Audio volume standards and calibration

[GettingTooOld]

I'm looking for a way to automate testing some of the aspects of 1 minute test recordings on my own computer system. It has audio tools available, which give some perhaps inconsistent results depending on which ones are used and how, but that's a different matter in a way. I'm sure I should be able to eventually find a tool that does a reasonably consistent job.

No tool I've tested so far is suggesting volume levels in decibels as high as the standards quoted for LibriVox (LV) which are somewhere around 90ish. My own way of doind my own audio volumes for my own needs has always been along the lines of visually and by comparison. The visual way is perhaps the best way, as it goes straight to the heart of the matter, that is, to maximise the volume of any source to just below the point that the peaks are clipped off. Clipping off the peaks, which we call distortion, makes sounds sound awful as there is missing information. The softer and quieter the sound however, the more noise there is from the recording system, the more hiss or hum or both or whatever. The louder the incoming wanted signal, the better it can drown it out. Lining this all up has been very easy using audacity since before I can remember and before audacity. I did the same thing when doing recordings to cassette tape by ear rather than eye and for all the same reasons.

Longwinded? sorry.

Looking at every decibel chart on the internet, they too all have different and somewhat inconsistent measures for all types of things around us, and LV standards are to be as loud as a chainsaw, snowmobile or hairdryer. Hairdryer is popular, along with large trucks. Conversation is consistently reported as less than LV standards.

Audio files and movies are turned up or turned down for the effects desired really. Quiet movies want the volume turned way up so they can scare the bejeebers out of you with loud sounds at some point. Audiobooks just want consistent volume the way I have done it since audiotape and reel to reel (I had a few reel to reel machines in my time long ago).

How do we arrive at 90ish db, it seems tied to the output of an arbitrary software tool set up on a system not compatible with mine. Is there a ffmpeg filter that will show something consistent and meaningful does anyone know? Some commandline linux tool, or is just using eyes and ears ultimately better, because the desired target is ultimately arbitrary as it's not related to real world sound levels, it's down to computer files on software measured by other software and they're all arbitrary.

When sounds are translated into the computer, the Db's lose all meaning in the end because we want our audiobooks recorded to the identical standards as explosions in the movies, that is, to the maximum volume without distortion in an effort to minimise noise. To say 90db is not sensible unless it is always qualified by saying 'using this checker', which is usually but not always does.

Are there other standards for LV file volume, for example, is it just fine to go back to the days before the latest checking software, to the older LV standards of checking which are to the point. Or I should just post example results from whatever I can find and compare to the existing checker (bad idea, photocopy of a photocopy) or just post results to solicit opinions perhaps ? or will people just run any new checkers results through the existing checker to pick up shortcomings of the existing software, and add those to the errors of my new software, because of course, nothing is perfect, and software (and hardware) inevitably changes over time.

[redrun]

Short version: because everybody else is, so that's what every device supports. Search for ReplayGain tools, and you'll find something that can test for you, even if it needs some adaptation.

Otherwise, prepare for a ChatGPT-style answer: fuzzy background knowledge now hastily researched, and explained with an oversimplification. But hopefully a useful one.

I think the first part to address is "why a specific standard in db?"
To answer that, let's pretend that "db" actually means "% of max volume". Just as you say, with an old-fashioned speaker where you can start with any volume, then turn up the gain as high as you want, that db number is truly arbitrary.
But most digital players start at their max volume, and get quieter when the file says to, or when the user turns down the volume. Having a file that says to play at 90% of max volume is just fine, because the listener can adjust down if it's too loud. Having a file that says to play at 10% is probably always going to be too quiet.

So why specifically 89db, and... how can we measure it?
Now, "db" doesn't literally mean "% of max", but there is a definite algorithm for calculating the db volume of a given file or audio selection. How to do it is beyond me, but suffice to say it's relatively standard.
Part of why we'd use 89db is that many of the tools that implement this algorithm assume 89db is what you want. Many of the tools that have the same algorithm and (mostly) same volume assumption are called "ReplayGain", including a handy plugin for Audacity and several command-line tools. Tools like this one, designed to add metadata tags to files. We don't want those tags, but you can use the "-n" option to skip that part and just tell you the dbs it's measured.

