; however, they are unnecessary: no-one is claiming that 44.1k is good (in practice) for any signal <22.05k (or that 8k is good for anything <4k). 8k came about from telephony where the target upper frequency was around 3.3k, far below the 3.9995k you are trying to squeeze out of it here.
Wow, thanks for pointing this out, it seems that Audacity has broken export.
Anyway, when I save the project instead, it saves the file correctly.
And, not surprisingly, it sounds the same.
Even 999.8 Hz is severely distorted, although admittedly that's purely about waveform preservation, just ABXed it and didn't hear the difference between 8000 and 48000.
Axon, I would like to understand your claim about your "minimum spot noise margin" measurement. Perhaps you would be willing to elucidate a little if it isn't something likely to be already obvious to everyone but me (or it hasn't already been covered between my reading earlier today and when I get back to post this).
I'm thinking you said you measured some noise from an LP surface at -85dBfs, but what is a "spot noise margin"? Is it over some very small duration? How is the measurement made? Is this in a system where the LP signal peaks can reach somewhere in the vicinity of 0dBfs or one that would need considerable analogue amplification to achieve 0dBfs?
My system is fixed gain. A few LP Orchestral recordings have almost reached 0dBfs peak but normalization to the amount of +6dB to +12dB for my finished product is very common.
The raw recordings generally measure about -52dBfs to -55dBfs average RMS on unmodulated grooves between tracks (regardless of maximum music peaks), occasionally much higher. A high pass filter at 30Hz, which I use on most recordings, generally lowers the between tracks floor to about -60dbFs to -62dBfs (before NR or other processing).
Since my system noise floor is about -80dBfs average RMS (-102dB for soundcard, -90dB for phono preamp), the above seems to me to be a fair value for actual LP background noise. No improvement would seem possible through a quieter system. While I do record in floating point format, for subsequent processing, nothing would seem to be lost from the LP by instead using a 16 bit format. Do you think I have a definite misunderstanding of the situation or that I am doing something wrong to get such higher noise measurements?
Basically, my experience suggest that -85dBfs from the disk surface is not possible, although this would of course depend on the cartridge output plus any analogue amplification between the cartridge and the soundcard. Using my setup, but lowering the phono preamp amplification factor enough to reach -85dBfs for the LP background noise seems like a poor reason for claiming that 16 bits is not adequate unless we are talking about something different from anythng I know about.
By the way, based on some of your other posts, I suspect you might find this of at least passing interest (most of us would have to take a pass -- on the cost).
http://www.celemony.com/cms/index.php?id=capstan (http://www.celemony.com/cms/index.php?id=capstan)
Regarding some earlier discussion of LP processing and Sound Forge: When the product was owned by Sonic Foundry, their Noise Reduction 2 plugin was a separate product. The facilities included with the editor were fairly primitive. After Sony bought the line, the NR 2 plugin was added to some versions of Sound Forge (at a lower total cost than purchasing the plugin alone).
Both the declicking and NR abilities of this product in batch processing mode are better than many other LP/tape restoration programs. It may fare less well compared to some of the really expensive packages, but if your version of Sound Forge has this plugin (or its processes integrated into the editor) you do have the means to getting good results from most LPs.
The 3999.5 Hz wave cannot be reconstructed perfectly because you have to give the bandlimiting low pass filter some room (like I've explained in my initial post).
You could quite easily reconstruct the 3999.5Hz waveform properly from 8000Hz sampled data using a decent filter. Proper brick wall filters are easy. Long, but easy.
They're not used because, in the actual audio systems in use, this phenomenon occurs at 22.05kHz, rather than 4kHz, and so is inaudible.
A more gentle filter doesn't have to let anything through that you don't want. It can simply start to cut off at a lower frequency, e.g. 20kHz.
There are loads and loads of threads about this. I can't believe we're re-hashing this topic!
Cheers,
David.
Wow, thanks for pointing this out, it seems that Audacity has broken export.
Anyway, when I save the project instead, it saves the file correctly.
And, not surprisingly, it sounds the same.
Even 999.8 Hz is severely distorted, although admittedly that's purely about waveform preservation, just ABXed it and didn't hear the difference between 8000 and 48000.
Audacity's export is not broken. You're just, well, wrong.
Do you have any *earthly* idea how pointless your test sample is? In order to cleanly reproduce at 3999.5hz sine at 8khz sampling rate, you need a reconstruction filter with a 0.5hz transition band. Translation: the filter must be at least TWO SECONDS LONG. In addition to being hideously inefficient, such a filter will also ring for 2s (1s preecho, 1s postecho). So it's really a "test" of the worst kind possible: in order to pass the test, a playback system has to be not only slow, but sound like ass too. And interpreting anything out of Audacity's waveform plots of this wav are precisely as meaningless, for precisely the same mathematical reasons.
To say that Shannon-Nyquist Theorem "requires" an infinite signal to work is equally as logical as saying "Computing 2+2 doesn't work, because I *know* the answer must be 5". The "theorem" doesn't really care whether you are dealing with bandlimited data, timelimited data, whatever. The general behavior of aliasing in the Fourier domain is utterly unambiguous, predictable, and easy to interpret, and Shannon-Nyquist is so easily derived from such principles that calling it a "theorem" really overstates the complexity of the proof. For bandlimited signals, there is precisely zero aliasing. For timelimited signals, of course there will be aliasing, but it will occur at predictable and controllable levels, which, of course, ADC engineers have firm control over.
