Freesound Forums

Hello there,
I don't know if this is in the scope of possibilities.
But an interesting feature would be a signal to noise ratio filter. Idea being each uploaded sound would be analysed for noise level and signal level, and if the noise is under a certain threshold it would be marked as a low-noise sound.

I would presume this is complicated to do as there's probably some blurry lines between noise and a river recording, or just the sound of wind going though trees...

But for some applications such as dialog, music instrument samples, some SFX etc... I think it would be a useful feature!

Many thanks for your time and hearing me out.

I don't believe there is any general solution for how to automatically determine which parts of a sound are "signal" and which are "noise". Consider, for example, a sound that is intentionally a recording of, say, the wind noise of a flute. It's noise. It's signal.

Hi,

This is an interesting question. We've been doing some research around this idea, but were not able to define a reliable way to separate the concepts of signal and noise. But the main problem is to agree on the definitions, not necessarily the technical solution which should be feasible. We also thought about simple/complex sound filters which could have a similar effect, or "foreground to background" ratio, but all definitions have conceptual problems at some point. I guess we could eventually come up with a definition that works with *some* sounds and assume this is good enough.

One could avoid having to decide what is signal and what is noise and do something like dynamic range, where get the rms in say 100ms chunks and then call the dynamic range the ratio of highest_rms/lowest_rms. (The dynamic range would be a lower limit on the signal-to-noise)

I'm not sure that would be useful though - I suspect one cannot really get away from the fact that there is no way of deciding what is signal and what is noise (or even that one woman's
signal is another man's noise!)