Hi Annabelle.
Have you tried Voxengo deconvolver?
It processes audio recordings that you've made and creates the impulse
responses from it.
A test tone generator (sweep sounds) is included.
I attach a sample sweep.
The demo version should be fully functional, I think. However, it is a
bit tricky to find it as the application is no longer maintained, if I
recall correctly.
The subject is certainly interesting. But remember that the impulse
you get is not necessarily the right one.
Here are some random thoughts of mine, without having gone deep into the matter:
1. The exciting sounds should cover the whole spectrum
2. they should be reproducible, i.e. identical each time
3. analogue sounds (claps, pistol shots etc.) are fulfilling point 1
as they are not subject to a limited frequency range and being
coloured by a player/speaker module.
4. playing sweeps satisfies point 2 but have to be processed afterwards.
5. short impulses have to be very loud. Imagine having one sine tone
at -20 dB. You combine it with another one which gives you -14 dB etc.
6. For the analysis you need hundreds of frequencies covered which
gives you naturally less amplitude for a given bin
7. sweeps drag the volume over time and you can thus go with less.
8. player and microphone have to be neutralized in order to get the pure room.
I find it curious that procedures for recording IRs rely on a single
volume setup.
I would therefore create test tones at different levels (compare
clapping with different strength, what reflections do you get at low
volume?)
The idea is that you will at first record only the direct source and
the colouring through playback and recording device. The subsequent
sweeps do excite the room more and more.
This allows us to get a trend and extrapolate a response for a
exciting tone at an otherwise not achievable level, especially for the
environment you want to aim for.
If I had to design a test tone, it would:
- consist of sine tones at FFT bins (e.g. 44100 Hz divided by 1024 or
2048 etc) and at 3 different levels. Naturally, this would be a very
long tone, however, some tones that are not correlated can be stacked.
Of course, this needs a custom processing module but Nyquist should be
able to do that.
Sorry for the lengthy email.
Robert
