First explain that 200kbs is only an approximate dividing line, not accurate. What I want to express is the current technical level of lossy audio. After the lossy audio is higher than a certain kbs, 99.9% of people have been unable to pass the blind ABX test of these lossy audio. I think the dividing line is about 200kbs.
At the same time, in terms of not affecting the subjective sense of hearing and saving storage space, it is better to be below 200kbs, and it is better to be lossless in terms of perfect audio preservation. So lossy audio above 200kbs is not the best choice in both respects, what general advantages still exist for lossy audio above 200kbs now and in the future? What general use values still exist?
Universal is to exclude special cases, such as having the money to buy the most expensive audio equipment, but not wanting to buy a hard drive, or deliberately finding and listening to killer samples, etc.
First explain that 200kbs is only an approximate dividing line, not accurate.
Some sounds are easier to compress than others. The variable bitrate formats adjust the bitrate moment-to-moment as necessary to get the requested quality.
Some people are better at hearing compression artifacts that other people wouldn't notice,
Some CODEFs are better than others (AAC can be better than MP3) or different CODECs may have different artifacts.
At the same time, in terms of not affecting the subjective sense of hearing and saving storage space, it is better to be below 200kbs.
You have to think about why you are compressing . If you have limited storage space you have to compress to fit. If you have plenty of space you can use a higher bitrate or you can use a lossless format.
and it is better to be lossless in terms of perfect audio preservation.
If you use lossless that's one less thing to worry about. And if you have a lossless format you can convert to any other lossless or lossy format in the future. If you have lossy files editing or converting to another lossy format in the future can cause additional degradation.
Probably irrelevant for end users, but for business...
[1] When bandwidth / cost matters.
[2] Many games still use ADPCM (e.g. CRI ADX). Games can have a lot of background music, sound effect and speech. Not to mention video, texture and such.
[3] As a preview to potential customers, only offer lossless version after they pay.
what general advantages still exist for lossy audio above 200kbs now and in the future?
Streaming. You get faster buffering and more stutter-proof playback on mobile phones when you stream music. Also less battery consumption because less network usage (even though decoding time may be higher).
Also, to my personal opinion, it's NOT better to be lossless if we talk about audio that doesn't contain any disturbing artifacts. I don't think that <200 kbps is a great idea even for portable usage (headphones and earbuds nowadays are sounding great), but at the same time I think that using lossy that is 99.9% transparent most of the time is a smart way to listen to music. Thus 200-300 kbps range is needed here.
For me the use case is local music on an Android phone to be played via bluetooth. I don't have the SD card space to store my entire collection in FLAC, so the next best thing is a high bitrate lossy encode that should still serve as a good enough source to be transparent even after a second lossy transcode to LDAC. I use CoreAudio AAC @ 320kbps CVBR, which is the highest possible quality with CoreAudio (even though foobar has options up to 512kbps in CBR/ABR mode, if you do "qaac -h", you'll see that 320kbps is the limit for all modes; bitrate can spike up to ~390kbps at most in CVBR).
I use CoreAudio AAC @ 320kbps CVBR,
That's an overkill. Bluetooth codecs don't give generation loss this huge that you would have to worry about bitrate redundancy.
200 is about right. Or the range of 192 to 224. I agree with Eurobeat_fan that
much below that isn't the best.
To me, the dividing line is more like ~130kbps. Using Apple's AAC encoder (qaac), I have to try hard to ABX it with most music. So, I don't mind using it when space is a concern and I doubt that I would hear artifacts during casual listening.
If I really want peace of mind, only lossless will do.
People with better hearing might see the point of higher bitrates lossy, of course. And you need it for lower quality codecs like MP3.
And streaming services have their own specific needs too, like avoiding complaints about the lack of high frequencies in spectrograms. :)
I'm actually confused why not a single music streaming service uses Opus. It's a hell of a codec at 256 kbps, + some bandwidth would be saved compared to all the alternatives (even AAC CVBR 256 kbps usually needs more bitrate than Opus VBR 256 kbps).
not a single music streaming service uses Opus
YouTube (Music (https://music.youtube.com/)) does.
