Why 24 bit audio and anything over 48k is not only worthless, but bad for music.

[71 dB]

32bit recording gives you sound levels you might want at a later date.

Yeah, if you had microphones with dynamic range of about 200 dB, but you don't have that, not even close!

 

[gregorio]

Sure!

Then I still don’t understand your point, we’ve been using synths for 70 years and some/many sampled/virtual instruments are indistinguishable from the real thing.

Gregorio (if I can speak for him) was pointing out that realism wasn't a goal in the miking and sound mixing of commercial music.

We can’t really say that realism isn’t a goal when miking sound or music, sometimes it is and quite often it’s not. However, when mixing it’s pretty much never the goal.

Or are we no longer really reading anymore?

The situation is different these days because what we’re reading is different. It’s usually pretty hastily written posts on social media rather than carefully considered and edited books and articles written by professional writers.

So why are there bad recordings then? The intent is one thing. Making it happen is another, and that seems to be the hard part.

Two reasons:
1. The skill of the artists/engineers. Some just aren’t very good and even the good ones can sometimes make mistakes or errors of judgment.
2. How do we define what are “bad recordings” to start with? EG. Are they bad because they really are bad or because we personally don’t like them? This isn’t an easy question to answer because the answer can change, even the same persons answer, due to time and the evolution of tastes/preferences for example.

Given that music is in a real environment, I'm not sure how it can sound better than real.

That would be true, if we take that as “Given”. However, it’s not true because it’s not “Given”. With sound for film for example, the real environment where the sound was recorded was a Foley studio, an ADR studio and whatever environment the car starting, bomb exploding or bird singing, etc., was recorded in. So all different real environments, all mixed together at the same time and probably none of them remotely like the environment we’re seeing on screen, which is probably a plywood “set” anyway! In music we’ve often got a similar problem. With pop/rock and other popular genres the vocalist was probably recorded in a vocal booth, the drums in a different booth backing vocals and other instruments in the main live room and of course synths and DI’ed guitars don’t have any “real environment”, their output is analogue or digital, not acoustic. And even those instruments that are acoustic are mostly close mic’ed so there’s little or no “real environment” recorded. The “real environment” isn’t real, for the last 40 years or so it’s been a digitally manufactured environment and before that it was analogue reverbs, plates, springs and “worldising”. Plus, it’s not even a manufactured environment it’s almost always a combination of manufactured environments, say a small room or plate reverb on the lead vox, a reverb/delay indicative of say a sports stadium on the lead guitar, a tiled/bathroom type reverb on some of the drum kit, etc. Even acoustic music in a real environment, such as an orchestra in a concert hall isn’t a “real environment” because we’d virtually always record an orch with 40 or so mics, a main array a few metres above the orch, spot mics a metre or so from certain instruments, outriggers and also various ambience mics/arrays up near the ceiling and/or the back of the venue. Although it’s the same “real environment” it’s not an environment that could ever be perceived, unless you know someone with 40 or so ears spread say 60 metres apart.

A sound engineer's intent is making something that sounds good for any intended medium (and that it sounds "natural" as far as hearing the artist's performance).

No, that’s true for a mastering engineer but not for a sound engineer, a recording and other engineers. It is usually a consideration for mix engineers but not the main intent. The intent to “sound natural” is completely variable, in some cases that’s true, in other cases that’s the exact opposite of our intent (no one wants to hear a natural electric guitar, what they want is nearly pure distortion and a heavily distorted/effected drum kit is almost always the intent, there are countless other examples).

But it does still take effort to do proper miking at conventional bit depths. Less so with 32 bit

That all depends on what we’re mic’ing. Often there’s no difference and where there is a difference it’s just more effort in terms of being more careful when setting the mic pre-amp, to avoid clipping. There is no quality difference (assuming clipping/overload is avoided), it’s only about recording headroom and 32bit is not less effort than say 20bit or 24bit because it doesn’t offer anymore headroom. The most dynamic range any ADC can provide is about 20bit, beyond that it’s thermal noise.

Well there's mic placement: but again, 32bit recording gives you sound levels you might want at a later date.