I don't know much about the LibriVox Checker tool, but it seems to use the same algorithm. The Audacity plugin will tell you how much to change the volume in the file to get to 89db, but Checker will tell you the db level it measured. Both measure in the same way, with pretty consistent results, but display the results differently.

[ej400]

because the desired target is ultimately arbitrary as it's not related to real world sound levels,

Before I apply any amplification to my recordings, they will be around 60 to 65 db (and sometimes even lower, which is cannot be picked up by computer speakers sometimes unless it is amplified). If I were to submit that, I'd have to turn the volume up on my recording quite loud; however, some of the sound waves wouldn't be loud enough for my speakers to pick that up either, so I believe that the lower the sound waves are the less chance they will also have of even being picked up as "sound" at that point.

[InTheDesert]

I had this too when I wrote my 1 minute test validator.

Here's the Checker code that I ported to do the volume.

[GrayHouse]

As others have said, it's 89 dB according to the ReplayGain standard - those last words are the crucial bit. If you're comparing to the 'chainsaw' or 'jumbo jet' scale then you're not comparing like with like.

[GettingTooOld]

Short version: because everybody else is, so that's what every device supports. Search for ReplayGain tools, and you'll find something that can test for you, even if it needs some adaptation.

"everybody else" is an auto-discouragement in my personal firmware, I simply don't support what everyone else does. 'To be well adjusted to a sick society is not a sign of good health'. However, ReplayGain looks promising, I'll need to seach install that as a vlc thing and that may do it. Vlc will spit out all sorts of things when launched from the commandline. I love VLC, Vlc is good. The fact that almost 'everybody else' uses it too I put down to following my lead

It appears to add tags to the mp3 files, with perhaps useful information. I can write bash code to get that, I've been doing similar things for months now. Looks promising.

.... from https://addons.videolan.org/p/1135632/

How to use
First you need to analyze all your song files using replaygain. The replaygain algorithm calculates a perceptual volume (gain) for each track (track mode) and album (album mode) and uses this information to correct the playback volume so all tracks or albums sound equally loud. In album mode the relative differences between tracks of an album are preserved, so more quiet tracks will sound more quiet, and louder tracks will sound louder. Album mode is better for live recordings and albums meant to be more than a collection of tracks, while track mode is better for heterogeneus mixes and compillations.
This analysis will not change your file's sound data, it will store the information in tags.

So why specifically 89db, and... how can we measure it?
Now, "db" doesn't literally mean "% of max", but there is a definite algorithm for calculating the db volume of a given file or audio selection. How to do it is beyond me, but suffice to say it's relatively standard.

I expect it's a rate of change thing rather than peak numerical value or dc value at any given point, as it takes the ac component to actually produce the sounds you hear, same as in speakers themselves. So I think even if I write something to read the raw file, that won't get me anywhere. I have had sucess with that kind of thing before, getting frustrated with lack of free open software for cnc, I wrote up something to turn an image file into a 3D gcode cutting file, I was happy with that, but it was all digital rather than analogue analysis, no rates of change stuff.

When I look at the ffmpeg analysis there are a half dozen db readings for a given 1 minute test file. None of them relate in any way to the result given in the thread regarding volume.

Otherwise, prepare for a ChatGPT-style answer: fuzzy background knowledge now hastily researched, and explained with an oversimplification. But hopefully a useful one.

I'm auto-allergic to that style of fluff. The debain site is truely a pain in the R section, I mean download section, they use bulletproof BS to keep people away from trying to download it. Just go try, click download and it's like 'let me tell you the story of computers, from the beginning, over the next 2-1/2 pages before I tell you that you can't download it the way you want. Maybe I'll tell you. Yeah, it's worse than listening to me dawdle trough a conversation. It's Nuclear level bs, and AI has a similar effect, I auto-detect and auto-dislike AI. AI pages are like chewing through hours of infomercials.

I think redrun answers it with ReplayGain.

Before I apply any amplification to my recordings, they will be around 60 to 65 db (and sometimes even lower, which is cannot be picked up by computer speakers sometimes unless it is amplified). If I were to submit that, I'd have to turn the volume up on my recording quite loud; however, some of the sound waves wouldn't be loud enough for my speakers to pick that up either, so I believe that the lower the sound waves are the less chance they will also have of even being picked up as "sound" at that point.

yes, well, we can turn things up, but then we turn up the noisefloor as well. There are a lot of things that are hard for me to hear, especially on the laptops speakers, but I can enjoy the music, but then plugging in earphones and listening to something important perhaps binaural(not that binaural, the other binaural) sound or something, then you'll notice the noise, so it needs to be taken out of all recordings before it gets in there in the first place.