I'm actually confused why not a single music streaming service uses Opus. It's a hell of a codec at 256 kbps, + some bandwidth would be saved compared to all the alternatives (even AAC CVBR 256 kbps usually needs more bitrate than Opus VBR 256 kbps).
I tried Amazon Music HD some months ago on my computer, and the formats described in the audio path were FLAC for lossless and OPUS for lossy.
YouTube (Music (https://music.youtube.com/)) does.
But it does that for low bitrates at 128 kbps (same for base YouTube). It uses AAC for 256 kbps.
Didn't know about Amazon Music though.
But it does that for low bitrates at 128 kbps (same for base YouTube). It uses AAC for 256 kbps.
Incorrect.
(https://i.imgur.com/DIC9MOV.png)
(There were no other audio formats/bitrates available.)
No other "audio only".
The 3gp format is even lower including video - indeed, audio is the biggest part of it.
25k AAC audio only, 40 including video (for this Hotel California video - that is not reported from the tool, I downloaded the entire 1.8 MiB and checked).
Not that it at all contradicts your point.
But YouTube had AAC at 256 at some time. And 192. I don't know why they scrapped it - whether they improved their AAC encoder or just decided they didn't need that high number for marketing or what. After all,
But YouTube had AAC at 256 at some time. And 192. I don't know why they scrapped it
A (wisely determined) unnecessary waste of bandwidth would be my guess.
(There were no other audio formats/bitrates available.)
Youtube-dl doesn't show other formats because browser player doesn't do so in most cases. You need Youtube Music client for your smartphone and Premium subscription to get access to format 141 which is AAC 256 kbps.
See here for the full list of format IDs:
https://gist.github.com/AgentOak/34d47c65b1d28829bb17c24c04a0096f
Youtube-dl doesn't show other formats because browser player doesn't do so in most cases. You need Youtube Music client for your smartphone and Premium subscription to get access to format 141 which is AAC 256 kbps.
I don't use youtube-dl but yt-dlp (https://github.com/yt-dlp/yt-dlp), but other than that I stand corrected. The initial claim that "not a single music streaming service uses Opus" has nevertheless been shown to be incorrect by this example, but its goalpost-shifted cousin ("only for low
* bitrates") hasn't.
* - To be honest, I wouldn't consider 128 kbps a "low" bitrate for either Opus or AAC.
A (wisely determined) unnecessary waste of bandwidth would be my guess.
But for comparison, the video sizes they introduced. Look at video id LhfNrsEghkA at format 313 - doubling the audio bitrate would increase the file size by less than a percent.
And now it seems that yeah, they have kept 256 - only moved it behind the paywall? (Another topic is multichannel sound at YouTube ... there were some news about that recently, what are they going to do there and in what format(s)?)
and it is better to be lossless in terms of perfect audio preservation.
This applies per definition. Lossy cant perfectly preserve audio no matter how you look at it. You are right in saying that going for high bitrate in lossy is pointless. You get the worst of both lossy and lossless. Big Files + no quality guarantee.
I have seen some people go for 500 kbps on ogg/vorbis/opus, which is just ridiculous. I dont even know where they think that bandwidth would go as opus blocks any frequency above 20 khz anyway. For reference you can have the full spectrum on aac. Does it matter at all? I dont know i cant hear this stuff at all. But at least i know that there is at least theoretically some improvement possible.
Having said that, this is abt the point where i just value my ability to just not give a damn about these sort of things with lossless.
This applies per definition. Lossy cant perfectly preserve audio no matter how you look at it.
Let me describe some philosophical issues with me about this topic.
