Thread Context - Full Analysis History
Thread Context - Full Analysis History
Received: 2026-07-30 10:59 GMT+1 via Telegram from John Source: External thread context document (covers Songs 1-6 + guide track diagnostics) Saved here for reference. Original in Telegram.
Documents in this thread:
1. Song 1 "The Rinse Light" A/B - COMPLETE 2. Song 2 "Pass of the Shuttle" A/B - COMPLETE 3. Guide v1 diagnostic - COMPLETE 4. Guide v2 "why heard as humming" - COMPLETE 5. Song 3 "The Weight of Absence" A/B - COMPLETE 6. Song 6 "The Quiet Room" v6g render diagnostic - COMPLETE (this is the latest)
Key Cross-Song Findings
Three-song trend (Version A - the key artefact):
- Chord vocab: 17 → 22 → 15 (fixed)
- Chorus vs verse harm rate: 0.00 → 0.00 → -0.08 (first differentiation)
- LRA: 7.1 → 6.5 → 10.1 LU (big gain)
- 500-2000Hz presence: 11.1% → 7.5% → 14.2% (fixed)
- 2-6kHz: 0.79% → 0.90% → 1.93% (doubled)
- Chorus ceiling lift: +3.4 → -0.7 → +6.2 st (best yet)
- Melody range: 20 → 17 → 28 st (best yet)
- Hook r: -0.080 → +0.480 → +0.165 (REGRESSED)
- Chorus centroid: -345 → +93 → -404 Hz (REGRESSED)
- V1→Ch1 vocal swing: -7.4 → -7.4 → -7.7 dB (NEVER FIXED, 6/6 versions)
Version B trend:
- Hook r: +0.079 → +0.747 → +0.970
- LRA: 7.1 → 5.8 → 13.2
- Suno deflates the ending in all three
Persistent failures across ALL songs:
1. Vocal buried in choruses, progressively worse toward end 2. No energy above 500 Hz anywhere (4 of 4 versions) 3. Chorus harmonic rhythm never slowed except song 1's B 4. Hook motif contour self-similarity near zero in both A and B 5. Both deflate in final chorus 6. Long outros (40-50s) in all versions
Song 6 (The Quiet Room v6g) - Detailed Diagnosis
Dakota's root-cause hypothesis REFUTED:
The "MuseScore piano patch masks everything" theory is wrong. Refuted three ways: (a) Differential audibility: bass +21.9 to +29.6 dB, strings +7.3 to +19.0 dB, harmonics +6.5 dB - all plainly present (b) Choruses are STRING-dominated (57-60% of chorus energy in 420-700 Hz, the strings' register) (c) Section arc EXISTS and is large: verse -31.6 dB → chorus -22.4 dB = +9.2 dB lift, 17.5 dB range, LRA 14.2 LU
Actual diagnosis: AUDITORY FUSION
All four source-separation cues defeated simultaneously: 1. Onset: In C1, piano+strings+bass+pad all start on beat 0 2. Envelope: All share one struck-decay envelope (nothing sustains - -33 to -40 dB by 1.2s) 3. Pan: 45/50/55/60/64 on 0-127 = all within ±15% of centre (effectively mono) 4. Timbre: Programs 48 and 49 are near-identical patches
Five concrete bugs:
1. Outro never rendered (duration_beats:144 but form ends at beat 152) 2. No chorus escalation: C1 -22.54, C2 -22.33, FC -22.35 = 0.21 dB spread. Strings+harmonics pinned at velocity 127 3. -24.1 LUFS = ~10 dB too quiet for upload 4. Pad inaudible: A3=220Hz in UNISON with piano A3, duplicate pitches in chords 5. Harmonics E6 +2.4 dB and C6 +2.9 dB below detectability; only A6 clears
JSON annotations contradict note data:
- v3_to_fc_jump claims guitar 35→90 (+157%); actual 30→35 (+17%)
- velocity_drop_from_c2 claims 78→50 and 75→50; those velocities appear nowhere
- Bridge caesura claims beats 12-16 total silence; bass runs to beat 15, pad to beat 16
- high_element_enters claims A5 880Hz; note is A6 1760Hz
- frequency_map claims 200-500Hz left clear; strum includes C4 262, F4 349, G4 392, A4 440
- GT-1 (no vocal in 200-500Hz) violated by its own notes
- duration_seconds ~133s but 152 beats at 65 BPM = 140.3s
Sustain problem:
Nothing sustains. All notes decay -33 to -40 dB by 1.2s. Either program changes aren't reaching FluidSynth (everything renders as piano), or presets load without sustain. Need the .mid or five single-channel renders to settle it.
Fixes ranked:
1. Stagger onsets (onset_offset_ms per part) 2. Spread pan across 18-110 3. Differentiate envelope/attack times 4. Stop using velocity 127 5. Restore piano to ~52/72/40/84 6. Compute duration_beats correctly (152 not 144) 7. Gain +9dB post-render 8. Fix pad voicing to [E3,A3,C4] 9. Make the Bridge caesura real 10. JSON format: add onset_offset_ms / reverb_send / attack / target_lufs
Pre-render gate problem:
The pre-render gate validates the annotations, not the music. That's why 10 checks pass and the render fails.
Guide v2 Key Conclusion:
Suno is not misclassifying the guide - it is faithfully reproducing a signal whose acoustic signature IS background humming. 13 metrics measuring 1.9x to 206x deficit vs a real lead vocal. Top recommendation: sing/hum the guide into a phone. Fixes all 16 metrics at once.
Suno Workflow Research:
- Audio Influence in Create mode is NOT a melody-transfer tool (Cover/Stem Cover are)
- Slider ~50 is "loose inspiration", 65-85 recommended to preserve uploaded idea. 32 is well below midpoint
- Keep Weirdness under 25
- Upload 15-45s (over 60s introduces structural confusion)
- WAV 44.1/16 not MP3; normalise to -1 dBFS
- Sparse/solo beats dense mixes for pitch extraction