Episode 1 — what your thirty answers changed

This is the same page you answered, now showing the result. All thirty were ruffles — not one sound effect, not one “nothing there”. Twenty-two spans were already being taken out; these thirty make fifty-two. The twelve you called ruffle under my voice were treated and then checked one at a time, and every one of them passed.

Pick a sound below.

The whole episode

Seventeen minutes each. Nothing seeks inside a long file anywhere else on this page — these three are the only long ones.

The new master, with the storybook music you chose. This is the version that would ship.

The new master, no music. Fifty-two spans treated.

The one you have now, for comparison — twenty-two spans, no music.

Your two categories, and what each one meant

Eighteen plain cloth ruffles. Standard treatment: the room under them is replaced with quiet room from this same recording, your voices never touched.

Twelve “ruffle under my voice”. You asked us to invent this one, and it is not a kind of sound — it is a condition on the treatment: “better to remove but only if doesn’t harm meaningfully my speaking… otherwise leave as is for all of these.” So all twelve were treated and then each was measured on its own, and each one was kept or put back on its own number.

You were right about the leak. The separator does put a little of your voice into the room channel, which is why an isolated clip of a ruffle under speech sounds like you. That is exactly why the check is done on the finished mix and not on the isolated clip — the isolate cannot answer the question you asked.

The check on the twelve, span by span

Two things have to hold. First, what is left in the moment has to sit on your voice’s own level — that is the mix − voice column, and it should be near zero. Second, in the frames where your voice leads, the mix must not lose more than this method costs everywhere else: the 33 other spans in this build that have enough of those frames to measure run +3.04 to +7.78 dB. Above that is the signature of speech being taken out with the room, and that span goes back. Physics puts the ceiling near 3 dB when the two are independent, so the numbers clustered at +3.0 are this method costing what it has to cost and nothing more.

at	verdict	voice-led drop	mix − voice	frames	length s	voice dBFS
2:08.25	kept	+3.07	+0.00	22/22	0.55	-21.2
4:04.80	kept	+5.98	+0.00	15/22	0.55	-51.7
4:10.11	kept	+3.10	+0.00	32/54	1.35	-29.0
5:13.67	kept	+3.26	+0.00	9/56	1.40	-45.9
5:32.45	kept	+3.06	+0.00	20/20	0.50	-24.2
6:28.79	kept	+3.14	+0.00	59/206	5.15	-42.8
6:32.77	kept	+4.73	+0.00	14/68	1.70	-43.8
8:31.72	kept	+3.03	+0.00	34/34	0.85	-31.5
9:09.13	kept	+3.01	+0.00	26/26	0.65	-24.0
9:28.23	kept	+3.01	+0.00	24/24	0.60	-25.3
12:25.10	kept	+3.02	+0.00	27/28	0.70	-23.3
13:22.98	kept	+3.03	+0.00	24/24	0.60	-26.5

The table scrolls sideways on a phone — the verdict and the number behind it are the first two columns, so nothing you need is off to the right.

Kept 12 of 12, put back 0. Two sit higher than the rest — 4:04.80 at +5.98 and 6:32.77 at +4.73 dB — and both are short sounds where your voice leads only the frames at the edges, 15 of 22 and 14 of 68. Measured over the whole region each one sits in, they cost +3.13 and +3.08 dB, which is what every other region costs. Whole regions in this build run +3.03 to +3.38 dB, against +3.03 to +3.35 on the twenty-two-span build — so thirty more spans did not make the method cost more.

The eighteen plain ruffles

Loudest first, as before. Each has the sound by itself — lifted the same 21.6 dB you heard it at when you answered — and the same moment in the finished episode at its real level.

