Kdenlive "Align Sound To Ref" nails it (almost)

I built a 4 minutes clip with the text to speech Apple say command. The voice continually counts up passing time like "1 hour 23 minutes 23 secondes 1 hour 23 minutes 24 seconds 1hour 23 minutes 25 seconds". Kdenlive succeeded to place a 5 secondes extract at almost at the right point:

The position is off be a mere 9 ms. Next experiment: adding more and more noise. Addendum: I changed the depth from 16 to a more noisy 8 bits and it still works! (and gives the same offset of 441 samples @22050 sps).

8 bit is great: I could play synthesized clips ON the SAMD21...