SOLVED: AudoSnap transient detection is still frustrating me

[dougalex]

In Cakewalk Sonar AudioSnap, the transient markers seem to detect all the obvious transients, and also also adds transients where no transient exists. If you try to eliminate non-existent transients using the threshold, then you lose a lot of the transients that are needed. It always seem to require a tedious time consuming manual process to get the transients correctly marked. How is that "I can detect transients properly" better than whatever algorithm AudioSnap is using?

Why would I have to go through do so many manual fixes (e.g. disable extra transients, insert missing transients).

Has anyone found any tricks to get reasonably good "transient detection" in AudioSnap?

Edited Wednesday at 12:11 PM by dougalex
"Soften" solved

[David Baay]

Depends on the situation and the goal.

If it's already in sync with the timeline, try using the Resolution setting as well as Threshold to disable false-positive markers.

If the material has a lot of sustained tones with relatively weak or 'smudged' attacks (e.g. strummed acoustic guitar) try gating a copy of the audio and using that for transient detection. EQing may also help.

If it's a mix and you just want to extract the beat, you might try using a stem separation tool (like the one in Next) to extract the drums and get the beat transients from that and Apply them to the mix.

 

[dougalex]

2 hours ago, David Baay said:

Depends on the situation and the goal.

I am testing AudioSnap by trying to quantize a melody played without click (a short mono synth lead with sharp attack, playing a clear repeating rhythm, mostly downbeats and some eighth notes).

I think some of the  key methods I am learning are:

1] Basic start: Align start of clip to beginning of measure and set tempo that is a good average, that hopefully at least works for the first bar or so. Turn on playback click. 

2] Resolution: Start with resolution "All"

3] Threshold: Given a choice between "having extra phantom transients", and "missing transients", go with "having extra phantom transients". (easier to see and disable extra transients, than it is to see and add missing transients)

4] Fix major drift: At the point where the free form playing starts to drift badly off the set average tempo, make a multiple selection of markers from there to the end, and move the set of markers to adjust for the drift. Listen forward and repeat that process of moving multiple markers forward or backward,  as necessary.

5] Quantizing: Quantize, and fix remaining notes that may have gotten got quantized to wrong beat, if any.

 

 

  

Edited Tuesday at 07:09 PM by dougalex
Numbering, spelling

[David Baay]

I'm sure you know that if it's a synth part, you would be better off just working with the MIDI, but I gather the synth audio is just for testing.

A couple of thoughts:

- If the timing is basically good and it's just about tightening up the slow drift in tempo from measure to measure, I would focus on just the down beats or at most the beat markers.

- After using the Threshold and/or Resolution to knock out the majority of false-positives, you can right-click the grayed out 'stubs' of disabled markers and re-enable them as necessary. If you're just doing this for downbeats, there shouldn't be that many. Similarly you can right-click and disable any markers that the Threshold didn't knock down. You can also Promote markers individually or in groups as you go so that that changes in the Threshold don't affect them.

 

 

[dougalex]

4 hours ago, David Baay said:

I'm sure you know

Thanks for the comments. I just started over and retried using the steps I distilled in my prior post, and I finally got that 8 bar synth phrase done very quickly.

It seems that my problem was that I was using a time resolution for the threshold... the eighth-note resolution, which would seem logical. HOWEVER, I found that was what made impossible to get the expected results when fine tuning the threshold. It seems that "All" is the way to go, and the other time resolutions only work well if the original material was played very close to a constant tempo (e.g. played listening to a click).

Overall, I am just trying to take the necessary baby steps, and learn this AudioSnap tool because, when I "have it down", I very curious to try and hear what it sounds like when used on a multitrack live recording that was NOT played to a click... i.e. find out what is really possible and practical.

Thank you for your input!

[David Baay]

15 hours ago, dougalex said:

he other time resolutions only work well if the original material was played very close to a constant tempo (e.g. played listening to a click).

It will work much better if the project tempo is matching the averge tempo of the clip. To achieve this, count out to 8:01, snap the Now time to that transient, and use Set Measure/Beat At Now (Shift+M) to set that measreand beat which will reset the initial tempo to the average of the clip.