32 float extremely useful for my pocket recorder. No need to worry about gain. There isnt even a knob. rain storm from the first soft initial patter to the loudest thunder clap. one file, no clipping, no riding gain.
this is a good use case for float. bear in mind though that the mics, convertors and preamps all have dynamic range limits, and if the upper limit were calibrated to 0dbFS for example, its unlikely the performance would be worse with dithered 24bit
Yes, that is definitely the best use case. Live recordings in areas where you can not control other sounds like if you are making interviews in the street with some one that you will only get one chance to do - think news reporter of CEOs on the run. Then you can save a take that clips.
Coming from an EE, audio engineer, musician, and composer . . . you nailed it. Nice job, I didn't know how the 32 float was implemented, using a portion of the bit pattern for offsetting, neat. I won't be using that anymore. Although I have mostly been in the 24 bit arena. Thanks, good luck with the channel.
Comments like this are evidence of the quality of the information on this channel. I'm working through Ethan Winer's book (found through his interview on this channel) to understand this more thoroughly. I hope your subscriber and view counts correlate with the factualness of your content on a long enough timeline.
Wow, I was just discussing this at work this afternoon, and I came home to see youtube recommending this video! It's nice having a video that goes into such technical detail, but it's more personal than a filmed lecture and has some useful images. It looks like a brand new channel, so I wish you the best and look forward to your future content!
Nice one, glad you got value from the video. Yep, my channel and entire social media presence is brand new, so let's see if the channel is able to get anywhere!
Thanks! I've just got a few subs because I only uploaded my first video just a few days ago. Hopefully my channel will grow. Hope you come along for the ride!
As a hearing care professional/audio enthusiast/hobby artist this is my first encounter with a realistic engineer. I couldn’t agree more. You are absolutely right about the stapedius muscle protecting our ears and to be honest: 120 dB is mostly the limit of our ears and as we get older we lose dynamics as well as frequencies in our natural ears. I’m not interested basically in recording above 16 bit… I record partially on a Tascam that allows 24 bit and other stuff in my PC that would allow 32 bit, but I never use it. Why? Because I basically can’t hear the actual difference and besides: I like to make music and with loudness wars and all I think back at my childhood days where I was perfectly happy with my limited tapes and LPs. In fact: as a kid of 9 or 10 I would only listen to a transistor radio with a monophonic earbud with very limited frequency range… but loved the music nonetheless. If you listen to the music from Gary Numan or Jean Michel Jarre from recent years compared to their late 70s or early 80s output you will undoubtedly notice a massive difference in mastering. Cleverly done: I’ll grant it immediately, but is it equally interesting from a musical perspective? Personal taste left aside, but somehow the music was a lot warmer in those days. Natural dynamics and I loved the (sometimes hopelessly out of phase) stereo from back then on my headphones. I just loved it.True, it’s an age thing as well. But the nonsense that’s going on these days ruins the creation of excellent music, which should come first. Unless you’re able to find creativity in all these new ideas of compression techniques and use extra headroom wisely, it won’t make the music necessarily better.
I agree, there is a difference between music and sound. A good song stay’s a good song even if the sound quality is bad, a mediocre song doesn’t get better with super high quality. Music is more a feeling than just sonic perfection. I wish a lot of musicians and engineers would embrace that part more.
32 bits floating point is great for the internal maths of DAWs and plugins, exactly because you can't clip it, and you won't by mistake add noise. But for storage it's just 33% larger for no benefit. 😀
Doesn't matter, the analog bits including the A/D has a noise floor and a max level that clips. So you still need to get the level as close to 0db but not over. What comes out of the A/D will be fixed point (like 24bits) and storage format will make no difference at all. Store it as 32bit or 32 bit floating point, it will just add eight zeroes to that data. So your files will be 33% larger. But with today's storage solutions, yeah, that's not a problem, you have basically infinite storage anyway. 😀
I always export in 24bits because I'm usually making samples and most plugins won't load a 32 bit sample. Omnisphere, for example, won't load a 32 bit wav file it needs to be 24 or less.
32bit float makes a lot of sense for many recording scenarios, especially where sudden peaks are hard to control or predict. Particularly true for production dialog recording. Some dual system audio field recorders used stacked ADCs to allow for extended headroom.
