Re: [SoX-users] Generating multiple spectrograms

[Ulrich K. <ul...@ch...>]

Peter Shute <ps...@nu...>:

> I need to generate a spectrogram for some wav files, but sox only  
> allows a maximum spectrogram width of 5000 pixels. The horizontal  
> scale I need means that this will only be good for about 3 minutes,  
> but my files are much longer than that.

> I've been generating spectrograms with commands like:
> sox test.wav -r 16k -n rate remix 1 spectrogram -S 0:00 -d 3:00 -m  
> -l -x 5000 -y 642 -o spectrogram1.png

> (I've cobbled those together from examples I've found. The intention  
> is to resample the 44.1kHz file to 16kHz because I only want the  
> spectrograms to show up to 8kHz, and I only want the left channel.  
> Are my commands correct?)

Generally, yes, but it would be more efficient to
- use "trim" to extract the relevant part (as you already suspected),  
as in "trim 0:00 3:00",
- reverse the order of remix and rate (no need to resample the  
channels that are discarded anyway), and
- unless you absolutely need a height of exactly 642 pixels, go to  
either 513 (= 2^9+1) or 1025 (= 2^10+1).
So a faster version is:
sox test.wav -n trim 0:00 3:00 remix 1 rate 16k spectrogram -m -l -x  
5000 -y 1023 -o spectrogram1.png
(The last point is the most important. In an example I ran, -y 642  
took 110 seconds, while -y 1025 only took 9. And I don't mean 109, but  
9.)

> I'll join these images together to form one long one.
> I.e. I have to manually create a command per 3 minute segment. Is  
> there a better way? ... Can I use the newfile and restart effects to  
> do it in one command without generating intermediate files?

Not directly, because the spectrogram output will use the same file  
name each time, overwriting the previous output. (You could write to a  
named pipe, though, or something like that.)

Probably the easiest way is to lift the width limit of 5000 pixels, as  
it appears to be arbitrary. I don't know if there is a real limit  
(imposed by PNG/libpng perhaps?), but I changed it to 20000 in the  
source code and that seems to have worked. How many pixels do you need?

Ulrich