How to get mind-blowing 110 dB dynamic range from CD-standard digital audio
Comments on this video
You can comment on this video at YouTube
You can comment on this video at YouTube
Thursday April 20, 2023
David Mellor
David Mellor is CEO and Course Director of Audio Masterclass. David has designed courses in audio education and training since 1986 and is the publisher and principal writer of Adventures In Audio.
Anyway, if I would ever hear the distortion at -96 dB it would be nice to hear, so I would know that I am listening the first lowest significative bit of the 16. I would find it as nice to know, rather than to hear the dithering noise, but this is a matter of taste.
Another reason is that you can't remove dither once you have added it. Somewhere I have read that if you apply it multiple times, it can make the sound worst.
Anyway since 16 bits are too few for the audiophiles and 24 bits are waste of storage space, a middle way format around 19-20 bit could be a modern successor for the CD format, with sampling rate between 48 and 96 kHz I would say.
But for the music I listen, I never exploit the whole limits even of the CD format.
@igorbeuk4068: Diether is last tool for use to fix noise. Invert Polarity Phase on one Audio Channel and lay it perfect with duplicated Audio will make noise cancelation and that's something worth trying to make less noise and killing or removing resonance to some extent can make sound more pleasant in the Baby cry frequency spectrum.
@EddyTeetree: One of the biggest B/S debates in audio is dynamic range. From Yamaha Audio:All music has some degree of level fluctuation, but some genres tend to have broader dynamic ranges than others. Recorded pop, rock, R&B, hip-hop and country music usually have a relatively modest dynamic range — typically around 10 dB, although there are exceptions. Electronic dance music (EDM) probably has the smallest dynamic range — often in the 6 dB neighborhood — but makes up for it by creating contrast with its almost infinite array of instrument colors and textures coming from synthesizers and samplers.
On the other end of the spectrum are jazz and classical music, which can have considerably large differences between their quietest and loudest parts. In jazz, uptempo songs typically go from loud passages played on brass and saxophone instruments to quiet piano and bass solos. Even in jazz ballads, the dynamic range is usually relatively wide. A study of dynamic range in different musical styles conducted in 2016 revealed that dynamic ranges in jazz generally varied from 13 dB to 23 dB.
As a group, classical recordings have the widest dynamic range of any genre. The same study cited above found that recorded classical music typically offers between about 20 dB and 32 dB of dynamic range. While that might seem like a lot, it’s still quite a bit smaller than that of a live symphony orchestra performance, which can be as large as 90 dB.
Therefore analogue recording which has a DR of @ 70db is more than adequate for music reproduction!
@trmus: Thanks for making this free, i'm mixing my own music and i find many of your videos are so useful. Thanks from Chile
@wb5plj: Ug, stop using that fake human please. but thank you for the video.
@colloidalsilverwater15ppm88: Zero decibels signal on cassette tape isn't the same on Yamaha deck comparing it with Pioneer. How can it be, nobody messed around with trimpots in there?
@StephenWorth: Commercially recorded music rarely exceeds 55dB in dynamic range. 96dB is already overkill. The average recording studio has a noise floor of 25dB. Assuming you know how to set levels, I don't see how you'd really need that much dynamic range. You certainly wouldn't need it for playback, which is what a CD is intended for. At normal comfortable listening levels, the difference between 16 bit dithered and not dithered is barely audible, if it's at all audible. At some point, you have to say that the technology is sufficient and move on to more important things.
@ENZEEVIDS: do you have your technical assistants # ?
@therealwolfspidertoo: Sounds like a tape with a very low signal input.
@pzboyz72: But what is the dynamic range of the human ear?
@Lx655321: I understand little, but i love the AI girls 😍
@kurt120032002: well actually the dynamic range is the difference between highest and lowest signal, and in theory yes the lowest signal should be noise. Problem is that higher dynamic range is part of producing immersion, it became a number's race and it is actually detrimental. I recently upgraded my streamer, I excluded all USB from the chain and went with pure glass cable, the end result being a jump in dynamic range, and while the sound is overall better, the system became annoying. If I set the volume to the highest note in the track it is annoyingly quiet, if I set it a little higher I got constant shock, and I am not investing a lot of money (like everyone else) just get shocked while I try to relax and enjoy music, so getting 110 db not sure I want to try it. I would rather have the noise in the track and let the brain filter it out, them stretch the track dynamic range and become annoying.
@spectrelayer: A great video on dithering. Wish you'd covered a little about noise shaping - but good enough. I work in 24-32 bits of resolution - but think that 16 is just fine for playback as long as dithering is used. Nice video!
@TooTallForPony: Although your example was based on digital processing and the impact of quantization noise, there's a similar phenomenon in physiology called stochastic resonance in which the addition of noise can boost our ability to detect low-level analog signals. There's some compelling evidence that the human ability to detect low-intensity sounds (at 0 dB SPL or even lower) intrinsically depends on this stochastic resonance phenomenon.
@timothystockman7533: ... and there is emphasis ...
