For those on a budget

[Slouching Raymond]

Now, I ask this in all ignorance Ray, how much does such accuracy matter for audio work? Since the human ear has a "resolution" no better than 2dB, how many iterations would even a complex audio file have to go through until a difference could be detected? Thousands? Billions?

The answer is 'Not a jot!' You are quite right Dave.

How much calculation needs to happen to convert a midi file to an orchestra, or to simulate a spacial reverb from an impulse response? How much error can you tolerate? Like you, I'm asking in all ignorance, but I have to assume that the people who are creating these program and plugins know what's needed, not just some marketing guy telling the engineering dept "we need to be twice as good as the competition. If they have 32 bits, I want 64 bits".

I think the goal is to just eliminate any possible way that an audio signal might be corrupted in the digital domain. Then it's all up to the analog system to keep up.




Raymond, I remember the days when we got an 80286 CPU and an 80287 FPU for some of our instrumentation at work. Computers were pricey and the FPU just added to the cost. The bean counters wanted to know why we needed it. $3000 for a computer was a TON of money in the mid 80s!

Little calculation for the midi to orchestra, just stitching samples together, with simple scaling for note volumes.
There are 2 flavours of Digital Filers: FIR (Finite impulse response), and IIR (Infinite Impulse Response).
The FIR variety are just about convolving, sliding your signal past the fixed impulse response pattern, multiplying each point, then summing to create an output sample.
It is the IIR variety that needs higher resolution. They are architecturally simpler, with less arithmetic being necessary, but because they feed the signal back, which churns round the filter architecture many times, greater accuracy is required.

When the FPU chip was not present on the motherboard, all its functionality was simulated in software routines built from ordinary CPU instructions, which were nowhere near as fast.
You could buy a 486DX, which had a built in FPU, or if you were poor, you bought a 486SX, which did not have an FPU.
They would test the 486DX chips, and the ones where the FPU failed, woud have the FPU disabled, and would be sold as 486SX.

 

[jamesperrett]

Now, I ask this in all ignorance Ray, how much does such accuracy matter for audio work? Since the human ear has a "resolution" no better than 2dB, how many iterations would even a complex audio file have to go through until a difference could be detected? Thousands? Billions?

Dave.

I'm not sure where you get your 2dB from Dave but we can certainly hear differences of less than 1dB although we may not necessarily perceive them as level differences. I've experienced level differences of around 0.2dB causing a difference in sound - I was scratching my head trying to find out why a file didn't sound as I expected it to sound until I noticed a tiny 0.2dB cut in the mid eq setting. You won't necessarily hear this on your average hifi but things like this are more noticeable on decent monitors in a well treated room.

Big Tannoys are very revealing and in my world bit for bit accuracy is important.

 

[ecc83]

I'm not sure where you get your 2dB from Dave but we can certainly hear differences of less than 1dB although we may not necessarily perceive them as level differences. I've experienced level differences of around 0.2dB causing a difference in sound - I was scratching my head trying to find out why a file didn't sound as I expected it to sound until I noticed a tiny 0.2dB cut in the mid eq setting. You won't necessarily hear this on your average hifi but things like this are more noticeable on decent monitors in a well treated room.

Big Tannoys are very revealing and in my world bit for bit accuracy is important.

Historical James. It used to be said that the smallest level difference that the ear could detect was 3dB so I went a bit better! Yes, that work was done a long time ago and, AFAIK with pure tones? I agree the detection of differences on complex musical sounds could be more sensitive*. Then there are 'special people'! A very few adult males can still hear 25kHz and beyond (there was a composer who could but his name escapes me) But, if you take an average of the population, not having a lifetime's exposure and training in these matters, I would aver '2dB' is about right?

But surely my question is still valid wherever you put the goalposts? How many copies of a music file does it take before YOU can detect a change?

*This is not the same as a 2dB shelf cut or boost on say a monitor at 2kHz. That changes the sound energy over a very large and important part of the spectrum. But then I don't know what form 'digital degradation' IF detectable would take. Does anybody?

Dave.

 

[ecc83]

Just to add to the 'budget' point? Amazon emailed me yesterday offering the EM91-C at £29. I have ordered one. This time in matte black. Should be here Tuesday and I shall post something done with it ASAP.

Son can then take it back to France and have a bash at stereo! Anyone know of a cheap device to mount two LDCs as a CO-I pair?

Dave.