Sorry, normally I'm at least slightly less short about topics like this, but I have zero patience for lazy CS grads who act like armchair signal processing engineers without bothering to understand the math. It's *really* not that hard.
It's a common nonsense. In short, no, unless you have an infinite stream of samples.
Longer answer (https://extrememoderate.wordpress.com/2011/05/25/dont-abuse-nyquist-shannon-theorem-like-this/)
What you're seeing is beating between the wanted signal (3999.5Hz in this case), and the first image (4000.5Hz in this case). The reconstruction filter should remove all content above Nyquist, but isn't doing so.
Now this is just example of what can happen, purely artificial, obviously. But you can experience different sampling artifacts. And not only with frequencies close to the maximum, with lower ones the problem is just smaller.
Any "problems" you see with "lower ones" will be because of images - frequencies as far above Nyquist as the "lower one" is below it, which should be being filtered out by the reconstruction/resampling filter, but aren't.
There are two take home points from this:
1. you can use as good a reconstruction filter as you want - audacity is not the only software in the world
2. in CD sampled audio,
even without a reconstruction filter, what you have is exactly what you want up 22.05kHz, and then a spectral mirror of the 0-22.05kHz content above 22.05kHz. If you don't filter it out, the waveform will look quite strange, but the part that your ears can actually hear is perfect - your ears are doing enough filtering to make it work just fine. The reason for putting the filtering in DACs is because lettings lots of ultrasonic junk through can upset some equipment downstream, and intermodulation distortion in amps and (even more) speakers can make it has an effect within the audible range again.
Back on topic: 16/44.1kHz is enough, but
1) if your sound card is poor and doesn't filter out content above ~20kHz properly when recording (i.e. you get aliasing), then you can record at a higher rate (e.g. 88.2, 96), and then downsample in software.
2) if you're hopelessly careless with your levels when recording, but you have a very good sound card, then 24-bits could just about have some tiny benefit. Note: with a poorer sound card, the bottom 8-bits or more of the "24-bit" signal is all noise: better to set the levels properly and record in 16-bits than set the levels too low and record in 24-bits in this case!
3) floating point (in some software) allows you to be careless with levels when processing too. Though goodness knows what kind of processing you'd have to do to make this relevant - far more EQ than I'd ever want to apply!
Hope this helps.
Cheers,
David.
Cheers,
David.
A vinyl record can contain frequencies of [some ultrasonic value, doesn't matter if it's right or consistent], probably even higher.
Based on what I've read here and elsewhere (anyone feel free to offer or point to contradictory info), and notwithstanding the rare occasions when someone reports seeing a faint line in the upper reaches of spectrograms made from vinyl ripped at a high sample rate, it seems to me that vinyl's ability to reproduce what are thought to be ultrasonic frequencies is only meaningful to the extent that 1. the source instruments produced audible (thus not really ultrasonic) harmonics that high, and 2. those harmonics were preserved, at an audible level and without getting buried in noise, by every piece of equipment and media used at every stage—from multitrack recording on through to supplying a stereo mix to the vinyl mastering house (something quite often done via ordinary audio CD-R), and then during vinyl mastering, previous playback (the highest-frequency content is said to diminish with each play of the vinyl), ripping, and the format conversion for storage. If you consider the frequency limitations of the typical pieces of equipment in those chains, you'll find it's virtually impossible to justify a sample rate greater than 44.1 kHz for ripping vinyl, other than the reason David mentions in the post above.
IIRC, I read long ago on some site that using high samplerate (192kHz) in recording improves the AD conversion quality at higher frequencies (inside Nyqvist range) because of there are more samples to present the analog data as digital then (i.e. how many samples represents lets say 12kHz tone if I have understood it correctly).
It wasn't this site - http://electronics.howstuffworks.com/analog-digital3.htm (http://electronics.howstuffworks.com/analog-digital3.htm) but, maybe this site supports what I tried to remember.
Juha
That's audiophile crap. 24/44.1 is more than enough to make needledrops.
It's so easy to throw out that kind of statements without argument ... can you (or anyone else here) explain why it's audiophile crap?
Juha
Axon, I would like to understand your claim about your "minimum spot noise margin" measurement. Perhaps you would be willing to elucidate a little if it isn't something likely to be already obvious to everyone but me (or it hasn't already been covered between my reading earlier today and when I get back to post this).
I don't have enough time to respond to this with the attention it deserves tonight but I'll try to give a bit more detail here.
Note, I am *exclusively* talking about recording bit depth here. If one assumes a 72db SNR for vinyl ref. +0db (reasonable? no?), assume +0db records at 0dbFS, but then add +12db of headroom for playing the loudest 12" singles, and perhaps another +3db for particularly egregious instances of inner groove distortion on used records, then you're already to -87dbFS. And -10db is right around the edge of audibility for narrowband noise differences, isn't it?
By "spot noise margin", I mean, find the quietest part of an LP, look at its power spectrum, then create a WAV containing 16-bit TPDF and look at its power spectrum, then looking at how close the two power spectra are. In other words, subtract the TPDF spectrum from the vinyl noise spectrum, resulting in the "margin" between the two. I use the term "spot" to indicate that this is a frequency-dependent quantity, and I call this "noise margin" rather than "SNR" because there really isn't a "signal" here, just two sources of noise that I want to compare. If the spectra are dominated by broadband noise (which they are) and the spectrum analysis parameters are held constant, then, I believe, such a relative measurement ought to be meaningful.