1. Lossless doesn't perfectly preserve source audio either, because source audio is analog, and we quantize it during analog-digital conversion during recording, mixing and mastering (we talk lossless compression against original PCM stream which contains already somehow "butchered" original audio);
2. Why do we still choose 16/44.1 for distributing and listening to music, aside from compatibility with 40-years old CD format, if it butchers original analog sound a lot more than, for example, 32/384? Because 16/44.1 is actually ENOUGH for everyone's hearing if the stuff is properly mastered;
3. Wait... I started talking about "everyone's hearing" rather than "perfect audio preservation"... So what's the point of lossless 16/44.1 then if 256-320 kbps AAC is enough for everyone's hearing as well?
Just two philosophical cents about why I'm strictly a lossy fan (I use FLAC only for archiving).
Since the purpose of music is to bring enjoyment to the humans who listen to it, and humans can't hear above 20 kHz, 16/44.1 is indeed a perfect representation of the relevant information.
I tend to be of the general mindset that with lossy audio it's largely about using roughly smallest file size while still maintaining a high enough level of overall sound quality when just sitting back and enjoying ones music and in this regard I generally default to the following (which should be pretty good minimal starting points for most people in general in my estimations)...
-MP3 = V5 (130kbps). based on public listening tests I am willing to bet most people would struggle to notice a difference here when just sitting back and enjoying their music on a typical set of speakers/headphones and the like.
-AAC(AAC-LC) = 96kbps. while TVBR is the default, if someone wants to sacrifice a little storage space for 'maybe' a hair better sound quality, then CVBR is probably a bit safer, especially since 96kbps is pretty much a safe minimum for AAC-LC if you ask me. so by using CVBR it can help force the bit rate a bit higher on average over TVBR, which makes it a touch safer/less risky.
-Opus = 64kbps. while 96kbps etc is still a good option, I feel at that point one is better off using AAC due to near-universal compatibility and once you get around 128kbps/130kbps on up, a person is probably better off with MP3 for universal compatibility. that's why if someone is going to use Opus, I feel 64kbps is a pretty good all-around setting and gives it a worthwhile advantage (in file size) over MP3/AAC. like I would say it's plausible that Opus @ 64kbps ain't much different than MP3 @ V5 (130kbps) to where I am fairly confident if someone is happy with MP3 @ V5, they will probably think Opus @ 64kbps is good enough to.
because I feel of someone is THAT picky on sound quality were they throw a excess bit rate at it to clean up what is likely a pretty small difference overall, they are better off going straight to lossless (FLAC etc) because the whole point of lossy is to get a large enough portion of the overall sound quality at the lowest file size. I never understood those who use MP3 at 320kbps CBR as it's pretty much a waste of storage space when something around V2 (190kbps) is probably roughly as good but is much more efficient. even if there are some improvements, it's got to be so small that it definitely don't justify the loss in efficiency.
And streaming services have their own specific needs too, like avoiding complaints about the lack of high frequencies in spectrograms. :)
Yeah, it's amazing you see people complain about 'visual' on something like audio as visuals mean absolutely nothing as you already know as it's all about whether a person can actually hear a difference or not.
Since the purpose of music is to bring enjoyment to the humans who listen to it, and humans can't hear above 20 kHz, 16/44.1 is indeed a perfect representation of the relevant information.
Exactly. because standard audio CD's (or equivalent) already exceed human hearing so it's pointless to increase any further on standard stereo sound. which is why I think it's pretty much totally pointless for people to want those "HD" audio files at higher settings as all it does is waste storage space for no real world gains. that's why it's not really possible to improve on what we have had for pretty much 40 years now (i.e. standard audio CD format) as it topped out and that's that as there is no point in striving for 'better'.
that's why I never understand why some people dump a ton of money on sound equipment as once you reach a certain point, differences likely become pretty small at that point to where it definitely don't justify spending a boat-load of money for such small gains (if any gains). like someone who spends hundreds of dollars tops (like less than $1k and I suspect even half of that tops will get your something well more than good enough) is probably very similar to those who spend thousands or more. even if I had money to burn, I still don't think I would spend beyond a reasonable figure just on principle.
p.s. hell, even visual stuff with TV's, which is easier to notice differences... even here I think things are pretty much at the point for a while now to where you can get a pretty strong TV at a fairly big size for not much money. so those spending a lot of money on a TV, it's probably not much better (even if it is on some level it's probably just smaller differences (like small differences in color etc) and probably largely comes back to personal preference) for double or triple the price etc.
