Chapter 10 · Reps

Creating with AI: Voice and Audio — Reps

← Back to Chapter 10

Chapter 10 — Reps

Eleven reps to turn Week 10 from something you read into something your ear and your judgment can do. They move from making audio, to signing for a voice, to catching a fake, to building the human agreements that no clone can forge. Your phone and your browser are the gym. Do not read these — do them.

Ground rules

  • Write it yourself. Type your own prompts, your own consent scope, your own reflections. No pasting a chatbot’s answer in as your own thinking. Your judgment is the thing being trained, and it does not train by copy-paste.
  • Run/try everything. Actually generate the audio, actually make the overview, actually listen. A rep you only imagined is a rep you did not do. Free tiers — plus the documented text-to-speech stand-in where a rep touches voice cloning (usually a paid feature) — cover every rep here.
  • Predict before you measure. Before you generate a voice or an overview, write down what you expect — how good, how convincing, what it will get wrong. Then compare. The gap between prediction and result is the learning.
  • AI usage note (every AI rep). End each rep that used a tool with one honest line: what you asked, what it got wrong or oversimplified, and what you verified yourself. You choose the tool; you own the verdict.
  • Where to work (CUW). BoodleBox — the AI platform Concordia licenses — is your default home for the writing and planning in these reps (scripts, prompts, memos, reflections): sign in with your Concordia account at box.boodle.ai, start a chat, and @-mention a strong model to draft with. BoodleBox is text- and image-first as of 2026, so the actual voice, music, and audio generation still happens in the specialized tools named in each rep. No campus access? Any free assistant does the drafting.
  • Keep a reps.txt. One short written reflection per rep, in a single document (Google Doc, Word, whatever you have). This is graded thinking, not busywork.

Reps 1–2: Text-to-speech and accessibility

Rep 1 — Turn a document into audio

Take one page of your own writing — an email you sent, a report intro, a lesson. In BoodleBox, first tighten it into a listenable script (spoken text wants shorter sentences and cleaner transitions than written text). Then paste that script into a free TTS tool — ElevenLabs’ free tier, ChatGPT voice, or any consumer TTS, since the voice generation itself is external — and generate spoken audio. Try at least two different voices. Listen to the whole thing.

Reflection: Was it good enough to actually send to a colleague? Where did it stumble — a name, an acronym, a number, an emphasis? Predict-vs-actual: was it better or worse than you expected? End with your AI usage note.

Rep 2 — The accessibility pass

Now reframe Rep 1’s audio for a specific person: a colleague with low vision, a teammate with dyslexia, or a commuter who cannot read at the wheel. Regenerate if needed so it is genuinely listenable (pace, pauses, clear pronunciation of key terms).

Reflection: Name one real person (or role) whose access to your work just widened. What did you have to fix to make it work for them rather than just “for anyone”? This is the safe, high-value center of the whole chapter — say why it matters.

Rep 3 — Read and fill the consent form

Open code/voice-consent-form.txt and fill it in for your own voice. Do not skim it — fill every field: whose voice, permitted uses, forbidden uses, disclosure line, retention, revocation. Save your completed copy.

Reflection: Which field made you stop and actually think? Most people trip on “forbidden uses” or “how consent is revoked.” What would have gone wrong if you’d skipped it? Note one thing a machine could never fill in here on your behalf.

Rep 4 — Clone (or high-fidelity narrate) your own voice

Using a consumer voice tool’s free or trial tier, create an instant clone of your own voice, or — if no cloning tool is available to you — pick a high-quality TTS voice as a documented stand-in. Then generate the provided script code/voice-script.txt in that voice, including the spoken disclosure line at the end.

Reflection: How convincing was it as you? What was missing — the breath, the hesitation, the little imperfections that make a real voice real? Predict-vs-actual on the quality. End with your AI usage note (name the tool and, if you can, note whether it applied a provenance credential).

Rep 5 — Spot the fake

Listen back-to-back to a synthetic voice (your Rep 4 clip, or a demo) and a real human recording (a voicemail, a podcast, a family video). Write down, before you know which is which if a friend can help you set that up, what tips you off to the fake.

