Two things were asked for on the voiceover: excitement in the delivery, and the Spanish read saying “Arden Fair” in English. Every option here is the approved Remembrance read over the identical picture — only the delivery changes. Start with the brand strip: it plays just the two words, read after read, so the ear can decide in seconds. Voice is The Insider · Arden Fair throughout.
V5X4GESESSXESTTwo routes. Tags are directions the newer eleven_v3 model reads with, not words it reads out — shown as chips on the line they direct. That model paces this voice more deliberately, so the tagged read carries trimmed copy to land on the same cards at natural pace: tags cost words, and every trim is shown. Lively keeps every approved word and window and opens up the delivery settings of the shipping voice instead.
D4multilingual_v2Names what Recuerdos means without naming the holiday. Warmest of the five, and the one that best fits the ofrenda imagery.
X4Wmultilingual_v2The approved lines and windows, delivery lifted by STYLE rather than by dropping stability: 0.44/0.64 against X4S's 0.45/0.55. Excitement without the warble that 0.32 introduced.
X4Vmultilingual_v2The approved lines and windows on the shipping model, delivery pushed past X4S: stability 0.32, style 0.72. Excitement in the tone, no tags, no copy given up.
X4Tv3The approved read on eleven_v3 with [excited] on the brand line and the call to action. v3 reads more deliberately than the shipping voice, so the two middle lines are trimmed to fit their cards at natural pace — tags cost copy.
X4Smultilingual_v2Excitement without tags: the exact approved lines and windows, the same voice and model as the shipping read, with the delivery opened up (stability 0.45, style 0.55). Fresh takes of all four lines, so it is one texture throughout — and it fits everything.
X4Gv3Shipping as EL15H-BThe client asked for the approved script on eleven_v3 in the Insider voice, giving nothing up. The repo said that was impossible - X4E overran 2 of 4 lines - but that verdict came from a pace figure measured on a TAGGED line. Re-measured per line off the wav files, v3 against the shipping v2 read: the brand line +26.8% and the call to action +8.4% (both [excited]), against +1.7% and +11.6% on the two UNTAGGED lines. The tag buys a performative beat worth ~0.4-0.5s; the model costs almost nothing. Drop the tags and every approved word fits.
ESmultilingual_v2The approved D4 read in Spanish. Spanish runs 20-25% longer, so the list in line 3 is trimmed rather than the read hurried.
ESVmultilingual_v2The Spanish twin of X4V: the approved Spanish lines with the same opened-up delivery, and "Arden Fair" cut in from the approved English take exactly as ESS does it - so the brand is English and the script is untouched.
ESSmultilingual_v2Shipping as EL15H-B-ESToday’s Spanish read with one change: “Arden Fair” is the English brand name lifted from the approved English take, cut in front of a fresh remainder read in the same voice and model. The other three lines are the shipping takes, untouched.
XESTv3The Spanish twin of the tagged read: eleven_v3 with [excited] on the brand line and the call to action, trimmed copy. The brand line is read whole — v3 says Arden Fair in English inside the Spanish sentence, so nothing is cut in.
D1multilingual_v2Sets a scale before the detail arrives. The plainest of the five and the closest in voice to the line that follows it.
D2multilingual_v2Leads with what the open actually shows — the drum being struck on beat one. The line the picture is already making.
D3multilingual_v2The only read that asks for a specific audience. The footage is full of children performing, so it is earned, not assumed.
D4multilingual_v2Names what Recuerdos means without naming the holiday. Warmest of the five, and the one that best fits the ofrenda imagery.
D5multilingual_v2Turns 1–4 PM into a reason rather than a detail — the only fill that adds a fact instead of a feeling.
These carry the original three-line structure, with the 0:03–0:06 gap still in them.
V1multilingual_v2Written to the voice's own persona - knowing, unhurried, a local telling you something. Opens on sound, not on the brand.
V2multilingual_v2Turns a spectator into a participant in the first six words. The warmest read, and the one that best fits JOIN US.
V3multilingual_v2Three nouns and a stop. Leaves the most room for the music and the drums to carry the open - the least VO of the five.
V4multilingual_v2The only read that names what the holiday is actually for. More feeling, less retail - best if the client wants heart over footfall.
V5multilingual_v2Brand, offer, when - the clearest of the five, and the one the client picked. FREE carries the middle line, and the close is a spoken call to action that says the same words as the card.
Where “Arden Fair” is in English on a Spanish read, it is the same two words from the approved English take, cut at the word boundaries with 40 ms handles, level-matched to within 1 LU of the Spanish that follows, and joined with a 75 ms breath. No phonetic respelling, no pronunciation dictionary — the brand name is never spelled differently from how the client writes it.
Each tagged take was transcribed back three times; no tag word appears in any transcript. The tags are English keywords even inside the Spanish lines, which is how the model expects them.
The tagged model reads this voice about a third slower than the shipping one, and it has no speed control. Rather than hurry the read, the two middle lines were shortened to fit their cards — the original wording is shown struck through under each trimmed line. The brand line and the date line are verbatim.
Same model, same words, same windows as the shipping read; only the delivery settings change (stability 0.45, style 0.55 against 0.62 / 0.15). All four lines are fresh takes so the texture is one voice throughout.
Every line is placed at its card and must clear the next graphic change. Nothing here is sped up or stretched: a read that did not fit was re-rolled at natural pace, and where it still ran long the card says runs long rather than hiding it.
Unchanged from the approved mix: the score is trimmed 9 dB for the whole spot and dips 4 dB further under the voice, so it reads as one steady bed. Every option here is mixed the same way and normalised to −14 LUFS.
Each generated line was transcribed back. Ofrendas, recuerdos, gratuita, súmate and octubre return intact. A recogniser is not an ear on a proper noun — the brand strip above is where that call is made.
The voice says the event is free; no graphic does. The cards stay as approved, and the word reaches viewers with sound on.