8Z-AUDIO
OriginIzvor Sprint Benchmark SongsGlasba RoadmapNačrt
Benchmark EvolutionEvolucija Appendix M Master Plan
Gemini GPT Grok
ENG
SLO
🌙
AIM³ Institute · Ljubljana · February 2026Februar 2026

Five Days
Against FLAC
Pet dni
proti FLAC-u

How a lossless audio codec was built from scratch in a single sprint — and why it beats the 25-year-old gold standard on nearly half of all test clips.Kako je brezizgubni avdio kodek nastal iz nič v enem samem sprintu — in zakaj prekaša 25 let star zlati standard pri skoraj polovici testnih posnetkov.

5
Days of DevelopmentDni razvoja
7/15
Clips Beating FLAC‑12Posnetkov boljših od FLAC‑12
10.3%
Best Single‑Clip WinNajboljša zmaga na enem posnetku
v0.1 → v1.5
Versions ShippedIzdane verzije
Chapter I · Cross-Domain TransferPoglavje I · Meddoménski prenos

Not Built in IsolationNi nastalo v izolaciji

8Z-Audio is the third domain-specific compressor in the 8Z-LO (Lossless Optimized) framework, following 8Z-FASTA (genomic DNA sequences) and 8Z-TIF (satellite imagery). The project didn't start from an audio textbook — it started from a radical question: what if the same mathematical architecture that decodes DNA compresses music?

8Z-FASTA · DNA

The DNA Compressor

Before any audio code was written, the team ran a sanity check: convert Pink Floyd's "Shine On You Crazy Diamond" into DNA bases (A, C, G, T) and compress it with gemZ, the FASTA compressor built for genomic sequences.

gemZ's DNA models found zero hits — 100% RAW blocks. The biological pattern-matchers understood nothing about waveforms. But the architectural skeleton — MDL model battles, per-block predictor selection, residual coding — was exactly what audio needed.

8Z-DCC · Adaptation

The Digital Claustrum Controller

The DCC was born in the TSP solver, migrated into the FASTA encoder, and landed in audio. It's an adaptive search governor: a learned probability model that narrows the encoder's candidate search space in real time, spending compute budget where the signal is hardest and coasting where it's easy.

In v1.5, DCC settled for the first confirmed time on a real audio file (Pink Floyd 192kHz, u=10→15), proving the concept transfers across domains.

The 8Z framework rests on one principle: Minimum Description Length (MDL) model selection. Every bit saved must correspond to genuine reconstruction capability — not statistical correlation, not heuristic intuition. Each frame of audio is a competition. Multiple predictors battle; MDL picks the winner. FLAC runs one predictor (LPC) on every frame, always. That's the gap.

Chapter II · The Five-Day SprintPoglavje II · Petdnevni sprint

v0.1 → v1.5

From an empty file to a compressor that beats FLAC at maximum compression. Every version had a lesson. Every regression had a diagnosis.

Day 1 · February 18

The FASTA Experiment & v0.1 — First Proof of Life

LPC Orders 1–32 DELTA Predictor Mid-Side Stereo SHA3-256 Verified No Rice Coding

The first working encoder was built in a single chat session. LPC via Levinson-Durbin autocorrelation, DELTA prediction ported from gemZ, MDL battle per frame, residual compression via int32 → LZMA, all four stereo decorrelation modes, bit-perfect verification.

9,417,674 bytes vs. FLAC 6,055,738 — architecture proven, Rice coding gap identified immediately
Day 2 · February 19

v1.0–v1.2 — Rice Coding & FLAC Territory

Golomb-Rice Entropy Partition Search QLP Precision Search 8–16 3 Window Functions

Implemented proper Golomb-Rice coding with partition search, QLP quantization precision search, and three apodization windows (Hann, Tukey, none). The encoder crossed into FLAC-5 territory for the first time.

Approached FLAC -5 compression on Pink Floyd 192kHz. The real FLAC gap isolated.
Day 3 · February 20

v1.3.1 — Multiprocessing & Best-Ever Compression

Python Multiprocessing Exhaustive Search ~600 Candidates/Frame 11-Hour Encode

Exhaustive combinatorial search across all candidate configurations — LPC orders, windows, QLP precision — evaluated per frame with MDL arbitration. Expensive. Brutally thorough. This became the permanent compression quality ceiling to beat.