32bit does not give you any more sound levels than 16bit when recording. It might give you more headroom in certain situations but of course that’s less sound level, not more and when recording, it doesn’t give you more of anything over 20 or 24bit.

G

 

[gregorio]

However, people think it is enough (in life performances). Why isn't the same enough with home play?

Because it’s home play. In live performances you see the real life performer/s in front of you. You spend effort deciding and planning to attend a live performance, you spend money on the tickets, a significant amount of time travelling to the concert venue, some time queuing and then sitting and waiting for it to start. All the while expectation is growing and, you’re surrounded by hundreds/thousands of others all with growing excitement/expectations. It’s near impossible for all that not to affect your perception. Even the musicians themselves, who do it week in, week out are affected by it.

Why do we have to make home audio less natural and real, but "greater"?

You don’t have to buy a ticket to listen to your home stereo, you don’t have to get a train and walk for 20 minutes to get from your kitchen to your sitting room, you’re not impressed by a world famous concert venue, you aren’t surround by hundreds of excited/expectant fans, the famous musicians aren’t in the same room as you, etc. We obviously cannot make a recording for home play that recreates all these factors so you have the same perception but what we can do is make it a little more dramatic than it actually was, to slightly compensate. The reason we “have to” make home audio greater but less natural/real is that history demonstrates lower appreciation and sales if we don’t.

How should a recording listened at home sound compared to life music? Not a 100 % trivial question in my opinion.

It is completely trivial in the sense that we commonly don’t ask or consider that question. We’re often just considering what to do to make it sound subjectively good/better, with no reference or comparison to live music. If anything, the issue is the opposite, how to make the live performance comparable/recognisable to the recording the audience are accustomed to. With acoustic genres, it’s a more valid question.

The bigger star he/she is the more his/her word matters in the production.

Again, that completely depends. Especially in the classical world, many artists have absolutely no interest in the production. They do the recording session, pack up and go home or the really big stars go off to a meeting with a team of agents/managers or to a press conference, etc. Some don’t even listen to any of what they’ve just recorded, let alone have any input to any of the editing, mixing or mastering. This is not true of all classical artists but the majority IME and in the rock world for example, there’s often at least one member of the band who takes an active interest beyond just their own performance.

The problem of bad recordings is relevant when it comes to more obscure composers and works.

Sure but that’s an issue of money, the recordings of obscure composers and works aren’t going to sell in large numbers so you can’t justify a large recording budget. Abbey Road Studio 1 costs about £16,000 a day, that could easily be 10 times more than the available budget for the whole recording, editing and mixing process, which will likely take several days/weeks. So, you’re not going to get world class equipment, world class recording, mixing or mastering engineers or a world class producer. What you’ll end up with is a decent recording if you’re lucky, a very good recording if your exceptionally lucky or a crap recording if you’re unlucky.

In the example you gave, what was the budget, maybe each of the sponsors gave a few hundred Euros? Maybe it was a budget big enough for a top class studio, although that would be surprising. Top class studios don’t produce bad recordings because they couldn’t justify their prices and would very quickly be out of business. There are exceptions to this general rule but IME they are always down to interference by the clients. IE. The clients giving recording, mixing or mastering directions against the advice of the engineers/producer or occasionally a client will insist on using a novice or incompetent producer.

G

 

[71 dB]

I won't be responding to the comments, because I wasn't arguing. I was thinking out loud + I don't feel good (health problems).

 

[Davesrose]

That all depends on what we’re mic’ing. Often there’s no difference and where there is a difference it’s just more effort in terms of being more careful when setting the mic pre-amp, to avoid clipping. There is no quality difference (assuming clipping/overload is avoided), it’s only about recording headroom and 32bit is not less effort than say 20bit or 24bit because it doesn’t offer anymore headroom. The most dynamic range any ADC can provide is about 20bit, beyond that it’s thermal noise.

32bit does not give you any more sound levels than 16bit when recording. It might give you more headroom in certain situations but of course that’s less sound level, not more and when recording, it doesn’t give you more of anything over 20 or 24bit.