I had this too when I wrote my 1 minute test validator.

Here's the Checker code that I ported to do the volume.

The code you point to looks like C, which I've long forgotten, but it looks like C because arduino code is some bastardised version of C I think, and i write some arduino code, or dabble at least. C is a bit much, and I don't think many distros are out of the box c enviroments, because the competetion who put together the distros are really scared of people writing their own code, so it is rare to find some distro ready to go. Not the C64 days, where switch on and programming language comes up. Over time, the whole world turns upside down.

wait a minute, that's java by the url, meh. I don't do java at all. don't hate me. Is java free n opensource? that may explain it.

C and me never really got along, I tried dating her for a while but it was never going to work despite my efforts. Python was invented and is now run by satan, so that's out even though there was that time I don't want to talk about, just change the subject ok? thankyou. (edit, Im not talking about python people, its about the brackets controversy)

Linux bash and I are getting along fabulous. I'm sure whatever I come up with will have a lot to do with her influence.

I'll try with the VLC addons first, because VLC is already installed across all the linux distros which I use freqently. (I'm 100% windows free)

[redrun]

Making an MP3 file so that it has the expected properties is no different than following HTTP protocols to post to a forum. Doing what everyone else does on this level facilitates communication, and doesn't harm or discourage you.

Sorry my answer wasn't to your liking, but I'll leave it in case it helps someone else. It sounds like you've found something that works for you!
I like VLC too. Truly a great Free Software utility!

[GettingTooOld]

Sorry my answer wasn't to your liking, but I'll leave it in case it helps someone else. It sounds like you've found something that works for you!
I like VLC too. Truly a great Free Software utility!

It was to my liking, I seem to be having too much trouble making myself understood, maybe I'm getting too old.

I really did try to make it clear that you had answered my question (and the whole thread) to my satisfaction. For example, here :

I think redrun answers it with ReplayGain.

C'mon, that was almost understandable, or am I really that bad at expressing myself. Ah, it could have used a smilie, and I really should have said thankyou.

btw, Thankyou.

I've been looking into the replaygain, I was reading a page which listed a lot of dependencies, but then later I saw most everything on the page was 14-18 years old, the comments and so on. Hopefully that won't be a problem. I do use old versions of linux. I'm still looking into it. I think as it is a vlc extension, I'll look at the versions of vlc which I have and then see what docs are available. perhaps it's been incorporated somehow over the years...

[GettingTooOld]

As others have said, it's 89 dB according to the ReplayGain standard - those last words are the crucial bit. If you're comparing to the 'chainsaw' or 'jumbo jet' scale then you're not comparing like with like.

ah!

still reading (and after dinner and many hours of study about to fall asleep)
haven't quite found out yet how to do it., Am looking through videolan docs, old ones, and howto install vlc extensions and addons. I've come across one of those wonderfully helpful docs which tells you to simply click a particular button _which_does_not_exist_on_my_version_ . helpful!ergh. giving up and sleep, maybe another day to do battle.

[sjmarky]

Having a little trouble following all this, so if my reply is off base, I apologize.

I don't know why Replay Gain and Checker use a loudness scale different from everyone else, but it is translatable.

90db in Checker equates to just about -23 RMS-1. This is based on the typical 0db (maximum) to -100db (silent) range measured in other analytics. it is also the scale used by Audacity.

So if I analyze audio using the Analyze function in TwistedWave for example, the acceptable range is about -25db to -18db RMS. This is the same standard used by Audible for professional recordings, and is about the same as 86 to 92db in Checker. The noise floor standard would be about -60db RMS. I get similar diagnostics when rendering audio out of Reaper.

So other DAW diagnostics work; you just have to translate the scales.

Again, if I missed the point, I do apologize.

[DACSoft]

...
I don't know why Replay Gain and Checker use a loudness scale different from everyone else, but it is translatable.
...

I can think of one (very good) reason why Replay Gain, Checker, and LV would use decibels (dBs) as an audio volume standard.

