Skip to main content

Understanding How the AI Works

AudioNova uses a large language model (LLM) — the same kind of AI that powers tools like ChatGPT — to interpret your mixing prompts.

Unlike traditional audio plugins, it doesn’t use fixed knobs or settings. Instead, it “understands” what you’re saying and decides what to do based on context.

🧠 How the Language Model Interprets Prompts​

When you send a prompt like:

Add reverb to the vocals, but keep them intimate and close

Here’s what’s happening under the hood:

  1. The model reads your text
  2. It parses your intent — not just keywords, but tone and phrasing
  3. It converts that into a high-level audio command
  4. That command is translated to backend DSP or routing logic (via code)

💬 What the AI is Good At​

  • Understanding adjectives
    → warm, punchy, dreamy, gritty, clean

  • Understanding musical context
    → club-ready, bedroom pop, film score energy

  • Sequencing logic
    → compress → EQ → delay

  • Handling nuanced comparisons
    → Make the mix feel like Bon Iver but with EDM drums

🧩 What the AI Might Struggle With​

  • Contradictions
    → Make the vocals dry and reverberant
    → too ambiguous, clarify your intent

  • Overloaded prompts
    → Make the snare louder, dry, wide, slower, then echo and more vintage but still tight
    → break it into steps

  • Uncommon slang
    → Use industry-standard terms when possible

  • Literal vs abstract
    → Make it blue and sleepy might work... or not. Add audio-specific phrasing.

🧪 Model Behavior: Things to Know​

  • Prompts are stateless within each message — but sessions retain some context
  • Prompts are interpreted by patterns, not by rules — it’s learning from language, not knobs
  • Parentheses () may emphasize or group weight
  • Brackets [] can add context or references

🧠 Prompting Principles (Recap)​

PrincipleDescription
Clarity over clevernessBe specific about the result you want
Break up instructionsStep-by-step is better than all at once
Use referencesArtists, genres, gear, or FX plugins
Think like a producerDescribe intention, not settings

🔬 Example: Prompt Deconstructed​

Make the vocals smoother, a bit warmer, with subtle reverb and compression

The AI will likely interpret this as:

  • Smoothing = compression + de-essing
  • Warmth = EQ boost at 200–600Hz
  • Subtle reverb = low mix level, short tail
  • Compression = medium ratio, soft attack

AudioNova is powered by language — and the more clearly you speak to it, the more it becomes your mix engineer. 🎚🧠