Topic: 24 bit > 16 bit

[jerryfreak]

the best equipment (mytek, benchmark, etc) is in the 112-120 dB range

They rate the SNR, that is, what is the ratio of converter noise to the maximum input (0dB FS).  The highest ones are, what, 108dB?  That is something like 18 bits, right?

  Richard

[Petrus]

  I'm just trying to dispell some myths out there.  Everyone thinks more bits/higher sampling rate/etc are better, when in reality there are lots of other things in the signal chain that are limiting us...

Limiting factors, in the order of typical importance

- listening room noise floor
- listening room acoustics
- reproduction system (speakers)
- microphone self noise
- recorder mic preamplifier noise
- human hearing capabilities
- 16/44.1k recording limitations
- digital edit system rounding errors
- 24/96k recording limitations

[boojum]

  I'm just trying to dispell some myths out there.  Everyone thinks more bits/higher sampling rate/etc are better, when in reality there are lots of other things in the signal chain that are limiting us...

Limiting factors, in the order of typical importance

- listening room noise floor
- listening room acoustics
- reproduction system (speakers)
- microphone self noise
- recorder mic preamplifier noise
- human hearing capabilities
- 16/44.1k recording limitations
- digital edit system rounding errors
- 24/96k recording limitations

I pretty much agree with the order.  But of all those things, which do we have control over???  Bit depth and sampling rate.  Therefore, the most easily manipulated in the list are the ones we want to attack first.  And if we do not get depth and samplig rate right the rest of the list will matter little as even the best of venues would not be captured well and the playback gear can only play back what it is fed.  And that is why I am grateful to have 24/48 and 24/96 as options.

I have a good playback system so that is well taken care of, good mics and recording hardware and failing hearing.  I am not sure hearing aids would be any more helpful than just cranking up the stereo.  And face it, if you read any of the old Stones record liners you remember the two things they said, "Play It Loud" or "Turn It Up."  Right, Mick.

L8R

edited: typo

[jerryfreak]

just because your hearing is failing doesnt mean that you shouldnt be making the best recordings you can. Your dog's gotta listen to it too!

  I'm just trying to dispell some myths out there.  Everyone thinks more bits/higher sampling rate/etc are better, when in reality there are lots of other things in the signal chain that are limiting us...

Limiting factors, in the order of typical importance

- listening room noise floor
- listening room acoustics
- reproduction system (speakers)
- microphone self noise
- recorder mic preamplifier noise
- human hearing capabilities
- 16/44.1k recording limitations
- digital edit system rounding errors
- 24/96k recording limitations

I pretty much agree with the order.  But of all those things, which do we have control over???  Bit depth and sampling rate.  Therefore, the most easily manipulated in the list are the ones we want to attack first.  And if we do not get depth and samplig rate right the rest of the list will matter little as even the best of venues would not be captured well and the playback gear can only play back what it is fed.  And that is why I am grateful to have 24/48 and 24/96 as options.

I have a good playback system so that is well taken care of, good mics and recording hardware and failing hearing.  I am not sure hearing aids would be any more helpful than just cranking up the stereo.  And face it, if you read and of the old Stones record liners you remember the two things they said, "Play It Loud" or "Turn It Up."  Right, Mick.

L8R

[gratefulphish]

Perhaps I should have read my post again a little more carefully.  I understand that the bit depth really only relates to dynamic range, and should have probably used 16/44.1 vs. 24/96, not just 16 vs. 24 bit.  I have a reasonable understanding of the Nyquist concept, so assuming that 48K is going to reasonably cover anything that we can possibly hear, is there any real reason to record in 24/96 vs. 24/48?  Are there more data points on that analog waveform at one vs. the other, in the range that we can hear, or are all the rest of the data points outside of human hearing and/or playback capability range, understanding that there are vast differences in playback systems?

[live2496]

I have a reasonable understanding of the Nyquist concept, so assuming that 48K is going to reasonably cover anything that we can possibly hear, is there any real reason to record in 24/96 vs. 24/48?

