About

MUNRA is a personal audio ML research lab exploring generative models for creative music production.

What We Produce

Audio Samples & Loops

A growing library of unique recordings: drums, voice, and synths. Used for training models and available for direct use.

In progress — Target: 2,000+

Generative Models

Tiny neural networks (<50M params) that generate drums, voice, and textures. Trained on our sample library, optimized for lo-fi aesthetic.

In development — First model: June 2026

Documented Process

Every experiment is documented publicly: architecture decisions, failures, breakthroughs, insights. Complete transparency.

Experiment #001 in progress

Why This Approach

Most audio AI optimizes for perfection: pristine, studio-quality output. MUNRA does the opposite.

We train on imperfect, characterful audio: tape hiss, vinyl crackle, timing drift. The goal isn't to replace musicians, but to create unique building blocks for experimental music.