3. Wait... I started talking about "everyone's hearing" rather than "perfect audio preservation"... So what's the point of lossless 16/44.1 then if 256-320 kbps AAC is enough for everyone's hearing as well?
Just two philosophical cents about why I'm strictly a lossy fan (I use FLAC only for archiving).
You asked the question and you answered it.
Not tampering with the signal is also idiot-proofing. (Some of us need that.)
In the age of remasters, I wonder how many original digital masters are actually around.
Of course, had one had today's codec technology in 1980, they would have used something else than uncompressed PCM on the "CD" or whatever it would have been. I wonder though, when would a "CD player priced" machine have been able to decode a FLAC-alike format.
source audio is analog
I'd say the source audio for a
music release is whatever was used during production, usually a 44.1-96kHz / 32-64bit float digital format. The "butchering" of various elements that were used is not just pretty much inevitable, but also desirable, if you want the music to sound right.
The final 44.1kHz/16bit release is therefore a product of lossy conversion, yes. But if it's audibly transparent (and I don't mind asking that), does it matter?
if 256-320 kbps AAC is enough for everyone's hearing as well?
It isn't always, as shown by killer samples.
I'd say the source audio for a music release is whatever was used during production, usually a 44.1-96kHz / 32-64bit float digital format.
Giving the fact that I talked strictly about philosophy, the source audio for the stems used in production is still an analog, continuous signal. The voice of a signer, live instruments, etc etc - these are all analog signals, not discrete. So no matter what bit depth and sample rate you used in production, you technically butchered the original signal during AD conversion because you made it discrete.
So no matter what you do in a digital sound, you work with butchered signal, and the philosophy is to find a compromise between how much you butcher it and how much data you're ready to spend on a result. Is lossless 16/44.1 really a good compromise comparing to a lot of lossy formats existing in a wild? I don't think so. I'm talking about final masters of course.
It isn't always, as shown by killer samples.
Music industry proved AAC 256-320 to be robust against 99.9% of music material in the world, even if there're extremely rare cases where AAC fails (they're a lot more rare than some labels failing to provide properly mastered versions of music to streaming services for example).
Giving the fact that I talked strictly about philosophy, the source audio for the stems used in production is still an analog, continuous signal. The voice of a signer, live instruments, etc etc - these are all analog signals, not discrete. So no matter what bit depth and sample rate you used in production, you technically butchered the original signal during AD conversion because you made it discrete.
This is nonsense. You need to educate yourself about how sampling works. All frequencies below 22 kHz are perfectly preserved; nothing is "butchered". Frequencies above 22 kHz are eliminated entirely, since humans can't hear them, and storing them would be a waste of space.
Xiph has an excellent video explaining how digital audio is able to perfectly recreate analogue signals:
https://www.xiph.org/video/vid2.shtml
This is nonsense. You need to educate yourself about how sampling works. All frequencies below 22 kHz are perfectly preserved; nothing is "butchered". Frequencies above 22 kHz are eliminated entirely, since humans can't hear them, and storing them would be a waste of space.
@Eurobeat_fan's point is that these operations are
lossy but good enough.
You are calling nonsense because it is good enough ... ?
Also, humans have in controlled experiments heard frequencies over 22 kHz (https://asa.scitation.org/doi/abs/10.1121/1.4788280). (Now look at the volumes they applied.)
I'd say the source audio for a music release is whatever was used during production, usually a 44.1-96kHz / 32-64bit float digital format.
Giving the fact that I talked strictly about philosophy, the source audio for the stems used in production is still an analog, continuous signal. The voice of a signer, live instruments, etc etc - these are all analog signals, not discrete. So no matter what bit depth and sample rate you used in production, you technically butchered the original signal during AD conversion because you made it discrete.
