When DJing it’s down to your own ear really, some people just can’t hear the diferance between 44.1/96,000 etc. and the people who can’t normaly stick with the smaller file size.

With regards to producing a high bit depth is allways good imo as your software synths will create the waveform much more acuratly to begin with and this is essentual if you’re going to be adding effects to it as the lower bit depths have a tendancy to fall apart a little and gain artifacts.

I offten use 32bit float when making tunes and record the final version in the same bit depth (as most daws are totaly rubish at converting the bit rate after) and then resample in soundforge pro 10 (pretty much the best dither algorythem out there) to 24 bit.

With regards to sample rates, it’s best to use what the end recording will be in as all you gain really is extra frequencies above and below what most people can hear. I used to work in 96Khz but recently have been working in either 44.1 or 48 and it’s not hurt the sound of my music noticably.

Depends what you’re doing to it. if you intend to do any processing (mastering for example) its worth putting things at as high a bit rate as possible regardless of the original quality. If you’re just recording, mixing and putting back onto 16 bit medium you might as well keep it at 16 bit

As for the sample rate I don’t think anyone bothers to go above 44

@DaftFader 454467 wrote:

With regards to producing a high bit depth is allways good imo as your software synths will create the waveform much more acuratly to begin with and this is essentual if you’re going to be adding effects to it as the lower bit depths have a tendancy to fall apart a little and gain artifacts.

That makes sense actually, I’ve never really thought about the combined effect of all the rounding errors you’d get when you stack tons of plugins on top of a synth.

@Clusterfrog 454510 wrote:

As for the sample rate I don’t think anyone bothers to go above 44

It depends what medium your music will be listened to on. I.E. CD is 44 standard, but DVD is 48 … so if it’s music for a film you should probably make it in 48 to start with to avoid having to convert it later.

@Clusterfrog 454510 wrote:

Depends what you’re doing to it. if you intend to do any processing (mastering for example) its worth putting things at as high a bit rate as possible regardless of the original quality. If you’re just recording, mixing and putting back onto 16 bit medium you might as well keep it at 16 bit

As for the sample rate I don’t think anyone bothers to go above 44

I think GL just wants to record some DJ mixes off his computer and wants to know if it’s really worth the hassle of using the oversampling/high bit-rate modes on the recorder when it will produce files that most media players won’t recognise anyway (whatever happens, the files will likely end up being converted to 44.1/16 wav or MP3 anyway). I.e. are there enough pros to outweigh the cons of having to convert the files.

44.1kHz is CD standard. 48kHz has been the ‘professional’ DAT standard for the last couple of decades. Oversampling at 96kHZ or higher (often using very expensive ADCs) is commonly used by recording engineers, particularly in the classical world, to get the best possible sound down ‘at source’, due to the benefits of oversampling which I tried to explain in a slightly half-arsed way earlier.

@cheeseweasel 458272 wrote:

I’ve often wondered what an MP3 would sound like to a dog or cat (in fact I remember asking this question in a lecture once and people thought I was taking the piss), as the way MP3 works is by exploiting redundancy in the human auditory system by using the masking effect. This requires a virtual model of the human ear to be incorporated into the encoder, which tells it which frequencies will be masked by the programme material (these frequencies are then re-sampled at a much lower bit-depth but the resulting quantising noise is masked). To a non-human listener it must sound like total junk.

Aye – I would imagine they sound pretty horrible (if you play them on a half decent size and quality rig that so called redundancy in the human auditory system becomes immediately obvious to most people)… They don’t encode anything over 16KHz (the last band of the 32 bank PQF only goes up to 16KHz as the “average” persons hearing doesn’t usually hear much over that). Cats, dogs and the rats mentioned probably hear well into the middle 20-30KHz range and maybe even further, so to them it’s missing a lot of it’s audio information – even a CD likely sounds stunted to them (which would explain why our cats always seem to prefer analogue sources to digital ones)…

The masking effect (specifically the limits thereof) where the theory says lower amplitude sounds are masked by higher amplitude to the point that the lower are not in fact heard at all if the difference is large was brought home to me pretty forcefully with the new Orb album recently. When I first got it, it was a 192Kbps mp3 version which I listened to for a while before eventually going out and getting a cd version. When I got the cd I was fairly shocked at the difference – the latest album by the Orb has very much gone back to their original sound where the sound is layered and extremely complex, built up over the course of the song, and in many ways felt as much as physically heard – the mp3 had stripped so much of that out that it was almost like listening to 2 totally different albums…

@General Lighting 454438 wrote:

My source material is currently (unfortunately) mostly 320k MP3s with vinyl transfers being recorded at 44.1K/16 bit (as I think thats the standard of my USB turntable and also the mixing software I am using). This is being fed through an analogue mixer and into the Tascam line in.