When I mentioned a 32bit workflow, I was thinking about this article: which does bring up the cases in which 32 bit float is used. From what I understand, it’s not needed in a studio environment in which you can easily control sound levels.

https://www.wired.com/story/32-bit-float-audio-explained/amp

It also seems to me that we’re arguing terms if I say you want to strive for a recording that naturally conveys the intent of a performer vs adequately capturing a distortion effect is “not real”.

 

[gregorio]

When I mentioned a 32bit workflow, I was thinking about this article: which does bring up the cases in which 32 bit float is used.

Yes, but it’s wrong! There are numerous points that are wrong or misleading. My guess is that it’s not deliberate just a lack of understanding. I’ll give just a few examples:

“24-bit audio recordings can capture a dynamic range of up to 144.5 dB. Meanwhile, 32-bit float audio can capture the absolutely ludicrous range of up to 1,528 dB.” - Theoretically 24-bit audio has a dynamic range of 144dB but we cannot capture/record 144dB dynamic range because there are no ADCs capable of recording it, most of the best ADCs have a dynamic range around 120dB (which is equivalent to 20bit). The limiting factor is thermal noise in the analogue section. 32bit float offers the values which would theoretically allow 1,528dB dynamic range to be numerically represented but of course there is no sound which has that range and we’re still limited to the same pesky laws of physics all ADCs are subject to, which dictate a thermal noise floor around -120dB so what we can record with 32bit in practice is a 120dB dynamic range, same as 24bit!

“For a bit more scale, a person whispering can be around 20 to 30 dB, while a typical conversation is around 60 dB. A motorcycle driving by would be about 90 dB, and a really loud concert can be in the area of 110 dB” - But that’s peak level, not dynamic range, you’re not going to get a motorcycle driving past you in an anechoic chamber (with a noise floor of 0dB), the motorcycle will be on a road and the noise floor would probably be 50dB or more, so the dynamic range is just 40dB and the noise floor of a really loud concert would probably be 60-70dB, so a dynamic range of 40-50dB. All the examples given are way within the limitations of 16bit, let alone 24bit.

“Now, typically you’ll set audio levels when setting up your equipment to avoid hitting that limit. Setting those levels involves applying gain to the signal from the mic, which is an irreversible step that crushes the dynamic range of even 24-bit recording.” - That’s just nonsense, it demonstrates again, they don’t understand the difference between peak level and dynamic range. Applying gain to the signal from the mic obviously changes the peak level but does not affect the dynamic range. Let’s say we have someone talking at say a level of -30dB and the noise floor of the room is say at -55dB so a dynamic range (SNR) of 25dB. What happens if we turn up the gain of the mic pre-amp by say 20dB? The talking is now at -10dB and the noise floor is at -35dB, quite different peak levels but the dynamic range is still 25dB. There’s no “crushing”, not at 24bit or 16bit.

“Setting levels on 24-bit systems can be tricky because of the noise floor. To oversimplify, no matter how quiet you make your recording space, there’s always some amount of noise from background objects, or even the electronics you record on. Adding gain to the signal while recording will amplify that noise, as well as your audio source, and once it’s baked into the recording, it’s there for good.” - And again, adding gain does not change the dynamic range and now they’re demonstrating they don’t know the difference between the acoustic noise floor, the digital noise floor or the analogue noise floor. If that’s not bad enough, they’re effectively suggesting that the 32bit float format can somehow magically reduce the acoustic noise floor.

There are in fact benefits to 32bit float and indeed to 64bit float but that’s only in the mixing process, under certain conditions and doesn’t require or benefit from the audio file format being 32 or 64bit, only the mix environment.

I could go on, this is all pretty basic stuff and it’s rather shocking that a Sound Supervisor/Dialogue Editor doesn’t know it, although knowledge of digital audio theory isn’t required for those particular roles. Instead of supervising or editing, he should actually try some tests with a mic an ADC without a mic pre-amp and see for himself what happens! Having said all the above, it could be a case of the article author not reporting accurately.

