Audio Ad Scripts and Voiceover in Four Languages

Nielsen's decomposition of what actually drives sales puts creative at close to half the contribution, ahead of reach, targeting and recency combined. In audio the share is higher still, because there is no logo, no product shot and no second look. So the script is written first and the media plan is built to fit it, rather than the other way round. Writing: A comfortable English read runs about two and a half words per second, which makes a thirty second spot seventy to eighty words and a sixty around a hundred and fifty. That is one idea, one proof and one instruction. Scripts arrive as a set of three to five with genuinely different openings, a question, a number, a sound, a name, because the only thing worth testing in audio is the first four seconds. Five music beds under identical words tests nothing. Vernacular scripts are written, not translated. isiZulu and Sesotho carry more syllables per idea than English, so a translated thirty second script overruns, gets rushed by the talent, and lands as a different and worse advertisement. We write to time in the target language and accept that it will hold fewer ideas than the English version. Clients who insist on parity of content across languages are asking for four spots that are each slightly wrong. Casting and recording: Casting happens on your script, not from showreels. Five candidates read the actual opening line and you hear all five before anyone is booked, because a showreel proves someone can read well for a different brand. Sessions record in Johannesburg with you dialled in, or remote-directed to talent elsewhere. English in South African and neutral international registers, plus Afrikaans, isiZulu and Sesotho as the brief requires. Direction in the room is the part that decides the spot. The difference between a read that sounds like an advertisement and one that sounds like a person is usually four takes and one instruction that has nothing to do with the words. Mix and delivery: Streaming platforms normalise loudness. Spotify plays back at around minus fourteen LUFS and podcast inventory commonly specifies minus sixteen mono, which means a spot mastered hot for terrestrial radio arrives turned down and sounds thin against the track before it. Masters are delivered per destination rather than as one file: WAV at 48 kHz alongside encoded versions at each platform's stated specification, with the tail trimmed to the exact duration the ad server expects, since a thirty point four second file gets rejected by systems that will not tell you why. Rights: Talent buyout terms are quoted before the session rather than discovered after it: term in months, named territories, media limited to digital audio, and the price of extension fixed now. South Africa has no residuals structure comparable to a union market, which means rights that are not written down are rights you do not hold. We write them down, and we will not record talent on terms we would be uncomfortable explaining to them. Synthetic voice: Used for exactly one purpose. Before anyone books a booth, the script set is rendered synthetically and put into a small live test so the winning script is chosen on data rather than on the most senior opinion in the review. Nothing synthetic goes to air with our name on it, and we do not clone an actor's read to avoid paying them for the next flight. Not this: Jingles and sonic logos. A weekly script subscription. Health, financial or wellness claims without written substantiation on file, which we ask for before writing a word. And rewriting a television soundtrack into a radio spot: the picture was carrying half the meaning, and removing it leaves a hole no voice actor can fill.