Speech can survive being flipped backward every twenty milliseconds. The same speech in backward pieces of 200 then 100 then 50 then 20 milliseconds. The original comes last. When do the words become clear? A listening demo from Oruk.

@OrukLabs First go round, I understood it at 20ms. Second go round, I understood it at 50ms.

@OrukLabs Better than me!

@OrukLabs This is dumb as a rock. That's because the smaller you go the more the individual pieces sound similar when reversed. Very smaht, maybe Yann LeCunn will find this mega interesting.

@OrukLabs isn't that simply because phonomes are generally significantly longer than 20/50ms, so you're causing discontinuities but not breaking the overall shape?

This is local time reversal: sound runs backward inside each slice, but the slices stay in order. Short slices can leave speech understandable. Longer ones disrupt more of the cues. There's no single cutoff for every listener.

Study: Human recording: We reconstructed 200/100/50/20-ms reversal from a short excerpt. This is a demonstration, not original trial data.

It was indeed not suitable

@OrukLabs the backwards part is irrelevant, could be almost any noise or silence. its like when you can see through a gap in a fence as you walk by. the fence doesnt have to represent the other side at all, you're just able to see/hear through the interference.

@OrukLabs It's not really about reversing the audio every 20mst though, is it? You could just insert any sound in 20ms increments inbetween the spoken sentence and it’s the same effect.

@OrukLabs I did this experiment in 1971 by physically splicing reel-reel tape. I cut the tape on phoneme boundaries instead of fixed intervals, however.

@OrukLabs makes sense

@OrukLabs first try?

@OrukLabs last😂

@OrukLabs But it’s pretty obvious. 20 milliseconds is short enough that often the sound him be quite constant over the duration. So reversing it has almost no effect

@OrukLabs 50ms first round is when I got it

@OrukLabs shorter ones lighten the clarity a bit more

@OrukLabs what's the minimum length?

@OrukLabs According to the video the 20 ms ones works but I think if you cpuld break it down to maybe like 10ms or 15ms it should be a bit more identifiable.

@OrukLabs the 200ms slice sounded german :D

@OrukLabs At 20 ms. Everyone else claiming they can hear at anything else for the first time is likely a liar. The people who claim they can hear it the second time with more reversal, either imagine it or have excellent audio processing in their brains.

@OrukLabs iff continuous, speech -> original as piece length -> 0

@OrukLabs