@AudioMasterclass replies to @timothystockman7533: Pre-emphasis is useful in places. I might cover this at some point. DM
@Wizardofgosz: Coming over from your other video.
This doesn't "prove" you need dither at all because no one is listening to the noise floor and normalizing it.
If you can't hear the difference with dithered vs. truncated in double blind tests with real music examples then you can't hear it.
@jjcale2288: ERASE your assistant! It's annoying!
@studioopinions5870: Betty is pretty, but, she seems to be an Avatar! So, Is She? The reason I ask, is there seems to be anonymous background. Terry
@dancemusicorganisation: See you soon 😄
@GCKelloch: I love your videos and your light-hearted approach, but I don't think it "proves" there's more dynamic range than 96dB in 16bits. It's not possible, but it does show that noise can shape the lowest 2 bits in a way that makes it sound like there's more range to the ear/brain. It's like how adding the harmonic series to a signal can make it sound like there's deeper bass than there actually is, or that beat frequencies make it sound like there's a note that isn't actually there.
@AudioMasterclass replies to @GCKelloch: There isn't any more signal-to-noise ratio than 96 dB, and 'dynamic range' is often used as a synonym for signal-to-noise ratio. As the video shows however, it is possible to hear signal that is lower in level than the noise and many would regard this as the true meaning of dynamic range. DM
@nickwallette6201 replies to @GCKelloch: @@AudioMasterclass It's interesting to think about that definition. If you consider "dynamic range" to be the difference between the loudest signal and the quietest signal, then surely you can't cherry-pick the component of the signal you care about?
Ergo, just because you can still make out remnants of a 220Hz sine below the noise floor, the noise itself can't be ignored when you're considering the level of the quietest signal that the medium can resolve. We've changed the shape of the signal and added broad-band noise to it instead of neatly aligned multiples of its fundamental. But the level of that signal -- whatever that signal ends up being -- is ultimately the same: The volume of the lowest bit of resolution changing its binary state.
Only if you push that noise entirely above the pass band of the lowpass filter would it make a difference, IMO, and even then we're really measuring the resolution of the processing chain -- not the binary data. (And that ends up in a debate about whether the definition of "DAC" includes the output filtering.) Which is probably why it's often stated as "effective" resolution.
@melaniezette886: Intellectually interesting. Today dac can achieve 115 to 120 dB, amps 80 90 dB, speakers 0.1, 3, 10% and more distorsion. 40 dB noise floor in home. And classic albums with 50 dB range...
When I could afford my first CD, what a revolution, I never came back to scratches, clicks, rumble, wow...
Thanks to digital audio creators 🙏
@GCKelloch replies to @melaniezette886: You know that early CD's were neither dithered nor originally recorded at high enough bit rates for the final pressing to utilize all 16bits. I believe the latter issue wasn't addressed until the early 90s, and the former untill much later.
@pedrocols replies to @melaniezette886: @@GCKelloch Still no scratches, clicks, rumble, wow, etc., etc.
@GCKelloch replies to @melaniezette886: @@pedrocols I'm not making a case for vinyl at all, but CD sound has improved over the years.
@jen3800: the robot and robot voice are extremely annoying. the dulcit tones of your own voice suffice. please stop
@AudioMasterclass replies to @jen3800: Sure I'll stop. Kidding. DM
@precisionsoundworksstudio: Loving these vids, David!
@MrPwnageMachine: Cannot watch and listen to that AI voice/animation.. disgusting
@nicc5122: I'd be interested in an analysis of the HDCD format from you. I have a couple of players and 3 DAC's (Cambridge ISODAC and Audio Alchemy) that have this functionality, and a handful of CDs (Mike Oldfield remasters) which are HDCD.
@6643bear: Hi , really great and interesting video. Regards mark
@NewGoldStandard: Amazing.
Great content, as always.
@editingsecrets: I stand corrected. I thought triangle dithering could only add 2 bits worth of additional encoding, giving effective 18 bit range for 108 dB theoretical maximum. If real world measurement is that signal is happening and noise isn't at -110 dB, that would imply another bit's worth of information... from the frequency shaping of the dither perhaps?
@patthewoodboy: you dont. the noise floor in your house wont allow a 110 dB range
@editingsecrets replies to @patthewoodboy: He didn't say it was a good idea to play back this signal through loudspeakers, only that the system could record and reproduce it accurately as an electronic signal.
@eddiewillers1: Betty's voice definitely needs to be a little more euphonious; it doesn't match her appearance.
@MostlyBuicks: So why is the effective dynamic range of music on CD is often only 10 or even 5db, when 20 or 30db is very pleasing and fun.
@sandersonstunes: Why do people care so much about the most quiet of sounds? The music people listen to doesn't come anywhere near these dynamics. This pursuit seems like a waist of effort when you really should be focusing on actual sound design skills.
