Signal Over Noise #28

November 12th, 2025

Dear Reader,

I took a gamble this week and decided to put my AI Writing Field Guide up on Product Hunt today. If you’d like to show your support, log in and give it an upvote or a comment/review if you’ve bought it.

Don’t forget that you can also get your own copy with €10 off until Saturday by following this link. As a Signal Over Noise subscriber, this should really help you on your journey using generative AI.

​Last week, I explained why perfect AI voice clones are a security liability — especially during vishing season (so did the FT, a few days later ;-)).

This week we’re going to look at how to build a voice agent that represents you without replicating you.

The goal isn’t to avoid AI voice tools — they’re far too useful at this point. The goal is strategic differentiation: building voice agents that are obviously yours when you want them to be, and obviously NOT trying to impersonate you when security matters.

And here’s how to do it.

Stop Building Clones, Start Building Named Agents

My rationale behind this shift is security-first: Your voice agent shouldn’t use your voice at all.

Your Voice Agent Has:

  • A name (not your name)
  • A different voice (from a platform library, distinctly NOT yours)
  • Your content (your scripts, your thinking, your strategic messaging)
  • Its own identity (obviously an agent presenting your work)

Think of it like having a professional narrator for your content, named “Mercury” (for example). Mercury doesn’t sound like you - Mercury sounds like, well, Mercury, a distinct voice that your audience recognises as your content delivery agent.

  1. Security through differentiation - Your voice = real-time you, Mercury’s voice = pre-recorded content
  2. No clone vulnerability - Scammers can’t use “your” voice because Mercury doesn’t use your voice
  3. Professional consistency - Mercury delivers content with reliable quality (even on sick days, or when you just can’t face dealing with other people)
  4. Clear boundaries - Your voice for conversations, Mercury’s voice for content

The difference matters. A scammer could clone YOUR voice from podcast audio. They can’t impersonate Mercury because Mercury isn’t trying to sound like you in the first place.

I’d like to see us moving towards a future where we are thinking “agents working on my behalf” instead of “recreating real things in AI” (although I often think I’d like to get my voice cloned to be used as a snarky satnav for my kids and grandkids, but I digress).

Voice Agent Implementation

What You Need:

  • A voice platform: ElevenLabs or similar with a voice library
  • Your written content (scripts, articles, podcast outlines)
  • A clear agent identity (important if you’re going to create multiple agents at any point)

Step 1: Choose Your Platform & Voice (1 hour)

You can also play around with AI voice generation on openai.fm​

​ElevenLabs Creator ($22/month):

  • Professional voice library (100+ voices)
  • 100,000 characters (~1.5 hours audio/month)
  • High-quality narration
  • Best for: Podcast narration, course content, professional content

​SUZA Voice Studio (Free, Windows)​

  • Free to use
  • Complete offline generation
  • No app tracking
  • No APIs required
  • No internet required

​Voices by Jordi Bruin (Pay what you want, Mac)

  • Use ElevenLabs, OpenAI and PlayAI locally (bring your own API key)
  • Generate locally with Kokoro (like SUZA, completely offline)
  • (My personal favourite, btw)

Key Decision: Select a voice from their library that:

  • Sounds professional and authoritative
  • Is distinctly different from your actual voice
  • Fits your content style (conversational vs. formal)
  • Your audience will recognize as “Mercury”

You’re NOT cloning your voice. You’re selecting Mercury’s voice.

Step 2: Define Your Voice Agent Identity (30 minutes)

Create your agent profile:

VOICE AGENT PROFILE: MERCURY

Primary Function: Content narration for [Your Name]

What Mercury Does:

  • Narrates podcast episodes (from your scripts)
  • Voices video content (from your outlines)
  • Reads articles as audio versions
  • Delivers training content you’ve written

What Mercury Is:

  • A professional narrator for your content
  • A distinct voice (not yours) that represents your work
  • Always identifies itself at the start
  • Only presents pre-written/approved content

What Mercury Never Does:

  • Make phone calls
  • Conduct live conversations
  • Make independent decisions
  • Handle urgent requests
  • Represent you in real-time interactions

Mercury’s Voice:

  • [Voice name/ID from ElevenLabs library]
  • Professional, clear, authoritative
  • Distinctly NOT your actual voice
  • Consistent across all content

Security Through Differentiation:

  • Your voice = real-time conversations, phone calls, meetings
  • Mercury’s voice = pre-recorded content delivery only
  • Scammers can’t clone “your” voice via Mercury because Mercury doesn’t use your voice.

