< Back to IRCAM Forum

Data filtering strategies in OM?

OpenMusic
#1

I want to work on some data sets generated from SDIF analysis files. In most instances predictably massive amounts of notes are generated to a CHORD-SEQ. I’ve been researching approaches to filtering that data in sensible ways but I could use some pointers to effective strategies for data reduction. What tools should I look at ? Any particular tutorials and/or videos ?

TIA!

Dave Phillips

#2

Dear Dave,

Unless you write your own filters, i recommend using AS->OM (for partial tracking sdif files). The method has more than one interesting ways to filter the data. You will find it in RepMus library and it is straight forward function.
Starting from here, you can of course filter the resulting chord-seq with your own method. If you need a particular example, i will be glad to provide you with one, like a register filter (mostly done with a loop).

Best
K

#3

Thank for the advice, Karim, I’ve been working with AS->OM and it appears to be what I was looking for. The Repmus examples have been helpful. I’d like to see any example file(s) you could provide, I need all the help I can get. :slight_smile:

Best,

dp

#4

Yes of course. In fact you pointed out exactly the lack of examples regarding SDIF transcription. Need to gather some patches and send them here. Will do ASAP.

Best

K

1 Like
#5

Dear Dave,

Here is a range filtering example for a chord-seq:

Screenshot_2020-09-20_19-18-19
Screenshot_2020-09-20_19-18-19688×510 33.1 KB

It allows to filter either pitch, dynamics or channels in pass and reject mode (optional input).

Here is the lisp code to be included in the init folder:
chord-seq_filt.lisp (2.2 KB)
Here is the patch:
chord-seq-range-filtering.omp (187.7 KB)

I’ve written it directly in lisp because it will be faster and more optimized. If you need an equivalent style in patch visual programming will send it also. But be warned, this will be rather messy… :slight_smile:

Test it and tell me if it worth including in OpenMusic’s Functions.

BEst
K

#6

Dear Dave,

before going to the data filtering issue maybe it worths also looking on your amplitude threshold in your Audio Sculpt analysis. It is a very important factor on the quantity of resulting data. The original setting, (-90dB) if I am not wrong, it is far too low… Try to experiment with some higher values, (-60 or -40dB) and see what you get. The issue is always filtering…

Best,

Dimitris

#7

Thanks a lot, Karim, your function would be a good addition to the OM Functions. It worked perfectly with some tests last night, I’ll play with it some more today.

Nice work ! :slight_smile:

Best regards,

dp

#8

Hi Dimitris ! Thank you for your suggestion, but I’m working on Linux so I’m not using AudioSculpt. I’ve been producing SDIF files with Michael Klingbeil’s SPEAR program and I do still have a lot to learn about tuning the analysis parameters. I’ll check the default amplitude threshold settings and make some new tests.

Best regards,

dp

#9

Hi Dave,

In Spear it is a similar process.
It is always a game to find a balance between Frequency Resolution, Minimum Amplitude Threshold and later if you want to filter unwanted or not important partials from your result to choose Select Partials Below Threshold in the edit menu and eventually eliminate.

Best regards,

D