|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
65.19.76.104
In Reply to: RE: -60dB posted by Artoa on July 12, 2014 at 17:39:52
Normally, I use a 65K FFT for spectrum averaging. But my software allows other sizes as well, plus different windows. (Soundforge 10). I also have iZotope RX and this includes a spectrum plot of time (X) vs. Frequency (Y) vs. intensity (Z, i.e. color). The size of the temporal resolution can be adjusted but there is the inescapable tradeoff of time resolution vs. frequency resolution.
With these spectrum plots it is very easy to see high frequency noise and to distinguish it from high frequency musical sounds. In addition, one can edit the spectral plots to easily remove unwanted noises, even to the point of removing notes out of a chord or removing guitar string squeaks.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
Follow Ups:
Unlikely your software will allow to do a FFT with 60*96,000 points. There's a bunch of averaging going on that's a function on how the analysis is setup (number of points in the window, window overlap,...) The loss of frequency resolution that comes with shorter windows does not matter since energy gets summed up in each frequency band. A delta f of 1Hz or of 10Hz will still tell the story of whether there's significant energy at higher frequencies.
It looks like you posted in the wrong place.
I agree that it's unlikely that Abe Collin's FFT averages over 60 seconds. Possible, but not likely. I know that my FFT averages over 65K points with my usual settings and this corresponds to about 2/3 of a second at 96 kHz. The points in the window are weighed with the central points counting more according to a formula that smooths out the plot. I generally use a Blackman-Harris window.
There are a lot of other issues involved with interpreting FFT plots, such as "FFT gain" which is a measure of how the noise floor varies in FFT plots according to the window size and window type. If Abe wanted to, he could calibrate his FFT gain by creating a test file with a 28 kHz tone and adjust the level of this tone to where the FFT shows it at the point on the suspect download file. A little knowledge can be a dangerous thing, but I'm confident that Abe Collins knows enough to figure these things out, if he hasn't already done so.
BTW, I have other software that I use that uses 8 million points. I've used this to do convolution by transferring to the frequency domain and multiplying. In addition, my player routinely does convolution at 2.8 MHz sampling rate with a time period of about 1 second (over 2,000,000 samples). I presume the software uses FFTs, to avoid the necessity of more than 10^14 multiplications per second using a simple time domain approach, which even if my hardware could do would run up a huge electric bill.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
Post a Followup:
FAQ |
Post a Message! |
Forgot Password? |
|
||||||||||||||
|
This post is made possible by the generous support of people like you and our sponsors: