Speech detection (formulation)

Hi all, i am currently doing speech detection in white gaussian noise.
I would take fft of a frame of noisy speech and if the maximum value is
above a certain threshold, i can say that speech is detected. Of
course, there is another part of the algorithm that does a more
thorough check.

My question is, i need to come up with a formula for this threshold.
meaning threshold=x^2 + y for example, rather than threshold =0.7. How
should i do it?

I have an idea here but i don't know if it can work. That is, for
example, if energy of frame=0.1, my threshold should be 0.5
if energy of frame=0.2, my threshold should be say 0.6
if energy of frame=0.3, my threshold should be say 0.63

Then i plot out these on cartesian plot and get the equation of the
curve,that will be the formula for the threshold. Will this solve my
problem? The curve will definitely not be linear. If so how can i find
out the equation of the curve? Any advice?

Thank you

 

For several years I have maintained that Digital Signal Processing,
for all its fancy maths, is an utterly subjective dicipline.

You have found the reason why.

In most practical applications of DSP, one needs to set a threshold
somewhere. In radar/sonar one sets the sensitivity of the reciever,
you have found another problem that requires a threshold to be set.
The performance of the systems depend on these thresholds, and
thus on subjective choise. High thresholds give low false alarm rates,
but don't catch weak signals. And vice versa.

There are very few applications where one can find methods to
computationally decide a thresshold, as opposed to some human
enetering it. The only application I can remember off the top of
my head, is Akaike's Information Criterion (AIC) and Rissanen
and Schwartz' Minimum Description Length (MDL) methods
for determining the order of AR predictors. There are other
methods for choosing these orders as well.

Nevertheless, the solutions from these types of methods are
still influenced by subjective choise, only on a level higher up.
Given the data, the numbers that come out of the AIC or
MDL order estimators are deterministic. But the user/designer
is still free to choose which of the AIC, MDL or other order
estimators to use. As the order estimators give different
answers with the same data, you are back to the dependency
on subjective choises.

Rune

 

Hi Rune, as i'm just a beginner to DSP, i really could not understand
most of your reply. I was thinking of doing "curve fitting" and getting
the equation of the curve. Do you think it will work? Or is there a
better way and simple way?

Thank you.