When I contribute to LV by recording sections here, I'm glad I do not have to be (nor want to be) a technical expert or programming guru to do what I so much enjoy doing ... reading aloud and making those readings available to the public for free.

I believe that dBs, volume and how it relates to "loudness" or "softness" is a concept it is very easily understood by those of us who are non-technical oriented people. As well as adjusting volume: increase the dBs if you want to increase the loudness; reduce the dBs if you want to make it softer.

Knowing what RMS is (and I don't) will not enhance my level of enjoyment of making audio recordings of books for LV. But for those who desire that increased technical knowledge, they can do whatever translations are necessary.

FWIW,

[GettingTooOld]

Knowing what RMS is (and I don't) will not enhance my level of enjoyment of making audio recordings of books for LV. But for those who desire that increased technical knowledge, they can do whatever translations are necessary.

RMS is, in non tech terms, just a way to average something out. For the electricity coming into your house, it changes constantly unlike a battery which is constant. It goes to +170volts 60 times a second, and -170volts just as often. It only stays there for a tiny moment before swinging back the other way. Averaging out the voltage mathematically in the 'usual' way is not the best way to work out the work the voltage can do, it turns out not to be very accurate, so Root Mean Squared (RMS) is a different way to say 'average' and a different way to calculate the amount of actual work the actual voltage will actually do in real life. So the 170v peak to peak waveform can actualy do the same work as 110 volts DC, so they call it 110v AC (RMS) but leave out the rms when speaking and sometimes the AC part too, just calling it 110, everywhere else it's just called 'silly american voltage' cause they use 220v AC RMS

The same 'strict math average doesn't work in the real world' idea is same for driving speakers, so RMS applies to them as well.

I believe that dBs, volume and how it relates to "loudness" or "softness" is a concept it is very easily understood by those of us who are non-technical oriented people. As well as adjusting volume: increase the dBs if you want to increase the loudness; reduce the dBs if you want to make it softer.

Knowing what RMS is (and I don't) will not enhance my level of enjoyment of making audio recordings of books for LV. But for those who desire that increased technical knowledge, they can do whatever translations are necessary.

true true, personally I think 'turning it up to 11' is an even easier scale than that. db's are not a linear scale and so they still confuse many people. Me included. even with my background I still think they are weird.

[GettingTooOld]

I do apologize.

don't worry at all. at least you didn't start hitting me or ban me. As I get older either people have a harder time understanding me or I them. Seems universal.

90db in Checker equates to just about -23 RMS-1. This is based on the typical 0db (maximum) to -100db (silent) range measured in other analytics. it is also the scale used by Audacity.

So if I analyze audio using the Analyze function in TwistedWave for example, the acceptable range is about -25db to -18db RMS. This is the same standard used by Audible for professional recordings, and is about the same as 86 to 92db in Checker. The noise floor standard would be about -60db RMS. I get similar diagnostics when rendering audio out of Reaper.

So other DAW diagnostics work; you just have to translate the scales.

Here, I'll post some output.
example file (that doesn't change over time like 1 min tests):
https://ia801900.us.archive.org/16/items/baypoems_2005_librivox/bay_02_lawrence_64kb.mp3

command used:

Code: Select all

ffmpeg -i  'bay_02_lawrence_64kb.mp3'   -af astats a.mp3 

output :

