Re: more accurate beat locations

[Hamish Allan <hamish.allan@gold.ac.uk>]

Hi Alex,

You might want to have a look at Simon Dixon's BeatRoot:

http://www.elec.qmul.ac.uk/people/simond/beatroot/index.html

He uses an alternative method for determining onsets which is  
documented elsewhere, but his lightweight multi-agent then post- 
processes those onset locations to determine the most likely tactus.  
From what I remember of his work, it shouldn't be hard to persuade  
some of those agents to report information for other layers of the  
metrical heirarchy (such as bars).

Best wishes,
Hamish

On 26 Mar 2007, at 18:53, Alex Norman wrote:

> Hello,
> I'm wondering if anyone here has some code available that processes  
> the
> beat/onset info that aubio gives to get even more accurate beat  
> location
> information?  I've found that if I calculate the median of the  
> distances between
> neighboring beats, I get a rather good estimation of the tempo, and  
> then I can
> remove beat locations that don't correspond to this tempo, but then  
> I have a lot
> of empty space without beat locations.  I'd like to intelligently  
> interpolate
> the beat locations so that I have a solid tempo indication over the  
> whole song.
>
> Anyways, if anyone has advice or code examples for this please send  
> them my way.
> I figure doing the auto-correlation of beat locations would help,  
> or simply
> finding beat locations that are close to whole number multiples of  
> the median
> distance from a long run of "solid" beat locations [and then  
> inserting the
> missing in-between locations].
>
> -Alex
>
>
>
> --
> To unsubscribe, send mail to aubio-unsubscribe@piem.org.

Attachment: smime.p7s
Description: S/MIME cryptographic signature