I suppose that in hindsight "spot noise factor" would be slightly more formal, but I'm not sure any more understandable. :F
I measured the 13db margin as being centered at ~1100hz, with a bandwidth of a couple hundred hz or so. IIRC it was probably under a bark. But the margin was 16db or less for a *much* wider bandwidth (like 5khz or so). The recording was (of course) flat eq, with the gain set up to record +0db at around -30dbFS, which gives me 6db of headroom IIRC for my torture tracks.
If you're evaluating noise/SNR on an RMS basis rather than a spot/spectral basis then you need to be a hell of a lot more aggressive about your lowpass filtering. Everybody knows that the SNR of vinyl below 30hz is a dirty dog, but it's still a dog at 200hz; IIRC on a lot of records it doesn't actually reach a local minimum until 500-2000hz. I highpassed my snipped of vinyl noise at 200hz.
Don't get me wrong -- I really don't think 16-bit dither would ever likely be audible in this context -- but I really do think that's a closer call than some might imagine.
Hopefully I'll have more time to reply tomorrow.
So it's really a "test" of the worst kind possible: in order to pass the test, a playback system has to be not only slow, but sound like ass too. And interpreting anything out of Audacity's waveform plots of this wav are precisely as meaningless, for precisely the same mathematical reasons.
That's about what I meant by saying that it's good that decoder's don't work this way.
To say that Shannon-Nyquist Theorem "requires" an infinite signal to work is equally as logical as saying "Computing 2+2 doesn't work, because I *know* the answer must be 5". The "theorem" doesn't really care whether you are dealing with bandlimited data, timelimited data, whatever. The general behavior of aliasing in the Fourier domain is utterly unambiguous, predictable, and easy to interpret, and Shannon-Nyquist is so easily derived from such principles that calling it a "theorem" really overstates the complexity of the proof. For bandlimited signals, there is precisely zero aliasing. For timelimited signals, of course there will be aliasing, but it will occur at predictable and controllable levels, which, of course, ADC engineers have firm control over.
Maybe I was not precise enough. By 'work' I meant 'be able to reconstruct the signal perfectly'. For any decoder there is a signal that's won't be restored perfectly. So no, it doesn't work and can't work and even Master Shannon can't change it.
What you're seeing is beating between the wanted signal (3999.5Hz in this case), and the first image (4000.5Hz in this case). The reconstruction filter should remove all content above Nyquist, but isn't doing so.
I don't get it. How should the waveform look after reconstruction filter?
1. you can use as good a reconstruction filter as you want - audacity is not the only software in the world
Actually I used foobar and VLC to run the wav. Any suggestions?
2. in CD sampled audio, even without a reconstruction filter, what you have is exactly what you want up 22.05kHz, and then a spectral mirror of the 0-22.05kHz content above 22.05kHz. If you don't filter it out, the waveform will look quite strange, but the part that your ears can actually hear is perfect - your ears are doing enough filtering to make it work just fine. The reason for putting the filtering in DACs is because lettings lots of ultrasonic junk through can upset some equipment downstream, and intermodulation distortion in amps and (even more) speakers can make it has an effect within the audible range again.
But the same is well visible on a waveform around 1/2, 1/3 so on up to 1/6 N too. And these are within human hearing range.
Now whether it's audible - 1/4 is not for me and I have no idea how about the others.
Thanks, 2Bdecided, you're helpful.
It's so easy to throw out that kind of statements without argument ... can you (or anyone else here) explain why it's audiophile crap?
Because audio reproduction at 44.1/16 is sufficient. The public has an appetite for higher definition (e.g. TV) but if they can't perceive a difference, they don't buy it: 30+ years on and, despite trying (e.g. with formats such as DVD-A), no-one has been able to convince that 44.1/16 is lacking in any way.
IIRC, I read long ago on some site that using high samplerate (192kHz) in recording improves the AD conversion quality at higher frequencies (inside Nyqvist range) because of there are more samples to present the analog data as digital then (i.e. how many samples represents lets say 12kHz tone if I have understood it correctly).
It wasn't this site - http://electronics.howstuffworks.com/analog-digital3.htm (http://electronics.howstuffworks.com/analog-digital3.htm) but, maybe this site supports what I tried to remember.
Juha
That's audiophile crap. 24/44.1 is more than enough to make needledrops.
It's so easy to throw out that kind of statements without argument ... can you (or anyone else here) explain why it's audiophile crap?
Because it shows the sample points joined by straight lines, saying "When the DAC recreates the wave from these numbers, you get the blue line shown in the following figure:"(http://static.howstuffworks.com/gif/cd-sample15.gif)
There's not a DAC in the world that joins the sample points with straight lines like that.
Cheers,
David.
But the same is well visible on a waveform around 1/2, 1/3 so on up to 1/6 N too. And these are within human hearing range.
Yes, the beating is visible in an unreconstructed waveform (i.e. sample points).
Let's use some numbers...
48kHz sampling (easier numbers, but sample principle applies at 44.1kHz, or any other value).
Nyquist = 24kHz.