Reminds me of the AVGN sketch where he says "64 bits, 32 bits, 16 bits..." to the point where there are no bits left, but he keeps going and the quality gets worse. And the really funny thing is that as the quality gets worse, he says it more and more aggressively. It reminds me of how, on the one hand, people can use the most expensive converters, but they're actually mastering their music down to -3 LUFS, using a lot of clipping, and not even using oversampling.
yeah, some people have expensive gear and no skills but I think they are outnumbered by the kids with poverty setups making hits. I know of someone who makes all his music exclusively listening through his built in Macbook speakers
For mixing and mastering this is true, but for sounddesign it isnt. Low bitdepth equals more digital distortion. Just how low bitrate causes more digital aliasing. When you significantly mangle a sound it starts to matter how true the dynamic representation is, cause even if the noise floor introduced by the bitdepth might currently be well below hearing range it definetly wont be after you significantly distorted it. Obviously this can also be used a wanted effect, but it does end up making an audible difference.
sure, when you are doing extreme stuff in sound design, like boosting 80db or something, of course you need more bits. I think I actually mention this later in the video, can't remember. but I agree
What you say is true for saving/storing audio files, but during processing things are a bit different. You can not talk about bit depth without involving sampling rate. All audio adc and most dac use a 1 bit delta sigma front end oversampled into the megahertz. It is then downsampled to 48000 using some clever maths to push the noise out of band. The microphone in your cell uses a 1 bit interface to the CPU, this solves crosstalk noise. Things like ANC headphones will directly process the 1 bit stream directly to reduce latency for noise cancellation. As far as audio in a daw, filtering is easier to implement in floating point, fixed point implementations, especially low pass or peaky ones, can be problematic in 32 bits.
You are right that sample rate and bit depth have an important relationship. In fact, dan lavry writes about that in the paper I mention in this video... I also mention DAW internal mixing bit depth is often 64bit float around 10m into the video but most people probably dipped by then tbf. In terms of 1bit... This is approaching the edge of my technical knowledge on this kind of gear, as I'm just a mastering engineer and not an electronics engineer but chips like the AK5394A used in benchmark ADCs use multi bit technology and not just 1 bit but this is getting into the weeds. The video was for the most part aimed at producers wanting to know what bit depth to bounce at ha ha.
In my daw (bitwig) i can set a bit rate for recording and a bit rate for bouncing audio. Contently i have set 32 bit floating for recording and 24 bit for bouncing audio. The reason because i use a modular and i want to have more headroom when recording it. What do you think about this settings?
Thank you for the great videos, AP Mastering! People who advocate for 32-bit have one main argument: they believe that with 32-bit, a distorted signal can be restored (although this is an amateur mistake!). So, is it possible to restore a distorted signal in a 24-bit environment using 24-bit?
its not about the number, its about whether its float or not and how it was distorted. if you run it through a distortion unit and then bounce it in float, you cannot restore it. if you exceed 0dbFS and then bounce as float, you can restore it. Theoretically in "24bit float" you could restore it too but 24bit float for all intents are purposes doesnt exist in audio and would sound terrible even if it did exist.
tbh, I don't know a lot about DSD. This isnt something that i really had a lot of contact with as a mastering engineer. I might still make the sample rate video but im not 100% sure if it will be boring / overly niche 🤔.
@@APMastering There is a lot of similar snake oil and arguments in the audiophile circle also. I'm looking forward for your content. Have a good one sir.
32 bit is only useful when recording. You just don't have to worry about clipping. Also, as I understand, you can record very quiet signals with the noise floor not being a problem.
What they do now in portable recorders now is combine the output of multiple ADCs to achieve the crazy numbers of dynamic range. I think the customer might fel better if 0 is around normal level rather than having a massive "wasted" headroom in integer.
@@APMastering 32 bit portable recorders have two or more parallel preamps with different gain levels as well as ADCs, after that outputs of ADCs are combined in DSP to high dynamic range/low noise floor signal and saved as 32 bit file.