[Parsed_astats_0 @ 0x564e68922cc0] Channel: 1
[Parsed_astats_0 @ 0x564e68922cc0] DC offset: -0.000001
[Parsed_astats_0 @ 0x564e68922cc0] Min level: -0.361933
[Parsed_astats_0 @ 0x564e68922cc0] Max level: 0.338709
[Parsed_astats_0 @ 0x564e68922cc0] Min difference: 0.000000
[Parsed_astats_0 @ 0x564e68922cc0] Max difference: 0.135556
[Parsed_astats_0 @ 0x564e68922cc0] Mean difference: 0.005232
[Parsed_astats_0 @ 0x564e68922cc0] RMS difference: 0.010488
[Parsed_astats_0 @ 0x564e68922cc0] Peak level dB: -8.827442
[Parsed_astats_0 @ 0x564e68922cc0] RMS level dB: -26.432712
[Parsed_astats_0 @ 0x564e68922cc0] RMS peak dB: -17.887296
[Parsed_astats_0 @ 0x564e68922cc0] RMS trough dB: -84.893475
[Parsed_astats_0 @ 0x564e68922cc0] Crest factor: 7.590380
[Parsed_astats_0 @ 0x564e68922cc0] Flat factor: 0.000000
[Parsed_astats_0 @ 0x564e68922cc0] Peak count: 2
[Parsed_astats_0 @ 0x564e68922cc0] Bit depth: 32/32
[Parsed_astats_0 @ 0x564e68922cc0] Dynamic range: 191.287960
[Parsed_astats_0 @ 0x564e68922cc0] Zero crossings: 179467
[Parsed_astats_0 @ 0x564e68922cc0] Zero crossings rate: 0.103366
[Parsed_astats_0 @ 0x564e68922cc0] Number of NaNs: 0
[Parsed_astats_0 @ 0x564e68922cc0] Number of Infs: 0
[Parsed_astats_0 @ 0x564e68922cc0] Number of denormals: 0
[Parsed_astats_0 @ 0x564e68922cc0] Overall
[Parsed_astats_0 @ 0x564e68922cc0] DC offset: -0.000001
[Parsed_astats_0 @ 0x564e68922cc0] Min level: -0.361933
[Parsed_astats_0 @ 0x564e68922cc0] Max level: 0.338709
[Parsed_astats_0 @ 0x564e68922cc0] Min difference: 0.000000
[Parsed_astats_0 @ 0x564e68922cc0] Max difference: 0.135556
[Parsed_astats_0 @ 0x564e68922cc0] Mean difference: 0.005232
[Parsed_astats_0 @ 0x564e68922cc0] RMS difference: 0.010488
[Parsed_astats_0 @ 0x564e68922cc0] Peak level dB: -8.827442
[Parsed_astats_0 @ 0x564e68922cc0] RMS level dB: -26.432712
[Parsed_astats_0 @ 0x564e68922cc0] RMS peak dB: -17.887296
[Parsed_astats_0 @ 0x564e68922cc0] RMS trough dB: -84.893475
[Parsed_astats_0 @ 0x564e68922cc0] Flat factor: 0.000000
[Parsed_astats_0 @ 0x564e68922cc0] Peak count: 2.000000
[Parsed_astats_0 @ 0x564e68922cc0] Bit depth: 32/32
[Parsed_astats_0 @ 0x564e68922cc0] Number of samples: 1736236
[Parsed_astats_0 @ 0x564e68922cc0] Number of NaNs: 0.000000
[Parsed_astats_0 @ 0x564e68922cc0] Number of Infs: 0.000000
[Parsed_astats_0 @ 0x564e68922cc0] Number of denormals: 0.000000

example file 2:
https://ia801900.us.archive.org/16/items/baypoems_2005_librivox/bay_14_lawrence_64kb.mp3

command :
same as above

output 2:

