[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Sc-devel] Machine Listening plans

2007/11/17, Click Nilson <clicksonnil@xxxxxxxxx>:
> > Well it could be called BeatTrack perhaps?
> I have an unreleased beat tracker called BeatTrack. I suppose I don't
> mind the name switch, but AutoTrack was to indicate the use of
> autocorrelation.

Well I don't mind much about the name but I think that motivation will
be lost on most people. Perhaps TempoTrack might be OK. I'll leave it
to you though.

> I'd need a consensus on name space (ML_Loudness) versus just naming
> UGens (Loudness) . Better to know before I add anything or write any
> more.

Yes, I agree. I don't want to be the only one throwing suggestions at
you from afar, maybe someone else can venture an opinion, even if
they're not strongly bothered?

> Perhaps I'd still need to add an extra plugin target for
> cleanliness; putting then in FFT2 might clash a bit. These UGens also
> tend to be large in code terms so usually have a .h and .cpp to
> themselves for each UGen.
> >> Just built an ML_MFCC by the way (in this case, the underscore
> >> helps to
> >> avoid too many capitals in a row?). Dan, we can swap code off list
> >> if you
> >> have a different implementation ;  )
> >
> > Yes I do, and it's all ready to go. I suppose we should have a
> > wrestling match and the winner can add their UGen to core. The loser
> > just has to add their to the end of a long slashdot thread or
> > something, for permanent obscurity.
> I'm not interested in wrestling


> but we could trade specs, code etc.
> Yours may be more authentic than mine; mine uses ERB bands and EQLC
> before the cosine basis transform. Runs about 3% CPU on 1.67 Mhz PPC
> G4. Could call mine EFCC and yours MFCC if you have an authentic Mel
> scale? ;  )

Mine is an authentic Mel scale, yes. On my laptop (same spec as yours)
it runs at about 3% CPU too. It inherits code from libxtract which
includes a couple of nice touches like the ability to choose either
equal-power or equal-area filters. No equal-loudness curve though.

If anyone wants to comment on MFCCs please do, otherwise I hereby take
this offlist...