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

Re: frequency to mel formula



It's funny how numbers get around.

There's a good summary/survey at Hartmut Traunmüller's page:
http://www.ling.su.se/STAFF/hartmut/bark.htm
With respect to Beranek 1949 he says:
"Ratio pitch m (in mel): m = 1127 ln (1 + f / 700)
 This is an approximation based on data tabulated by Beranek [8]."
but doesn't say who published the approximation.

The wikipedia article uses the factor 1127.01048 to get exactly 1000
mel at 1000 Hz.  Searching for this number in Google scholar, the
earliest source I find for it is in a 2005 paper that acknowledges
that it came from wikipedia.  Since that time, it appears in many
papers.  It's another example of unsourced wikipedia information
being copied and published, likely to later be sourced to someone
who copied it from wikipedia.

The number was added to wikipedia in Jan 2003, here:
http://en.wikipedia.org/w/index.php?title=Mel_scale&diff=prev&oldid=571444
by this guy in Eindhoven: http://en.wikipedia.org/wiki/User:Ap

This page: http://users.utu.fi/jyrtuoma/speech/Mel2Hz.html
suggests that the editor may be J.R. de Pijper of IPO (just because he
provided the formula and he's in Eindhoven).

Since that time the excessively accurate number has crept into a few
2007 and 2008 books:
http://books.google.com/books?as_q=1127.01048

I recommend we stick with 1127, though I still don't know where it
comes from; does anyone?  Give me a source and I'll fix wikipedia...

Dick