Re: Aubio

[Paul Brossier <piem@piem.org>]

Hello Kevin,

Thank you for your mail, I'm glad you are successfully using aubio. The
reply to your question is rather long. What you want is optimising the
parameters (or even modify the actual algorithm) to a specific
collection of music pieces.

This depends on why you wnat to do that: specify your application to a
given type of music, propose different presets to an end-user, or use
external informations (musical genre, recording year, or other audio
descriptors) as a hint for automatic selection of these settings.

Depending on this reason, the next step would be to set up a test
database with annotated data to evaluate the performance of the
algorithm with different settings. The database should reflect as well
as possible the set up of your final application.

As you may have read in my thesis, I used a database to work on various
musical genres, but the details of the result clearly show it is
'easier' to detect the tempo of electronic music than classical pieces.

The best approach from where you are is to set up the largest database
you can get, and evaluate as much as you can. Using n-folds techniques
to validate the robustness of your parameters should help. I did not use
n-fold myself, but it's definitely worth trying.

Let me know your further findings!

Best wishes, Paul

On Mon, Mar 24, 2008 at 11:52:32PM -0400, Kevin Coletta wrote:
> Hey,
> 
> My name is Kevin Coletta and I am currently doing a class project.  We 
> are utilizing your Aubio library, specifically the aubiotrack function.
> 
> It is a great piece of software, but we were wondering if you could give 
> us any tips on how to use the settings of program. For instance, if we 
> were to input a rock song or a rap song, are there any specific settings 
> you've found should be changed?
> 
> Any help you could give us would be greatly appreciated, and again, 
> thank you for this wonderful library.
> 
> -Kevin Coletta
> 


-- 
To unsubscribe, send mail to aubio-unsubscribe@piem.org.