Import a voice recording (of yourself or something found, say on Internet Archive) into an audio editor, and using only the scissors/splice tool, drag and drop, and copy/paste, play with parsing it over a timeline, slicing it and moving the pieces around.
Some useful verbs are disperse, repeat, reverse, re-order, restore.
Perhaps play with the transition from the speaker speaking normally (where the fact of the recording disappears to a degree into the communicative function of speech) to the arrangement of the speech in more or less musical or rhythmic ways (so the fact of the recording foregrounds itself and the communicative flow recedes into the background).
As an additional layer, add a track with underscoring or ambient sound, and play with the ways that sound layer makes its own transition (and from what to what? ) in relation to the shifting function of the speech. For example, something rhythmic being looped under normally progressing speech might strike your ear as background, but if, in parsing the speech, you place the little speech samples in a rhythmic relation to the loops, then the loops come forward to become controlling or stabilizing elements.