G

 

[Davesrose]

Well I'm not sure to beleaguer a point: but it seems you're talking about the realized levels of one microphone, whereas I think the article is talking about the project file: that has a combination of tracks with microphones set at different levels. It's similar to why 3D artists always use 32bit files. Even though there are scenes that don't actually realize 32 stops of light, 32bit float can be faster at processing, and there's no loss in saving the data in a source.

 

[gregorio]

Well I'm not sure to beleaguer a point: but it seems you're talking about the realized levels of one microphone, whereas I think the article is talking about the project file: that has a combination of tracks with microphones set at different levels.

Of course I’m talking about one microphone, when recording what else would you have in an audio file and how could you mix anything if it were all on the same track/audio file?

The “project file” (or session file) doesn’t contain any audio, it’s a file that contains descriptions of all the tracks individually with links to all the referenced audio files. When you load the session file a (64bit float) mix environment is created in RAM and all the audio files at 16 or 24bit are loaded into that mix environment (or read off hard disk as needed, if there isn’t enough RAM). But of course now we’re talking about mixing, not recording and none of the article is applicable because there has never been any 24bit mix environments.

It's similar to why 3D artists always use 32bit files.

I have no idea what 3D artists do but it does not sound similar. Mixing audio today and for many years is always “virtual” and even in the analogue days it was effectively virtual.

32bit float can be faster at processing, and there's no loss in saving the data in a source.

Faster processing is one of the reasons why 32bit was used but 64bit is almost ubiquitous these days. I’m not sure what you mean by “saving the data in a source”. We don’t save the audio data until we’re finished mixing, it’s all virtual and you can’t save a 64bit float file anyway.

G

 

[Davesrose]

The “project file” (or session file) doesn’t contain any audio, it’s a file that contains all the tracks individually with links to all the referenced audio file. When you load the session file a (64bit float) mix environment is created in RAM and all the audio files at 16 or 24bit are loaded into that mix environment (or read off hard disk as needed if there isn’t enough RAM). But of course now we’re talking about mixing, not recording and none of the article is applicable because there has never been any 24bit mix environments.

I have no idea what 3D artists do but it does not sound similar. Mixing audio today and for many years is always “virtual” and even in the analogue days it was effectively virtual.

I know you're an experienced engineer that uses professional studio software. Is there really no applications that don't utilize "linked" tracks, but actually embed them in a saved format? I can also understand how 24bit is all you ever need for mixing in a studio setting when you have control of what you're recording. As a layman with sound engineering (who often tries to draw the similar analogies to digital imaging), I'm trying to consolidate your perspective vs other engineers saying they're using 32bit workflows.

Now when you bring up "virtual", I think we're now back to differences in terminology. You've brought up the dynamic range of a microphone. That is similar to the dynamic range of a camera sensor (which has a realized range based on your exposure settings and the noise floor/saturation point of that particular sensor). These days the best cameras can capture 16 stops of light at base ISO (or 16bit). However, you can have a working image of 32 stops of light that's a merging of separate exposures. If we are to harken back to the analogue film days, there was also a "virtual" editing in the darkroom that you could selectively expose different areas of a print with a different exposure (often needed as a film negative could have 14 stops of light, while a print would have 8 stops).

 

[gregorio]

Is there really no applications that don't utilize "linked" tracks, but actually embed them in a saved format?

I’m not sure what you mean by “linked” tracks or a saved format? Presumably you’re not you talking about a stereo file?

I can also understand how 24bit is all you ever need for mixing in a studio setting when you have control of what you're recording.

No, for mixing 24bit is not enough and there has never been any commercial 24bit mixers as far as I’m aware. The first digital mixer on the market was Yamaha’s DMP7 (in 1987) and if I remember correctly it had a 27bit mix environment, the first large format mixers in the early/mid 1990’s were 32bit float. ProTools when I first used it in about 1997 was 56bit (fixed).

I'm trying to consolidate your perspective vs other engineers saying they're using 32bit workflows.ex

Depends what you mean by workflow, I edit, mix, process and master using a 64bit workflow.

Now when you bring up "virtual", I think we're now back to differences in terminology.