Just to 100% avoid misunderstanding: my point was that the analog stems are not the music product. They're merely building blocks and additional processing (mixing...) is required, with some rare exceptions maybe.
Now, that additional processing could be done in the
analog domain and in some ideal world it would add no additional noise and distortion. And then it would be delivered to end users on an ideal analog medium. But 1. those analog stems would still be butchered by the production processes and 2. we don't live in a world with ideal analog media. The best we can do to minimize noise and distortion is to use
digital processing and delivery.
So, while I think I get the philosophical point, I don't see how it applies to the real world or why that's an argument for not using a lossless over lossy format, if high fidelity is required. It might be just 100% vs 99.9%, but that's still something.
Whether 44/16 is in fact always transparent or not is a different can of worms. It could turn out that we need more than that in some cases. But I don't see any evidence that digital audio encoding in general is not good enough.
Wait... I started talking about "everyone's hearing" rather than "perfect audio preservation"... So what's the point of lossless 16/44.1 then if 256-320 kbps AAC is enough for everyone's hearing as well?
There's a
very substantial difference in the amount of assumptions and modelling simplifications/compromises involved in ascertaining that lossless 16/44.1 is transparent to everyone's hearing and that a lossy encoder at any bitrate is transparent to everyone's hearing.
For the former, the
only two assumptions you need to make about human hearing are the audible frequency range and the audible dynamic range. That's it.
As for the latter, well... the book Psychoacoustics: Facts and Models (https://books.google.com/books/about/Psychoacoustics.html?id=eGcfn9ddRhcC&redir_esc=y) has over 400 pages detailing the various assumptions as well as modelling simplifications/approximations and compromises involved. Here's just one example of such a modelling approximation, from page 159 of the book:
(https://i.imgur.com/QMrPObj.png)
Just two philosophical cents about why I'm strictly a lossy fan (I use FLAC only for archiving).
Yup, same here, FLAC as a best possible source for good encoding (here SoX resampling to 48k, then Apple LC AAC at tvbr 100/109 - already a bit much security). 48 because native on my devices. Direct listens to flac only on PC, on crappy desktop speakers. I love my m4as, created and tagged with ❤️
@Eurobeat_fan's point is that these operations are lossy but good enough.
You are calling nonsense because it is good enough ... ?
None of the frequencies that you can possibly hear are lost, so the operation is lossless. This nonsense about digital audio "mangling" analogue audio is a major factor in why the snake oil about needing 192 kHz sampling rates continues to persist. If 44.1 kHz is mangling the continuous analogue sound, then surely 192 kHz will mangle it much less and provide a better listening experience! ::)
None of the frequencies that you can possibly hear are lost, so the operation is lossless.
Lossless is not about what YOU hear though, it's about ideally preserving source information. When you start talking about what YOU hear and what's enough for YOU as a human, you enter the lossy area. (which is not bad to me if you ask)
Lossless is not about what YOU hear though, it's about ideally preserving source information.
Except that's physically impossible, so it makes no sense to define the term that way. I also refer you to my previous message.
The Monty video is correct about the sampling theorem, but it does not mean every ADC has the same performance (e.g. noise, distortion, decimation filter specs etc) when using the same recording format.
Before things get overly complicated, consumers don't have access to unprocessed raw audio data and DAW project files. In the simplest form, digital source can be from a standalone recorder (e.g. Tascam, Sony , Zoom...) going through a preamp or analog console without using any DAW or intermediate analog storage medium (e.g. tape).
In either case, consumers cannot control any upstream process, apart from worry about things like conversions between DSD and PCM, hi-res and 16/44, and the more destructive perceptual formats <- the original topic.
You know, this feels like arguing that images can never be stored losslessly, because you must pick a resolution and colour depth to do so, so might as well always use JPEG.
Yes. It's like complaining that cameras are not lossless, because they don't capture the infrared and ultraviolet portions of the spectrum.