bigshot
Headphoneus Supremus
Some people believe “more is always better”.
What’s the point of upsampling?
- Thread starter gaex86
- Start date
There were a couple of guys on a Facebook group spamming they’re cd ripped higher claiming “huge sound improvement”.
Maybe they are reading between the pits
I’m wondering, something like DAR (digital audio remaster) offered on some high end Astell kern DAP, is some kind of upsampling?
You’re welcome, glad it was useful.
This hobby isn’t a hobby, it maybe a hobby to you but of course it’s actually not just a profession but a whole bunch of professions. Obviously the musicians, composer and arranger are professionals but so are the recording, mixing and mastering engineers, the producer, the record labels, the recording studios, the scientists and engineers who invented, developed and standardised analogue and digital audio, the manufacturers and retailers of all that recording and reproduction equipment and unfortunately, also all those who market all these products. We’re covering an awful lot of ground/fields here: Physics, mathematics, acoustics, psychoacoustics and digital/computer science just to list the main scientific fields and that’s before we get into all the specialist hardware and software engineering fields, plus musical and business/management fields. In other words, even if you do nothing other than study this “hobby” for a few decades, you’ll still only really know/understand a relatively small portion of it.
Yes. If you have an original sample rate, say 44.1kFs/S, and you increase (“up”) that sample rate, then that’s up-sampling.
More (but empty) data. If we record something at say 88.2kFs/S then we can in theory potentially record frequencies up to somewhere relatively close (depending on the decimation filter transition band in the ADC) to the Nyquist Point of 44.1kHz, while with a sampling rate of 44.1kFs/S everything above 22.05kHz has been filtered out. The ONLY difference between say 88.2kFs/S and 44.1kFs/S is therefore the frequency band between 22.05kHz - 44.1kHz. This fact answers your question because if we record or convert to a sample rate if 44.1kFs/S then a filter must have been applied at 22.05kHz and the sole advantage of higher sample rates (other than the relaxed analogue filter requirements when converting to/from digital) of potentially capturing frequencies above 22.05kHz no longer exists, because they’ve already been filtered out.
While the sample rate effectively defines the frequency range we can capture, bit depth defines the dynamic range. With 16bit we effectively have a dynamic range of ~96dB (about 6dB per bit) which is already overkill for music recordings which almost always have a dynamic range of 60dB or less. So how do we convert from 16bit to 24bit? It couldn’t be simpler, we just add 8 zeroes to the end of every sample and baring in mind that consistent zeroes across samples is digital silence, then we’ve added literally nothing. So, there’s literally nothing to be gained from ripping a 16bit CD to 24bit! The story isn’t quite so obvious with records though, LPs very rarely have a dynamic range greater than about 50dB, so even further within our available 96dB dynamic range with 16bit. However being analogue, exactly where that 50dB dynamic range window is, is not defined relative to the 96dB dynamic range window of 16bit. Obviously there shouldn’t be a problem, with so much excess dynamic range to play with, it should be trivially easy to fit the 50dB of the record into the 96dB of 16bit without overlapping the edge but 24bit gives us an even bigger excess dynamic range (144dB in theory and about 120dB in practice) so if we’re totally ham fisted or don’t know what we’re doing, there’s even less chance of overlapping the edge. In other words, there could potentially be a difference using 24bit if the person re-recording the LP doesn’t know what they’re doing or isn’t being careful but not when ripping a CD.
In which case, “literally nothing” somehow magically makes a “huge sound improvement”! We commonly hear differences where there are none. If our brains know there is a difference then it will often alter our perception to perceive/hear a difference, to avoid the cognitive dissonance of knowing there’s a difference but not hearing one. This is even more likely if you have almost an entire (audiophile) industry constantly throwing advertising at you that more bits/higher sample rates are high definition, while 16/44.1 is standard definition and obviously high definition is higher definition than standard definition. Advertising works, if it didn’t companies wouldn’t be spending tens of billions on it!
Sure, so “in your experience” adding literally nothing results in “better sound quality”! What does that say about your experience?