The mp3’s are a waste of time recording over 44.1, due to the above mentioned upper cap on the highest frequency. Even when mp3’s are recorded with a 48K rate, the compression algorithm essentially removes most of the benefit of the higher frequency by focussing on the theoretical human audible response – essentially the fletcher/munson curve of between 2 and 5KHz and stripping out much of the data at the extremes in order to do this (even a 320Kbps mp3 is running at less than 1/2 the bandwidth required for 44.1KHz and 16bit – increase the frequency and more has to be stripped out – ditto the bit rate. Some of that compression is done losslessly, but not that much)…

The vinyl is a different matter, and would benefit from as high a rate as you can get away with – I tend to like recording at 96KHz and 24bit if I can, as the resulting recording sounds far better at that rate (and even taking into account the need to dither for changing bit depth, and anti aliasing for the sample rate change, the resample to cd quality still usually comes out with an acceptable sound)…

As for the DV, sony stuff tends to use their proprietary codec for the audio – either the pca lossless one, or the lossy one that I cannae remember the name of off hand… They are both (though it pains me to say anything good of the evil empire :cry:) pretty good quality, but are going to need downsampling (and therefore anti aliasing) to put on a cd (and will sound better at the 48KHz original rate if you can use it)….

Noname wrote a really excellent guide to audio fundamentals a while ago. It explains all of those things:

http://www.partyvibe.com/forums/sound-engineering/27792-how-build-sound-system.html

nyquist Is a formula that tells you how to get the best reprisentation possible by making the sample rate twice as big as the differance between the lower and upper most frequency you want to use. (At least that’s how I understand it … I’ve not 100% got to grips with nyquist my self). It’s got other uses as well and is in alot of other formula’s used when working with sound reprisented as “n”..

Aye – pretty much what cheeseweasel and Daftfader said. When I talked about having to use anti aliasing when you downsample – ie from 48K to 44.1K or such it is for the reason explained above (anti aliasing attempts to remove any aliasing errors by filtering everything above the Nyquist limit) Another process called interpolation is used to make the waveform more closely follow the higher rate waveform by altering the placing of some sample points in this manner:

[ATTACH=CONFIG]81023[/ATTACH]

Not 100% sure, but I believe this is still covered in the Nyquist theorem, although it’s called Shannon-Whittaker interpolation after some of the other folk who are responsible for the theory (that only Nyquist seems to generally get his name on for some reason)… Much of the original mathematical proofs are written by Claude Shannon, and Whittaker came up with the same idea independently of them.

Anyways to explain the diagram:

The absolute path for a halfed sample rate (the dotted red line) is much less dynamic than the original at the higher rate – therefore we move some points (the green line) to more faithfully represent the source (obviously it isn’t perfect, but the green line is markedly more dynamic than the dotted red where the signal fails to get out of negative Voltages for 6 samples in the middle – the original does so 3 times during the same period, and the green manages it twice).

Thats a very basic explanation of the theory. The actual process of anti aliasing and downsampling usually involves running a low pass filter to filter out any frequency changes that are above the Nyquist limit (in the same way as converting from analogue to digital – although a band pass is sometimes used in AD conversion because the analogue source is continuous and band pass can help avoid some quantisation errors. Downsampling can’t use a band pass filter, and must use low pass (AFAIK – I could be wrong here though – there are bits of Nyquist that are just too much maths for my brain to cope with…)

Then I believe some form of inverse Fourier is used to try and rebuild the original source as far as possible, and this is then quantised back to the digital waveform required (this could be complete nonsense though – as I say, maths makes my head melt – especially when greek letters and such are involved :crazy:)

As for bit depth, it dictates the resolution of the representation – 8 bit gives 512 levels of resolution, 16 bit 65,536, and 24 bit 16,700,000 or so. This dictates the dynamic range (the difference between the smallest and largest sound representable), and as a consequence of that it also dictates the signal-noise ratio. When audio is quantised (the process of squaring the waveform – necessary because it’s being represented by a series of discrete points instead of a continuous wave as it is as an analogue signal), it introduces repeated quantisation errors (which manifest as harmonic distortion at the 1st and subsequent harmonic frequencies – not generally very nice sounding – think someone planking a flat palm on a piano keyboard and hitting 6 or 7 keys together). So the process of “dithering” tends to be used (where the repeated errors in the quantisation function are randomised so removing the build up around the harmonics).