→ all versions of this pipeline

Story 6 — the rustles, isolated again

Your labels split the problem in two, and only one half was broken. Twelve of the twenty-four windows really do contain the rustle — you heard it in the run-in — but only five of the spans I cut pointed at it. So the windows were useful and the isolation was wrong, which are different faults.

What I had backwards. I assumed cloth is a high-frequency sound and looked for a short, bright spike. Measuring your labels says otherwise — a window with rustle in it sits above the room in every band at once:

band40–120120–300300–800800–2k2k–5k5k–9k
rustle above the room, dB+2.6+3.4+5.7+6.2+7.6+9.5

The short bright spikes I was cutting were sibilance — ש, ס and ח are high-frequency, have no pitch and sit next to pauses, so they passed every test I had. The fix is to require the weakest band to be lifted, not the loudest: one loud band proves nothing, all of them at once is the signature. On several of these the low-mid is now the strongest part, which is the phone body being touched rather than air.

These are 10 distinct events, not twelve — three of your windows turned out to be the same rustle heard three times.

What I still cannot do is find these on my own. Across the whole nine minutes this measurement flags about 150 stretches, which is a list of everything, not a detector. It works here because you already told me the rustle is in these windows, so the question is only which moment rather than is there anything. Whether other rustles exist elsewhere in the story is still unknown.

nothing playing
1 0:10.29 245 ms · 17 dB above the room
low-mid dominated — contact against the phone body · was round-one #6
2 0:24.14 245 ms · 15 dB above the room
even across the spectrum · was round-one #11
3 0:27.32 245 ms · 14 dB above the room
low-mid dominated — contact against the phone body · was round-one #7
4 1:53.57 245 ms · 12 dB above the room
low-mid dominated — contact against the phone body · was round-one #20
5 2:22.16 245 ms · 8 dB above the room
low-mid dominated — contact against the phone body · was round-one #22
6 6:57.30 245 ms · 17 dB above the room
even across the spectrum · was round-one #1
7 7:00.73 245 ms · 5 dB above the room
low-mid dominated — contact against the phone body · was round-one #4
8 7:45.63 245 ms · 10 dB above the room
low-mid dominated — contact against the phone body · was round-one #3
9 9:08.87 245 ms · 28 dB above the room
even across the spectrum · was round-one #2, #10, #23
10 9:16.47 245 ms · 5 dB above the room
even across the spectrum · was round-one #18
Your answers will appear here — copy the block and send it to me.