log energy of audio frame

Started by adamchapman June 11, 2009
Hi,

sorry to post 2 questions in the same day but Ive been searching all
day and got nowhere.

Im trying to compute the log energy of an audio frame. I am aware that
the result of my calculation should be a sinle value for the whole
audio frame, so it is not the log of the spectrum.

I know that to find the energy of an frame you get the frame from the
audio stream with a window function, then sum the square of the
windowed samples. Now for log energy do you:

1. take the log of that sum

or

2. log the coeffieients before summing them?

Sorry if this question seems daft, I'm currently converting from
iomage processing into audio work
adamchapman wrote:
> Hi, > > sorry to post 2 questions in the same day but Ive been searching all > day and got nowhere. > > Im trying to compute the log energy of an audio frame. I am aware that > the result of my calculation should be a sinle value for the whole > audio frame, so it is not the log of the spectrum. > > I know that to find the energy of an frame you get the frame from the > audio stream with a window function, then sum the square of the > windowed samples. Now for log energy do you: > > 1. take the log of that sum > > or > > 2. log the coeffieients before summing them? > > Sorry if this question seems daft, I'm currently converting from > iomage processing into audio work
Energy is related to power, which is not amplitude ( but is related to amplitude ). I think what you want to do is take the RMS of the window of samples, then log that, but.. to be honest, I am not sure. It depends on how the samples are to be expressed. I think there may be a dBm versus dBV problem with your posting. http://en.wikipedia.org/wiki/DBm vs http://en.wikipedia.org/wiki/Decibel#Voltage But you really need to figure out how you're going to test this first. You really need to know what the boss wants. I would ask him or whoever is wearing the systems engineer shoes. If you can't do that.... then wow. Mind your savings. Since it's audio, if you're Tcl literate, you can download Snack and use it as a basis of comparison. But this says nothing about how you are going to sell off your implementation to the people you work with. -- Les Cargill
On Jun 11, 10:43 am, adamchapman <adamchapman1...@hotmail.co.uk>
wrote:
> Hi, > > sorry to post 2 questions in the same day but Ive been searching all > day and got nowhere. > > Im trying to compute the log energy of an audio frame. I am aware that > the result of my calculation should be a sinle value for the whole > audio frame, so it is not the log of the spectrum. > > I know that to find the energy of an frame you get the frame from the > audio stream with a window function, then sum the square of the > windowed samples. Now for log energy do you: > > 1. take the log of that sum > > or > > 2. log the coeffieients before summing them? > > Sorry if this question seems daft, I'm currently converting from > iomage processing into audio work
Just work out the average of the sum of squares - this is Power and then take logs. Hardy
On Jun 11, 10:43 am, adamchapman <adamchapman1...@hotmail.co.uk>
wrote:
> Hi, > > sorry to post 2 questions in the same day but Ive been searching all > day and got nowhere. > > Im trying to compute the log energy of an audio frame. I am aware that > the result of my calculation should be a sinle value for the whole > audio frame, so it is not the log of the spectrum. > > I know that to find the energy of an frame you get the frame from the > audio stream with a window function, then sum the square of the > windowed samples. Now for log energy do you: > > 1. take the log of that sum > > or > > 2. log the coeffieients before summing them? > > Sorry if this question seems daft, I'm currently converting from > iomage processing into audio work
Just work out the average of the sum of squares - this is Power and then take logs. Hardy
adamchapman wrote:
> Hi, > > sorry to post 2 questions in the same day but Ive been searching all > day and got nowhere. > > Im trying to compute the log energy of an audio frame. I am aware that > the result of my calculation should be a sinle value for the whole > audio frame, so it is not the log of the spectrum. > > I know that to find the energy of an frame you get the frame from the > audio stream with a window function, then sum the square of the > windowed samples. Now for log energy do you: > > 1. take the log of that sum > > or > > 2. log the coeffieients before summing them? > > Sorry if this question seems daft, I'm currently converting from > iomage processing into audio work
Energy is related to power, which is not amplitude ( but is related to amplitude ). I think what you want to do is take the RMS of the window of samples, then log that, but.. to be honest, I am not sure. It depends on how the samples are to be expressed. I think there may be a dBm versus dBV problem with your posting. http://en.wikipedia.org/wiki/DBm vs http://en.wikipedia.org/wiki/Decibel#Voltage But you really need to figure out how you're going to test this first. You really need to know what the boss wants. I would ask him or whoever is wearing the systems engineer shoes. If you can't do that.... then wow. Mind your savings. Since it's audio, if you're Tcl literate, you can download Snack and use it as a basis of comparison. But this says nothing about how you are going to sell off your implementation to the people you work with. -- Les Cargill
Hi,

sorry to post 2 questions in the same day but Ive been searching all
day and got nowhere.

Im trying to compute the log energy of an audio frame. I am aware that
the result of my calculation should be a sinle value for the whole
audio frame, so it is not the log of the spectrum.

I know that to find the energy of an frame you get the frame from the
audio stream with a window function, then sum the square of the
windowed samples. Now for log energy do you:

1. take the log of that sum

or

2. log the coeffieients before summing them?

Sorry if this question seems daft, I'm currently converting from
iomage processing into audio work