Launching today
PHARMAKON
Studio Audio Antidote & Hardware-Grade Mastering Console
2 followers
Studio Audio Antidote & Hardware-Grade Mastering Console
2 followers
Universal Studio-Grade Audio Restoration, De-Harshing & Stem Mastering Pipeline by Lex Gerek
Launch Team
AWS Activate CreditsCredits that fuel your startup’s next milestone
Promoted
Hi Product Hunt! I'm Lex Gerek, Sound Architect behind PHARMAKON.
Over the last year, generative audio (Suno, Udio) has made massive strides musically, but it constantly suffers from severe acoustic pathologies: harsh 5 kHz sibilance, phase-smeared low end, and rigid digital clipping.
Most existing tools try to treat this with bloated 4 GB neural networks that destroy phase coherence and take minutes to render.
We built PHARMAKON 2.0 as an ultra-fast hardware-grade mastering console running on pure mathematical DSP (NumPy & SciPy):
- 15 MB standalone footprint (zero PyTorch, zero CUDA bloat).
- 4-stem separation in 3 seconds flat via HPSS and Mid/Side center-purity coherence filters.
- Analog tape warmth (tanh), 16 kHz harmonic silk synthesizer, and 4X oversampled true-peak brickwall limiter.
- Real-time 60 FPS CRT phase scopes with delta residue auditioning.
The core repository is open on GitHub for the community to test through this weekend before we transition into private licensing.
Would love to hear your thoughts, feedback, and questions from producers, audio engineers, and developers!
A quick technical deep-dive for the audio engineers and producers here:
The #1 question I get asked is: "Why can't I just slap an equalizer on my Suno/Udio track and boost a +8 dB high-shelf at 16 kHz to let it breathe?"
Here is the acoustic reality of why equalizers fail on generative audio:
1. You cannot multiply zero:
An EQ is a linear scaling filter ($Output = Input \times Gain$). If there is no acoustic signal at a given frequency, the math is simple: $0 \times 2.51 (+8\text{ dB}) = 0$.
2. The 1600 dB/octave Brickwall Cliff:
Generative vocoders (trained at 32 kHz) and distribution codecs (128 kbps MP3) hard-truncate frequencies at 15.5 - 16.0 kHz. The spectrum drops by 60-80 dB across just 500 Hz. Above 16 kHz, useful musical signal is literally zero.
3. The Noise Floor Trap:
What DOES exist above 16 kHz is residual quantization noise and vocoder phase jitter (-85 dBFS). When you crank a high-shelf EQ, you don't resurrect cymbals: you simply amplify that digital noise floor by +8 dB, turning dullness into ear-fatiguing hash.
How we solved this inside PHARMAKON 2.0:
Instead of trying to EQ dead air, we synthesize it from scratch:
- We isolate the intact donor band (8 - 12.5 kHz).
- We apply 4X polyphase oversampling (pushing Nyquist to 96 kHz) to mathematically eliminate 3rd-harmonic aliasing foldover (which would otherwise reflect down to 6.6 kHz).
- We generate genuine 2nd & 3rd order physical harmonics via asymmetric polynomial synthesis and pre-filter tanh saturation.
- A steep 4th-order zero-phase highpass filter isolates pure air above 15.5 kHz and blends it back into the master bus.
Stop boosting quantization noise. Reconstruct the lost air with signal physics.
Curious to hear how other producers here handle the top-end cutoff on AI stems!