Great quality Audio converters need 3~4 bit for self noise (every hardware device has self noise embedded into the converted signal, usually inaudible in typical listening scenarios where you would have average full mix signal not exceeding 85~90dB SPL, sound pressure level as a reference to lets say digital clipping at 0dB) in the signal so for useful Music (audio) you would technically use not more than 20 bits. In most consumer market units that would be 18 bits of those 24 bits. Floating point allows for internal (digital realm, which we can not hear, although it can be visualised if needed on screen, but not all Peak\RMS\LUFS meters get to display it really) "headroom" spread of all kind of digital audio processing. It is useful to get detailed reverb tails, low level dynamics acoustic audio (if recorded properly with high sens. mics, ADC\DAC and such equipment).
The floating point shifts with every sample in increments of 6 db. You get something like 31 useful bits if the range over 1.0 is not used. That is if you actually want to turn something down really low and steganographically hide it. Adaptivity to the level is part of how data reduction algorithms like mp3 achieve their performance, but there the sliding window affects a longer time segment.
I'm a ME and I always advise plain old 24bit. In my experience, most people who mix too hot do so with compressors and limiters and the "overcooking" which happens is not normally clipping. Most people know not to clip but you get slammed mixes in various ways that do not straight clip but are nonetheless slammed. Here 32bit doesn't save the mix because the artists aren't clipping. Maybe 1 in 1000 mixes I get is actually just straight clipping nowadays. The bigger problem is the dynamically harmful stuff people put on their master buses.
ok cool. what area do you work in? also a mastering engineer? Ive found that different scenes have different common technical challenges. The vast majority of my clients are techno, house and disco, in that order. In techno, you get a lot of slammed mixes and various stuff including hats too loud. In disco you get vocal sibilance, not enough sub on the kicks etc. Maybe your particular niche for some reason loves to clip. I wouldn't be surprised if that was a scene wide trend too.
@@APMastering im not a mastering engineer but im an audio recording/mixing engineer. Im from iceland so ive done all genres (in iceland we need to do everything because the scene is so small if you specialize in only a couple of genres you will not get the dough$) Im no expert but ive been doing this for 25 years and have many icelandic mastering engineer friends... what ive seen the most in terms of errors is too loud mixes with no proper gain staging. My mastering engineer friends always ask for 32bit float due to this issue. Again, im not an expert.
Wow, Iceland is so cool. Lots of great music has come out of a country which has as many people in total as a medium sized city. In terms of 32float, if that works for your clients then that's great. Like I say, in the scene I operate in, I'm mostly pulling my hair out over compression rather than clipping. That said, it used to be a big trend in electronic music when I was a teenager in the late 90s to clip into the DAT / CD recorder because of the soft clipping on the way in on higher quality gear.
Useful in recordings as other mentionned. but its a lot like the snake oil in the audiophile community. 500$ cable sounding much better and that they can know if the dac is ess akm ti brown or cirrus logic by ear lol. There is always a point where we cannot physically hear a difference. Sample rate does make a difference with plugins like reverb and delays again in mixing or recording not actual playback. Listening in bit perfect sound better on a android phone tought because of the crappier android sound engine. there's also some exceptions to things of course. I use a JDS Labs Atom 2 stack for the headphones and monitors and the specs are so good for cheap price these days that unless we need a certain feature or I/O , its more then good enough. The audio world is just not the same now and the tools are totally different.
32 float extremely useful for my pocket recorder. No need to worry about gain. There isnt even a knob. rain storm from the first soft initial patter to the loudest thunder clap. one file, no clipping, no riding gain.
this is a good use case for float. bear in mind though that the mics, convertors and preamps all have dynamic range limits, and if the upper limit were calibrated to 0dbFS for example, its unlikely the performance would be worse with dithered 24bit
Yes, that is definitely the best use case. Live recordings in areas where you can not control other sounds like if you are making interviews in the street with some one that you will only get one chance to do - think news reporter of CEOs on the run. Then you can save a take that clips.
Coming from an EE, audio engineer, musician, and composer . . . you nailed it. Nice job, I didn't know how the 32 float was implemented, using a portion of the bit pattern for offsetting, neat. I won't be using that anymore. Although I have mostly been in the 24 bit arena. Thanks, good luck with the channel.
nice one! glad you enjoyed the vid!
