SUMMARY

STEPS TO REPRODUCE
1. Import an AV clip from Camera (here e.g.: *.MTS; AVHCD; Full-HD 50 fps + "original" Audio 48kHz stereo)
2. Import a separately recorded "new" audio file (here: e.g. 44.1kHz or 48kHz stereo AIF)
3. Try to manually sync-align the "new" audio file with the "original" audio track by looking at the displayed waveforms in the timeline.

(Maybe cut out some unneeded sceenes from both sources, and/or maybe try to compensate for slightly different sample clocks or for an initial problem in the waveform display generator that may require manual alignment of the two tracks further down.)

(If you want me to, I can also try it again using an aiff copy of the original camera soundtrack as "new" audio file. My error report does *not* refer to problems resulting from different sample frequencies or unsynchronized sample clocks on different recorders.)

OBSERVED RESULT
After a few minutes, the waveform displayed in the timeline preview for the "new" audio file will be out of sync with its actual audio content.
When I go e.g. 5 minutes into the material, and try to align the "new" audio track to the "original" audio by looking at both waveforms in the timeline at that position, and then play them both together at that position - I'll get a distinct echo or delay effect - even though the two displayed waveforms may be nicely aligned. The offset between displayed waveform and actually played audio content may exceed a few tenths of a second, i.e. a total desynchronization between piano playing video and audio.
-
Thus, it is rather impossible to use the waveform display to align separately recorded audio tracks to camera-recorded audio/video.
-
This is true, no matter whether the two audio tracks use the same or a different sample rate.

EXPECTED RESULT
The waveform display in the timeline should accurately represent the audio in this track at any position down into the audio material.
Thus, when aligning two audio tracks by looking at their waveform previews at some given position in the timeline, the preview-played and the rendered audio of these two tracks at that position in the timeline should also be aligned.

SOFTWARE/OS VERSIONS
Windows: 
macOS: 
Linux/KDE Plasma: kdenlive 18.x; 19.x; 20.07 
(available in About System)
KDE Plasma Version:
KDE Frameworks Version:
Qt Version:

ADDITIONAL INFORMATION
I suspect that the waveform display uses a waveform preview generator that goes out of sync with the actual waveform content - OR some built in synchronization that ties camera recorded audio to the video frames is unavailable for separate audio tracks, so that audio positions from the timeline do not refer to the expected position in the audio file. Possibly because of some integer/rounding problem or the like that adds up after processing a large number of samples.
-
I know that kdenlive has some master/slave/audio-auto-align functionality, but I have not succeeded in trying that out. I actually doubt that this will work when I only want to retain multiple individual scenes from the original recording, resulting in multiple audio-and-video snippets to be aligned and then crossfaded.
-
Additional problem, less severe: kdenlive apparently allows positioning of an audio clip only in full frame increments.
-
Additional problem: When GPU acceleration is *dis*abled, the preview playback of Full HD 50 fps video is so slow (on an older i7 cpu) that it is impossible to judge whether Audio/Video are in sync. Additionally, simple audio fade-in/fade-out effects are lost. - When GPU acceleration is *en*abled, preview play is sufficiently fast, and fade-in fade-out are there - but: *rendering* a clip will never finish. And transition between both settings is ... difficult.
-
Therefore, it's almost impossible (for me...) to replace camera recorded audio with externally recorded audio tracks in kdenlive. Simply, because the waveform displays are unreliable, and even quickly changing between a working video preview and working test renderings is impractical.
-
I've tested that with various versions of kdenlive on various machines, and even converted audio recordings externally to the same sample rate etc. It's a very sad restriction, because otherwise, kdenlive looks rather nice.
-
Simply searching for "waveform" in the kdenlive bug tracker returns a few errors that *might* be related. But their status is mostly "repo" (??) and/or they are old, or might describe different problems - so I don't know whether this problem is being handled. I can provide any additional testing if that is required.
-
Kind regards and thank you for looking into this - and for the otherwise very interesting work with kdenlive - js