Audio IP Whitepaper — parksy-audio

Version: 2.0 (Three-Layer Model)

Date: 2026-02-11

Category: Audio (Recovery Kit)

1. Three-Layer Copyright Model

Audio copyright has independent layers. parksy-audio secures all three:

1. Composition	Melody, harmony, structure	Composer (or public domain)	Public domain only (composer dead 70+ years)
Layer	What	Who Owns It	How We Secure It
2. Performance	Specific interpretation/expression	Performer	MIDI Pre-Processor generates new digital performance
3. Sound Recording	Actual audio waveform	Producer	Independent rendering (our SF2 + FluidSynth)

Layer 1: 누가 작곡했냐?  → Public Domain (Chopin, Bach, etc.)
Layer 2: 누가 연주했냐?  → Our code (humanize-midi.py Pre-Processor)
Layer 3: 누가 녹음했냐?  → Our engine (SF2 + FluidSynth)

All three layers are independently secured. No third-party rights at any level.

2. Why MIDI Pre-Processing = New Performance (Layer 2)

Downloaded MIDI files may carry the original transcriber's "performance signature":

**CC64 (Sustain Pedal):** Pedal patterns = transcriber's interpretive habit

**CC11 (Expression):** Dynamic curves = sequencer/DAW characteristics

**CC7 (Volume):** Volume automation = production intent

**Velocity distribution:** Attack patterns = performer's touch fingerprint

**Timing micro-deviations:** Rubato patterns = interpretive style

Our MIDI Pre-Processor (scripts/humanize-midi.py) removes all source-specific signatures:

[Original MIDI] — carries transcriber's performance signature
       ↓
[CC Stripping] — removes pedal/expression/volume patterns
       ↓
[Velocity Redistribution] — replaces original touch fingerprint
       ↓
[Timing Re-jitter] — replaces original rubato with new micro-deviations
       ↓
[Processed MIDI] — new, independent digital performance

"All MIDI sources are normalized and re-processed through an internal

MIDI preprocessing pipeline prior to rendering. This process removes

any source-specific performance signatures and produces a new,

independent digital performance."

3. Why MIDI Rendering = New Recording (Layer 3)

MIDI is instruction data, not audio. It contains:

Note number (pitch)

Velocity (volume)

Duration

Channel

MIDI has no sound. Rendering MIDI through a SoundFont/VST produces a new sound recording, legally equivalent to a musician performing from sheet music.

Sheet music → Pianist plays → New recording (pianist owns it)
MIDI file   → FluidSynth renders → New recording (we own it)

The rendered WAV/MP3 is our original sound recording.

4. Public Domain Composition Rules

A composition enters public domain when copyright expires:

**Most countries:** Life of composer + 70 years

**USA (pre-1928):** Published before 1928 = public domain

**Some countries:** Life + 50 years (check per jurisdiction)

Safe zone: Composers who died before 1956 (70-year rule covers 2026).

See COMPOSERS_PUBLIC_DOMAIN.md for the verified list.

5. SoundFont/VST Licensing

Rendering requires instruments. Each instrument has its own license:

Some are open-source (CC0, MIT, GPL)

Some are commercial with redistribution restrictions

Some prohibit use in competing sample libraries but allow musical use

Every SoundFont/VST used in production must be logged in LICENSE-INSTRUMENTS.md.

6. MIDI Source Tracking

MIDI files also have copyright considerations:

**Self-composed MIDI:** We own it

**MIDI transcription of public domain score:** We own the transcription

**Downloaded MIDI:** Original transcriber may have arrangement copyright

Every MIDI source must be logged in MIDI-SOURCES.md.

7. Content-ID Risk (Platform Lane)

YouTube's Content-ID system may flag audio even when legally clean:

Another recording of the same public domain piece exists in Content-ID database

The flag is on **their recording**, not the composition

Our rendering is a **different recording** — the claim is invalid

Response protocol documented in CONTENT-ID-MONITORING.md.

8. Defense vs Recovery

		Defense (Prevention)	Clean process + monitoring	Pipeline design + CONTENT-ID-MONITORING.md
		Type	What	Documents
		Recovery (Post-incident)	Evidence for dispute	This whitepaper + LICENSE-INSTRUMENTS.md + MIDI-SOURCES.md + COMPOSERS_PUBLIC_DOMAIN.md

The real defense is the process itself:

1. Only use public domain compositions (verified)

2. Pre-process all MIDI through signature removal pipeline (logged)

3. Only use properly licensed instruments (logged)

4. Only use tracked MIDI sources (logged)

5. Render independently (our equipment, our output)

The documents are recovery tools — evidence to present if a claim occurs.

9. Two-Lane Operation

Risk	Content-ID auto-scan	None
	Lane 1: Platform (YouTube)	Lane 2: Off-Platform (Local)
Required defense	Full Recovery Kit + monitoring	LICENSE-INSTRUMENTS.md only
Operating cost	Monitoring + dispute response	Near zero
Composition rule	Same (public domain only)	Same
SoundFont rule	Same (licensed only)	Same

Both lanes share the same clean pipeline. The difference is monitoring overhead.

10. One-Line Summary

parksy-audio pre-processes and renders public domain compositions

through an internal pipeline into original sound recordings.

All three copyright layers (composition, performance, recording) are independently secured.

These documents exist to prove it if challenged.

Built between shifts.