Detecting audio volume for creating waveform

Question

Hello all,
I need to display waveform from a sound file.
I know to use Graphics in java for drawing lines, etc
But, I don't know how to get the rate volume of a sound (in miliseconds) for creating a text log looks like:
23,40,54,32,2,43,4,45With these numbers (coordinates) I will draw a waveform.
In conclusion: How can I draw a waveform from a sound file in java?

Answer 1

sabre150 wrote:
AndrewThompson64 wrote:
sabre150 wrote:
..The magnitude is easy ( Math.abs() ) and the running average is relatively easy.. It is as difficult as it is useful. (I.E. not difficult and not useful).I disagree - I have done it this way for 40+ years. The average of some function of a time series is the average taken over the last N values. The running average is the application of this continuously so the running average at time 't' is is the average taken over the last N samples (t,t-1,t-2 etc) so this has a smoothing effect on the time series. Whether one use abs() on each value or any other function does not change the concept. Maybe not the concept, but certainly the end 'volume' number. If you do not use the abs value, the end volume (in the situation I outlined, where the buffer size covers a single wavelength) would be 0 for sawtooth wave spanning -1 <-> 1, -0.5 <-> 0.5 or -2 <-> 2.
..For what the OP is trying to do, square() is probably better than abs() but that is for the OP to decide.
Further consideration is normally given as to whether or not to do some band pass filtering of the time series prior to taking the running average. I would do so but the OP may decide not to. You lost me with the band pass filtering. I thought that required Fourier transforms, and whenever I have looked up FFTs on the net, the formulae made my eyes cross.
As to the the calculation of running average. All one needs is a ring buffer containing the last N sample values. Uh-huh. I'm with you there. A buffer of values is also required for RMS calculations.
..Care must be taken to stop the build up of rounding errors. ..O..K I was not even considering built up rounding errors when I commented.
..An approach to this is to recalculate the mean of the ring buffer.. 'ring buffer'? The group of the last 'n' instantaneous values?
(quick Google later, confirming same)
Um.. OK. Forget I ever writ that and pretend I knew it all along. ( Thanks ;)
..from scratch every now and again but there are many many approaches.
..(the incremental approach can be used where for each sample that arrives one adds it in to the average and removes the oldest from the average). ..Let us assume, for simplicity, that the signal is a sawtooth wave that spans from -1 to 1. The average when you happen to be keeping as many samples as fit in a single wavelength will be exactly 0.5. Assuming that you mean the average of the abs() then yes but the average of the unprocessed signal will be zero and not 0.5.
Now shift the sawtooth signal to span from 0 to 2 and the average is exactly 1, even though the end user would hear exactly the same signal and volume as the original signal that spanned from -1 to 1.Sorry but I don't see your point. My point was that to an end user, listening to the output, both sawtooth waves would sound identical, indistinguishable, yet the values for the average of the absolute values would be different, while the RMS values would be ..identical.
Compare that to the [RMS volume|http://en.wikipedia.org/wiki/Root_mean_square]. The RMS value would be identical for the -1 <-> 1 signal, and the 0 <-> 2 signal (both values greater than 0).
Now you have really lost me. I understand 'mean' . I understand 'root mean square'. Both are 101 concepts. Are you just trying to say that one must perform high pass filtering first (i.e. get rid of the mean prior taking the abs() or square() or whatever)? Again. Huh? I was not ever considering any form of filtering.
..With sound recordings this occurs naturally since microphones tend not to be able to measure absolute pressure but I would certainly filter out anything below the standard human low frequency cut-off point.Um.. OK.
I pointed to some code earlier. That code is used in renderings of sound signals. Its primary purpose is to plot the audio signal itself (real perty like), but I was also interested in offering left/right channel 'volume'. After much successful research (and unsuccessful experimentation), I settled on the RMS as the way to calculate the instantaneous volume.
You can view the results in both Tracker (designed for individual tracks compatible with Java Sound as well as MP3s) and DukeBox (that is designed to parse/play M3U playlists of JavaSound + MP3 playlists). On the lower left/right of the sound trace rendering, often 'visually boosted' (since RMS values tend to be low, there is an option to display the square root of the actual calculated RMS volume, measured from 0-1, which is the default for most of the the preconfigured renderings) in a real time situation.
I mention this only for the fact that the OP can use either of those apps. to see the end effect of the RMS style calculation.
..Where did the OP go, BTW?