Reflection: What was the tell — or was there none? Be honest: if you couldn’t reliably tell, say so, and write one sentence on what that means for “I recognized their voice” as proof. This is the point of the whole chapter; do not flinch from it.

Reps 6–7: Audio overviews and grounding

Rep 6 — Make a NotebookLM audio overview

Gather 3–5 documents on one topic (articles, a report, meeting notes — nothing confidential for this rep). Upload them to NotebookLM and generate an Audio Overview. Listen to the whole thing.

Reflection: What did the two hosts get right about your sources — and where did they oversimplify or over-agree? Rate how much you understood vs how much you just enjoyed. Note the format limits (pre-styled voices, caps). End with your AI usage note.

Rep 7 — Fact-check one claim against the source

From your Rep 6 overview, pick one specific factual claim the hosts made — a number, a date, a “the report says.” Now open the actual source document and try to find the sentence that backs it.

Reflection: Did you find it exactly, find it approximately, or fail to find it? Tie your answer to §10.4: an audio overview drafts your understanding; it does not replace reading where the stakes are real. What would you not be willing to act on from the audio alone?

Reps 8–9: Music and its unsettled questions

Rep 8 — Generate a short track

In BoodleBox, draft the prompt (and any lyrics) first — @-mention a strong model to shape the mood, tempo, and words. Then paste that prompt into Suno or Udio (free tier) to generate a short piece of music — a jingle for an imagined small business, or background music for a video. Iterate the prompt at least once.

Reflection: How close did it get to what you imagined? Now the important part: could you legally use this in a paid client project today? Write what you actually know (unsettled; active litigation; read the terms) versus what you’d need to confirm — and who you’d ask. Do not fake certainty.

Rep 9 — Read the terms

Open the terms of service / usage or licensing page of the music tool you used in Rep 8. Find what it actually says about commercial use and ownership of what you generate.

Reflection: In two sentences, summarize what the terms say you may and may not do commercially. Where were they vague or conditional? This is the habit that keeps you out of trouble: read before you sell, not after.

Reps 10–11: Ethics, security, and disclosure

Rep 10 — Run the vishing drill

This is the highest-return rep in the chapter and it involves no AI at all. Design an out-of-band verification protocol for your family and one for your team/workplace: (a) a pre-agreed code word for emergency calls, (b) a call-back rule (call a number you already had, never the one that called you), and (c) a dual-approval rule for any money movement. Write it down. If you can, actually have the five-minute conversation with your family.

Reflection: Which of your protocols would have stopped the cloned-grandchild scam? The deepfaked-executive scam (recall the Arup case, ~US$25.6M)? Why does verifying through a second channel beat “listening harder”? Connect it to §10.6.

Rep 11 — Write your disclosure line

Draft the exact one-line disclosure you would attach to a synthetic-voice clip you publish — spoken, written, or both. Then draft a second version for a context where a full sentence is awkward (a short social clip) and you have only a few words or a label. (In BoodleBox you can rough out a few candidate wordings; the one you would actually stand behind, you pick.)

Reflection: Why is disclosure worth the small awkwardness? Name the trust you’re protecting and the person on the other end. Tie it to §10.7 and to the Eighth Commandment framing in §10.9 — what does undisclosed synthetic voice quietly spend?

Done? One Last Thing.

A miniature of Project 10, end to end, in one sitting. Produce, in a single folder or shared doc:

  1. A signed consent form — code/voice-consent-form.txt, filled in for your own voice, with a real scope and a revocation clause.
  2. A short disclosed clip — the provided script code/voice-script.txt, rendered in your cloned or high-fidelity voice, ending with the spoken disclosure line.
  3. A three-sentence ethics note — one sentence each on consent, disclosure, and misuse boundaries for this specific clip.

If you can do those three tonight, you can do Project 10. That is the whole discipline in miniature: the tool makes the voice; you supply the consent, the disclosure, and the verdict.

Up next: Project 10 — Project P10: Your Voice, With Consent.