Maybe.

There are some designers that design audio circuits capable of recording ruler flat up to 100kHz. Rupert Neve is one of them. Why? Well, be believes that even though we can't hear beyond 20kHz, ultra high frequencies have some affect upon our perception of other sounds. He also cites that one researcher in Japan has discovered that our brains don't like bandwidth to be limited at 20kHz. Let the reader decide.

Have a listen to a recent interview from March 2007...
http://www.gearslutz.com/board/videos-podcasts-interviews-newsflashes-subcribe-so-you-dont-miss-out/115552-rupert-neve-interview-march-19th-2007-a.html

Here's a transcript from 2002...
http://www.prosoundweb.com/chat_psw/transcripts/rupert.php

I believe that there is a distinct benefit in recording 24-bit vs 16-bits. And most of us leave enough headroom when recording so that 16-bit quantization might lose some of the detail available from the preamp.

About the claims that Rupert is making. It certainly is an interesting theory. I must say that I lean toward believing him more than not. Some of my customers record whales, birds and other wildlife. So, I'm certainly very supportive of recording at up to 96kHz and beyond anyway.

To answer your question... I guess it depends upon your playback system, or in what you plan on doing with your audio recordings in the future.

[Petrus]

The audio waveform, be it pressure or voltage fluctuations, consist only of varying frequences. Smaller details = higher frequences. As there is a limit of what we can hear, there is no point in recording (and trying to reproduce, which is another matter) those frequences. There is nothing else to this, no hidden detail or such, it is just waves.

Sample rate determines the frequency range (detail), bit depth the dynamic range. Very simple to understand.

Aural capabilities of us humans have been studied for hundreds of years and the upper limit has been fixed to around 20kHz, only now, that cheap recorders capable of more are available, people start to hear things. Is it a scientific fact or just a rationalization for new toys? I vote for the later. These are hard times for hi-fi tweaking, recording systems and media are almost perfect, cannot tune turntables and cartritges anymore, now it is to braiding silver cables... (when loudspeakers and room acoustics are the weak point of amost every system, I guess soldering crossovers and glueing boxes & acoustic treatments is not sexy enough).

Still, using 96 kHz does no absolutelly harm. Just the reasons should be rational.

[Brian Skalinder]

Haven't looked for the link yet, but I remember an article posted here at some point suggesting the advantage of > 48 kHz sample rates lies not in revealing higher frequencies but better time coherency.  If I recall, the gist of it was the human ears and brain are very precise at distinguishing very small time differences.  The greater time precision of higher sample rates provides better time coherence to the listener.  Or something like that.

[nihilistic0]

hmm, and I think I recall reading somewhere about if 2 identical sounds are played in succession within say 2ms of eachother, that we cannot distinguish any pause between them or some shit

in short, our hearing has limitations, similar to how much detail the eye can resolve

[Petrus]

Those time coherence things might refer to the effects of the low cut filters needed before and after the AD/DA conversions. It is not possible or easy to design a brickwall filter which does not have some time domain effects on the signals just below the cut-off frequency. With 44.1 systems there might, just might, be some anomalies in the 19+ kHz area, specially with old/cheap systems. With 48 and 96 those effects move up past any possibility of us hearing them.

Still, good 16/44.1 classical CDs sound truly amazing... Even played loud, around 110 dB in peaks, there is no backround hiss (it is masked even by low room noise).

[Nicola Fankhauser]

hmm, and I think I recall reading somewhere about if 2 identical sounds are played in succession within say 2ms of eachother, that we cannot distinguish any pause between them or some shit

what the original poster might have tried to say is: if time consistency / resolution between (stereo) channels is below a certain level, the human brain detects very soon anomalies, since the room image it tries to build gets too little and (even worse) conflicting data. mp3 for example has very sloppy attack envelopes (try hi-hats or castagnettes, cembalo etc.) and coherence between channels in general which makes you feel tired listening to even high bitrate lossy compressed music.