I don't know why but I read this comment in Dan Worrall's voice
Comments like this are evidence of the quality of the information on this channel. I'm working through Ethan Winer's book (found through his interview on this channel) to understand this more thoroughly. I hope your subscriber and view counts correlate with the factualness of your content on a long enough timeline.
Wow, I was just discussing this at work this afternoon, and I came home to see youtube recommending this video! It's nice having a video that goes into such technical detail, but it's more personal than a filmed lecture and has some useful images. It looks like a brand new channel, so I wish you the best and look forward to your future content!
Nice one, glad you got value from the video. Yep, my channel and entire social media presence is brand new, so let's see if the channel is able to get anywhere!
nice video, i hope you make it on youtube, your production value is top notch. cheers!
thanks!
Great explanation! Can't believe you still have only a handful of subscribers.
Thanks! I've just got a few subs because I only uploaded my first video just a few days ago. Hopefully my channel will grow. Hope you come along for the ride!
As a hearing care professional/audio enthusiast/hobby artist this is my first encounter with a realistic engineer. I couldn’t agree more. You are absolutely right about the stapedius muscle protecting our ears and to be honest: 120 dB is mostly the limit of our ears and as we get older we lose dynamics as well as frequencies in our natural ears. I’m not interested basically in recording above 16 bit… I record partially on a Tascam that allows 24 bit and other stuff in my PC that would allow 32 bit, but I never use it. Why? Because I basically can’t hear the actual difference and besides: I like to make music and with loudness wars and all I think back at my childhood days where I was perfectly happy with my limited tapes and LPs. In fact: as a kid of 9 or 10 I would only listen to a transistor radio with a monophonic earbud with very limited frequency range… but loved the music nonetheless.
If you listen to the music from Gary Numan or Jean Michel Jarre from recent years compared to their late 70s or early 80s output you will undoubtedly notice a massive difference in mastering. Cleverly done: I’ll grant it immediately, but is it equally interesting from a musical perspective? Personal taste left aside, but somehow the music was a lot warmer in those days. Natural dynamics and I loved the (sometimes hopelessly out of phase) stereo from back then on my headphones. I just loved it.True, it’s an age thing as well. But the nonsense that’s going on these days ruins the creation of excellent music, which should come first. Unless you’re able to find creativity in all these new ideas of compression techniques and use extra headroom wisely, it won’t make the music necessarily better.
I agree, there is a difference between music and sound. A good song stay’s a good song even if the sound quality is bad, a mediocre song doesn’t get better with super high quality. Music is more a feeling than just sonic perfection. I wish a lot of musicians and engineers would embrace that part more.
32 bits floating point is great for the internal maths of DAWs and plugins, exactly because you can't clip it, and you won't by mistake add noise. But for storage it's just 33% larger for no benefit. 😀
Great channel, quality and factual content. You just got another subscriber.
thanks mate
32bit float when recording live performances 👌
Doesn't matter, the analog bits including the A/D has a noise floor and a max level that clips. So you still need to get the level as close to 0db but not over. What comes out of the A/D will be fixed point (like 24bits) and storage format will make no difference at all. Store it as 32bit or 32 bit floating point, it will just add eight zeroes to that data. So your files will be 33% larger. But with today's storage solutions, yeah, that's not a problem, you have basically infinite storage anyway. 😀
I always export in 24bits because I'm usually making samples and most plugins won't load a 32 bit sample. Omnisphere, for example, won't load a 32 bit wav file it needs to be 24 or less.
Excellent video thank you!