Sample a 17kHz tone. Just looking at the sample points, it looks a little ragged. You're seeing the effects of sampling - you get images at higher frequencies. It's exactly what would come out of the DAC is you did no reconstruction filtering at all, but just set the output voltage equal to the sample value at each sampling instant - no interpolation or anything.
So 17kHz becomes 17kHz + 31kHz + 65kHz + 79kHz + ... (infinite series, in theory).
It's the extra 31kHz + 65kHz + 79kHz + ... which make the sample points look more ragged than the 17kHz you start with.
It's the 17kHz tone, and only the 17kHz tone, that you ear can hear (if you're quite young) - i.e. exactly what you started with. Nothing more, nothing less.
Hope this helps.
Now, click that FAQ button top left, and read some useful threads!
Cheers,
David.
I don't get it. How should the waveform look after reconstruction filter?
Try this example (with slightly less arduous filter steepness):
sox -r 2k -n 1.wav synth 5 sine 996
sox 1.wav 2.wav rate -b 99.7 48k
then look at the 2 waveforms in audacity: beating in the first, smooth sine in the 2nd.
Because audio reproduction at 44.1/16 is sufficient. The public has an appetite for higher definition (e.g. TV) but if they can't perceive a difference, they don't buy it: 30+ years on and, despite trying (e.g. with formats such as DVD-A), no-one has been able to convince that 44.1/16 is lacking in any way.
I understand your argument if we are talking 'bout D/A conversion (because of human hearing isn't sensitive enough and equipment (speakers) for to reproduce audio aren't lossless) but, wasn't it A/D conversion in question here as well ... where might be some post-processing involved? Wouldn't it still be best to use as good accuracy as possible in recording/post-processing tasks?
Juha
Because it shows the sample points joined by straight lines, saying "When the DAC recreates the wave from these numbers, you get the blue line shown in the following figure:"
There's not a DAC in the world that joins the sample points with straight lines like that.
Cheers,
David.
Hmm... 'bad' graphical illustration is enough reason ... ?
Juha
Because audio reproduction at 44.1/16 is sufficient. The public has an appetite for higher definition (e.g. TV) but if they can't perceive a difference, they don't buy it: 30+ years on and, despite trying (e.g. with formats such as DVD-A), no-one has been able to convince that 44.1/16 is lacking in any way.
I understand your argument if we are talking 'bout D/A conversion (because of human hearing isn't sensitive enough and equipment (speakers) for to reproduce audio aren't lossless) but, wasn't it A/D conversion in question here as well ... where might be some post-processing involved? Wouldn't it still be best to use as good accuracy as possible in recording/post-processing tasks?
Juha
24-bit is good for post-processing (to reduce the effect of accumulated truncation error) but a typical DAW will process at >= 24 bit automatically, regardless of the input bit-depth. Recording at >48kHz usually just fills your disk with useless information. Exceptions include recording bats etc. e.g. for spectrograms or for slowing down into the audible range.
Hmm... 'bad' graphical illustration is enough reason ... ?
You asked why the article was nonsense. The article says we need higher sample rates because lower sample rates don't give a good-enough representation of the original data back, and proves this by showing the data you
do get back. Except what's shown is not what you get back at all!
What you get back (with CD, for example) contains all the frequencies that you can hear, reproduced faithfully. Any different in shape between the original waveform and what comes out of the CD player is due to changes above the range of human hearing.
This has all been discussed here before.
I'll even post the link to the FAQ...
http://www.hydrogenaudio.org/forums/index....7516#entry74075 (http://www.hydrogenaudio.org/forums/index.php?showtopic=7516#entry74075)
Here's a much simpler argument: it's virtually impossible to ABX.
Not entirely impossible, because if you use faulty equipment (i.e. equipment that adds its own distortion) then ultrasonics can cause an effect within the audible range.
Cheers,
David.
Don't get me wrong -- I really don't think 16-bit dither would ever likely be audible in this context -- but I really do think that's a closer call than some might imagine.
Good enough!
IIRC, I read long ago on some site that using high samplerate (192kHz) in recording improves the AD conversion quality at higher frequencies (inside Nyqvist range) because of there are more samples to present the analog data as digital then (i.e. how many samples represents lets say 12kHz tone if I have understood it correctly).
Juha
What you gain by having more samples representing a 12 kHz tone is the ability to reproduce it's harmonics (24 kHz and up) if it's distorted. If 20 (or 22) kHz is the limit of what you're trying to reproduce, then the extra samples are unnecessary.
Good enough!
Not for me.
We need to be discussing SNR, not dynamic range. What is the "spot noise" of 16-bit LPCM?
It's very simple, really.
If the analog signal contains NO content above Nyquist, then it can be reproduced EXACTLY by sampling at 2x and then converting back to analog. End of story!
I'm sorry that I oversimplified (a lot), but I get really frustrated at seeing the same misconceptions stated over and over again.
Because it shows the sample points joined by straight lines, saying "When the DAC recreates the wave from these numbers, you get the blue line shown in the following figure:"
There's not a DAC in the world that joins the sample points with straight lines like that.
Cheers,
David.
Hmm... 'bad' graphical illustration is enough reason ... ?
The reason is that its complete nonsense. The author made up some bullshit, put it on the internet, and got ad revenue because people linked it. Is that more clear?