all in all I think this argument is solid when advocating higher sampling rates - but it is not directly related to frequency range, but timing resolution.

regards
nicola

[gratefulphish]

I am still not sure that I got a direct answer to what I was trying to ask, so I will try asking another way.  Assuming that we are recording in 24 bit, and assuming that we are just recording a single guitar, all of the frequencies of which are well within the 20-20K range, if we are recording at a 96K sample rate, will there be more data points representing the guitar, within our audible range, than if we were recording at 48K, or is all of the other additional data made up of other points above and below the audible range?  I am just trying to discern whether we are truly getting more data points on the audible portion of the waveform, than just recording a lot of additional inaudible data at higher and lower frequencies.

[Nicola Fankhauser]

I am still not sure that I got a direct answer to what I was trying to ask, so I will try asking another way.  Assuming that we are recording in 24 bit, and assuming that we are just recording a single guitar, all of the frequencies of which are well within the 20-20K range, if we are recording at a 96K sample rate, will there be more data points representing the guitar, within our audible range, than if we were recording at 48K, or is all of the other additional data made up of other points above and below the audible range?  I am just trying to discern whether we are truly getting more data points on the audible portion of the waveform, than just recording a lot of additional inaudible data at higher and lower frequencies.

a very difficult question. however I think you can say (theoretically) you'll get more available dynamics (since it has 24 bit resolution) and better stereo image (because of the 96'000 samples per second).

regards
nicola

[gratefulphish]

I am still not sure that I got a direct answer to what I was trying to ask, so I will try asking another way.  Assuming that we are recording in 24 bit, and assuming that we are just recording a single guitar, all of the frequencies of which are well within the 20-20K range, if we are recording at a 96K sample rate, will there be more data points representing the guitar, within our audible range, than if we were recording at 48K, or is all of the other additional data made up of other points above and below the audible range?  I am just trying to discern whether we are truly getting more data points on the audible portion of the waveform, than just recording a lot of additional inaudible data at higher and lower frequencies.

a very difficult question. however I think you can say (theoretically) you'll get more available dynamics (since it has 24 bit resolution) and better stereo image (because of the 96'000 samples per second).

regards
nicola

This should be a straightforward scientific/mathematical answer, IMVHO.  It is the crux of the question to which I have been trying to get an answer.  There have to be either more, less or the same number of points, sample wise, to describe the same exact musical note.  I am just trying to determine, one way or the other, whether or not we are really getting more data, within the audible realm, as opposed to adding additional data above and below that range.

[live2496]

I am still not sure that I got a direct answer to what I was trying to ask, so I will try asking another way.  Assuming that we are recording in 24 bit, and assuming that we are just recording a single guitar, all of the frequencies of which are well within the 20-20K range, if we are recording at a 96K sample rate, will there be more data points representing the guitar, within our audible range, than if we were recording at 48K, or is all of the other additional data made up of other points above and below the audible range?  I am just trying to discern whether we are truly getting more data points on the audible portion of the waveform, than just recording a lot of additional inaudible data at higher and lower frequencies.

Whatever frequencies are present in the audio, all of it is sampled 96000 times per second. Whether it be a wave from a guitar or a hit on the cymbals. The state of the electrical circuit is measured at even intervals and digitized to a number.

Think of graphing a sine wave. Let's pick a frequency of 1000 Hz. That wav undulates 1000 times per second. A 2000 Hz sine wave will have a wavelength that is half the size of 1000. So it goes up and down twice as fast. This makes the peaks closer together. The higher the frequency, the faster the up and down movement occurs on our graph. If a harmonic from a cymbal was present at 20000 Hz, it would be fluctuating on our graph 20 times more often than the 1000 Hz wave.

While I am discussing this, I will mention the limits of sampling at 96000.
Taken to an extreme, a 48000 Hz sine wave would not be able to be graphed correctly by our system. Because it takes at least two sample points to represent the frequency and the sampling is not occurring quickly enough to support this. These frequencies, therefore are filtered by an audio circuit prior to digitizing.