Slow learner audio is comfortable for me now, and I can read the same material without much trouble. Put on a normal conversation and half the word boundaries disappear.
I do not think the issue is simply vocabulary. If I read a transcript afterward, I usually know most of the words. What I missed in the audio is that familiar words get shortened, linked together or pronounced differently from the neat version in my head. By the time I identify one phrase, the speaker is already two sentences ahead.
I have started using very short clips instead of full podcasts. First listen with no text, second listen trying to catch only the main idea, then I check the transcript and replay the parts that collapsed into noise. Sometimes I shadow one sentence, mostly to force myself to notice the rhythm rather than to imitate an accent perfectly.
This is much more tiring than textbook listening, so I only do 10–15 minutes at a time. The interesting part is that after a few repeats the clip stops sounding fast; it sounds normal, which makes me think segmentation is a big part of the problem.
Did native-speed listening improve for you mainly through huge amounts of exposure, or did deliberate work on short pieces make the difference?