@editingsecrets replies to @sandersonstunes: Dither affects the lowest used bit. If you have a full scale blast, the least significant bit is at -96 dB. If you are recording to a 12 dB dynamic range standard, then the least significant bit is only at -12 dB down and you can definitely hear what happens there.
@focuspixvideoaudioservices68: Sorry call me stupid but to clarify.. When delivering to a 16bit service in 24bit does the codec they use dither with any grace or is it preferable to pre dither...Also with a 24bit service in mind as well - you should have different masters depending on the service. I guess that's my point...
Thanks if you answer. And none if you don't..
@editingsecrets replies to @focuspixvideoaudioservices68: Ask the service. When they drop the lowest 8 bits, they might just truncate the file to 16 bits (chop off the last 8 and leave the top 16 unchanged), or they might dither the bottom bit or two of the 16 bit output to encode some of the contents from the lower 8 bits. Depending what types of dither your system can produce, the result might be better or worse than what their system would produce. This would be a good conversation for you to have with a technical expert at your service.
@MonguzTea: 90s cd players applied dither in the dac to boost dynamic range. It didnt make them sound better though.
@editingsecrets replies to @MonguzTea: It's only useful if it encodes a combination of noise and signal below the nominal cutoff, in this case from bits 17 and beyond of a greater than 16 bit wide sample. On playback, there's no additional information available to the player other than what's already encoded in the 16 bits.
@Seiskid: What happens if you apply a 220hz pass band filter to your dithered signal. Obviously you can't do that for music. But I imagine it cleans it up considerably. Likewise running the non-dithered signal through the same filter would remove the harmonics. Interesting concepts.
@editingsecrets replies to @Seiskid: The term is noise shaped or frequency shaped dither, it's a real thing to try to push the noisy artifacts of the process out of audible frequency range.
@Kalvinjj replies to @Seiskid: If you add the band pass filter to the square wave distorted 200 Hz you'll also recover a perfect 200 Hz signal as well. Kinda how tuning a radio works.
@casualintrovert207: Ironically, I have my airpods max wirelessly connected to my PC with a usb filter inserted into the chain before the bluetooth receiver, and if I listen with noise cancelling on, I can just barely make out the tone at -90db. I was expecting the bluetooth stream compression along with the youtube codec to pretty much get rid of it, but I could actually hear it, albeit with my headphones at max volume.
@artisan002: This brings up an issue I've been debating lately, and I definitely want your take on it. I constantly find people arguing about ripping CDs to higher bit depth and sample rates. Huge recitations of things people have read. Blah blah, copy-paste Nyquist to get the spelling right. But, opinions change or go silent when one mentions the high frequency data sometimes discovered when resampling. There's an opinion my best friend shares that the new wave data is harmonic distortion having come from the resample. But, peripherally, that would threaten the "math is pure" stance that materializes when resampling arguments come up. Now to the part I'm concerned with: I recently started wondering if that resampling phenomenon isn't creating harmonic distortion, but is instead revealing a particular dithering method/algorithm used on the album.
@AudioMasterclass replies to @artisan002: I would find it hard to believe that ripping a CD to anything other than a 16-bit, 44.1 kHz PCM file would provide any benefit. DM
@artisan002 replies to @artisan002: Right. But, I'm not concerned with questions of benefit.
Rather, it's the matter that there's a change to the high frequency audio content, and I have yet to find answers on why or how this is happening. The fact that it's inconsistent is all the more fascinating to me. (I'm ignoring older lossy formats with this, btw, and thinking more about FLAC and .wav.
But, yeah.) I've seen it happen a fair number of times. When it does turn up, it's usually inaudible, but shows up via spectrogram/spectrograph.
Obviously it's largely irrelevant at a common user level if it's typically inaudible. But, it doesn't change the fact that it's happened, and I'm just intensely curious to know how this happens. In 10 years of noticing this, nobody ever seems to have an answer.
@editingsecrets replies to @artisan002: For a perfect clone, write one output bit exactly for every one input bit. 16 bit linear 44.1 kHz PCM in, 16 bit linear 44.1 kHz PCM out. If you want to change the output to have different qualities than the input, such as pitch shifting or equalizing it, processing it at higher resolution will help reduce artifacts from the processing step. But this is internal to the processing step only. It has no more source information to work with than is already present in 16 bit linear PCM words at 44.1 kHz.
@editingsecrets replies to @artisan002: @@artisan002 The CD player might have some small random jitter, when the reading laser doesn't perfectly catch every bit at exactly the right time, or the output buffer clock doesn't always increment to read out the sample at exactly the right number of nanoseconds between words. If you're using a personal computer to measure a live audio stream, then occasionally an operating system interrupt or hardware timing glitch could make this happen.
@editingsecrets replies to @artisan002: True for a realtime capture. Audio pros usually tweak a system so as little is possible might block kernel audio processing. If a block at a time is read from the CD into a buffer, then written from the buffer to a drive, and that doesn't have to be realtime, a stray overtime interrupt will delay appending that block to the output file. But the file will be a perfect clone of consecutive samples of the source. On the input side, the CD format's error correction algorithm should be able to use the checksums and redundancies in what's on the disc to recover from an occasional misread pit.
