Re: [Audacity-devel] Better EQ curves

[Richard A. <ri...@au...>]

On Mon, 2010-03-15 at 00:48 +0000, Martyn Shaw wrote: 
> This doesn't seem like a good way forward (modifying the current FIR 
> algorithms to give good/accurate phase response).  If we want to get 
> something-like the correct phase response as well as amplitude 
> response then we should be looking at using IIR filters shouldn't we? 
My courses never used the FIR/IIR jargon so I'm relying on wikipedia,
but yes, all real RIAA/IEC etc preamps use IIR analogue filters, that's
why they have non-flat phase responses. The same would be true of almost
any analogue tone control system, varying from mild shifts on hifi-type
controls to quite large ones on the high-Q filters found in audio
production consoles and effects pedals.

At the core of this is the fact that there are an infinite number of
watys to design a filter with a given frequency response curve. Even
allowing for the fact there are fewer good ones, there will always be
more than one possible interpretation on any interface that only
provides a frequency response input.

Providing a phase and frequency response input removes all ambiguities -
the two curves _completely_ define the filter's response, any filter
that implements the two curves correctly will have no measurable
difference from any other (That's an over-simplification, but anything
that is outside it isn't really EQ, certainly in hardware). This is why
I was suggesting adding one extra set of parameters (hidden by default
I'm sure) as a simple way to provide almost any EQ function that could
ever be desired.

> We haven't got those at the moment in EQ but could make it if we 
> wanted.  We could have an interface that's lets us set those break points.
> 
> I'm no 'golden ear' but I don't believe that people can hear a 
> difference in phase, and I have tried this a number of times.  But you 
> can hear a change in phase of one component to a constructed waveform. 
>   I'm sure that we could get into 'heated debate' here about phase, 
> but let's not.

Do you mean by this you can't hear an overall phase shift, but you can a
frequency-dependent one? That would be a reasonable statement to me as
well.

The usualy illustration of this is filtering the high harmonics of a
square (or nearly square) wave with a filter with linear phase response
and with the RC-type response - one has ripples on it's waveform where
as the other looks rounded off, and this is claimed to be audible.

Richard