5,104,935 bytes — beats FLAC -8 by 2.1%. First time a handwritten encoder surpassed FLAC's maximum default setting.
Day 4 · February 21

v1.4 — DCC Port from HYB4

Digital Claustrum Controller CodecLearner 7-bit DCC Events 1.53% Regression vs v1.3.1

The DCC search governor was ported from the HYB4 FASTA encoder. 7-bit event encoding was too fine-grained — the DCC never found a stable pattern to exploit. Encode time dropped to 102 minutes (from 11 hours), but compression regressed slightly. Lesson: DCC event granularity matters enormously.

Encode: 11 hr → 102 min. DCC concept proven but not yet settled. 7-bit events too noisy.
Day 5 · February 22 — The Benchmark

v1.5 — DCC Settles. Benchmark at Scale.

Blocksize 16384 4-bit DCC Events DCC Settled First Time 15-Clip Parallel Corpus 8 Workers · 75 min

Switching DCC from 7-bit to 4-bit events was the breakthrough: the controller finally settled on Pink Floyd 192kHz (u=10→15). Block size optimized to 16384. Scanner v1.2, clipper v1.2, and parallel pipeline v1.2 built in the same session. 15 clips across 10 songs encoded simultaneously.

v1.5 beats FLAC-12 on 7 of 15 clips. Best win: Lady Gaga "Die With A Smile" — 10.3% smaller than FLAC's best. By March 2026, avFLAC v1.2 with vFLAC v1.5 MDL block probe and ACMD v1.9 MDL frame probe: beats lax_t6 on Abyssal by 14,667 bytes, wins 15/15 clips vs FLAC-12, beats OptimFROG on 3 Radiohead clips by 31–40%. All files lossless-verified via SHA3-256.
Week 3–4Teden 3–4
v1.6–v1.9 + xFLAC Suite
MDL Frame Probe (v1.9) MDL Block Probe (vFLAC v1.5) 4-Encoder Orchestrator

Two parallel codec families consolidated: 8Z-AC (native .8za format) with MDL-driven frame size selection, and xFLAC (aFLAC + vFLAC + avFLAC) producing valid .flac files via per-segment arena competition. vFLAC v1.5 introduced MDL block probe — empirically testing 3 block sizes per block instead of heuristic assignment. avFLAC's MDL final comparison picks the best of arena stitch, AF_whole, and VF_whole per file.

avFLAC beats lax_t6 on Abyssal by 14,667 bytes. First full-track win over the best single FLAC encoder. Beats OptimFROG on 3 Radiohead clips (up to 9.2% smaller). avFLAC wins on 8 of 18 test files vs best baseline. avFLAC invariant holds on all 18 files — never worse than the best component.
Chapter III · ArchitecturePoglavje III · Arhitektura

What Makes It DifferentKaj ga naredi drugačnega

FLAC uses one prediction model for every frame of audio. 8Z-Audio uses up to 600 candidates, evaluated by MDL, with an adaptive search governor that learns which candidates work for this particular signal.

8Z-Audio v1.5 Frame Pipeline
WAV Frame 16384 samples
DCC Governor Budget allocation
Candidates ~600 configs
MDL Battle True total cost
Best Predictor Rice-coded residuals
.8za Frame SHA3 verified
📐

LPC Orders 1–32

Vs. FLAC's 0–12. Higher orders capture longer-range correlations in complex harmonic content.

🪟

Three Window Functions

Hann, Tukey, and no apodization — evaluated per frame. FLAC uses no apodization at its default settings.

🎯

QLP Precision Search

Exhaustive search over 8–16 bits of quantization precision per frame. FLAC uses a fixed precision per level.

🧠

DCC + CodecLearner

The Digital Claustrum Controller learns per-signal statistics, adapting search depth to where the signal is hardest.

⚖️

MDL Model Selection

Every bit counts. MDL sees the true total cost including predictor overhead — no bit is "free."