@ScotPeacock: Was the dither in this example simple dither? If so, I would have liked to see the results with noise-shaping. Apparently, with noise shaping you can get something like 120db dynamic range with 16bit.
@AudioMasterclass replies to @ScotPeacock: The simplest dither Pro Tools has to offer. I felt that anything other would overcomplicate this demo but I'll consider it for a future video. DM
@josephkosak1675: I don't even know if doing that makes sense. If a device's amplifier doesn't have a signal to noise ratio of at least a few dB more than that, the quiet parts will be masked by the amp's background noise. Also, I wonder what the dynamic range of an average person's hearing.
Hmmmmm...
@editingsecrets replies to @josephkosak1675: You're right that the weakest link in the chain limits the whole system. Amplifiers can easily have noise and distortion levels WAY better than most recording formats. The recording is usually the weak link.
For most people the threshold of discomfort to pain is around 100 to 120 dB SPL, permanent major hearing damage above that. It's very rare to find a room with such sturdy construction or such a quiet setting that the background noise inside is less than about 30 dB SPL. Eddie Van Halen and some other producers have remarked on easily noticing a less than 1 dB difference in mix levels.
Unless you live out in the country or have massive walls, playback at over 100 dB SPL inside, in addition to being painful pressure for many people, might be enough to get a visit from the cops. So around 70 dB dynamic range is actually about as much as actually useful at home, which hi-fi vinyl through a good system can provide. Studio multitrack recorders, 2" 24 track, can provide that for the raw tracks. 16 bits gives some extra headroom when tracks are combined, the design goal intended to be able to accurately capture a full symphony orchestra blasting away without distortion.
@josephkosak1675 replies to @josephkosak1675: I'm surprised EVH could say that, unless it was early on in his career. As time went on, I wouls say their onstage SPL would have destroyed at least some of his hearing.
@nononsenseBennett: I love that Karen!
@editingsecrets replies to @nononsenseBennett: She's going to ask to speak to the manager in a little bit.
@julesc8054: Question: in a 16bit recording he says 6db/bit in 24bit audio are is this smaller?
Are there more steps per decibel?
@AudioMasterclass replies to @julesc8054: 6 dB per bit both in 16 and 24, approximately but near enough. But then there will be other sources of noise. DM
@julesc8054 replies to @julesc8054: So the advantage is having the system playback at higher bit depths to accommodate volume changes of the output rather than having the files at 24bit to accommodate dynamic range within a normalised file.
I’m thinking both in the DAW control room volume as well as the average listener with digital volume on their listening device or hifi.
Thank you for this its been missing from my understanding for some decades. I honestly thought there were more steps per db. lol
@ToddSauve replies to @julesc8054: @@julesc8054 The problem is that so much of this is theoretical and not actually usable for your average listener. Dynamic range in a recording is wonderful if you are the only listener in a soundproof room. And if your amplifier can actually supply the incredible amount of power to your speakers that reproducing the 24bit recording will require. A robust recording utilizing all 24 bits of dynamic range will demand thousands of watts! This is a practical impossibility for more reasons than one. And who wants to be the neighbour of someone who has a stereo like that? Not me! 🤷♂
@julesc8054 replies to @julesc8054: @@ToddSauve thanks this makes sense. So many thinks don't though.
I don't think I'm understanding this too well. There are 16,777,216 in 24 bit in and in 16 bit only 65,536 steps. In 16 bit there are 68 steps per db and in 24 bit there are 116 508 steps per db if the db scale was linear. If I the last 96db (-96 to 0 db) only uses 65 536 steps that means that the first 48 db (-144 to - 97db)uses 16 711 680 steps. Firstly this is inverse to the exponential scale as the louder we get the more steps per db we should need as the graph curves up on the db scale.
I also don't grasp sample rates. At 48khz Niquest states all frequencies can be recreated perfectly up to 24khz. Do all the frequencies need to start at a zero crossing to be recreated correctly? What if the timing of sub harmonics do not line up with what is allowed in Niquest therom?
@ToddSauve replies to @julesc8054: @@julesc8054 Math like that is not my strong point either Jules.
@maidsandmuses: You are only looking at 16 bit PCM correct? Whilst not exactly the same, I think a similar theory underlies the workings of 1-bit DSD. There you get a high dynamic range out of only a 1 bit signal but at a very high pulse-modulation frequency, effectively integrated over time with noise shaping. This works because the binary pulse train frequency is so very much higher than the highest audio frequency that needs to be reproduced.
I suspect in this demo the 220Hz can still be reproduced to some extent at the low level here as the dithering used can employ the 44.1kHz sample frequency, effectively doing a similar job to that of the binary pulse modulation used in 1 bit DSD. Thus the noise dithering can be shaped such that it has an integrated residual of the 220Hz sine wave buried in the noise.
