Sylpheed/docs/re/structures/intro-audio-decomposed.md

✅ The boot intro's audio: the movie's own 5.1 bed at 0.600, plus three streams in 5.1

Classification: measured. Xenia Canary, 2026-08-30, one boot, no input. This settles the port's ask #4 and confirms the corpus's leading hypothesis from the output side, where it had been recorded as "not established".

🔴 First: ADV.wmv is not three XMA streams. It is one WMA Pro 5.1 track.

ffprobe /disc/dat/movie/ADV.wmv
  Stream #0:0(jpn): Audio: wmapro, 48000 Hz, 5.1, fltp, 384 kb/s
  Stream #0:1(jpn): Video: wmv3, 1280x720, 30 fps

One audio stream, 5.1, decoded by Xenia's WMA path — not the XMA path the three probe contexts come from. Any framing of the intro's audio as only "which of three voice streams to ship" was missing the bed entirely.

The decomposition

Aligning the 148 s capture against that track and solving capture = g × movie + residual (../data/intro-audio-decomposition.txt):

ch	gain	r	capture rms	residual rms	residual/capture
FL	0.600	+0.905	−21.86	−29.28	−7.43 dB
FR	0.600	+0.935	−20.28	−29.29	−9.02 dB
FC	0.597	+0.146	−25.18	−25.27	−0.09 dB
LFE	0.600	+1.000	−43.66	−115.73	−72.06 dB
BL	0.600	+0.956	−24.89	−35.46	−10.58 dB
BR	0.600	+0.964	−24.02	−35.49	−11.47 dB

The gain is 0.600 on every channel — a uniform −4.44 dB, which is a mixer setting, not a fit artefact. LFE is reproduced to −115.73 dBFS, 72 dB below the signal: at that residual the two decoders agree essentially exactly, which is what rules out "the leftovers are just codec differences".

🔴 And FC is the exception that carries the answer. The movie explains nothing of the capture's centre channel — the residual is the whole signal (−0.09 dB). The movie's own FC is 91.6 % silent; the capture's is not.

The residual is three signals, not one

	FL	FR	FC	LFE	BL	BR
FL	1.000	0.918	0.017	0.001	0.009	0.004
FR	0.918	1.000	0.034	0.001	0.015	0.014
FC	0.017	0.034	1.000	0.000	0.385	0.378
LFE	0.001	0.001	0.000	1.000	0.000	−0.000
BL	0.009	0.015	0.385	0.000	1.000	0.929
BR	0.004	0.014	0.378	−0.000	0.929	1.000

Three coherent groups: a front pair (0.918), a rear pair (0.929), and a centre whose partner LFE is empty. The FC residual's 100 ms frame levels span 34 dB (median −53.9, p90 −19.9) — bursty, not steady noise.

✅ This confirms the 5.1 hypothesis, and predicts the silent channel correctly

voice-three-streams-are-concurrent.md recorded "three concurrent stereo streams is six channels" as the obvious reading and marked it not established, with a specific piece of supporting detail: that ADV stream 2 is mono-in-stereo, "a centre paired with a silent LFE looks exactly like that".

That is exactly what the residual shows — a live centre whose paired channel is empty to −115 dB. Measured from the output, with no access to the stream contents:

XMA stream	lands in
one	FL, FR
one	FC, LFE silent
one	BL, BR

So both things are true and the port needs both: the movie's own 5.1 WMA Pro track and the three streams mixed over it in 5.1.

🔴 Correction to the census page, and to the recipe page's channel order

intro-audio-output-census.md labelled its channels using the permutation [0,1,4,5,2,3] that audio-capture-alsa-file-tee.md records for ALSA. That permutation does not apply to this capture. The 6×6 correlation matrix was computed without assuming any order, every row's maximum falls on a distinct movie channel, and the result is the identity.

So the census's "BR is 82 % silent" was really LFE — which also reconciles it with the movie, whose LFE is 80.64 % silent. ⚠️ The recipe page's permutation was measured on a different chain and is not wrong there; what is wrong is assuming it travels. Measure the channel order per capture; a 6×6 matrix that comes out a clean permutation is its own control.

Reach

⚠️ One boot, one movie. ADV only. ✅ The assignment is now determined — see the section below. It was open when this page was first written. ⚠️ --gpu=null, so no video cross-check. ✅ The 0.600 gain is measured on this run; whether it is a fixed mix constant or a volume setting is not established.

✅ Which stream is which (2026-08-30, later)

The three chunks were dumped from the resolved voice region (examples/adv_voice_dump.rs) and decoded: ../data/adv-stream-assignment.txt.

chunk	byte_size	probe ctx	L rms	R rms	R silent
0	806 912	ctx0, clipped tail (full 1 294 336)	−24.79	−24.81	53.1 %
1	1 118 208	ctx1	−20.33	−inf	100 %
2	1 171 456	ctx2	−30.67	−30.68	53.6 %

🔴 Two instruments failed first, and both look convincing

• Envelope correlation cannot discriminate here. A per-pair lag search returns 0.86–0.95 for every chunk against every channel, because all six residual channels share the dialogue's activity timing. A number that high reads as a result; it is the instrument having no resolving power in this regime. ⚠️ Corrected 2026-08-30 — do not generalise this. The port agent controlled the same estimator on a single dialogue track and got r = 1.0000 at zero offset and −0.08…+0.08 everywhere else: it localises sharply. The saturation here is specific to concurrent streams sharing timing at zero lag, which a lag search over one track never encounters. This page said it flatly for a day after that correction was known, which is the failure METHOD.md calls a correction that does not reach the artifact.
• Sample-level correlation returns ≈ 0. The chunks do not start with the movie and the XMA decode's framing offset is unknown.

✅ Level settles it, under the same 0.600 gain

chunk	level	× 0.600	nearest residuals (error, dB)
0L	−24.79	−29.23	FL 0.05 · FR 0.06 · FC 3.96
1L	−20.33	−24.77	FC 0.50 · FL 4.51
2L	−30.67	−35.11	BL 0.35 · BR 0.38 · FR 5.82

Each stream lands within 0.5 dB of exactly one residual pair and misses the others by ~4–6 dB. The same 0.600 that scales the movie bed also scales the voice — which is itself worth having: it is one mixer gain, not two.

✅ Ratio test, immune to chunk 0 being clipped: chunk0 − chunk2 = +5.88 dB against FL − BL = +6.18 dB, agreeing to 0.30 dB; swapped, the ratio would be wrong by 11.76 dB.

✅ Structural confirmation. Chunk 1 is the only chunk with a digitally silent channel, and LFE is the only output channel with an empty residual (−115.73 dBFS). One to one. And the internal L/R correlations track: chunk 0 +0.932 against the FL/FR residual's +0.918, chunk 2 +0.962 against BL/BR's +0.929.

stream	→
ctx0 · 1 294 336	FL, FR
ctx1 · 1 118 208	FC (LFE silent)
ctx2 · 1 171 456	BL, BR

🔴 And the identifier this is indexed by does not resolve against the disc. The port refused to apply this assignment because the three contexts sum to 3 584 000 B against a resolved region of 3 114 352. It was right to: the region start is wrong, by 238 packets for ADV and on 8 of 10 multichannel regions disc-wide — voice-region-starts-late.md. The assignment above still stands (the ratio test was chosen to be immune to the clipping), but chunk 0's absolute level was measured over 62 % of its stream.

⚠️ Reach. Levels, not waveforms — this is an argument from three numbers agreeing to 0.5 dB and a 1:1 structural match, not from a matched waveform. One movie, one boot. And chunk 0 is a clipped tail, which is why the ratio test is quoted alongside the absolute match.