But why go through this elaborate last step? If at some point you start using Mercury as a text-based agent (which we’ll get to next week), you’ll want some kind of crib sheet to make sure you know what it’s personality sounds like.

If you’re just testing things out, you don’t need to do this step. But you may want to think about some variations of a voice agent that might work for you:

  • one that narrates your TikTok / Insta content
  • one that reads back meeting summaries and gets sent round to your team as a podcast
  • one that sends voice messages to the team etc

Step 3: The Critical Protocol

Mercury MUST:

  1. Always identify itself at content start: “Hi is Mercury, Susie’s voice agent.”
  2. Use consistent professional delivery:
    • No filler words or casual patterns
    • Structured, clear narration
    • Maintains professional tone throughout
  3. Only present your pre-written content:
    • You write the script/outline
    • Mercury narrates it
    • No independent content generation
  4. Never appear in real-time contexts:
    • Pre-recorded content only
    • Never on phone calls or live meetings
    • No interactive conversations

Why This Works:

Scammers trying to impersonate you face a problem: They can clone your voice from public audio, but “Mercury” doesn’t use your voice so they can’t “fake Mercury” to scam your team. And your team knows: Your voice = real-time, Mercury’s voice = content only.

Real World Example: Podcast Production

Traditional Process (Without Voice Agent):

  1. Write podcast script (2 hours)
  2. Set up recording equipment (15 minutes)
  3. Record episode with multiple takes (1-2 hours)
  4. Edit for mistakes and pacing (1-2 hours)
  5. Export and process (30 minutes)

Total: 5-6 hours

With Voice Agent (Mercury):

  1. Write podcast script (2 hours)
  2. Generate audio with Mercury (15 minutes)
  3. Review and approve (15 minutes)
  4. Minor edits if needed (15 minutes)

Total: 2.5-3 hours

Time Saved: 2.5-3 hours per episode

Early readers may remember my automated podcast experiment from 2023, which was perfect when I had a head cold or lost my voice — the content was able to keep going and I didn’t need to worry about extra takes and over-editing.

Why Agents Beat Clones (Beyond Security)

​The vishing threat is real, but that’s not the main reason to start seriously thinking about using named agents instead of perfect clones.

We’re moving from “AI that pretends to be you” to “AI that works for you.” That’s not just semantics - it’s a fundamental rethinking of how humans and AI systems should interact.

Perfect clones have three problems:

  1. Security vulnerability (what we covered last week)
  2. Accountability confusion (who said what? who made that decision?)
  3. Scaling limitations (you can’t be in five places at once, even digitally)

Named agents solve all three:

  1. Security through transparency - Everyone knows they’re interacting with your agent, not you
  2. Clear accountability - “Mercury” handles content production, you handle real-time decisions
  3. True scalability - Your agent can handle content production while you focus on strategy

When your team gets comfortable with “Mercury narrates our podcasts, but Susie handles all real-time communication,” they’re not just learning security protocols. They’re learning the boundaries that will define how we work with AI for the next decade.

This is about building the right architecture from the start — one that’s more secure, more accountable, and actually scalable — instead of chasing the perfect clone that creates more problems than it solves.

A named voice agent with clear scope and protocols is: - Easier to verify than hidden AI pretending to be you - More scalable for content production - More defensible against social engineering - Better preparation for the agentic AI future

Take the time this week to consider how you, your team and execs are putting themselves out there on the internet. Start thinking about how you could create your own in-house agent and what roles and personality it/they would have.

Your clearly-identified voice agent isn’t a compromise on authenticity. It’s your content production tool AND your security layer.

Until next week,

Jim

€9.00

€0.00

The AI Writing Field Guide

Stop Sounding Like Everyone Else Using AI
The 5-step system for teaching AI what your voice actually sounds like — not… Read more

Get it now!

​

​

Signal Over Noise is weekly, reader-first publication on AI “without the hype” published by Jim Christian. If you’ve been forwarded this issue, you can subscribe for free: go.signalovernoise.at. You can also Join the free Signal Over Noise Community on Skool where I go into much more detail.

Made with ❤️ in Valencia by Jim Christian. For feedback, please reach out to [email protected].

Built with Kit​