The reason is that its complete nonsense. The author made up some bullshit, put it on the internet, and got ad revenue because people linked it. Is that more clear?
? Now, that reply should need some reasoning.
Juha
Man so many people came out of the wood work over this, so many different opinions. I think I am going to have to just experiment and try everything
That would be a remarkably ineffecient thing to do.
Do the transfer at 24 bits for sure -- you lose nothing, and you gain peace of mind. Your editing/declicking software is going to do everything in high-bit domains anyway (I hope).
Use a 88.2 or 96 kHz sample rate if you have a nagging feeling that 44.1 is
merely adequate and again want peace of mind. As Dan Lavry (who builds pro ADCs for a living) has written (http://www.lavryengineering.com/documents/Sampling_Theory.pdf),
"In fact all the objections regarding audio sampling at 44.1KHz, (including the arguments relating to pre-ringing of an FIR filter) are long gone by increasing sampling to about 60KHz."
Sampling a vinyl record at 192kHz is
utter absurdity.
So do your transfer at 88.2/24bit. Your files will be bigger than the really need to be, but not idiotically so.
Do the transfer at 24 bits for sure -- you lose nothing, and you gain peace of mind. Your editing/declicking software is going to do everything in high-bit domains anyway (I hope).
Absolutely, but he should also know that once it all the editing is finished and the data is normalized, converting to 16-bit isn't going to do any harm.
Now, that reply should need some reasoning.
Not the part about it being bullshit.
We lately had a similar discussion in another forum and someone linked to that pages that imho do a nice, accurate summary of some numbers regarding LPs.
http://www.audiomisc.co.uk/HFN/goodresolutions/page2.html (http://www.audiomisc.co.uk/HFN/goodresolutions/page2.html)
I don´t know it was already linked there, so i hope it is of some use. Don´t forget to read the other pages over there.
This is a good read why 16 bit are enough:
http://www.hydrogenaudio.org/forums/index....c=61758&hl= (http://www.hydrogenaudio.org/forums/index.php?showtopic=61758&hl=)
Sometimes I wonder why people consider 16 bit audio has not got sufficient resolution.
Think of that: with 16 bits you've got 2^16 possible amplitude values the audio signal can have that are equal to 65536.
If you want to paint such a signal on a wall and each one step equals 1 millimeter you need a wall that is almost 66 meters high!
Seen that way 16 bit is quite a decent resolution.
We lately had a similar discussion in another forum and someone linked to that pages that imho do a nice, accurate summary of some numbers regarding LPs.
http://www.audiomisc.co.uk/HFN/goodresolutions/page2.html (http://www.audiomisc.co.uk/HFN/goodresolutions/page2.html)
I don´t know it was already linked there, so i hope it is of some use. Don´t forget to read the other pages over there.
Interesting link, thank you.
-k
Sometimes I wonder why people consider 16 bit audio has not got sufficient resolution.
Think of that: with 16 bits you've got 2^16 possible amplitude values the audio signal can have that are equal to 65536.
If you want to paint such a signal on a wall and each one step equals 1 millimeter you need a wall that is almost 66 meters high!
Seen that way 16 bit is quite a decent resolution.
However, our ears a kind of like zoom lenses. Imagine zooming to wide angle to take in the full 66 meters, then to telephoto to examine 1 mm details. Your camera can't see all 66 meters at 1 mm resolution, just as your ears can't hear a -96 dB detail and 0 dB level sound simultaneously.
The reason is that its complete nonsense. The author made up some bullshit, put it on the internet, and got ad revenue because people linked it. Is that more clear?
? Now, that reply should need some reasoning.
Juha
How so?
24-bit files use 50% more space than 16 bit. With storage at less than 10 cents per gigabyte, that certainly is not a massive concern.
Assuming a constant sampling rate, of course. Perhaps my reply was ambiguous regarding this point, but that assumption was by no means implicit.
WRT storage space, TB drives are indeed fantastically inexpensive. Given finite resources, one
might conceivably wish to refrain from purchasing additional HDD space, particularly if one is going to the trouble of digitizing old media that is inherently self-destructive (barring laser turntables) upon playback, rather than buying new (presumably better mastered) versions. I have roughly 6.5TB of storage space in my principal desktop, at 85% usage with no free 3.5" internal mounting points. Many people have MicroATX systems and laptops with zero free HDD slots and limited USB ports.
Likewise, a "massive waste of space" is not the same thing as a "massive concern". If time and space is free, it's not a concern at all.
I don't think we'll get anywhere trying to determine the inherent resolution of an analog medium. What we do know is that usable resolution of a decent ADC is greater than 16 bits and no more than 24 bits. To ensure no loss of data, 24-bit is a reasonable choice.
Really? I must be misunderstanding something. It seems like you're suggesting that if we ignore the actual source being captured, then more resolution on the capture side
must be advantageous! But we do know the source media and the limitations of the format are well established. Standard vinyl LPs are not Platonic ideals of analog media.
Could one not make the same argument ad infinitum as better ADCs are developed?
That said, there is no loss in recording at 24bit
save for the loss of disk space and the
loss of immediate portability associated with 24 bit word recordings (or 16 bit at >48khz), which cannot be played back by most embedded electronics. Manipulation of larger files requires more power; additional HDDs pull additional power. Editing/denoise/declicking larger files will require more time to open/process & save such files - this could be compounded into a real issue with three hundred hours of material and an Atom system with a 4200rpm laptop drive. It's no issue at all with an i7 running fast solid state storage and three hours of material.
