Sound/Silence Finder enhancements

[steve]

Thanks for the comments Peter.

@PGA

Please don’t take this personally but, isn’t that taking a rather narrow view of things? “Recordings” might not have instantaneous sound but use of Audacity isn’t restricted to recording. Audacity is also used extensively to edit sound files. If I were to carefully edit a sound file so that it began with a waveform peak, would that not be “instantaneous sound”?

That’s what I meant by “not so illogical”…
The obvious “logical” response to the statement that “all sounds start with silences” is “but what about sounds that do start at the very beginning?”. Points (a) and (b) were intended to explain why the statement is not as illogical as it may first appear;

(a) This one is pretty simple. Most recordings start at or near silence. This is by far the most common user case for Silence Finder.

(b) This is the tricky one that falls into the category of “ALL reasonably foreseeable eventualities”.
One way of looking at this is to consider the physics of loudspeakers. When you start playing a sound, no matter how rapidly the sound starts, the speaker cone accelerates from stationary. There is no sound from a stationary speaker cone, so it is accurate to say that the sound starts from silence.

OK, but isn’t that just the limitations of clunky mechanics? I’m looking at the waveform, not at a device for pushing air around.

Another way of looking at it is from the mathematics of sound. Sound is not a matter of “speed” but of “acceleration”. If a sound begins suddenly at time zero (the start of the selection) then there is an enormous acceleration from the first sample to the second sample. You can see this in the spectrogram track view. If you create a split in a sine wave, a red vertical line appears. This is because the split sets a “start” point for the sound. The “sound” is starting from nothing, so there is a huge acceleration from the start point. This happens even if the split is exactly on a zero crossing point.

Both speed and acceleration can only be relative to some reference point (and according to Einstein this applies to everything else as well). In our case the reference point is “time zero”, the start of the selection (or the start of the track if the entire track is selected). At time zero there is no speed. Whether we measure speed in km/h or m/s or samples per second or any other units speed is defined as “distance over time”. During the first sample period (the smallest unit of time in digital audio) there is only one fixed sample value so the rate of change during the first sample period is zero. In other words that first sample period is always silence.

OK so that’s fine for the digital world, but what about the real (analogue) world? In the real world there are no “minimum units of time” (at least not before the quantum level) so doesn’t that change everything? No it doesn’t, because then we are back into the realms of dropping needles onto spinning grooves and making speaker cones vibrate from a standstill.

No matter how suddenly a sound starts it is reasonable to think of it as starting from silence, but more importantly it is very useful (in this case) to think of it in this way. It means that no matter how rapidly the sound starts, whether there is three minutes of silence, or 0.000010417 seconds of silence, we can treat all cases in the same consistent way. It enables us to handle ALL reasonably foreseeable eventualities. Even more important than that, it means that when the tool is being used (as is it’s intended purpose) to split a recording into tracks, the first track does not get missed, even if it starts “instantly”.

[waxcylinder]

@Steve - am I right in assuming that this thread does not need transferring to Wiki>Pending Feature Requests?

Have you got all you need from it? Should I delete this thread or do you want it transferring elsewher on the forum (Audio Processing say) for retention and reference?

Thanks,
Peter.

[steve]

Topic moved to the Nyquist plug-in archive section.

[Gale_Andrews]

@steve

Just a thought, but if we have a “Silence Marker” that always marks within detected silences, how about if the label placement (currently set at x seconds before the end of the silence) was set as a percentage of the silence?

Thus,
50% = middle of the silence,
10% = near the start of the silence
100% = as close as possible to the end of the silence.

Considerations:

• Unless the silences are all about the same time, the distance before the next sound will not be constant. Does this matter? Is this a disadvantage or an advantage?

I think it is probably a disadvantage for understanding, particularly as there would always be a label before the first song even when you could not “see” any silence. And almost certainly there will be people who just simply want [x] seconds as a fixed “lead in” to the next song.

I still suspect “[x] seconds before the end of the silence” is more intuitive than “retain [x] seconds of silence after the preceding sound” (though that is more intuitive for the final label).

Can you “limit” the “[x] seconds before the end of the silence” as “up to [x] seconds before the end of the silence”, so the label is constrained from going into the preceding song? I think that would be understandable by most.



Gale