“LΞX” GEREK

Treating Generative Audio: Pure Mathematical DSP vs Neural Vocoder Fatigue. Direct A/B and Delta phase-null tests inside the PHARMAKON 2.0 monitoring bay. Generative AI music from Suno and Udio sounds impressive for the first thirty seconds... until your ears fatigue from metallic 5 kHz sibilance and blurred sub-bass phase. We built the antidote: - 15 MB standalone architecture (pure NumPy & SciPy) - 4-stem separation in 3 seconds flat - Tanh analog tape warmth & 16 kHz harmonic silk - Real-time 60 FPS CRT Lissajous phase scopes The core repository is open on GitHub through this weekend before we move into private licensing: github.com/alex-gerek/PHARMA… Stop accepting digital plastic. Treat your audio. #Suno #SunoAI #Udio #UdioAI #AudioEngineering #DSP #MusicProduction #StemMastering #LexGerek

My GitHub account (alex-gerek) was temporarily flagged by an automated system. I have submitted an appeal to GitHub Support to resolve the issue. Development on PHARMAKON 3.1 continues as planned.

Sometimes the best soundtrack is no soundtrack at all. Baltic Sea. Rügen.

Spent three days investigating who killed the bass. Turns out it was the bass. Great. Now I need to investigate the limiter. Hahaha

A couple years ago we started putting little audio processors inside toy cars. We'd buy model cars, gut them, stuff our circuits inside and attach the whole thing to microphone stands. Why cars? Fuck knows. We thought it was funny. And those fancy black aluminum boxes were boring as hell. The idea was to deal with some of the usual live vocal bullshit. Sibilance, stage rumble, sudden peaks. Even a $3,000 microphone won't magically fix all that. I remember this singer walking into the studio, staring at the little Mustang hanging off her mic and asking what the fuck that thing was. So I started explaining the signal processing. About thirty seconds in I realized I was giving a fucking electronics lecture to someone who'd asked about a toy car. Most vocalists actually liked those stupid cars. Pretty sure a few thought we were complete fucking idiots. Whatever. We were having fun. Did we stop buying expensive microphones? Fuck no. I keep an engineering logbook. All the crazy shit we build or plan to build goes in there. Schematics, calculations, designs. Pretty much everything. Found this drawing while going through it. Still think putting a DSP inside a Mustang was a damn good idea. #AudioEngineering #StudioHumor

PHARMAKON 2.0 — Core DSP Patch & Smart Auto-Update Build (v2.0.2) is live. What’s new in this release: • Isolated DELTA POISON (.WAV) Export — solo and inspect the exact AI harshness, vocoder whistle, and phase mud removed from your track. • 4X Oversampled Asymmetric Tape Stage — global asymptote normalization (-38.05 dBc H2 warmth) + 10 Hz recursive DC-blocker (Whitepaper v2.6). • Full-Track 4X True-Peak Audit — ITU-R BS.1770-5 scanning (-0.3 dBFS ceiling) with soft-knee tanh safety protection and calibrated Audiophile mode. • Seamless Online Patch Sync (AUTO-SYNC) — starting with this build, PHARMAKON automatically checks and applies future DSP & console patches whenever you’re online (while remaining 100% functional offline). If you’re on the initial release, download this updated build once — future patches will arrive automatically: github.com/alex-gerek/PHARMA…

Most mastering tools operate as black boxes: you hear the result, but you never see what was stripped away under the hood. Today we upgraded the PHARMAKON 2.0 Mastering Report with full Delta Telemetry. The report now visualizes and audits the exact acoustic delta: the difference between your raw track and the final master. You can clearly inspect that only harsh sibilance, MP3 phase smear, and unmusical resonances were eliminated, while your musical core and transients remained completely untouched. Complete transparency over faith. Full audit report pictured below. Download & Source: github.com/alex-gerek/PHARMA…

Why do headphone mixes fall flat in the club? It’s not mystery—it’s phase cancellation and acoustic fatigue. An engineering deep-dive into club revenue, dancefloor physics, and our deterministic C++20 signal conditioning solution.