Minor losses to be sure, the degree of which varies depending on available resources. They are still losses unless the source file is destroyed post processing. I'm just unclear as to what benefit you're positing. Incidental losses associated with file size bloat versus what gain?
In the case where the number of records to be digitized is quite limited (many vinyl collectors own hundreds, if not thousands of platters - hence massive waste of space), and the soundcard ADC is flawed in some way at 16 bit, clearly 24 bit is the way to go. That case is not remotely universal, suggesting that exceeding 48khz sampling for this purpose is paranoia, while exceeding 16 bit word length is compounded paranoia.
edited for brevity
However, our ears a kind of like zoom lenses. Imagine zooming to wide angle to take in the full 66 meters, then to telephoto to examine 1 mm details. Your camera can't see all 66 meters at 1 mm resolution, just as your ears can't hear a -96 dB detail and 0 dB level sound simultaneously.
If you are using the amplifier volume knob to "zoom" then you can zoom in a lot if the material contains long enough quiet enough passages. By most, this is considered "cheating" or an irrelevant reason to demand better audio specs.
If you are talking about the perceptual system, we have non-linear normalization for sight as well as sound (try driving into a tunnel at a bright sunny day to see how the adaptation to brightness works). But one could just as well see this as a limitation in instantaneous sensing precision, I believe.
-k
In the case of vision there are dynamic range limitations in both resolution and intensity. Dynamic range of intensity is quite large due to features such as changing pupil size. Dynamic range of resolution is much more limited.
I was specifically referring to how our ears are able to, unaided, adapt to a wide range of sound level. Use of the volume control knob is something else entirely.
I was specifically referring to how our ears are able to, unaided, adapt to a wide range of sound level.
I do believe that 16 bits is usually enough to cover this range if used properly.
-k
I was specifically referring to how our ears are able to, unaided, adapt to a wide range of sound level.
I do believe that 16 bits is usually enough to cover this range if used properly.
-k
I agree completely.
Here's a little experiment to see what's going on:
I took a high fidelity recording of a percussion duet, recorded in 24 bit / 96 kHz. The track has a high dynamic range, the new ReplayGain scanner reports a track gain of +4.7 dB (for comparison: lots of my loudness war metal tracks have a gain of -11 dB!).
Here's the waveform of the chunk I've analyzed:
[a href="http://img19.imageshack.us/img19/4895/drum24vs16wl.png" target="_blank"] the level of the midrange would blow your ears). In this scenario the 16-bit noise would be around 40 dB SPL (about the same level as residential ambient noise!).
Same can be done with different sample rates, but above 20 kHz the level is already very low and drops very fast from there. I tried to listen to the content from 22.05 to 48 kHz but - as expected - couldn't really hear anything.
I do my rips to a standalone cd-r for ease, but if I were using a computer I would do 24/96.
There is no need for 24 bits for playback. I have a phono preamp that Stereophile said was extraordinarly quiet (Linn Linto) which means s/n of 85 dB (a weighted), 66 dB broadband. I also have an Ortofon test record which has recorded bands of -40dB, -50dB, -60dB, -70db, and "greater than -70dB"). Based on that, even with very fine equipment (VPI deck, SME arm, Lyra cartridge) I can't hear a difference with the last band, implying that the effective s/n of my vinyl playback/amps is between 60 and 70 dB. That's no better than 12 bits.
However, when setting levels, there is always the chance of clipping. The s/n for SIGNAL may only be 65 dB, but scratches and other defects can be greater in amplitude than the highest signal. Plus you can always underestimate what the peak level will be, especially with classical (esp. if you're not familiar with the piece). Everything's going peachy, peaking at -6, then boom, it gets louder and you get clipping. With 24 bit, I could set levels with hardly a care, say at -35 dB, and not worry about clipping. Also, when you store in flac, if you didn't use the last 6 bits, you don't store them.
I would keep the files in 24/96 flac b/c storage is cheap and because I understand post-processing is more transparent in hi-rez.
That's just what I would do.
Axon, while I may miss some of the finer points of your explanation that 24 bits may be necessary to capture everything from LPs, if I'm not misunderstanding completely, a simplified statement of your hypothesis is that, at higher frequencies, the noise inherent in vinyl is low enough that distinguishable music signal below the 16 bit level might exist at those higher frequencies. This signal would be lost if one records from the LP at 16 bit.
A frequency distribution graph is an easy way to see that the unmodulated groove noise is below the 16 bit level, if considered in a narrow enough frequency range. Making such graphs on between track selections of a few of my recent transfers shows the noise level dropping below -100dBfs somewhere between 600 and 1000 Hz.
Since the noise we perceive is the sum over a more extensive range, I doubt, but don't really know, that the SNR of the recording we listen to would permit hearing any music signal more than 16 bit down anyway, even in a 24 bit integer or a floating point capture. Even without the disk noise, I think any such low level signal would be swamped by any "normal" level signal, say at -70dBfs, regardless of its frequency.
However, I question the hypothesis that, if such signal exists, it would not be captured in a 16 bit recording.