🔁

All Stereo Modes

Mid-Side, Left-Side, Right-Side, and Independent — selected per frame by MDL, not globally per file.

Chapter IV · ResultsPoglavje IV · Rezultati

Benchmark

15 clips across 10 songs — selected by the 8Z Audio Scanner v1.2 to cover all difficulty categories. Encoded in parallel with 8 workers. All results verified lossless (SHA3-256). Compression ratio: lower is better (fraction of original size).

8Z avFLAC v1.2 (ACMD v1.9 + vFLAC v1.5 + aFLAC v1.3) · vs Best Baseline per Clip
Clip
Category
avFLAC
Best Base
OFR
vs Baseline
Rammstein — Du Hast (diverse)
diverse
0.4279
0.4713
0.7255
▲ 9.2%
Rammstein — Du Hast (hardest)
hardest
0.3418
0.3636
0.5162
▲ 6.9%
Rammstein — Du Hast (60s DCC)
dcc_best
0.4146
0.4455
0.6008
▲ 6.8%
Lady Gaga — Die With A Smile
diverse
0.2154
0.2329
0.1911
▲ 3.3%
Pink Floyd — SOYCD (30s dynamic)
dynamic
0.2049
0.2066
0.1802
▲ 0.7%
Pink Floyd — SOYCD (10s tonal)
tonal
0.2532
0.2654
0.2353
▲ 0.4%
BD — Fractured Harmony (tonal)
tonal
0.2560
0.2670
0.2319
▲ 4.1%
Metallica — Lux Æterna (buildup)
buildup
0.7904
0.7938
0.7734
▲ 0.1%
BD — Fractured Harmony (easiest)
easiest
0.3394
0.3508
0.3044
▲ 3.2%
BD — Awakening AI (tonal)
tonal
0.3696
0.3780
0.3338
▲ 2.2%
BD — Between the Strings
dcc_stress
0.4642
0.4709
0.4378
▲ 1.4%
Metallica — Lux Æterna (transient)
transient
0.6363
0.6437
0.6158
▲ 1.1%
BD — Where Do I Go
dcc_stress
0.5082
0.5157
0.4793
▲ 1.5%
BD — Echoes of the Shore
dcc_stress
0.5247
0.5284
0.4901
▲ 0.7%
BD — Lacrimosa Requiem
diverse
0.5509
0.5583
0.5189
▲ 1.3%

Total — avFLAC wins: 15/15 clips vs FLAC-12 · 8/15 vs best baseline · 3/15 vs OptimFROG (Radiohead)

Full-Track Arena Results · avFLAC v1.2 (March 2026)Rezultati arene za celotne skladbe · avFLAC v1.2 (marec 2026)

The avFLAC arena splits tracks into acoustically distinct segments, encodes each with 43 candidates (including vFLAC v1.5 with MDL block probe), then MDL picks the smallest final output from arena stitch, aFLAC whole, or vFLAC whole.Arena avFLAC razdeli skladbo na akustično razločne odseke, vsakega kodira z 43 kandidati (vključno z vFLAC v1.5 z MDL sondo za bloke), nato MDL izbere najmanjši končni izhod med arena stitchem, aFLAC celodatotečnim ali vFLAC celodatotečnim.

Abyssal WINS −14,667B

Best whole-fileNajboljši celodatotečnilax_t6 (17,320,224B)
avFLAC outputavFLAC izhod17,305,557B (arena stitch)
Arena winnerZmagovalec arenevFLAC v1.5 (21/22 segments)
Signal typeTip signalaDynamic, varied (d=0.020–0.120)Dinamičen, raznolik

Ethereal Arc +7,042B

Best whole-fileNajboljši celodatotečnilax_t6 (7,951,589B)
avFLAC outputavFLAC izhod7,958,631B (VF_whole)
Arena winnerZmagovalec arenevFLAC v1.5 (16/16 segments)
Signal typeTip signalaTonal, uniformTonalen, enakomeren
Chapter V · Key DiscoveriesPoglavje V · Ključne ugotovitve