Again, because there’s a difference you believe you hear one but the actual differences are so minuscule they can’t possibly be heard. In fact, during the recording, mixing and mastering process the audio will have gone through at least several and possibly as much as several dozen up and down resampling processes. If just one resampling process is so bad/different that it’s a “no return process”, then a dozen or several dozen of them should make the end result almost unrecognisable as a music recording. So according to your assertion, most digital music recordings are therefore almost unrecognisable as music recordings … OR, your assertion is false!
G
Last edited:
Thanks again G., this was awesome.
Anyway I didn’t want to diminish this world calling it a hobby; I was just talking about the most commercial part of it, where lot of hobbyist spend thousands chasing the last iem, dac, dap or whatever, believing and spreading fake knowledge.
Very similar to music, where hobbyist, rather than professionals, are the one spending thousands for the last overdrive, amp, cable whatever, chasing, again, the holy grail.
And I know that this happens basically in every hobby; the consumer part, fueled by a dishonest industry, fake beliefs and spectacular claims, often overshadow the enjoyable part. I see with the audiophile world, with played music, even sports, if nowadays a runner without the last super shoes, sunglasses, smartwatch, hr band, socks, shirt, tights, earbuds, can’t run.
Then obviously everyone is free and should spend his money on what think is good, useful , beautiful, whatever.
I just find peculiar in the audiophile world (I call it this way for simplicity), where as you said there’s a lot of science involved, from physics, to engineer, up to neuroscience, if we really want to consider the human part of it, there still so much disinformation and fake claims.
Sorry for the useless rant. Thanks anyway for what has been written so far, it was really interesting and useful.
Anyway I didn’t want to diminish this world calling it a hobby; I was just talking about the most commercial part of it, where lot of hobbyist spend thousands chasing the last iem, dac, dap or whatever, believing and spreading fake knowledge.
Very similar to music, where hobbyist, rather than professionals, are the one spending thousands for the last overdrive, amp, cable whatever, chasing, again, the holy grail.
And I know that this happens basically in every hobby; the consumer part, fueled by a dishonest industry, fake beliefs and spectacular claims, often overshadow the enjoyable part. I see with the audiophile world, with played music, even sports, if nowadays a runner without the last super shoes, sunglasses, smartwatch, hr band, socks, shirt, tights, earbuds, can’t run.
Then obviously everyone is free and should spend his money on what think is good, useful , beautiful, whatever.
I just find peculiar in the audiophile world (I call it this way for simplicity), where as you said there’s a lot of science involved, from physics, to engineer, up to neuroscience, if we really want to consider the human part of it, there still so much disinformation and fake claims.
Sorry for the useless rant. Thanks anyway for what has been written so far, it was really interesting and useful.
Hobbies vs. professions is a bit of a hot potato where a field of interest can be both. It is often the playground for gatekeepers to the hobby; what is deemed an inappropriate or ill-advised practice from a professional perspective, may be acceptable or indeed desirable when enjoying the field as a hobby (and vice versa).
It depends on one's objectives and IMO those need to be made very clear when participating in forum discussions so there can be no misunderstanding.
I also participate on a forum dedicated to the use of (often vintage) manual focus camera lenses, and there you can find endless (and often pointless) arguments between amateurs and professionals re. the appropriateness of using optical means to achieve an effect vs. simulating those effects in digital PP (post-processing). Much of that discussion overlooks the fact that professionals work on commission under time pressure whereas amateurs often have all the time in the world to faff about with optical filters/lenses and/or PP as they please.
Science should be a common language that ties at least the technical and theoretical aspects of a profession and a hobby, but often the disputes extend here also thanks to scientific ignorance on part of either one or both parties involved.
It depends on one's objectives and IMO those need to be made very clear when participating in forum discussions so there can be no misunderstanding.
I also participate on a forum dedicated to the use of (often vintage) manual focus camera lenses, and there you can find endless (and often pointless) arguments between amateurs and professionals re. the appropriateness of using optical means to achieve an effect vs. simulating those effects in digital PP (post-processing). Much of that discussion overlooks the fact that professionals work on commission under time pressure whereas amateurs often have all the time in the world to faff about with optical filters/lenses and/or PP as they please.