32bit float makes a lot of sense for many recording scenarios, especially where sudden peaks are hard to control or predict. Particularly true for production dialog recording. Some dual system audio field recorders used stacked ADCs to allow for extended headroom.
stacked ADCs is an interesting idea but you will ultimately have a noise floor from the preamp and other points in the signal path
@@APMasteringas I understand the idea is to have 2 or 3 AD and choose the one which fits the best
Reminds me of the AVGN sketch where he says "64 bits, 32 bits, 16 bits..." to the point where there are no bits left, but he keeps going and the quality gets worse. And the really funny thing is that as the quality gets worse, he says it more and more aggressively. It reminds me of how, on the one hand, people can use the most expensive converters, but they're actually mastering their music down to -3 LUFS, using a lot of clipping, and not even using oversampling.
th-cam.com/video/VUcI0p04XJw/w-d-xo.html
yeah, some people have expensive gear and no skills but I think they are outnumbered by the kids with poverty setups making hits. I know of someone who makes all his music exclusively listening through his built in Macbook speakers
For mixing and mastering this is true, but for sounddesign it isnt. Low bitdepth equals more digital distortion. Just how low bitrate causes more digital aliasing. When you significantly mangle a sound it starts to matter how true the dynamic representation is, cause even if the noise floor introduced by the bitdepth might currently be well below hearing range it definetly wont be after you significantly distorted it. Obviously this can also be used a wanted effect, but it does end up making an audible difference.
sure, when you are doing extreme stuff in sound design, like boosting 80db or something, of course you need more bits. I think I actually mention this later in the video, can't remember. but I agree
What you say is true for saving/storing audio files, but during processing things are a bit different.
You can not talk about bit depth without involving sampling rate. All audio adc and most dac use a 1 bit delta sigma front end oversampled into the megahertz. It is then downsampled to 48000 using some clever maths to push the noise out of band. The microphone in your cell uses a 1 bit interface to the CPU, this solves crosstalk noise. Things like ANC headphones will directly process the 1 bit stream directly to reduce latency for noise cancellation.
As far as audio in a daw, filtering is easier to implement in floating point, fixed point implementations, especially low pass or peaky ones, can be problematic in 32 bits.
You are right that sample rate and bit depth have an important relationship. In fact, dan lavry writes about that in the paper I mention in this video... I also mention DAW internal mixing bit depth is often 64bit float around 10m into the video but most people probably dipped by then tbf. In terms of 1bit... This is approaching the edge of my technical knowledge on this kind of gear, as I'm just a mastering engineer and not an electronics engineer but chips like the AK5394A used in benchmark ADCs use multi bit technology and not just 1 bit but this is getting into the weeds. The video was for the most part aimed at producers wanting to know what bit depth to bounce at ha ha.
trout mask
In my daw (bitwig) i can set a bit rate for recording and a bit rate for bouncing audio. Contently i have set 32 bit floating for recording and 24 bit for bouncing audio. The reason because i use a modular and i want to have more headroom when recording it. What do you think about this settings?
16bit is the best and 12bit on some sound such leads / pluck to make the sound more crunchy
nice
Great channel. Watched everything so far. Could I ask what is the name of the track @1.18 please?
nice one. it's this: th-cam.com/video/jBBJdmx5MdE/w-d-xo.html
Its one of my own tracks
Thank you for the great videos, AP Mastering!
People who advocate for 32-bit have one main argument: they believe that with 32-bit, a distorted signal can be restored (although this is an amateur mistake!).
So, is it possible to restore a distorted signal in a 24-bit environment using 24-bit?
its not about the number, its about whether its float or not and how it was distorted. if you run it through a distortion unit and then bounce it in float, you cannot restore it. if you exceed 0dbFS and then bounce as float, you can restore it. Theoretically in "24bit float" you could restore it too but 24bit float for all intents are purposes doesnt exist in audio and would sound terrible even if it did exist.
DSD vs PCM should be your next after sample rate video ;)
DSD is louder and that confuse the people about sounding better thats what i can hear as a difference.
tbh, I don't know a lot about DSD. This isnt something that i really had a lot of contact with as a mastering engineer. I might still make the sample rate video but im not 100% sure if it will be boring / overly niche 🤔.
@@APMastering There is a lot of similar snake oil and arguments in the audiophile circle also. I'm looking forward for your content. Have a good one sir.
32 bit is only useful when recording. You just don't have to worry about clipping.