However, I would bet that trying the same with a 10kHz wave wouldn't work so well as it is getting so much nearer the 44.1kHz digital quantisation noise frequency (ignoring oversampling); then filtering out the 44.1kHz quantisation noise from the 44.1kHz dithering/modulation that tries to integrate to a 10kHz wave becomes nigh impossible. Thus this increased dynamic range for 16bit 44.1kHz PCM would only work for low frequencies like the 220Hz used here, you couldn't do it to the same extent for much higher frequencies. Am I correct?
@editingsecrets replies to @maidsandmuses: "There you get a high dynamic range out of only a 1 bit signal but at a very high pulse-modulation frequency, effectively integrated over time with noise shaping. " Exactly!
@melaniezette886 replies to @maidsandmuses: When I look at frequency response graphs, Dsd noise is closer to 20Khz than PCM. Some argue that it's the reason why some prefer dsd sound. I have an sacd player and frankly I've never been able to tell any difference
@editingsecrets replies to @maidsandmuses: @@melaniezette886 DSD is just the raw oversampled bitstream without it being binned into 16 bit Linear Pulse Code Modulation words. With only one bit measurement, theoretically every sample could be 50% quantization error noise relative to the signal, as opposed to 16 bit theoretical quantization noise down at -96 dB, 1/64-thousandth of the signal. But with that quantization noise at 16 times the frequency of CD, theoretically the fuzz is at such extremely high frequency that it either won't get passed through the electronics, won't get pushed out of the speakers or headphones into the air, or won't be perceivable by humans.
As you mention, in practice most people can't hear any of the theoretical difference. As the math is harder than with LPCM and requires different algorithms throughout the mixing process, it never really caught on in either production or consumer levels beyond a tiny number of audiophile enthusiasts.
@morbidmanmusic: Your assistant may need to see a doctor. She shows signs of hypertension
@editingsecrets replies to @morbidmanmusic: It's because her boss keeps her dithering just a bit
@AllenCavedo: I loved the “See you soon” from the fake gal.
@Konspiration100: "Wikipedia the source of all knowledge that is righteous and true" I hope this comment was ironic cause nowadays wikipedia is quite far away from that.
@markitoxi: Anyone knows which piece of software/web page is he using to bring virtual AI assistant Betty ?
@markfischer3626: I have a collection of over 3000 CDs mostly classical music, some with exceptionally wide dynamic range. I have yet to find a single where any part challenged the dynamic range of a CD. Here is a disc that is a good test for dynamic range capability. The Disney Soundtrack Pirates of the Caribbean Dead Man's Chest. It includes a symphony orchestra, a pipe organ, and a men's chorus all at the same time. Try first the very soft opening of Track 1 and then track 2 The Kracken. This will test not only the limits of these musical instruments but the dynamic range and frequency limits of a sound system.
@editingsecrets replies to @markfischer3626: Not sure if it's true but I read (don't have citation for this) that CD format designers used a recording of Mahler's 9th Symphony as the dynamic range reference.
@GumonX: Assistan is AI generated? 😔
@AudioMasterclass replies to @GumonX: Yes. In the future, I will be too. DM
@GumonX replies to @GumonX: @@AudioMasterclass indeed it is, a sad future were everything is fake, sorry for being an antiquated guy! And thank you for the awesome video…
@AudioMasterclass replies to @GumonX: @@GumonX The scary part is that these kinds of things will look 100% real within a couple of years or so. Maybe they are already... DM
@eddiecucumber5342: The redook dynamic range is all very well and good, but unfortunately most music commercial music is compressed to between 10 - 15DB dynamic range. The same goes for SACD. This is a commercial mastering practice. So dynamic range on media is a moot point in the real world.
@ToddSauve replies to @eddiecucumber5342: I have found that the mix from the studio that goes to the mastering company will often sound quite a bit better than what the mastering company can do. The purpose of mastering, it seems to me, is to provide an acceptable listening experience for those who don't really have a good stereo system.
@editingsecrets replies to @eddiecucumber5342: @@ToddSauve Also to make minor changes to blend the mixes into a consistent sound across an album, and perhaps major changes to modify that blend to conform to popular trends or the demands of executives.
@ToddSauve replies to @eddiecucumber5342: @@editingsecrets From the mixes I have heard that come straight from the studio, they sound much better than the mastering job. Particularly in the lower frequencies, but all across the hearable spectrum. Instruments are much clearer. For instance, a bass guitar will sound very real to the point where you car hear the plucking of the individual string. Once it comes out of the mastering process this is much diminished. 🤷♂🤨
@editingsecrets replies to @eddiecucumber5342: @@ToddSauve "You want the truth? You want the truth? You can't handle the truth! I have a greater responsibility than you can possibly fathom. You have the luxury of not knowing what I know. That my distortion's existence, while grotesque and incomprehensible to you, sells songs. You don't want the truth because deep down in places you don't talk about at parties, you want me on that brick wall limiter... you need me on that brick wall limiter. Hell yes I ordered the V.U. Red!" - Col. Mastering Engineer Jessep, "A Few Good Media"
@ToddSauve replies to @eddiecucumber5342: @@editingsecrets I do hope I have not offended you in some way! I did not intend to disparage you. I was only making a personal observation on the clarity of mixes straight from the studio compared to what normally comes out of the mastering process. I also realize that mastering must generally serve the lowest common denominator, and be done to still sound acceptable on boom boxes and AM radio.
