I think the thing that caught me most off guard was an idea that came out of all the psychoacoustics reading I do. Most phonemes in the English language have a length somewhere between 100 and 500 ms. I found that you can slice up a piece of dialog into segments longer than 100ms, batch reverse them and it will be completely unintelligible. You can still identify it as a voice, but no one will understand it. It’s great because it doesn’t just sound exactly like a piece of dialog that’s been reversed. The individual slices change the cadence a bit.
The threshold of comprehension, where it sounds weird but can still be intelligible is somewhere between 50 and 100 ms. It all depends on the speaker, the listener, and the content of the lines. The smaller the slices, the better understood it will be.
Whether you’re going for recognizable voice with no intelligible dialog, or just a weird vocal effect, that slice and reverse can be a great starting point! Here’s an example from a presentation I did, large slices to small (400ms, 200ms, 100ms, 50ms, and 25ms followed by the original unprocessed recording).
Posting this again because I got a database error, apologies if it ends up coming through twice.
I love the way some animals sound when they’re reversed, very other-worldly and creepy.
I also like how different explosions, fireworks and gun-shots sound when turned backwards, making great build-ups and whooshes.
I’ll have to try that slicing trick too. Should be easy with Reaper being as you can split items at grid, reverse them and then you could even randomize their order to disrupt the cadence even further.
Something that worked very well in a dream sequence I worked on. The sound supervisor paused playback and asked ‘this low sound, it is beautiful…. what is it?’
Low rumbling from a retreating thunderstorm (without the strike/clap) reversed, you get a sense of danger or threat but you can’t place it. And the subtle movements/tremblings are so weird and organic that it fits very well with other sounds.