What the Data RevealedKaj so razkrili podatki

1.9×

AI Audio Is More Predictable

Scanner data across 10 songs proves AI-generated audio (Producer.ai) has mean difficulty 0.25 vs. human recordings at 0.50 at matched sample rates. AI music is structurally simpler — more sustained tones, less transient chaos. Business implication: a specialized codec for AI audio platforms could achieve significantly better ratios than general-purpose lossless codecs.

u=10→15

DCC Settled — First Time

The Digital Claustrum Controller reached a stable learned state on Pink Floyd 192kHz (30s, 563 frames). The critical parameter: 4-bit events instead of 7-bit. The finer granularity of 7-bit encoding produced too much noise for the controller to find a pattern. Coarser events → cleaner signal → stable adaptation. Two-pass architecture will eliminate the sequential bottleneck entirely.

+39%

OptimFROG's Genre Weakness

Industrial music (Rammstein) consistently catastrophically fails in OptimFROG — 39.3% worse than FLAC on the hardest 10-second clip, 26–28% worse on the others. 8Z-Audio also loses on Rammstein, but far less badly (+3–8%). This is a specific architectural opportunity: if 8Z can handle industrial music better, it becomes a competitive differentiator in an otherwise mature field.

∀ domains

Cross-Domain Transfer Works

DCC: TSP solver → FASTA encoder → audio encoder. Scanner concept: DNA pipeline → audio difficulty pre-screener. MDL framework: 8Z image encoder → audio frame selection. The same architectural patterns — MDL arbitration, per-block model competition, adaptive search governors — produce gains across every domain they've been applied to.

100%

Perfect Category Win Rates

On tonal content: 3 for 3. On dynamic content: 1 for 1. On the "easiest" category: 1 for 1. On buildups: 1 for 1. 8Z-Audio's multi-model approach was purpose-built for structured content — and the benchmark confirmed it. The remaining losses are concentrated in industrial/transient content where LPC's weaknesses are known and targeted for v1.6.

≡ Frozen

The Field Is 18 Years Stale

Every major lossless audio codec — FLAC (2001), WavPack (2002), TAK (2007), OptimFROG, Monkey's Audio — uses LPC as its primary or sole prediction model. No production codec performs per-frame multi-model MDL selection, harmonic modeling, or periodic template matching. The field has been architecturally frozen since ~2007. 8Z-Audio is the first systematic attempt to break out of that constraint.

−6.23%

Per-Segment Beats Whole-FilePer-segment premaga celodatotečno

Abyssal Arena v2.1 proved that per-segment encoding decisively wins on dynamically varied music. The track ranges from near-silence (d=0.004) to full climax (d=0.392) — acoustic variance that a whole-file encoder cannot exploit. lax+tukey won 10 of 12 segments with blocksize 8192 and LPC-32. The raw payload beats the best whole-file encoder by 1,078,771 bytes. Uniform tonal music (Ethereal Arc) shows no gain — the approach pays off precisely where real production music has diversity.Abyssal Arena v2.1 je dokazala, da per-segmentno kodiranje odločilno zmaguje pri dinamično raznoliki glasbi. Skladba se razteza od skorajšnje tišine (d=0,004) do polnega vrhunca (d=0,392) — akustična varianca, ki je celodatotečni kodirnik ne more izkoristiti. lax+tukey je zmagal na 10 od 12 segmentov z velikostjo bloka 8192 in LPC-32. Surovi zapis premaga najboljši celodatotečni kodirnik za 1.078.771 bajtov. Enakomerna tonalna glasba (Ethereal Arc) ne kaže dobička — pristop se obrestuje natanko tam, kjer ima prava produkcijska glasba raznolikost.

