Voices, Music Beds and Sound Design

A strong script read by the wrong voice is a weak ad. A great read buried under loud music is a weak ad too. This lesson covers the two halves of what listeners actually hear: the voice and everything around it.

Human talent or AI voice

You have two broad choices for the voice, and both are legitimate.

Human voice talent gives you judgment. A good actor finds the warmth in "two dollars a month" without being told, adjusts to a note in one take and sounds like a real neighbor. Talent costs money: a session fee, and often a usage fee that depends on how long and where the ad runs. Local talent for a small-market radio spot is often affordable, and many will negotiate for a cause they like. Get the usage terms in writing: which media, which markets, how many weeks.

AI voices give you speed and control. You can rewrite a line at midnight and hear it in seconds, try ten deliveries, or build a two-voice dialogue without booking two people. Jeff's free tools do this: Voice Studio turns text into speech, and Dialogue Editor builds multi-voice scripts with ElevenLabs voices, which suits a testimonial or a two-person scene like the Walt story in lesson 2.

Where AI voices struggle: long emotional arcs, comic timing, place names and local pronunciations. Every town has a street everyone says wrong to outsiders. An AI voice will say it the outsider way. Listen for it, and respell the word phonetically in the text until it comes out right.

A sensible approach for a small campaign or nonprofit: draft and time the spot with an AI voice, use that draft to get the committee's approval, then decide whether the final runs with the AI voice or a human. The AI draft is also the best timing reference you will ever give a voice actor.

Consent and disclosure

These rules come from honesty first and law second. Follow all of them for every ad.

• Never clone or imitate a real person's voice without their informed, written consent for that specific use. That includes the business owner, the candidate, your own volunteer. ElevenLabs' use policy prohibits cloning or impersonating the voice of any living person without their explicit and informed consent, and it bars political campaigning without the company's prior written consent, so get that approval before using its voices in a levy or candidate ad. Other platforms have similar rules.
• Never put words in a real person's mouth with a synthetic voice, even with satire as the excuse. In an election ad, this is exactly what Idaho's synthetic media law, Idaho Code 67-6628A, targets: a candidate deceptively shown saying or doing something they did not can sue. Lesson 6 covers its required disclosure.
• A generic AI narrator (a stock synthetic voice that does not imitate anyone) is different from a clone. No real person is being represented. Still, the clean habit is to say so somewhere the audience can find it: a line in the video description, the website or the end card. Some ad platforms require disclosure of synthetic content in political ads. Lesson 6 has specifics.
• Testimonials need a real person. If an ad presents someone as a real neighbor telling their story, it must be a real person, saying things they actually believe, with their permission. An AI voice reading an invented "testimonial" as if it were real is a fabricated endorsement. Do not do it. If you use actors to dramatize, make it obviously a dramatization.
• Phone calls are a separate world. The FCC ruled in February 2024 that AI-generated voices count as "artificial" voices under the Telephone Consumer Protection Act, so robocall consent rules apply to them. Do not take a radio spot and turn it into an automated call without learning those rules.

Directing a read

Whether the voice is a human or a text-to-speech model, you are directing. Good direction is specific and gives the voice something to play.

Before the session, send the talent the script, the brief's audience line and a reference: your AI draft or a spot you like. Tell them the exact length.

During the session, useful notes look like this:

• A situation: "You're telling a friend at the hardware store, not announcing to a crowd."
• One word to hit: "Lean on 'Saturdays.' Throw away 'about.'"
• A pace target: "That was 31 seconds. Same feel, find me two seconds in the middle."
• A contrast: "Try one drier, like you've said this a hundred times and still believe it."

Notes that waste a take: "more energy" (usually gets louder and faster), "make it pop," "better," "more natural." Say what you hear and what you want instead.

Always record three or four complete takes even when the first is good. Then record pickups: single lines again, with a slightly different read, so the editor has choices. Mark your favorite takes on the script as you go.

With an AI voice, the same notes translate into changes in the text and settings. Punctuation is your pacing tool: a period makes a full stop, a comma makes a short one, and a line break in the script often adds a breath. Generate several takes, compare them on the same speaker, and keep notes on which settings gave the read you liked.

For cleaning up a recorded voice (de-noise, EQ, compression), the AV craft chapter making a voice sound right and the SoundStudio theory sections on EQ and dynamics go into detail.

Music: license it or make it

The single most common rights mistake in local ads is a popular song under the voice. The station pays ASCAP, BMI and SESAC, so it feels covered. It is not. Those licenses cover the station playing music. Building a song into a recorded commercial needs a sync license from whoever controls the composition, and if you use a specific recording, a master use license from whoever owns that recording. Broadcast attorneys have been warning stations about this for years. There is no free 10 second rule.

Your practical options:

• Production music libraries sell tracks licensed for ads. Read the license: some exclude broadcast, some exclude political use, some charge more for TV.
• Original music from a local musician, with a short written agreement that you may use it in ads, in these media, for this long.
• Music you make yourself, including with an AI music tool such as MusicStudio. Check the tool's terms on commercial use and ownership before using its output in a paid ad, and keep a record of when and how you made it.

Sound effects follow the same logic. Record your own, or use a library whose license allows advertising.

Pick music for the tone in the brief, even when your own taste runs elsewhere. A gentle acoustic bed suits the library levy. A driving drum loop under a library ad sounds like a monster truck rally is coming to story time.

Building the mix

The voice is the message. Everything else supports it.

1. Place the voice first. Clean, consistent level, breaths trimmed where they distract.
2. Bring in the bed. Start it full for a second or two if there is an intro, then duck it under the voice. A starting point is the music sitting 15 to 20 dB under the voice while words are spoken, coming up a little in the gaps. Adjust by ear.
3. Add effects at a level that sounds natural for the scene, with the voice clear over their tails.
4. End clean. Fade or resolve the music so nothing is cut off at the spot's end. A note chopped mid-sustain sounds like a mistake, because it is one.
5. Check on bad speakers. Play the mix on a phone speaker and in a car. If any word gets lost under the music, pull the music down. Studio headphones lie about this every time.
6. Measure loudness. TV spots in the United States must match the station's program loudness, which is built on a -24 LKFS target. Radio stations have their own targets. Lesson 7 covers delivery loudness. The SoundStudio loudness section explains how LUFS and LKFS work.

0:00  Library date-stamp SFX, music bed enters at full level
0:01  Voice starts, music ducks down about 18 dB over half a second
0:13  Two-beat pause in the read, music rises about 6 dB
0:14  Voice resumes, music ducks again
0:26  Sponsor line, music stays down
0:28  Music resolves on its final chord
0:29.6  Silence