It is clear that the playback software shifts the pitch downward so it matches the 1x pitch, but there's still some "artifacts" that show up when you go beyond 1.5x.
Is there a precise word for this kind of distortion? besides "choppy"?
Also, I wonder how far you could get in time compression by merely removing spaces between words without actually compressing the words?
That is interesting, but a good speaker uses spacing to better convey their message. In order for it to work, it would have to somehow take into account rhythm. Perhaps just cutting longer spaces instead of spaces between words. Maybe even starting with the creator, or users tagging unnecessary bits such as coughing, or reaching for a drink.
And here's the thing I realised earlier - most interesting podcasts have usually two speakers, discussing something. Not much room for rehearsal and/or retakes I guess.
Some audio time-stretching libraries do exactly that, reducing the amount of samples in silent periods compared to samples of non-silent audio.
It's a really difficult problem and actually all the on-the-fly things don't do it really well (at some speed vocals blur in to consonants and consonants get clipped. good software that has no simple FIFO algo can do it much better (conserving consonants etc.).
This is Overcast's USP.