What bit rate do you record/mix at?

IF you must use higher bitrates that will be dithered, I suggest using multiples of the final bitrate. Something like 88.2 is much easier to dither down than something like 96 or even 48. It's entirely possible that you lose more audio information going from something like 96K than 88.2K even though you *technically* had more information available with 96Khz sampling rate. It's all in the math. Doing fractional multiplication and division in the digital domain will always lead to bit truncation as the resulting numbers are too large. Making the math more simple for the dithering process will result in less loss.

Besides, dithering is adding noise. It's pseudo-random noise applied to a signal that is being bit decimated in order to mask distortion from the decimation process.

So why on earth would you record at a higher rate, just to add noise to mask inherent distortion later? That's why I stick with the same bitrate as my final product and avoid dither, the possible algorithm issues, adding noise and distortion that this process all entails.

Thinking along those lines, wouldn't the ideal situation be to record and mix at a higher sample rate the pitch the final mix to another computer/converter setup recording that mix at 44.1/16?

Just thinking out loud here.

I record at 24/44.1. But that's just me. However, I also mix down through a console into a completely different sound card. In this case, I could track at something higher, say 88.2K, and just set the mix-down sound card to record at 44.1k. That would be cool if you can do it.

I suppose I should be more clear, if you are mixing ITB and everything stays ITB, then I would prefer to keep the tracks at the same bitrate as the final product. However, if you still want to record at a higher bitrate, I'd use a multiple of 44.1k. 88.2K would be fine.

Also, as stated in my earlier post, I think the analog circuitry can make more difference to the sound than the sampling rate does. DACs work by "stepping" voltage/current(depending on how they are designed) into a "reconstruction filter". It's essentially a super LPF that rejects higher spurious noise and higher images. It does this by "rounding" the edges of the DAC's output signal which typically looks like a rough staircase going up and down. The "staircase's" rough edges are harmonics. The filter rejects these higher harmonics BUT it's a very thin line between the cutoff point of the filter, the speed of the filter itself and the frequency range. If this isn't done very well, you'll have problems. Same goes for the A/D side, it also has an input LPF for similar reasons. Also, any outboard gear will add more noise and distortions to the system than fidelity you'll gain from higher bitrates..

A/D and D/A converters are cheap commodity parts these days. If you use the Cirrus Logic parts, the TI parts, the Analog Devices parts, or whomever, chances are, you could never tell the difference between them. The places where the designers go cheap are in the filters, the power supplies and the layouts of the boards.

That's primarily why I say "does it sound good" first, before looking at numbers on paper. An Mbox(or equally cheap equivalent) running at 24/88.2 might still only actually have 18/38K worth of usable audio, the rest being noise and distortions, while the Burl or something equally good might get 21/44.1 worth of usable audio from 24/44.1k. Catch my drift?

Of course, I could be completely off my rocker.

18,000 CYCLES???

I doubt I'll ever hear 18khz again. The results of my last hearing evaluation showed I could hear up to 16.4khz, granted not as well as I could hear from 15k down. Most of us lose 20khz at around 16 years old.

Isn't SRC(sample rate conversion) the rate of digitally converted samples per second? Don't higher rates only equate to higher resolution, and not have anything to do with the audible frequency of actual audio signal? The conflating of the two, never makes sense to me?

I appreciate any clarification.

Sample rate is how many points per cycle are sampled in time (think X axis).  88.2K is twice as many points as 44.1k per second.  This *in theory*, should result in twice the resolution.

But again, as I mentioned before, the filters could limit the disparity between the numbers.


Now, bit depth is another thing.  Bit depth defines the amount of points that can be sampled on a volume/level scale (think Y axis).  If you take a 60db possible volume scale and divide it by 16, that's 3.75db per bit of level resolution.  if you take 60db and divide by 24, that's 2.5db per bit of resolution.  The smaller the dB/bit the more precise the math that can be done on the signal, the more precise the reconstruction that can be done by the DAC and the less stringent the filters have to be.

Agreed!

I get the feeling if all atmospheric sounds above my limited tangible hearing range disappeared, there would be a imminent, tangible and foreboding feeling of doom for the world as i know it! just saying lol!

And yes,
the practically modified formula for technically applied shannon-nyquist theorem uses the factor 2.2 for samplerate frequency to be able to fully reconstruct the maximum signal frequency, thanks, ward, for pointing that out.

I'm going to say it again.. Just because the sampling rate is 88.2K doesn't mean that the input and output filters *allow* 44Khz to be sampled. That was kinda my point all along. 88.2K (or whatever) is just a number on paper. The quality of the filters and the quality of the digital manipulation can mean much more than just having a high sampling rate.  Those filters still might be set to 20Khz or so even though you have the ability to sample at 88.2khz..  Get my drift?

The caps in the power supply are not really the same situation. Having poor decoupling can mean that your amplifiers are pulling the rails around and not robustly slewing as they should. They are also generating noise on the rails that seeps into the other amplifiers. The result is noise added to the audio or outright distortion.

I'm not sure the sampling rate thing even matters as much as people let on. Most commercially available CDs are hard LPF during mastering and production around 16Khz anyway. MP3s are hard cut even lower during encoding, somewhere around 12Khz.

I think the next point is that most people fall prey to "hearing detail" and immediately think it's high frequencies that define the detail. It's really just fast transients, harmonics and lack of muddy low mids that psycho-acoustically fool us into thinking that we hear like dogs.

Tony, I may be misunderstanding what you're saying, but definition or resolution is all about the number of bits per sample, each additional bit representing about 6 db dynamic range.  The sample frequency relates only to the maximum frequency recorded, and has nothing to do with resolution.  So in a perfect system, the highest frequency that can be recorded is a bit less than half the sampling frequency.

On a standard tuned guitar, for example, the high E at the 24th fret is about 1280 Hz.  So a sampling frequency of 3 kHz would be sufficient, except it would exclude any harmonics, which extend out to as far as you care to measure them.  But the quality or resolution of the sound would be determined by the number of bits.

Although this is a source of contention and debate amongst some, there were white papers done on this in the late 90s (during the old debates on sample rates and frequency response) that the human body could "feel" up to 30khz and that certain phonograph needles were capable of reproducing this, which is why analog had a technical edge over digital. The human body can also feel as low as the 5-10 cycle octave, even though most ears can't perceive pitch below 25 cycles.

I'm sure with some extensive digging, one of us could find the published studies on this... if we had someone amongst us who demanded peer-reviewed scientific proof. There are those who scream "BS" without it. You all know the types.