Chapter VI · What's NextPoglavje VI · Kaj sledi

What's Next: Two Parallel TracksKaj sledi: Dve vzporedni smeri

Development follows two parallel tracks toward OptimFROG-level compression. The xFLAC track (X1–X3) optimizes within the FLAC format. The AC Classical Max track (C1–C5) builds the native .8za format beyond FLAC's limits. Both share the MDL+DCC philosophy: let cost decide, not heuristics. Razvoj poteka po dveh vzporednih smereh proti kompresiji na ravni OptimFROG. Smer xFLAC (X1–X3) optimizira znotraj FLAC formata. Smer AC Classical Max (C1–C5) gradi lastni format .8za onkraj FLAC omejitev.

xFLAC · X3 (Next)

Stitching Elimination

Eliminate segment header overhead that turns raw compression wins into file-size losses. Frame-level competition or zero-overhead stitching.

Flip Ethereal Arc from +7 KB loss to win
AC · C1 (Next)

rANS Entropy Coder

Add rANS to MDL arena alongside Rice + LZMA. Per-partition selection. Captures non-Laplace residual distributions that Rice wastes bits on.

+2–4% on mixed/transient content
AC · C2

OLS+NLMS Predictor

Replace Levinson-Durbin LPC with OLS (actual sample matrix, no stationarity assumption) cascaded with NLMS adaptive filter. The biggest single gain.

+2–3% (closes 60% of OFR gap)
AC · C3

Joint Stereo Prediction

Cross-channel OLS taps — each channel uses past samples from both channels. Captures delayed and frequency-dependent stereo correlation that M/S misses.

+1–1.5% on stereo content
xFLAC · X1

Transient Detection

Onset-aware block splitting in vFLAC. 1024-sample blocks for drum attacks instead of wasting 4096-sample blocks on 50-sample transients.

+0.1–0.3% on mixed/transient clips
Critical Fix

24-bit Block Probe Bias

vFLAC v1.5 over-splits on 24-bit content (1.47 MB behind lax_t6 on LG-DWAS). Needs bit-depth-aware overhead model in MDL cost function.

Unblock 24-bit wins

Current Scorecard (March 2026)Trenutni rezultati (marec 2026)

avFLAC tested on 18 files (3 full tracks + 15 clips). All max compression, lossless verified. avFLAC testiran na 18 datotekah (3 celotne skladbe + 15 posnetkov). Vse max kompresija, brezizgubno preverjeno.

old: 7/15 (v1.5)
15 / 15
Clips beating FLAC-12
old: 10.3%
9.24%
Best win vs lax_t6
vs lax_t6
8 / 18
Files beating best baseline
vs OptimFROG
3 / 18
Files beating OFR (Radiohead)
The MusicGlasba

10 Songs, 6 Playable10 pesmi, 6 predvajalnih

Six original compositions by Bojan Dobrečevič, created via Producer.ai — used as the AI half of the 8Z-Audio benchmark corpus. Scanner data shows AI-generated audio is 1.9× more compressible than human recordings, making them ideal test subjects. Play the 10–30 second benchmark clip (WAV) or the full song (MP3).

▶   Select a song below to start listening▶   Izberite pesem spodaj za začetek poslušanja
Fractured Harmony
easiest tonal
8ZA vs FLAC-12: +5.1% win · +2.9% win
Where Do I Go
dcc_stress
8ZA vs FLAC-12: −2.0%
Awakening AI
tonal
8ZA vs FLAC-12: +1.5% win
Echoes of the Shore
dcc_stress
8ZA vs FLAC-12: −1.8%
Lacrimosa Requiem
diverse
8ZA vs FLAC-12: −0.2%
Between the Strings
dcc_stress
8ZA vs FLAC-12: −1.3%
Reference Recordings — Copyright protected, no playerReferenčni posnetki — Avtorsko zaščiteni, brez predvajalnika
Pink Floyd
Shine On You Crazy Diamond
192 kHz · 24-bit progressive
8ZA: +3.0% (dynamic) · +5.2% (tonal)
Metallica
Lux Æterna
thrash metal 44.1 kHz
8ZA: +0.03% (buildup) · −0.2% (transient)
Lady Gaga
Die With A Smile
pop 44.1 kHz
8ZA: +10.3% ← best win overall+10,3 % ← najboljša zmaga
Rammstein
Du Hast
industrial 48 kHz
8ZA: −2.7% to −8.0%
OFR catastrophic: −26 to −39% on industrialOFR katastrofalen: −26 do −39 % pri industrijskem