Updating Sound/Silence Finder

[steve]

It does give these false positives in music with "railroad track" silences (Doo Wop, Concert Band etc.) and, unlike Audacity, there is no user control of the variables.

I would guess that it uses something like "less than -30 dB for more than 2 seconds".
I have a cassette player/recorder that uses something like that for counting tracks.

[steve]

Looking through the suggestions, here's version 0.2.
We do not need to implement all of these features and we may want to add others or make other changes, but here are all of the suggestions so far in one graphical interface:

All features

v02.png (42.34 KiB) Viewed 2165 times

A breakdown of these features to follow....

[steve]

The original "basic" version 0.1:

v01.png (22.58 KiB) Viewed 2167 times

Given that the current new behavior in the Sound finding mode is to place a Point Label and the old behavior was to place a Region Label, and given that both of these behaviors might be useful, I think it needs a switch to allow the user to choose.

I did not like the original Silence Finder's behavior of placing the label at a specified distance in front of the end of the silence; I rewrote the effect in C++ and have it so that it places the label in the center of the silence. I think the user should have this as an option as well.

Version 0.1a provides a choice menu for the type of label. The first three options produce point labels at the start, middle or end of the detected region respectively. The 4th option creates a region label.

If "Sounds" are selected in the first menu, then point labels will be at the start, middle or end of the detected sounds. Region labels will encompass the detected sounds.

If "Silences" are selected in the first menu, then point labels will be at the start, middle or end of the detected silences. Region labels will encompass the detected silences.

v01a.png (30.19 KiB) Viewed 2167 times

[steve]

So far, no one has mentioned that version 0.1 creates far more labels than the Silence Finder or Sound Finder.

I did not tested with real-world audio. We have a piece of hardware in the studio which automatically detects the silence between tracks on a tape, record, CD or other input and splits it into separate tracks burned onto a CD;

Specifying the duration of sounds/silences is unfortunately a necessary evil.
A minimum duration of sound/silence must be specified in order to avoid labelling every zero crossing point (which could be hundreds of times per second).

This could be a fixed amount, probably somewhere between 0.5 seconds and 4 seconds, in which case there is no GUI element, or it could be a user option:

v01b.png (26.52 KiB) Viewed 2134 times

There may need to be some discussion and testing of the way that this works, but at present the way that I would envisage it is that a sound or silent region only "count" as a "regions" if they are have the specified minimum duration. In effect this is a "time threshold" much like the "level threshold".

[steve]

Silence Finder: Place a Point Label a specified distance in front of the "end" of the silence.

Strange isn't it
Let me read that again;
"A specified distance in front of the end of the silence" -
In other words, "a specified distance before the sound"?

I've always presumed it was done that way to provide a lead-in ( 2 secs ? ) prior to the "song" starting .

So we want a way to "adjust" the position of the labels relative to the detected region.
This is where it starts to become a bit tricky. We can't really just have "before end of silence" because this effect may be detecting sounds or silences.

Inspiration for terminology came from the world of "audio recording" and "audio processing" (quite appropriately )
A few terms came to mind such as "pre-roll", but I think that is quite hard to grasp in terms of labels.
Particularly in audio processing we have the term "lookahead".

"Lookahead" is a period of anticipation before an audio event happens. For example in a Noise Gate, If there is a "lookahead" of 10 milliseconds, then that means that the gate will open or close 10 milliseconds before the audio crosses the threshold level. The same thing could be applied to labels. If placing a point label at the end of a silence, then "lookahead" would determine how much before the end of the silence the label would be placed.

Here the selected region shows the "lookahead" when marking a sound with a region label.

lookahead

lookahead.png (13.65 KiB) Viewed 2138 times

Thinking of a complementary word for the period "after" an event, brought to mind the word "overshoot".
here is a point label at the "start" of a silence with an "overshoot" of 0.3 seconds:

overshoot

overshoot.png (9.83 KiB) Viewed 2138 times

Lookahead and Overshoot controls

lookahead-overshoot.png (29.25 KiB) Viewed 2138 times

[Edgar]

I am +1 on this new GUI; "look ahead" and "overshoot" work fine for me as long as I also have "center" for Silence Finder. Just so I could test this new effect I picked up a few hundred pieces of good quality vinyl in two genres which were almost completely unrepresented in my library – doo-wop and classic country. Their conditions range from excellent to fair so I will be putting Audacity to work on these. The last two days I processed 10 of these, a random selection of excellent and fair; these are now sitting on my hard drive is raw audio cleaned up and ready to be split into tracks. Eagerly awaiting the next version…

[steve]

I am +1 on this new GUI; "look ahead" and "overshoot" work fine for me as long as I also have "center" for Silence Finder.

To clarify, what I had in mind was that:

• Mark Start: (point label) will be set "early" (before the start of the detected sound/silence) by "Lookahead" (seconds).
• Mark Middle: (point label) is always dead centre of the sound or silence.
• Mark End: (point label) will be set "late" (after the end of the detected sound/silence) by "Overshoot" (seconds).
• Region Labels: The start of the labelled region will be set "early" by "Lookahead" and the end of the labelled region will be set "late" by "Overshoot".

This means that:
a) There is no way to "adjust" the position of point labels when using "Mark Middle".
b) Region labels are always at least as big as the detected sound/silence, and the sound/silence is entirely "within" the range of the label.

@Edgar: are you happy with regards the proposed behaviour for marking the middle of silences?

[Gale Andrews]

Can you post links to the old topics discussing Sound Finder and Silence Finder?

I seem to recall we got a long way with a new Sound Finder. I didn't think that was a complete dead end?


Gale

[steve]

This was (one of?) the previous thread(s): http://forum.audacityteam.org/viewtopic ... 44&t=65678

[steve]

BTW, I have an idea that may dramatically improve "false positives". This new feature (if it works as I hope it will) requires no user facing changes, it's just a small change to the detection algorithm. When I get the programming right I'll post a demo version. Hopefully it will cause few surprises for users, other than "working better".