@6643bear: Very interesting and informative video, I understand dynamic range on comms and radio equipment. Regards mark
@RocknRollkat: Interesting presentation, I've done these tests also, without the dither.
I personally prefer the term 'usable' or 'listenable' when discussing S/N ratios or any other audio phenomenon.
A sine wave buried in white noise is actually more annoying than the square wave of 1 bit sampling.
Maybe.
Both are pretty unusable or otherwise unlistenable.
Also, at a pleasantly loud 80 dB SPL listening environment, a signal at 2 bits quantization (-84 dBs). is inaudible.
But being 1/4 Welsh and 75 years old, these tests make perfect sense to me.
All the best,
Bill P.
@gblargg replies to @RocknRollkat: You wouldn't normally have signals that low. The dither ensures that any which get close to this region don't become distorted into awful square waves with their extra harmonics over the original wave.
@RocknRollkat replies to @RocknRollkat: @@gblargg Correct.
At those low levels, room tone is usually louder, making those signals inaudible.
Headphones are another issue.
Even so, a 110 dB 'down' is still pretty darn quiet.
@nickwallette6201 replies to @RocknRollkat: @@RocknRollkat I would agree. This is almost entirely an academic debate. Am I really worried about the harmonics of a signal whose fundamental is about 6-10dB below the point I'm going to be able to hear it under the most critical listening scenario? Who even cares? :-) Assuming the first harmonic is 3dB below the fundamental, it would be dwarfed by an ant farting.
@RocknRollkat replies to @RocknRollkat: @@nickwallette6201 Nick, if I understand you correctly, any signal that is 6 to 10 dB quieter than your critical 'quiet' level, then the signal is pretty much inaudible, harmonics or no harmonics.
As an aside, the first harmonic is the fundamental frequency, the second harmonic is the first multiple of the fundamental, or first harmonic.
It confuses me too.
Bill P.
@nickwallette6201 replies to @RocknRollkat: @@RocknRollkat HUH. Well... today I learned something new, and I stand corrected. Thank you!
@jtaylor0727: I really like Betty, I have no idea what we are talking about or the relevance of this noise but I enjoy your delivery!!
@melaniezette886 replies to @jtaylor0727: 😃
@georgeogrady449: Better noise free
@morbidmanmusic replies to @georgeogrady449: Noise is your friend. Truth.
@georgeogrady449 replies to @georgeogrady449: @@morbidmanmusic thank you sir
@georgeogrady449: 18k tweeter 20k no good don't work even in 31 band
@georgeogrady449: Perfact sound 320 bits audio and in peak
@sherrillshaffer579: I notice that the distorted 200Hz tone is much louder than the dithered tone when both are boosted by the same amount (doubtless because the dither noise contributes its own acoustic power to the total, in the dithered case). So, although the original un-boosted tone was at -90dB, the tone that remains in the dithered case must be softer than -90dB, implying that dithering not only removes quantization distortion but also reduces the level of the remaining signal below its original level - thus introducing a type of dynamic nonlinearity. More tradeoffs... I haven't seen this discussed anywhere.
By the way, what happens if you apply noise reduction in your DAW to the dithered sample? I suppose, if you had boosted the dithered soft tone back to a normal level before noise reduction, you could recover a clean tone.
@AudioMasterclass replies to @sherrillshaffer579: Yes this is interesting. The peaks levels are the same but if I measured the subjective levels in LUFS I imagine I'd get something different and I might try that. As for noise reduction in these very low levels, maybe it will work but I'd more likely suspect some new kind of hell that we would be better offer staying out of. DM
@iainisbald replies to @sherrillshaffer579: I think the issue is with the normalisation. Adding dithering probably adds some peaks higher than the sine wave peaks. This means that the original level is not restored by normalisation.
@Kalvinjj replies to @sherrillshaffer579: The distorted tone reaching the same amplitude as the original 200 Hz tone will have a TON of square wave harmonics, those all adding up to the sound, hence the perceived added loudness.
Now, I'm too lazy to calculate (integrate the waves and compare) but I would also assume that the distorted one also does indeed carry more energy on it by being at the same maximum amplitude.
@gblargg replies to @sherrillshaffer579: @@Kalvinjj Yes, the distorted version that's a square wave is louder than the sine wave should have been. The dithered version is the correct loudness. Dithering preserves linearity of volume.
