README.md

metadata

title: Neural Analog Engine
emoji: 🎛️
colorFrom: blue
colorTo: purple
sdk: gradio
app_file: app.py
pinned: false

🎛️ NEURAL-ANALOG ENGINE

Turning Large Language Models into Analog Synthesizers.

Most AI audio tools use black-box diffusion models to generate messy, artifact-heavy audio. The Neural-Analog Engine takes a radically different, elegant approach:

1. We prompt a 1.5B parameter LLM (Qwen 2.5).
2. We use PyTorch Forward Hooks to intercept the hidden state activations of its 15th neural layer in real-time as it generates text.
3. We pipe these raw, floating-point "thoughts" directly into a digital signal processing (DSP) oscillator.

The result is pure, artifact-free, analog-style ambient drones driven by the actual cognitive processing of an AI.

---

🎵 Powered by Livadies

Created to explore the intersection of machine cognition and acoustic art.

• 🟢 Spotify
• 🔴 YouTube
• 🟡 Yandex

🔥 Active Project Baseline: «RUSSIAN WINTER 26»