overlapping acquired audio dataStarted by 5 years ago●3 replies●latest reply 5 years ago●245 views
I was just wondering if is there any benefit on windowing and overlap an acquired audio frame if i don't have to perform spectral analysis?
In other words, if I'm going to acquire audio frames (let's say of 256 sample) from a sound card and i'm going to perform some operations in time domain (like gain and compression), there will be any difference if I use "rectangular windowed" audio frames versus windowed&overlapped audio frames (besides changes in the actual sampling time)?
thanks in advance for any insight.
Say you make the gain on frame one different from frame two, which produces a jump between them. You probably will want a way to smooth-out this discontinuity. A typical method is to use overlapping frames, then gradually blend one into the other. Some authors use "windows" in describing how this blending can be accomplished. But most of it is just common sense and you can understand it without referring to "windows." The other stuff you will see written about windows (Spectral Leakage, Spectral Smoothing, etc,) doesn't apply to your case.
What analysis are you looking at, if not spectral that would require windowing ? Windowing is pretty much a spectral processing thing (FFT, wavelet, Gabor, etc).
You can ignore the rest of this ... I thought a simple spectral example might still be helpful.
Windowing will reduce wideband noise at the expense of "spreading" frequency content across adjacent bins. For example here are a couple of 2-D spectrograph pics showing a 1024 pt sawtooth wave with / without windowing. The first shows a 1024 pt FFT on each 256 pt frame, with a Hanning window and 50% overlap applied. The second has the same FFT/frame parameters but with a Rect window and no overlap. As you can see, in the 2nd spectrograph frequencies (main + harmonics) are sharply defined (to extent a 1024 pt FFT will allow), but there is considerable noise from the rectangular window "edges". For each frame, the FFT is seeing those edges as waveform discontinuities.
Yes, if the changes are sufficiently abrupt to warrant it, you could for example use triangular windows with a 50% overlap (to keep it simple). You would then of course have to process twice as many samples compared with the case without overlapping.
However, if you can control the rate of change between frames then you would not need this complexity at all.