Science should be a common language that ties at least the technical and theoretical aspects of a profession and a hobby, but often the disputes extend here also thanks to scientific ignorance on part of either one or both parties involved.
Last edited:
No worries, I didn’t take it that way. It’s not uncommon to see the assertion that “it’s just a hobby” and I was just trying to contextualise that it may just be a hobby to most consumers but of course it’s the product of a lot of professionals, including the scientists whose principles are being employed and unfortunately also of the professional marketers and this is where so many of the issues arise.
Being hobbyists in an area that encompasses so many scientific (as well as engineering and artistic) disciplines means the marketers can have a field day because there’s any number of claims they can make based on scientific/engineering details that consumers and even serious hobbyists simply won’t understand well enough (or at all) to recognise it as nonsense/BS. Commonly, this marketing is based on a “lie of omission”, IE. Most or even all of what they state is true but they omit just one crucial fact that entirely invalidates their claims. Jitter and Skin Effect are two of the most pernicious examples. Skin Effect is real, a great deal of research has gone into it and considerable engineering to mitigate it and the audiophile press/marketing has mentioned it frequently in regard to analogue cables. There’s just one thing they don’t mention, Skin Effect only comes into play with very high frequencies (starting around 100kHz), so it’s a serious consideration for cables carrying signals in the radio freqs (MegaHertz) range but none at all in the audio freq range.
Another common one is going into great detail about some practical problem with digital or even analogue and how their expensive bit of gear solves that problem. All well and good except that most hobbyists wouldn’t know one critical fact, they might have heard of the problem and possibly even correctly understand that it is a real problem, what they probably don’t know is that it was already solved 30 years ago and has been built into even the cheapest models as standard for the past 25 years!
G
Something I wondered about:
I don't know much about the perceptual masking phenomena exploited in MP3 compression, and I know even less about the algorithms used for the implementation.
In the past, based on a layman's engineering intuition more than any hard facts, I always assumed that to minimise resampling artefacts WAV PCM 44.1kHz material is best kept at 44.1kHz when converting to MP3, and that WAV PCM 48kHz source material is best kept at 48kHz sample rate when converting to MP3.
But I could be wrong; I have no idea whether my suspicion there has any foundation given that I don't know how the MP3 compression algorithm actually works. Anyone have any insights?
(FWIW, on occasion I have accidentally converted 48kHz source material to 44.1kHz compressed 320kbps MP3, and I couldn't hear anything wrong on the (mostly rock) music where I had done that.)
I don't know much about the perceptual masking phenomena exploited in MP3 compression, and I know even less about the algorithms used for the implementation.
In the past, based on a layman's engineering intuition more than any hard facts, I always assumed that to minimise resampling artefacts WAV PCM 44.1kHz material is best kept at 44.1kHz when converting to MP3, and that WAV PCM 48kHz source material is best kept at 48kHz sample rate when converting to MP3.
But I could be wrong; I have no idea whether my suspicion there has any foundation given that I don't know how the MP3 compression algorithm actually works. Anyone have any insights?
(FWIW, on occasion I have accidentally converted 48kHz source material to 44.1kHz compressed 320kbps MP3, and I couldn't hear anything wrong on the (mostly rock) music where I had done that.)
Last edited:
I have a little insight but not enough to answer your question definitively. I’ve always assumed it was probably better to maintain the input sample rate but that was purely a guess and I never tested thoroughly enough to be sure, FWIW, I couldn’t audibly detect any differences in the little testing I did. Baring in mind the MP3 encoding process, I’m not sure it’s even a relevant question.
MP3 encoding is not very simple, partly due to the fact that it is deliberately not fully specified in the ISO specs, which is what allows different encoders (LAME, Vorbis, etc.). Fundamentally, the encoding process creates frames and sub-frames (called granules) which are data blocks containing a variable number of samples, it then performs a Modified Discrete Cosine Transform (MDCT) on those granules, then applies a Fast Fourier Transform (FFT) to convert from the time domain to the frequency domain, applies another MDCT and splits the resultant freq spectrum into 32 bands, then a psychoacoustic model (including masking) is employed to remove unnecessary (inaudible) data according to a bit bucket/reservoir assigned to each of the bands and finally the bitstream is formatted in accordance with the ISO/MPEG specs so it can be decoded by any MP3 decoder.