@nickwallette6201 replies to @sherrillshaffer579: I suspect ianinisbald is probably on the right track here... The square wave can be normalized to 100% where peak and avg are about the same. If the signal-in-the-noise of the dithered version is still retaining anything of its original shape, its average level will be below peak level ... since it's not a square wave and all. :-)
@brianobrian9334: This stumbled across my feed and you so remind me of Paul McCartney, like his cousin or something or uncle
@AudioMasterclass replies to @brianobrian9334: I just wish I had his talent. DM
@artyfhartie2269: The best sounding cds are the ones marked AAD meaning they were produced using an analogue source, edited and mastered using analogue technology and put on cds as the final digital medium. Loud does not mean better. Huge dynamic range does not mean better. The best music reproduction is by using tape decks and reel to reel or cassette tapes. The equipment must be cleaned and aligned meaning the heads must be cleaned using isopropyl alcohol with q tips demagnetized amd lubed and the heads aligned. Not for lazy or people with preconceived ideas. The sound is like the sound of a live studio, music club or concert hall played using good well kept tapes.
@AudioMasterclass replies to @artyfhartie2269: Back in the 1980s, DDD was meant to be the best, ADD second, AAD worst. DAD would have been possible but an odd thing to do at the time. But, as you say, AAD might not be such a bad thing after all. Indeed, some people today listen to nothing but AAA. DM
@artyfhartie2269 replies to @artyfhartie2269: @@AudioMasterclass DDD sounds too sharp, artificial for my taste. Like the audio signals been purged of all ambience and natural sounds that human ears are geared to hear. The process of turning audio signals into binary codes, compress them, then go through clocking and jitter correction and conversion to analogue sounds too like computer generated sound to me. I do like the convenience. Digital video is very good though
@artyfhartie2269 replies to @artyfhartie2269: @MF Nickster Listen buddy, I dont listen to music with your ears. I dont like people pontificating to me. Piss off.
@morbidmanmusic replies to @artyfhartie2269: No. Tapes are almost as bad as vinyl. Dynamic range does mean better. Life is dynamic. If I want compression I. Will do it,.. not the tape and it's randomneess. I did 4 decades of tape. I know all about it. You're wrong.
@artyfhartie2269 replies to @artyfhartie2269: @@morbidmanmusic I don't like fools
@eaustin2006: Was there a point?
@AudioMasterclass replies to @eaustin2006: Yes. Use dither if you're mastering to 16 bits. DM
@editingsecrets replies to @eaustin2006: There was, but it was 100 dB down so nobody could hear what it was.
@tomekichiyamamoto2177: So, it becomes a digital soup.
Nice 👍
@marleypumpkin4917: Betty… I love you ❤
@adammachin: For me it didn’t sound any louder than the dialog.
@jeffchristian6798: Love your assistant, very cool.
@walthaus: I dunno, sounds to me like you've moved the goalposts a bit here. If a system's noisefloor, being louder than the actual signal is not considered to be the lower end of the signal-to-noise range because the signal is still audible in some form or another, then it follows that the S2N ratio of cassette, tape, vinyl or wire recordings is much larger than previously assumed, because somewhere in that noise there is signal that can possibly be heard, if it's the right kind of envelope, attack, pitch or whatnot. I'm not sure I'll buy that, but to each his own.
@AudioMasterclass replies to @walthaus: Your comment is correct if you substitute 'dynamic range' for 'signal-to-noise ratio', using dynamic range in the sense of including signal that is lower in level than the noise. So yes, even a wire recorder will have audible signal that is lower in level than the noise. DM
@editingsecrets replies to @walthaus: Eventually the signal level is so small that it can't magnetize distinct clumps of rust or individual spots on the wire. Analog recording actually is digitized at the microscopic level: a magnetic domain is magnetized, or it's not. But as manufacturing processes don't spread magnetic particles perfectly evenly, every millimeter has a slightly different effective sample rate. I don't have the reference handy right now but I seem to recall a paper explaining there are about a thousand domains per millimeter. Multiply that by tape speed and you get the equivalent sampling rate, then by track width and you get the equivalent bit depth.
@mwdiers: So this is really the whole point of dither, and it does not, under any circumstance, increase the dynamic range of your material, but it does prevent quantization errors. You cannot get a 110db range out of 16-bit in any circumstance. That's simply impossible. The reason there is any discernible signal in the dither at all is because you have retained that -110db signal in the 32-bit float internal processing of your DAW, and when that signal is converted to 16-bit with dithering, the signal is quantized into the dither noise. The result is still a 16-bit file with 96db of dynamic range, with the formerly -110db signal quantized into the lowest 4 bits of the dither-induced noise floor. Because of dither stochastically smoothes out the quantization errors that would otherwise result, the signal is still detectable. But the resulting file is still limited to 96db. In fact, because of the dither, it's actually closer to 84db of dynamic range. The dither eats up some of that range.