10:49.56 treated-40.9 dB · 2.15 s
room -24.3 dB · voice-led drop +3.43 dB

12:57.20 treated-43.1 dB · 0.55 s
room -26.5 dB · voice-led drop +3.11 dB

1:00.05 treated-44.4 dB · 1.05 s
room -28.0 dB · voice-led drop +3.04 dB

11:13.00 treated-46.4 dB · 0.90 s
room -21.5 dB · voice-led drop +3.32 dB

15:19.60 treated-48.7 dB · 1.35 s
room -21.1 dB · voice-led drop +3.16 dB

11:44.15 treated-49.1 dB · 2.55 s
room -21.8 dB · voice-led drop +3.08 dB

14:53.54 treated-49.5 dB · 5.30 s
room -16.5 dB · voice-led drop +7.78 dB

11:37.04 treated-51.7 dB · 0.50 s
room -20.5 dB · voice-led drop +3.95 dB

2:21.51 treated-51.8 dB · 0.55 s
room -20.2 dB · voice-led drop +3.04 dB

12:30.26 treated-53.3 dB · 1.35 s
room -17.1 dB · voice-led drop +3.38 dB

8:09.77 treated-53.9 dB · 2.70 s
room -19.3 dB · voice-led drop +6.56 dB

15:50.71 treated-54.8 dB · 0.55 s
room -15.8 dB · voice-led drop +3.09 dB

0:29.62 treated-54.9 dB · 1.80 s
room -17.6 dB · voice-led drop +3.05 dB

2:41.27 treated-54.9 dB · 0.85 s
room -16.6 dB · voice-led drop +3.04 dB

5:07.51 treated-55.8 dB · 4.05 s
room -15.8 dB · voice-led drop +6.13 dB

14:21.05 treated-55.8 dB · 4.90 s
room -16.7 dB · voice-led drop +3.30 dB

5:53.93 treated-56.4 dB · 0.70 s
room -14.8 dB · voice-led drop +3.04 dB

4:53.03 treated-56.6 dB · 4.50 s
room -14.7 dB · voice-led drop +5.57 dB

The twelve under your voice

Same four buttons. On any that was put back, the before and after are the same audio, and that is the point of showing them.

4:04.80 treated, and it passed the check-53.5 dB · 0.55 s
room -22.1 dB · voice-led drop +5.98 dB

4:10.11 treated, and it passed the check-54.4 dB · 1.35 s
room -21.3 dB · voice-led drop +3.10 dB

9:28.23 treated, and it passed the check-56.6 dB · 0.60 s
room -12.5 dB · voice-led drop +3.01 dB

5:32.45 treated, and it passed the check-57.2 dB · 0.50 s
room -16.4 dB · voice-led drop +3.06 dB

2:08.25 treated, and it passed the check-57.7 dB · 0.55 s
room -12.5 dB · voice-led drop +3.07 dB

13:22.98 treated, and it passed the check-57.9 dB · 0.60 s
room -10.9 dB · voice-led drop +3.03 dB

12:25.10 treated, and it passed the check-58.3 dB · 0.70 s
room -12.7 dB · voice-led drop +3.02 dB

9:09.13 treated, and it passed the check-58.5 dB · 0.65 s
room -7.7 dB · voice-led drop +3.01 dB

6:32.77 treated, and it passed the check-58.6 dB · 1.70 s
room -11.1 dB · voice-led drop +4.73 dB

6:28.79 treated, and it passed the check-59.1 dB · 5.15 s
room -12.6 dB · voice-led drop +3.14 dB

5:13.67 treated, and it passed the check-59.2 dB · 1.40 s
room -14.5 dB · voice-led drop +3.26 dB

8:31.72 treated, and it passed the check-59.2 dB · 0.85 s
room -14.5 dB · voice-led drop +3.03 dB

The numbers on the delivered file

	22 spans (now)	52 spans	52 + storybook
loudness, on the file	-16.50 LUFS	-16.50 LUFS	-16.20 LUFS
startle margin, dual-mono	+3.90 LU	+3.90 LU	+4.10 LU
true peak	-1.8 dBTP	-2.1 dBTP	-1.8 dBTP
duration	1013.438 s	1013.438 s	1013.438 s

Target −16 LUFS measured on the file; the bedtime rule is that nothing may jump more than 5 LU above the average, measured dual-mono on both sides. The technical audit scores the new master 93.0 out of 100, the same as the one you have — the room floor in the pauses went from −49.4 to −50.5 dBFS and nothing else moved.

The quiet room dropped into the widest gap sits 10.2 dB below the room going into it, at 5:53. On the twenty-two-span build that figure was 10.7 dB, so adding thirty spans did not make the worst seam worse. Across all 34 regions the median is -0.6 dB.

Nothing was published and nothing has been accepted. Both of those are yours, separately.