Regular audio artifacts with wav to wav export

[mltframework]

Regular audio artifacts with wav to wav export

[ddennedy]

See also https://sourceforge.net/p/mlt/bugs/98/
A patch for this is welcomed.

[ddennedy]

Since you know more quickly how to analyze the problem (I have bad hearing and cannot hear the glitch), can you please make a quick test? Edit the .mlt XML file to remove lines from <transition> through </transition> and re-render it. That will help narrow down some things. Also, I see you are upsampling: 44.1 -> 48 KHz. Another quick test is to output 44.1 KHz to see if the problem goes away or is different: add "frequency=44100" somewhere after "-consumer" in the render script.

Once you do that, there is another couple of tests. Where MLT is installed, first make a backup copy of $prefix/share/mlt/core/loader.ini. Next, comment out the audio filter lines, one at a time: "channels=..." and "resampler=..." to see if that eliminates the problem.

With all of that done, it almost a pass-through as much as possible. However, I just noticed another thing. Your input audio is 32-bit floating point while the output is 16-bit signed integer. It is not enough to simply set the audio output sample format: "sample_fmt=flt" somewhere after "-consumer" in the render script. You must also change MLT's internal format: add "mlt_audio_format=float" somewhere after "-consumer".

[hamoid]

Hi @ddennedy Thank you for the quick reply! The glitch is only audible with good headphones. I don't notice it with laptop speakers.

I tried the first thing you suggested (removing the transition) and it seems to have fixed the issue! I can't hear it nor see it on the spectrogram.

I wonder why is Kdenlive creating that transition. If you open my Kdenlive project you can see there's really nothing. No effects or transitions or multiple tracks. Just one track with the wav at the beginning.

Should I still try the other things you suggested? I don't know where to finde the "-consumer" you mentioned.

[ddennedy]

Closing this as will not fix. This is a side effect of using combine=1 with the mix transition. This was an attempt to provide a mixing algorithm with a simple, automatic approach to prevent attenuation of levels, which turns out to be a difficult problem in DSP. This simple approach inherently introduces artifacts. The workaround is to not use combine=1 and compensate for undesired attenuation levels with audio dynamics filters. Shotcut is moving away from using combine=1.