7 Common Mistakes When Choosing a Voice for Your Promo Video

Short answer: Choose a promo voice by listening, not by reading names. Test it with your real script, in your audience’s language, against your real video and music. Time it, check every brand name, and confirm you are allowed to use the audio commercially before you publish.

A good voice can carry a 20-second video. A poor match can make a good product sound cheap or confusing. The trouble is that most mistakes happen before anyone presses play.

Below are seven we see often, each with a simple fix. Where it helps, we point to a demo clip on voiceover.my so you can hear the point for yourself. As of 3 November 2026 the site has 169 demo clips (95 Malay, 74 English) across 12 situations, made with Google Gemini and ElevenLabs.

Mistake 1: Choosing a voice by its name

Voice names sound helpful, but they tell you very little. On our site you will find voices called Leda, Fenrir, Abang Khai and Cikgu Aisyah. None of those names says how the voice will sound reading your promo.

Here is a simple example. “Cikgu” means teacher, yet the Cikgu Aisyah cafe promo carries a Human-reviewed note that it fits the advert. You would never guess that from the name alone.

Fix: Shortlist three or four voices by ear. Play them one after another, on the same speaker, at the same volume. Ignore the names until you have a favourite.

Mistake 2: Testing with a different script from the real one

Many people audition voices with a sample sentence, pick one, and only then paste in the real script. The real script often has longer sentences, numbers, a price, a phone number or a call to action. A voice that sounded lovely on “Hello, welcome” can stumble on “Book your table today.”

On voiceover.my, clips in the same situation and language usually read the same script. The Malay advertising clips, for example, all read the same 55-word weekend cafe promo. Compare Leda and Umbriel to hear how much the same words change with the voice.

Fix: Write your final script first. Audition every voice with that exact text, including the ending and the call to action.

Mistake 3: Ignoring the language your audience actually uses

Malaysia is a mixed-language market. A promo that works in English for one group may feel distant to viewers who would rather hear Bahasa Melayu, and the reverse is also true.

Listen to the same voice, Despina, reading the same made-up product ad in Malay and in English. Same voice, same message, quite a different feel.

Two more things to keep in mind:

• “Supported” is not the same as “sounds local”. Google’s speech generation documentation lists Standard Malay among the languages its speech models support. That does not prove a voice will sound Malaysian to your customers. Ask a few of them to listen.
• Check the accent of English voices. Several English cards on our site carry a Human-reviewed note that the voice sounded American to the reviewer. Decide whether that suits your brand.

Fix: Pick the language first, then the voice. If you sell to both groups, make two separate versions.

Mistake 4: A tone that fights your visuals and music

A calm, slow voice over fast cuts and loud music feels like two different adverts. A punchy, excited read over soft, slow shots feels pushy.

Tone also comes from the script. Two of our Malay social media ad demos promote the same made-up product. The Abang Khai clip reads a casual “don’t scroll yet” script. The Leda clip reads a more formal version. When we made these demos, the casual script was judged too relaxed for the style we wanted, so a more formal version was written as well.

Fix: Play the voice over the actual edit with the actual music, not on its own. If the voice and the pictures pull in different directions, change the voice, the script or the music until they agree.

Mistake 5: Too many words for the length of the video

This one is easy to get wrong. Someone writes 80 words for a 15-second slot, and the voice either races or the video has to be stretched.

Word count alone is also misleading across languages. Here are the running times of real clips on our site.

Times as shown on the demo cards, checked 3 November 2026. Your voice and script will differ.

Notice that the Malay ad has fewer words than the English one but runs longer. Malay words are often longer to say. If you translate a script, time the translation separately.

Fix: Generate a draft and check its running time against your video. Leave a second or two of breathing room at the end for your logo or offer. Cut words, not pauses.

Mistake 6: Not checking how brand and product names are said

Brand names, product names, place names and mixed English-Malay words are common places for AI voices to slip. A wrong stress on your brand name in the very first second is hard to forgive.

Our own demos use made-up names such as “Kafe Taman” and “Smart Key Hook”, which is why each card shows the full transcript. When you test, listen for each name in context.

Fix:

• Put every brand and product name in your test script.
• If a name sounds wrong, try spelling it the way it is spoken, then listen again.
• Ask someone who did not write the script to listen. Writers tend to hear what they expect.

Mistake 7: Forgetting about usage rights and the licence

A great voice is no use if you are not allowed to run it in a paid advert. Rules differ between providers and plans, and they change.

Two examples from the providers behind our demos:

• Google’s Gemini API terms say Google won’t claim ownership of content you generate, and that you are responsible for how you use it and how others you share it with use it.
• ElevenLabs’ pricing page lists a “Commercial License” among the features of its paid plans, starting from the Starter plan.

Also think about the other parts of the video. Music, stock footage and fonts have their own licences. And never copy a real person’s voice without their clear permission.

Fix: Before you publish, read the current terms for the plan you actually use, save a copy with the project, and note the date. The clips on voiceover.my are samples for comparing voices; generate your own audio under your own account.

A note on the labels you will see

If you use our demo cards to compare voices, you will see two labels. “As requested” is what was asked for, or what the provider reports, such as a style instruction like “upbeat, clear, friendly”. It is not proof of how the clip sounds. “Human-reviewed” means a person listened to the clip and recorded what they heard, for example “Fitting for this scene”. Trust your own ears over both.

The seven mistakes at a glance

Checklist before you approve a promo voice

• I listened to at least three voices reading my final script.
• The language matches how my viewers talk about the product.
• I played the voice over the real video and music.
• The read fits the video length with room to spare.
• Every brand and product name sounds right.
• Someone who did not write the script has listened to it.
• I have read and saved the current usage terms for my plan.

Start by playing a few clips side by side. Ten minutes of listening now can save you from re-editing a finished video later.