Very possibly, I mean that it only exists in RAM, it’s not committed to a file. In the analogue days, we had a tape recorder with 24 tracks, these 24 tracks were fed into 24 channels on the mixer, here we adjusted the levels, EQ, compression, etc. of each channel, the audio from these channels were not saved or committed. We reproduced/replayed this mix by replaying the tape of the original recording through those same channels without changing the channel settings. The mix is effectively virtual, it only exists in the mixer, until we record it to another tape machine when we’re finished. This workflow is essentially the same with digital/DAWs. Except we’ve got audio files instead of tape tracks the whole mix environment is in RAM instead of on a mixing desk, the audio files are played/loaded into the mixer and we can save the state of the mix environment, close the computer down and reload it all. While in analogue mixing we had to leave the desk switched on and couldn’t “reload it” if we changed the settings (although later analogue desks offered a limited “Recall” ability).

You've brought up the dynamic range of a microphone.

I didn’t even mention the dynamic range of mics, I mentioned the actual acoustic dynamic range (peak sound levels and the noise floor of the recording location), I mentioned the analogue dynamic range of ADCs and the theoretical digital dynamic ranges. If we’re talking about mic’s in the context of film sound though, Shoeps CMIT 5 is the cream of the crop, about $4k with the required accessories ($2.4k just the mic) which is the most expensive shotgun mic I’ve seen/used. A great mic, with great off axis rejection, freq response, great specs in general and great features, it would definitely be my first choice for production sound recording if money were no object. It also has a very good dynamic range compared to competitors, at 81dB. That’s 15dB less dynamic range than 16bit and about 40dB less than the best 24bit ADCs. So if you’re using 16bit, recording to the dynamic range limits of a top mic and need more than 15dB headroom, then a 24bit ADC would offer an advantage, although I can’t think of any situations where you’d need both 81dB dynamic range and 40dB of headroom. A 32bit file format offers no benefit at all over 24bit when recording.

That is similar to the dynamic range of a camera sensor (which has a realized range based on your exposure settings and the noise floor/saturation point of that particular sensor).

I’ve no idea how any of the above relates to camera sensors.

G

 

[Davesrose]

I’m not sure what you mean by “linked” tracks or a saved format? Presumably you’re not you talking about a stereo file?

No, for mixing 24bit is not enough and there has never been any commercial 24bit mixers as far as I’m aware. The first digital mixer on the market was Yamaha’s DMP7 (in 1987) and if I remember correctly it had a 27bit mix environment, the first large format mixers in the early/mid 1990’s were 32bit float. ProTools when I first used it in about 1997 was 56bit (fixed).

Depends what you mean by workflow, I edit, mix, process and master using a 64bit workflow.

Very possibly, I mean that it only exists in RAM, it’s not committed to a file. In the analogue days, we had a tape recorder with 24 tracks, these 24 tracks were fed into 24 channels on the mixer, here we adjusted the levels, EQ, compression, etc. of each channel, the audio from these channels were not saved or committed. We reproduced/replayed this mix by replaying the tape of the original recording through those same channels without changing the channel settings. The mix is effectively virtual, it only exists in the mixer, until we record it to another tape machine when we’re finished. This workflow is essentially the same with digital/DAWs. Except we’ve got audio files instead of tape tracks the whole mix environment is in RAM instead of on a mixing desk, the audio files are played/loaded into the mixer and we can save the state of the mix environment, close the computer down and reload it all. While in analogue mixing we had to leave the desk switched on and couldn’t “reload it” if we changed the settings (although later analogue desks offered a limited “Recall” ability).

I didn’t even mention the dynamic range of mics, I mentioned the actual acoustic dynamic range (peak sound levels and the noise floor of the recording location), I mentioned the analogue dynamic range of ADCs and the theoretical digital dynamic ranges. If we’re talking about mic’s in the context of film sound though, Shoeps CMIT 5 is the cream of the crop, about $4k with the required accessories ($2.4k just the mic) which is the most expensive shotgun mic I’ve seen/used. A great mic, with great off axis rejection, freq response, great specs in general and great features, it would definitely be my first choice for production sound recording if money were no object. It also has a very good dynamic range compared to competitors, at 81dB. That’s 15dB less dynamic range than 16bit and about 40dB less than the best 24bit ADCs. So if you’re using 16bit, recording to the dynamic range limits of a top mic and need more than 15dB headroom, then a 24bit ADC would offer an advantage, although I can’t think of any situations where you’d need both 81dB dynamic range and 40dB of headroom. A 32bit file format offers no benefit at all over 24bit when recording.