[Parsed_astats_0 @ 0x563475d1c7c0] Channel: 1
[Parsed_astats_0 @ 0x563475d1c7c0] DC offset: -0.000001
[Parsed_astats_0 @ 0x563475d1c7c0] Min level: -0.345712
[Parsed_astats_0 @ 0x563475d1c7c0] Max level: 0.324965
[Parsed_astats_0 @ 0x563475d1c7c0] Min difference: 0.000000
[Parsed_astats_0 @ 0x563475d1c7c0] Max difference: 0.181596
[Parsed_astats_0 @ 0x563475d1c7c0] Mean difference: 0.005561
[Parsed_astats_0 @ 0x563475d1c7c0] RMS difference: 0.011276
[Parsed_astats_0 @ 0x563475d1c7c0] Peak level dB: -9.225707
[Parsed_astats_0 @ 0x563475d1c7c0] RMS level dB: -26.332681
[Parsed_astats_0 @ 0x563475d1c7c0] RMS peak dB: -17.323431
[Parsed_astats_0 @ 0x563475d1c7c0] RMS trough dB: -82.592850
[Parsed_astats_0 @ 0x563475d1c7c0] Crest factor: 7.167186
[Parsed_astats_0 @ 0x563475d1c7c0] Flat factor: 0.000000
[Parsed_astats_0 @ 0x563475d1c7c0] Peak count: 2
[Parsed_astats_0 @ 0x563475d1c7c0] Bit depth: 32/32
[Parsed_astats_0 @ 0x563475d1c7c0] Dynamic range: 190.613929
[Parsed_astats_0 @ 0x563475d1c7c0] Zero crossings: 130733
[Parsed_astats_0 @ 0x563475d1c7c0] Zero crossings rate: 0.107795
[Parsed_astats_0 @ 0x563475d1c7c0] Number of NaNs: 0
[Parsed_astats_0 @ 0x563475d1c7c0] Number of Infs: 0
[Parsed_astats_0 @ 0x563475d1c7c0] Number of denormals: 0
[Parsed_astats_0 @ 0x563475d1c7c0] Overall
[Parsed_astats_0 @ 0x563475d1c7c0] DC offset: -0.000001
[Parsed_astats_0 @ 0x563475d1c7c0] Min level: -0.345712
[Parsed_astats_0 @ 0x563475d1c7c0] Max level: 0.324965
[Parsed_astats_0 @ 0x563475d1c7c0] Min difference: 0.000000
[Parsed_astats_0 @ 0x563475d1c7c0] Max difference: 0.181596
[Parsed_astats_0 @ 0x563475d1c7c0] Mean difference: 0.005561
[Parsed_astats_0 @ 0x563475d1c7c0] RMS difference: 0.011276
[Parsed_astats_0 @ 0x563475d1c7c0] Peak level dB: -9.225707
[Parsed_astats_0 @ 0x563475d1c7c0] RMS level dB: -26.332681
[Parsed_astats_0 @ 0x563475d1c7c0] RMS peak dB: -17.323431
[Parsed_astats_0 @ 0x563475d1c7c0] RMS trough dB: -82.592850
[Parsed_astats_0 @ 0x563475d1c7c0] Flat factor: 0.000000
[Parsed_astats_0 @ 0x563475d1c7c0] Peak count: 2.000000
[Parsed_astats_0 @ 0x563475d1c7c0] Bit depth: 32/32
[Parsed_astats_0 @ 0x563475d1c7c0] Number of samples: 1212795
[Parsed_astats_0 @ 0x563475d1c7c0] Number of NaNs: 0.000000
[Parsed_astats_0 @ 0x563475d1c7c0] Number of Infs: 0.000000
[Parsed_astats_0 @ 0x563475d1c7c0] Number of denormals: 0.000000

I put two as one will be louder than the other. Actually I tried to use my ears Vs ffmpeg and eyes(audacity) Vs ffmpeg to tell which of two samples was louder and I don't think it always gets it right.

[GettingTooOld]

I'm glad I do not have to be (nor want to be) a technical expert or programming guru to do what I so much enjoy doing ...

I'm not either when it comes to this sort of stuff, I'm just aiming to find anything that can give a scale, any scale, consistently. Something I already have on my machine or can easily add. Then it will be trivial enough for me to write a script and make automated comparisons with a standard. I can't do C code and python ships me to tears but bash script is ok and still fresh in my bwain from the last few weeks.

So far I have ffmpeg which I use a lot for many other things, so it would be good to use. So far I have not been able to get vlc to add any kind of plugin or extension or work out how to do that for it, but it's early days. ffmpeg seems to have somewhat inconsistent output, so it's perhaps not suitable.

I am aiming for something not too technical see, something that can take a forum index and read out all the newly uploaded 1 minute tests and perhaps go so far as to either preload text to a reply editing window or have the text ready otherwise for cut-n-paste. I like doing automation lately in linux bash scripting files.

Then maybe I can wrap it up into a kind of automated scanner that reads the forum every ten mins or so looking for new files to check. That way if I'm on the computer it can play an audiofile to alert there is some new 1 minutes test to look at and perhaps reply to. That way the me generation can get instant satisfaction to their postings from anyone who does the linux thing perhaps. Something like that.

If you're going to post a hundred, they should post 'emselves. (with human oversight and a human touch)

[annise]

I'm moving this to off-topic, it isn't necessary for readers to know but you are free to discuss it. All LV needs is a file we can use and any advice given is to help people get this result, it's no use us having files that can't be listened to. And especially with group projects, who wants to have to turn the volume up fully for file 1 and then have your ears damaged by the next one?

Anne