@AudioMasterclass replies to @mwdiers: Respecting your point of view, I believe I covered this very early in the video. Wikipedia gives two definitions of dynamic range, one down to the noise floor, the other down to what is discernible as signal below the noise floor. I'm happy with either definition, preferably stating which one. Of course, perhaps a higher authority - AES, EBU, BBC whatever - may have a more definitive definition, if that's not a tautology. Whether one prefers distortion or noise, I guess that's a personal choice. DM
@Jacob-hl6sn replies to @mwdiers: My understanding is that you can get far more than 16 bits of dynamic range out of 16 bit audio with noise shaping. Specifically in audible frequencies.
@gblargg replies to @mwdiers: I think this is a good point. Even signals louder than his example will have some of that square-wave type distortion. The dither eliminates that from creeping in more and more as signals get closer to the noise floor.
@gblargg replies to @mwdiers: @@mfnickster9754 I wonder how much all of this frustrates audio compression algorithms. They'd much prefer the original 24-bit recording without dither.
@AudioMasterclass replies to @mwdiers: @@gblargg Apple specifies exactly this. 24-bit, no dither. DM
@EgoShredder: I think she knows more than she is letting on. Just sayin' 😉
@andrewbrazier9664 replies to @EgoShredder: 🤣
@alanm.thornton4055: 🤯
@Chunksville: We are talking of an audible sound even at 96db that cannot be heard unless you dare to turn up your amp way beyond usable listening levels, to over complicate it your analogue circuitry chain and components connected will have more noise being created than any digitally converted sound/noise anyway
@nickwallette6201 replies to @Chunksville: @@GladeSwope Oh yeah. Particularly with all the love for tape, and tube mics, and vintage ribbon mics with 60dB of gain applied to them, etc. Nothing in the input signal chain has 100dB of resolution. The only way we can push that envelope is to manufacture and manipulate signals in the digital domain. Digital gain, EQ and modulation, synthesis, and so on. Anyone mixing "out of the box" and bouncing to 2-track tape to create the master need not worry about what's happening at the noise floor of CD. They're well, well above that.
@DrBroncanuus: I didn't understand a word but would like to meet a real life Betty...
@editingsecrets replies to @DrBroncanuus: If you get to know her more closely, you realize that at a low level she's really a bit square
@DrBroncanuus replies to @DrBroncanuus: @@editingsecrets so she has the Right Angle on things ?
@editingsecrets replies to @DrBroncanuus: @@DrBroncanuus Some says she's obtuse but David thinks she's acute.
@DrBroncanuus replies to @DrBroncanuus: @@editingsecrets hopefully Betty , will help me get over 7 of 9...
@editingsecrets replies to @DrBroncanuus: @@DrBroncanuus She's kind of busy right now with all that pressure to come up with a new catchphrase
@CitizenOfEverywhere: So, dither is removing the distortion on something I can't hear anyway? Ok, I appreciate this example is not representative of "normal" content, but now I'm going to have to set up test - how audible is the improvement made by dithering on regular content? I suspect I won't be able to tell the difference.
@zbyszekolko3998 replies to @CitizenOfEverywhere: Undithered tails auditioned loud sound like a torn paper. Dither is used along with noiseshaping what further improves listening experience.
@editingsecrets replies to @CitizenOfEverywhere: Dither affects whether the lowest level bit that is used sounds clean or distorted. If you record a full scale 16 bit signal, the lowest level bit that is used is at -96 dB and you're right that you probably won't hear it. If you have an 18 dB dynamic range, fairly wide for modern recording of an expressive singer or trumpet player, you're only using the top 3 bits of the format and it's all zeroes for the remaining 13 bits. With the least significant USED bit at only -18 dB, you probably can hear a difference if that drops off smoothly or with quantization noise distortion.
@basspig: I did manage to get 86 decibels of dynamic range squeezed into a 16-bit audio on a Blu-ray disc. It's my ultimate fireworks Blu-ray which is a fully uncompressed natural sound recording of fireworks from the launch location we had to have special access passes and sign in Insurance waiver with all our recording crew but we made one heck of a recording. Only 1% of the sound systems in the world can play it effectively.
@carminedambrosio7 replies to @basspig: The remaining 99% of sound systems will blow up their speakers !😁
@editingsecrets replies to @basspig: @@carminedambrosio7 Fireworks show included for free with the audio track!
@nickwallette6201 replies to @basspig: Oh cool, you also get a sense of the recorded smell then, too!
@basspig replies to @basspig: @@carminedambrosio7 On a moderate power system, the explosions sound like static ticks or pops.
@basspig: Sure you can just use a dbx4bx in the signal chain after the CD player.
@martineyles: Is the noise shaped into a particular frequency band, or is it plain white noise. Does that make any difference?
@editingsecrets replies to @martineyles: It does, frequency shaped dither is a thing.
@SwishaMane420: Now put an entire track below the noise floor, then use AI to remove the noise, then normalize.
@gblargg replies to @SwishaMane420: A whole new level of weird audio compression artifacts. I remember when mp3 came out decades ago and bandwidth was low so things were over-compressed.
@TheYuhasz01: Thanks. Chesky records produced test cd in 1994(volume 3 in test series) showing impact on positive sound quality of using dither as well.