PHARMAKON PRO: Native DAW Plugin Sneak Peek. Here is a first look at the real-time C++20 DSP engine running live inside Cockos REAPER. We chose REAPER for this demonstration to show zero-latency bridging, instant docking, and seamless integration into professional master bus workflows: - Native VST3 / AU / CLAP architecture with sub-millisecond processing. - Real-time 60 FPS CRT vector scopes (phase correlation, dynamic crest factor, and stereo field). - Zero GPU overhead, zero external dependencies, and instant parameter recall. - Hardware-modeled analog tape saturation and 4-band stereo matrix mastering. Our full engineering whitepapers and deep-dive technical articles will be published on tech media platforms mid-week (Tuesday / Thursday). For now, here is a preview of the audio engine processing "Pure Voltage" live on the master bus. Native plugin release lands this coming week. #ReaperDAW #VST3 #AudioEngineering #DSP

To the 134 brave souls who actually ran `git clone` in their terminal and fought Python virtual environments this week: we see you, we respect the dedication, but we decided to show mercy. Let's be realistic: when you are working on a track, the last thing you want to do is install Python runtimes, manage pip dependencies, or debug terminal path errors. We packaged the entire PHARMAKON 2.0 studio console into a standalone, portable Windows edition: - Zero installation, zero Python, zero command line. - Download the zip (139 MB), extract it anywhere, and launch. - Drag and drop your audio file directly into the browser console. - Instant 4-stem DSP separation in ~3 seconds on an ordinary laptop CPU. - Built-in analog tape saturation, sub-bass phase alignment, and True-Peak brickwall ceiling. Direct download link: github.com/alex-gerek/PHARMA… (And for those waiting for the real-time C++20 DAW plugin with 60 FPS vector scopes: the commercial PHARMAKON PRO release is landing this week).

SPECTRAL GEOMETRY VS GIGABYTE WEIGHTS: FAST DETERMINISTIC DSP FOR VOCAL ISOLATION Author: Lex Gerek (Sound Architect / Audio Engineer) Studio: Lazuli Pictures × PHARMAKON Studio Console 1. THE PROBLEM: INSTRUMENT BLEED & NEURAL BLOAT Modern stem demixing has predominantly migrated to heavy deep learning architectures (HTDemucs, MDX23, BS-Roformer). While they produce impressive results in offline studio environments, they introduce severe production friction: - Downloading 2 to 6 Gigabytes of PyTorch weights. - Prohibitive GPU requirements (CUDA or Apple Silicon). - Pervasive phase artifacts (diffusion warble, smudged drum transients). The main headache for producers working with generative audio (Suno v3.5/v4, Udio 1.5) is acoustic bleed: rhythm and lead guitars aggressively leak into the vocal stem because distorted electric guitar overtones share the exact same frequency zone as human vocal formants: 300 - 3800 Hz. Does clean stem separation truly demand a multi-gigabyte server cluster? No. By leveraging classical time-frequency geometric distribution and inter-channel statistical coherence, stem demixing executes via pure deterministic mathematics in ~3 seconds on a standard laptop CPU. 2. THE 4-STEM DSP PIPELINE In PHARMAKON 2.0, stem separation operates as a zero-phase pipeline across the STFT complex plane: Stage 1. Infrasonic & Sub-Bass Isolation (< 120 Hz): Steep 4th-order Butterworth crossover. All acoustic energy below 120 Hz is consolidated into a pure True-Mono stream: M = 0.5 * (L + R), guaranteeing 100% mono punch and zero subwoofer phase cancellation. Stage 2. Harmonic-Percussive Source Separation (Ono 2D Median Filter): Decomposes spectrogram magnitudes into: - Continuous horizontal harmonic formants (vocal sustains, guitars, chords). - Broadband vertical transient spikes (snare attack, kick punch, cymbal hits). Stage 3. Wiener Mask & Phantom Center Vocal Dome (300 - 3800 Hz): Isolates the phantom stereo center and applies a smooth Wiener formant dome covering 300 to 3800 Hz. Off-axis guitars and stereo width instruments are cleanly filtered out. Stage 4. The Residual Stem (Other): Strict phase-null subtraction: Other = Full_Mix - (Bass + Drums + Vocals) Summing all 4 stems reconstructs the master file with exact floating-point precision to 0.0 dB without phase distortion. 3. BENCHMARK: PURE DSP VS DEEP NEURAL MODELS - Processing Speed: ~3 Seconds (PHARMAKON) vs 45-120 Seconds (Demucs). - Download Footprint: 0 MB (Embedded DSP) vs 2-6 GB PyTorch weights. - Hardware: Standard Laptop CPU vs High-End CUDA GPU. - Transients: Crisp, punchy drum attacks vs Blunted, smeared attacks. - Artifacts: Zero Neural Hallucinations (Deterministic) vs Diffusion warble and chirping. 4. EXPERIENCE THE ALGORITHM - Free Standalone Console (PHARMAKON 2.0): Download portable zero-install package (139 MB): github.com/alex-gerek/PHARMA… - Professional Real-Time DAW Plugin (PHARMAKON PRO): VST3 / AU / CLAP for Ableton, REAPER, Logic, and FL Studio with 60 FPS CRT scopes (Launching Tuesday).

