The Narration Is Written In Her Voice

A narration is 2 artefacts inside 1 file: the words, and the instruction for how those words arrive. Most teams write the first and carry the second in their heads, which means it arrives differently on every take and differently again in the edit. The 2 minute Brazilian narration we built carried 5 lines of direction above the script: unhurried; breathe between sentences, because the silence is part of it; Portuguese place-names carried with a soft local lilt; the grandmother's line lower and warmer with a faint smile inside it; and the final line ending on the smile rather than on a period. Those 5 lines, written before the first take, took the booth to 2 passes instead of 11. A written tone is a decision. An unwritten tone is a mood that the room has to guess. 16 years of field work from rural Nigeria to post-conflict Serbia to Kazan to Fukushima taught me the same rule about every brief I have ever signed: what is not written is not shared, and what is not shared gets rebuilt by whoever is standing closest to the work.

A default voice is not a neutral choice. It is a choice nobody made. The register is a decision, so it gets 1 named voice on 1 named model, printed in the spec next to the script: a warm, unhurried female read on a named fast model, cast for the tone rather than for the price. A read gets cast, it does not get configured. When 4 female voices sit on the shelf and the brief says female, the brief has not been written yet. 1 name in the file means 1 argument fewer at the 3rd pass, and it means a re-record 12 months later can find the same instrument rather than approximating it. I grew up inside that discipline: an 85 per cent scholarship to Mannes College of Music and a Carnegie Hall performance taught me that a room full of capable players still needs 1 named part. The probability that an unnamed voice lands the same way twice is low, and it drops further the moment the take is handed to a different studio.

Copy written for the eye carries clauses a mouth cannot finish. Read every line aloud with a timer before it goes anywhere, because 1 number decides more than the paragraph around it: the read has to fit the cut. The narration we wrote runs about 2:05 at an unhurried pace and sits over an 8 shot 2 minute cut, faded so the last line lands on the held smile rather than after it. That is 2 numbers written against each other, and it means the narration and the edit were written against each other instead of reconciled in post. Entrade built 200 wood-gasification machines across the UK and Europe on the same discipline: the drawing and the delivery date lived in 1 document, so the field crew never had to re-cut a design to change a finish. A take that runs 11 seconds long is not a rounding error at the mix. It is a rewritten narration or a re-cut of 48 frames of picture, and both of those cost a day nobody budgeted for. 1 room, 48 kHz, 1 file per take.

The strongest line in that script was not written in the office. 4 sentences of source notes sat next to 1 sentence the subject remembered her grandmother saying, and the remembered line did the work of a whole paragraph: when you smile at someone, you are giving them back to themselves. A person's own sentence carries rhythm a writer cannot fake, and it survives translation, compression, and 40 takes. Write her voice, not a description of her voice. The distinguishing detail is almost never the adjective; it is the specific object in the specific room, mango on the table, salt in the air, hands that keep moving. I lost my savings in the 2013 Cyprus bail-in, and it taught me a version of the same lesson about what actually holds its value: the specific line survives, and the paraphrase evaporates. Spell out the register at the top of the file so the actor, the editor and the client are all reading the same instruction rather than 3 versions of it.

A script with no direction is a script every reader performs differently.

A direction that cannot be counted cannot be reviewed. So the passes get numbered and the notes get timed: 5 lines of direction, 2 read passes, 1 review gate, and 0 notes arriving after the mix. 210 energy systems were deployed across 11 countries in Africa and Asia on that shape, the design fixed once and the finish tuned per site, and the crews never re-argued the design to change a finish. The same split holds for a narration: the words are the design and the read is the finish, and they are tuned in that order, never at the same time. The probability that a script drifts into a tone nobody chose is not small, and it never announces itself. It arrives as 1 take that sounded fine at the time. When the notes are numbered and the passes are logged, the 3rd pass is a cheap decision rather than a fresh argument, and the file ends up auditable a month later.

The last check is 1 minute of listening to the whole thing at speed, out loud, with the picture in front of you. Does the breath land where the direction said it would. Does the local lilt survive the 40th take. Does the last line finish on the smile. In 2008 I signed a EUR 20 million personal guarantee on a company that folded on Christmas Day, and the lesson I kept was about reading the whole document rather than the summary of it. A narration earns the same treatment. If the read is 2:05 and the cut is 2:00 and the fade covers the gap, ship it. If the register slipped at the 3rd beat, fix the direction and re-read 1 section rather than re-cutting the picture to hide it. Write 5 beats in her voice, name the voice, count the passes, and finish on the smile. The file stays good because the direction stayed written.