Harkening to my recent post in the other 16 bit vs 24 bit thread,
http://www.hydrogenaudio.org/forums/index....9843&st=200 (http://www.hydrogenaudio.org/forums/index.php?showtopic=49843&st=200)
on page 9, made on 05/30/2011 at 01:01 AM (unfortunately, I don't know the coding to link directly to that post), I wrote that I could not distinguish between the 32 bit floating point and 16 bit integer versions when I "properly" converted to 16 bit. Where I could tell which is which, it was, I suspect, only because the non noise shaped dither I used is not only audible, but significantly louder than the signal. Perhaps if there is someone who can hear the mostly high-frequency dither in the noise shaped conversion, both 16 bit versions could be easily enough distinguished from the higher bit depth original.
These tests were (1) with a 3500Hz sine wave generated at -115dBfs, and (2) with the same tone at -100dBfs. Without dithering, a proper conversion to 16 bit of either signal level produces a flat nothing. There is no signal left to hear. With dither, and enough analogue amplification, I can hear the tone easily in both 16 bit conversions, although neither test signal is at a high enough level to contribute anything audible if mixed with any real music.
When recording from vinyl, one is never without dither. It is there in the broadband noise from the disk and in the broadband noise in the phono preamp. It isn't optimum dither, but neither is the non-shaped rectangular stuff in my tests. My hypothesis is that, because of the inherent noise, if any such low level, higher frequency signal exists, it will be captured as well at 16 bits as at 24 bits.
Hi All -
I recently stumbled upon this very informative and mostly enlightening topic, though I must confess, it is a tad beyond me when the deep technical details are conveyed. My sole interest here is listening to music via my pc and the Winamp player, either thru headphones or a standard 2-channel mini-plug out connection to my stereo system. I have been in the process of collecting audio files, many of which are only available as 24-bit and either 96 or 192 kHz. Now, after my reading of this topic (and please correct me wherever wrong), I am getting the understanding that the investment in time & materiel for efforts to create the hi-res files are a worthwhile endeavor for the extraction of the audio from the source material. However, I also seem to learning here that for the purpose of just listening (playback), audio files of 16-bit/44.1kHz, which I believe it was mentioned is the standard CD numbers, is more than adequate in light of one's understanding of our human anatomy specification. If my understanding is thus far on target, it seems that I can convert all the hi-res audio I have collected to the CD standard, thereby gaining additional disk space, and not be the wiser for anything lost. From the files I have collected thus far, on average, using 24-96 example, I am calculating a 3:1 difference in file sizes. I appreciate and welcome anyone's thoughts as to whether I should embark on this task.
thnx, -Marcos
Harkening to my recent post in the other 16 bit vs 24 bit thread,
http://www.hydrogenaudio.org/forums/index....9843&st=200 (http://www.hydrogenaudio.org/forums/index.php?showtopic=49843&st=200)
on page 9, made on 05/30/2011 at 01:01 AM (unfortunately, I don't know the coding to link directly to that post),
In the upper right corner of every HA post there's a clickable link that says Post
- -- in this case, Post #224. Clicking on that gives you a pastable direct link to the post, e.g.
http://www.hydrogenaudio.org/forums/index....st&p=757580 (http://www.hydrogenaudio.org/forums/index.php?s=&showtopic=49843&view=findpost&p=757580)
Thanks for the posting information
Hi All -
I recently stumbled upon this very informative and mostly enlightening topic, though I must confess, it is a tad beyond me when the deep technical details are conveyed. My sole interest here is listening to music via my pc and the Winamp player, either thru headphones or a standard 2-channel mini-plug out connection to my stereo system. I have been in the process of collecting audio files, many of which are only available as 24-bit and either 96 or 192 kHz. Now, after my reading of this topic (and please correct me wherever wrong), I am getting the understanding that the investment in time & materiel for efforts to create the hi-res files are a worthwhile endeavor for the extraction of the audio from the source material. However, I also seem to learning here that for the purpose of just listening (playback), audio files of 16-bit/44.1kHz, which I believe it was mentioned is the standard CD numbers, is more than adequate in light of one's understanding of our human anatomy specification. If my understanding is thus far on target, it seems that I can convert all the hi-res audio I have collected to the CD standard, thereby gaining additional disk space, and not be the wiser for anything lost. From the files I have collected thus far, on average, using 24-96 example, I am calculating a 3:1 difference in file sizes. I appreciate and welcome anyone's thoughts as to whether I should embark on this task.
thnx, -Marcos
Well, this must either be a low-traffic forum or my post is of minimal-to-no interest, or something else. Nevertheless, I have decided as follows :
The low cost of storage is not really relevant and I apologize for originally pointing out the file size differences and inadvertently giving the impression that it was of significant concern to me. While the inexpense of storage nowadays is welcome, in my case I will forever be wanting for more, given the amount of files (audio, video, artwork/images, documents etc.) I collect and, thus, must commit to external storage.
All authoring is not always equal and it is therefore possible for certain hi-rez files to be inferior. This I conclude from an understanding of authoring competence, as well as equipment quality, configuration and usage. Unless logical reasons to the contrary surface, it now seems my best approach to this task is to convert each file, do an a/b listen and then decide which to keep and which to discard. In the event of a tie, the choice seems obvious.