I’ve no idea how any of the above relates to camera sensors.

G

I think we're back to mixing up concepts. So in your previous post, you said the individual tracks are either 16bit or 24bit sound files (I'm not talking about the processing bit depth). So the difference with that vs digital imaging is that the universal fixed formats are 8bit, 16bit (integer or float), or 32bit (float) per channel (be it R+G+B, or additional greyscale data). There's also RAW sensor formats that are specific to that camera's dyanmic range (or video codecs that save space and are limited to 10bit, 12bit, 14bit).

I think one difference of concept we're crosstalking is realized dynamic range vs what the file format is capable of. You're talking about the realized dynamic range of the recording after you have set the microphone's sound level (and what you had issues with my linked article's talk about theoretical dynamic range). With digital imaging, I'm saying the best ADC can realize 16bit with the right environment and exposure settings. But for quite a few years, we've had access to 32bit image file formats that gives us more exposure latitude (or realistically simulate light levels in daylight for 3D animation). I'm seeing that you're quick to dismiss "32bit workflow", and I can understand in your professional environment in a studio, it's irrelevant. As a layman, I've been trying to consolidate and understand your perspective vs others saying they're now using 32bit float audio files.

 

[chef8489]

I think we're back to mixing up concepts. So in your previous post, you said the individual tracks are either 16bit or 24bit sound files (I'm not talking about the processing bit depth). So the difference with that vs digital imaging is that the universal fixed formats are 8bit, 16bit (integer or float), or 32bit (float) per channel (be it R+G+B, or additional greyscale data). There's also RAW sensor formats that are specific to that camera's dyanmic range (or video codecs that save space and are limited to 10bit, 12bit, 14bit).

I think one difference of concept we're crosstalking is realized dynamic range vs what the file format is capable of. You're talking about the realized dynamic range of the recording after you have set the microphone's sound level (and what you had issues with my linked article's talk about theoretical dynamic range). With digital imaging, I'm saying the best ADC can realize 16bit with the right environment and exposure settings. But for quite a few years, we've had access to 32bit image file formats that gives us more exposure latitude (or realistically simulate light levels in daylight for 3D animation). I'm seeing that you're quick to dismiss "32bit workflow", and I can understand in your professional environment in a studio, it's irrelevant. As a layman, I've been trying to consolidate and understand your perspective vs others saying they're now using 32bit float audio files.

I think you need to stop trying to apply photography to audio. It's not in the same world and doesn't work the same and terminology transfer over.

 

[Davesrose]

I think you need to stop trying to apply photography to audio. It's not in the same world and doesn't work the same and terminology transfer over.

I've been saying there's an easy application for why 32bit float file formats is used with digital imaging. I've seen folks say that they're now using audio formats that are 32bit: I'm the one asking why?

 

[chef8489]

I've been saying there's an easy application for why 32bit float file formats is used with digital imaging. I've seen folks say that they're now using audio formats that are 32bit: I'm the one asking why?

Because it's marketing just like people pushing the sales of 24bit audio. Because higher has to be better right. At least that is what the uneducated person thinks, so the industry pushes it to make sales to the gullible.

 

[Davesrose]

Because it's marketing just like people pushing the sales of 24bit audio. Because higher has to be better right. At least that is what the uneducated person thinks, so the industry pushes it to make sales to the gullible.

We're talking about sound editors: not a gullible audiophile consumer pool.

Edit, when I look up 32bit audio files, I get the consensus that it's for new devices that don't need gain control (vs traditional recording technology where you do set levels to set the recorded range). https://www.pro-tools-expert.com/pr...t-float-audio-when-is-the-best-time-to-use-it