djadhav

thesis thoughts
Ad 0:
Digital Ocean
Providing developers and businesses with a reliable, easy-to-use cloud computing platform of virtual servers (Droplets), object storage ( Spaces), and more.
2003-02-19 21:05:53 (UTC)

I ve just read that pattern..

I've just read that pattern recognition is something that
can be used to classify speech into voiced and unvoiced
speech. Will have to read more on the topic.

I have doubts on whether using the Mel scale (rather than a
linear frequency scale), results in any noticable
improvement in performance, so I will refrain from using it
in any code for now. However, I will add it later to see if
it makes any difference.

The decision parameters I have worked on so far
are : time energy, frequency energy and autocorrelation. I
intend to work on the following : zero crossing rate, LPC,
pattern recognition, and maybe bispectra (when I get more
insight on the topic).

Right now I am working on using the zero crossing rate. So far, I
have realised that the ZCR does not indicate the level of background
noise. Under fixed background noise, the ZCR increases where there is
speech. Under variable background noise, the ZCR decreases where
there is speech. Right now, the ZCR seems to be the least robust of
all decision parameters tried so far.


Ad:1