-thnx all
Your lack of readership was perhaps because you posted in an existing thread that was only superficially related to your question, as discouraged by Term of Service #2 (http://www.hydrogenaudio.org/forums/index.php?showtopic=3974). I’m sure there is much discussion to be had on the topic, but this particular thread may not be the ideal venue.
Your lack of readership was perhaps because you posted in an existing thread that was only superficially related to your question, as discouraged by Term of Service #2 (http://www.hydrogenaudio.org/forums/index.php?showtopic=3974). I’m sure there is much discussion to be had on the topic, but this particular thread may not be the ideal venue.
You and/or the mods are most welcome to re-locate my posts to where appropriate. Thank you for enlightening me.
I assume you're using lossless compression (e.g. FLAC)? If not, use that.
As for downsampling, use 44.1/16 if you think you'll ever burn to CD or need to use on something that only supports 44.1kHz (e.g. some legacy DAPs, some versions of flash), and probably 48/16 or 48/20 if you think you won't.
lossyWAV (used without noise shaping, and with highly conservative settings) can throw away the LSBs when they're not needed, but keep those extra bits (i.e. more than 16) in the rare moments when they are / may be. This can bring the FLAC bitrate down dramatically, without resampling.
Cheers,
David.
^
thnx David
Hello,
I am currently using an EMU 1616M to rip vinyl and using Sony Sound Forge 10 to record with. I am wondering what the best settings are to use for exporting the .WAV files. I have the option to record in 24bit or 32 bit at 192 Khz and save the
.WAVs in using IEEE Float or PCM. I am wondering if it would be best to record in 32bit and bounce down to 24 when I export to .WAV, or just record in 24 if there won't be any difference. No matter what I will need a 24 bit file in the end to
convert to .flac as it is the highest bit rate the format supports.
Furthermore, I am confused as to whether to use IEEE Float or PCM. I understand the IEEE float was developed for the broadcast industry while PCM is used in the red book CD standard. I'm guessing when I convert to .flac it will end up as PCM
anyway and am not sure if I would get any benefits from utilizing IEEE Float. If anybody can shed some light on this I would be most appreciative and believe this sort of information should be included in a wiki somewhere as I have been unable to
find any useful information as to the pros can cons of IEEE Float and PCM .WAV files with regards to audiophile needs.
I would recommend you record in either 24-bit / 96kHz if you need to save some space, or in 24-bit / 192kHz if you want the best quality at the expense of extra space. Recording in PCM is fine.
Or you could read the rest of the thread, which may leave various technical issues open for debate but seems to be converging on the opinion that a sample-rate of 192 kHz is definitely a needless use of disk-space and that one of 96 kHz may well be the same. There are unlikely to be any such high frequencies, those that exist probably aren’t musical and therefore won’t be worth saving, and good luck
hearing any of them anyway; we’d all be very interested in how you did it.
By the middle of the first page, the OP was already becoming convinced:
I had orignially assumed that a higher sampling rate would result in a more accurate copy of an analog signal given the increase in samples per second, yet it seems as you say 44.1khz is more than enough to replicate everything in the 20hz-20khz bandwidth according to Nyquist. This then makes me wonder why we even have 24/192 dacs in the first place except to sell stuff or make use of some psychoacoustical effect of higher frequencies or if there is a secret audiophile community of dogs somewhere...
While we’re re-treading old ground, a previous post:
Do the transfer at 24 bits for sure -- you lose nothing, and you gain peace of mind. Your editing/declicking software is going to do everything in high-bit domains anyway (I hope).
Use a 88.2 or 96 kHz sample rate if you have a nagging feeling that 44.1 is merely adequate and again want peace of mind. As Dan Lavry (who builds pro ADCs for a living) has written (http://www.lavryengineering.com/documents/Sampling_Theory.pdf),
"In fact all the objections regarding audio sampling at 44.1KHz, (including the arguments relating to pre-ringing of an FIR filter) are long gone by increasing sampling to about 60KHz."
Sampling a vinyl record at 192kHz is utter absurdity.
So do your transfer at 88.2/24bit. Your files will be bigger than the really need to be, but not idiotically so.
I have transferred about 100 of my 2000lps to hardrive so far.
From some direct a/b comparisons I can only say that I was not able to discern a difference between the rip and the direct from lp playback. (one has to cue very precise to get gapless switching correct). I convert using a customized ex-studio phono preamp feeding an m-audio 1010lt soundcard.
I transfer using 16/44.1, and then converting the wav. file to flac for playback in foobar. I keep the wav.f files for later editing in a separate folder.
I use a small program called "spin it again" for transfer (I have trouble using the audacity program) tagging the tracks, and mp3 tag for album art tagging.
I do not see any reason to use higher sampling rates than 16/44.1
As for me, I do my rips in a 44.1/16 format and have never claimed that higher resolution and sampling frequency may be beneficial for vinyl rip quality.
I'm convinced, as an electronics specialist and scrupulous listener, that properly performed digital conversion absolutely doesn't worsen the original analogue signal, otherwise I wouldn't spend time and efforts to rip all 300 LPs of my vinyl collection.
I don't ignore post-conversion processing, although prefer to do clicks elimination manually.
Excessive vinyl disc noise, whether originated from a noisy master tape or acquired in the course of vinyl pressing, isn't a problem, I cure it with the help of an original noise reducer just after the phono preamplifier, operation of this device is demonstrated at the mentioned above site.