It’s quite a complex process and I don’t claim to have much insight beyond the superficial. It’s a very specialised and highly developed field but due to the fact it’s maintained and ratified by several large international bodies (ISO, IEC and of course MPEG) and implemented by numerous manufacturers, there’s probably tens of thousands of scientists/engineers who have an in-depth understanding of it but outside that specialist field there’s no need for other types of scientists or audio engineers (or consumers of course) to have an in depth understanding, and I don’t!
If you want to delve deeper, you’ll need a decent grounding in fairly high level applied math (at least it seems high level to me!). Auditory Masking is worth investigating further though, this Wikipedia Page on the subject is quite good.
G
Thanks for the reply Gregorio, and for the Wiki link, an interesting read.I have a little insight but not enough to answer your question definitively. I’ve always assumed it was probably better to maintain the input sample rate but that was purely a guess and I never tested thoroughly enough to be sure, FWIW, I couldn’t audibly detect any differences in the little testing I did. Baring in mind the MP3 encoding process, I’m not sure it’s even a relevant question.
MP3 encoding is not very simple, partly due to the fact that it is deliberately not fully specified in the ISO specs, which is what allows different encoders (LAME, Vorbis, etc.). Fundamentally, the encoding process creates frames and sub-frames (called granules) which are data blocks containing a variable number of samples, it then performs a Modified Discrete Cosine Transform (MDCT) on those granules, then applies a Fast Fourier Transform (FFT) to convert from the time domain to the frequency domain, applies another MDCT and splits the resultant freq spectrum into 32 bands, then a psychoacoustic model (including masking) is employed to remove unnecessary (inaudible) data according to a bit bucket/reservoir assigned to each of the bands and finally the bitstream is formatted in accordance with the ISO/MPEG specs so it can be decoded by any MP3 decoder.
It’s quite a complex process and I don’t claim to have much insight beyond the superficial. It’s a very specialised and highly developed field but due to the fact it’s maintained and ratified by several large international bodies (ISO, IEC and of course MPEG) and implemented by numerous manufacturers, there’s probably tens of thousands of scientists/engineers who have an in-depth understanding of it but outside that specialist field there’s no need for other types of scientists or audio engineers (or consumers of course) to have an in depth understanding, and I don’t!
If you want to delve deeper, you’ll need a decent grounding in fairly high level applied math (at least it seems high level to me!). Auditory Masking is worth investigating further though, this Wikipedia Page on the subject is quite good.
G
It looks like your intuition and your experience are aligned with my own at least.
Seeing the FFT and discrete Cosine Transforms are involved my gut feel tells me the practical approach is to play it safe and maintain the input sample rate, but lose no sleep over it if I sometimes forget to check when I do the conversion...
Due to the involvement of FFT, MDCT and splitting into 32 bands, my gut tells me there’s probably no impact and indeed I’ve never noticed any but because I don’t really know/understand what’s going on and there are no circumstances where I need to change the sample rate of an MP3 anyway, then I “play it safe” and just leave it as it is.
However, my intuition and experiences, even if they align with yours, is no basis to make a claim of fact, especially in a science discussion forum, so I’m not! I’m only adding a little to the pile but I’m not aware of any studies or other reliable evidence regarding the audibility of changing the output sample rate. I’d be very surprised if there isn’t any, either I’ve missed it or it may not be publicly available, maybe someone else has some?
G
Whazzzup
Headphoneus Supremus
Just try different rates, pcm, dsd and see what happends in one’s own rig. The source, dsp software, dac conversions, make it better maybe maybe not, depends.
It’s just a cheap easy experiment to conduct in one’s own privacy
It’s just a cheap easy experiment to conduct in one’s own privacy
What reliable evidence do you have that upsampling with any of those things makes any audible difference, be it better or worse?
No, it’s not. How do you isolate and test upsampling against not upsampling using a properly functioning DAC?
G
Whazzzup
Headphoneus Supremus
so what did you find in the privacy of your home. ? I found my preferences, thanks
Users who are viewing this thread