Also, as I understand, you can record very quiet signals with the noise floor not being a problem.
this is not actually true because your preamps and ADC are the limiting factor
What they do now in portable recorders now is combine the output of multiple ADCs to achieve the crazy numbers of dynamic range. I think the customer might fel better if 0 is around normal level rather than having a massive "wasted" headroom in integer.
it still needs a preamp, so much of it is marketing because the preamp is not going to have zero noise floor across 200db dynamic range
But what if you cool it in liquid nitrogen to lower heat noise in the circuit? 😂
@@APMastering
32 bit portable recorders have two or more parallel preamps with different gain levels as well as ADCs, after that outputs of ADCs are combined in DSP to high dynamic range/low noise floor signal and saved as 32 bit file.
Great quality Audio converters need 3~4 bit for self noise (every hardware device has self noise embedded into the converted signal, usually inaudible in typical listening scenarios where you would have average full mix signal not exceeding 85~90dB SPL, sound pressure level as a reference to lets say digital clipping at 0dB) in the signal so for useful Music (audio) you would technically use not more than 20 bits. In most consumer market units that would be 18 bits of those 24 bits.
Floating point allows for internal (digital realm, which we can not hear, although it can be visualised if needed on screen, but not all Peak\RMS\LUFS meters get to display it really) "headroom" spread of all kind of digital audio processing. It is useful to get detailed reverb tails, low level dynamics acoustic audio (if recorded properly with high sens. mics, ADC\DAC and such equipment).
The floating point shifts with every sample in increments of 6 db. You get something like 31 useful bits if the range over 1.0 is not used. That is if you actually want to turn something down really low and steganographically hide it. Adaptivity to the level is part of how data reduction algorithms like mp3 achieve their performance, but there the sliding window affects a longer time segment.
32bit float is a common thing mastering engineers want from amateur mixing "engineers"
Its because many people mix too loud, too near the peak.
I'm a ME and I always advise plain old 24bit. In my experience, most people who mix too hot do so with compressors and limiters and the "overcooking" which happens is not normally clipping. Most people know not to clip but you get slammed mixes in various ways that do not straight clip but are nonetheless slammed. Here 32bit doesn't save the mix because the artists aren't clipping. Maybe 1 in 1000 mixes I get is actually just straight clipping nowadays. The bigger problem is the dynamically harmful stuff people put on their master buses.
@@APMastering Ive seen hundreds of mixes being too loud without anything on the master.
ok cool. what area do you work in? also a mastering engineer? Ive found that different scenes have different common technical challenges. The vast majority of my clients are techno, house and disco, in that order. In techno, you get a lot of slammed mixes and various stuff including hats too loud. In disco you get vocal sibilance, not enough sub on the kicks etc. Maybe your particular niche for some reason loves to clip. I wouldn't be surprised if that was a scene wide trend too.
@@APMastering im not a mastering engineer but im an audio recording/mixing engineer. Im from iceland so ive done all genres (in iceland we need to do everything because the scene is so small if you specialize in only a couple of genres you will not get the dough$)
Im no expert but ive been doing this for 25 years and have many icelandic mastering engineer friends... what ive seen the most in terms of errors is too loud mixes with no proper gain staging.
My mastering engineer friends always ask for 32bit float due to this issue.
Again, im not an expert.
Wow, Iceland is so cool. Lots of great music has come out of a country which has as many people in total as a medium sized city. In terms of 32float, if that works for your clients then that's great. Like I say, in the scene I operate in, I'm mostly pulling my hair out over compression rather than clipping. That said, it used to be a big trend in electronic music when I was a teenager in the late 90s to clip into the DAT / CD recorder because of the soft clipping on the way in on higher quality gear.
Useful in recordings as other mentionned. but its a lot like the snake oil in the audiophile community. 500$ cable sounding much better and that they can know if the dac is ess akm ti brown or cirrus logic by ear lol. There is always a point where we cannot physically hear a difference. Sample rate does make a difference with plugins like reverb and delays again in mixing or recording not actual playback. Listening in bit perfect sound better on a android phone tought because of the crappier android sound engine. there's also some exceptions to things of course. I use a JDS Labs Atom 2 stack for the headphones and monitors and the specs are so good for cheap price these days that unless we need a certain feature or I/O , its more then good enough. The audio world is just not the same now and the tools are totally different.
agree
The large on screen numbers read out in a monotone voice are a bit Monty Python tbh... 'NUMBER 3...The Larch'