Why EQ Fails on Generative Audio: The Physics of Dead Air Above 16 kHz The most common advice given to producers trying to fix dull audio generated by Suno or Udio is: "Just slap an EQ on the master bus, boost a high shelf at 16 kHz by +8 dB, and your mix will instantly breathe." It seems intuitive: the track sounds claustrophobic, so boosting high frequencies should open it up. In practice, it sounds terrible. Instead of silky air and vocal shimmer, you get harsh metallic screech, digital grain, and abrasive hiss. Why does an equalizer fail on generative audio? The answer lies in signal mathematics. The Fundamental Principle: An Equalizer Only Multiplies An equalizer is a frequency-selective multiplier. It multiplies the amplitude of energy already present at frequency f by a gain factor K: Output(f) = Input(f) × K(f) A +8 dB shelf represents an amplification multiplier K ≈ 2.51. If the acoustic signal at frequency f is physically zero (Input = 0), the equation yields zero: 0 × 2.51 = 0 An equalizer cannot synthesize a single missing harmonic. It can only increase the volume of information already present in the recording. The 15.5 kHz Brickwall Cliff To optimize compute budgets and model sizes, generative vocoders render audio using low sampling rates (often 32 kHz with a 16 kHz Nyquist ceiling) or apply steep anti-aliasing cutoffs at 15.5 kHz. Viewing raw generative audio in a spectral FFT analyzer reveals an unnatural vertical cliff: • Full acoustic energy below 15 kHz. • At exactly 15.5 kHz, energy plunges off a precipice with a slope exceeding ~1400 dB / Octave, dropping into a -85 dBFS digital void. Above 15.5 kHz, there is ZERO acoustic signal: no cymbal overtones, no vocal breath, and no room reverberation decay. Only low-level quantization noise and diffusion phase jitter remain. Boosting a high shelf at 16 kHz simply pulls that -85 dBFS noise floor up to -77 dBFS, converting dullness into ear fatigue. The PHARMAKON 2.0 Solution: 4X Polyphase Harmonic Synthesis Instead of equalizing dead air, PHARMAKON 2.0 physically synthesizes replacement harmonics from intact donor bands: Donor Band Selection (8 - 13 kHz): We isolate the upper-mid acoustic band using a zero-phase 4th-order equivalent bandpass where musical information remains untouched by the vocoder cutoff. 4X Polyphase Upsampling (192 kHz): Generating 3rd-order harmonics on a 12.5 kHz donor yields frequencies up to 37.5 kHz. At standard 48 kHz, this creates severe foldover aliasing at 10.5 kHz (|48 - 37.5| = 10.5 kHz), and down to 6.6 kHz on 44.1 kHz (|44.1 - 37.5| = 6.6 kHz). Running in a 192 kHz polyphase domain extends the Nyquist limit to 96 kHz, mathematically eliminating aliasing. Asymmetric Taylor Polynomial: Signal is routed through: y = x + 0.65x² - 0.25x³ The quadratic term generates musical even 2nd harmonics (2f), while the cubic term produces odd 3rd harmonics (3f), reconstructing a continuous dense air taper across 16 - 22 kHz. Soft pre-filtering tanh(2.2y) × 0.45 tames high-frequency crest spikes. 8th-Order Equivalent Zero-Phase Isolation (filtfilt): A 4th-order Butterworth highpass filter processed forward and backward via filtfilt yields an 8th-order equivalent response (-48 dB/Octave) with zero phase rotation, preventing comb-filtering cancellations when blended into the master bus. Headroom Notice: Harmonic air injection introduces a minor peak energy rise (+0.2 to +0.5 dB). In the PHARMAKON pipeline, this is naturally handed off to our 4X True-Peak Limiter stage (Article 05) to ensure absolute broadcast ceiling compliance. Key Takeaway: We do not believe in fighting AI audio artifacts with another heavy, hallucinating diffusion model. Clean physics. Pure deterministic DSP. Zero latency on CPU. Stop multiplying zero. Treat the signal. Open-source repository & studio console: github.com/alex-gerek/PHARMA…