Subject: Re: [linux-audio-dev] extending LADSPA, it's not that easy ......
From: Benno Senoner (sbenno_AT_gardena.net)
Date: Fri Oct 27 2000 - 14:45:05 EEST
On Fri, 27 Oct 2000, Tom Pincince wrote:
>
>
> I woke up this morning thinking the same thing. This 128 step
> resolution reveals itself sonically in many places. The solution that I
> imagined was also for linear interpolation within each block. Thinking
> as generically as possible I imagined two entry fields for each plugin
> control parameter, one for the beginning of a block and one for the
> end. If only one field was used there would be no interpolation. If
> interpolation did not apply to a particular parameter then the second
> entry field would be ignored.
this works, but for the host app it would be easier to simply provide one
control value which is then sloped using the one supplied previously,
(which the plug remembers internally). But this has disadvantages too:
to start with a value different from zero, when processing the first block,
you have first to disable interpolation, set the blocksize to 1 and set the
control value to the desired amount,
So it's probably better to use your model ? thoughts ?
> required me to do 50 or more of these microscopic fades per day. In
> these situations the end points of the fade regions need to be defined
> with single sample accuracy. Working with blocks of even 32 samples is
> much to crude for this.
Not sure if I misunderstood you, but I'd like to see the person which can
distinguish between a fast pure exponential volume envelope, and the
same made of linear segments with 32sample len.
The ear can distinguish only very few volume levels , so as Olli would
say the exponential aproximation would be "perceptually perfect".
But if you are paranoid, you can always use a gain plugin which takes
(as Richard said) two audio streams as input: one for the audio data and
the other for the volume control data.
This way you can supply arbitrary volume curves.
(the plug would be a bit slower due to a multiplication involved for each sample
but that's not a problem for today's CPUs).
>
> Prior to following lad my mental image of digital audio was entirely
> based on single sample frames, not blocks. Now that I am thinking in
> blocks it is challenging to create a mental image that allows single
> sample accuracy for anything, even though I use Digidesign software
> every day that does so.
The blocks can vary arbitrarily in its size (to suit your needs), the important
thing is to avoid to use a blocksize of <20-30 all the time.
eg. if you need to do precise cuts, fades etc, you can easily go down to 1sample
blocks every now and then (let's say at the beginning and end of each
segment of an EDL) since it will non add significant overhead to the whole
thing.
Benno.
>
> Tom
>
----------------------------------------
Content-Type: text/html; name="unnamed"
Content-Transfer-Encoding: 7bit
Content-Description:
----------------------------------------
This archive was generated by hypermail 2b28 : Fri Oct 27 2000 - 13:36:02 EEST