Cost of audio for 20 ad variants: voiceover, music and a check

Audio for 20 ad variants costs about $3.13 on Sume: $0.43 for 20 voiceovers of 450 characters, $2.50 for 20 music tracks at $0.125 each and $0.20 for a one-minute transcript check on each. Audio is a small share of an ad budget, and music is the largest line within it.

This post prices an audio pass per variant using the catalog rates read on 2026-10-02, and shows where the savings are. The assumptions are stated in each line, because a different script length or a shared music track changes the total.

None of this includes the video itself. Audio is the cheap half of an ad, and the visuals are priced separately by the second or by the clip.

What does one variant cost?

Each variant gets a voiceover, a music bed and a quick check that the spoken words match the script. The check is optional, but at a cent it is cheap insurance.

The voiceover line rounds to a fraction of a cent per variant, so the total is dominated by whichever step has a flat per-item price. That is a useful pattern to look for in any batch budget.

Should every variant get its own music?

Probably not. Music is 80 percent of the audio cost in this example. If your test is about the hook or the script, generate three or four tracks, reuse them across variants, and keep the other 16 for something more useful. Four tracks cost $0.50 instead of $2.50, bringing the total to about $1.13.

Reusing a track also removes a source of noise from the test: if every variant has a different bed, you cannot tell whether a result came from the words or the sound.

Is the transcript check worth a cent?

For voiceovers, yes, when the audio goes into a pipeline that nobody listens to. A transcript lets you compare the spoken words with your script and catch a mispronounced brand name or a skipped sentence before you pay for the video around it.

The reserve for a transcript is one cent for a clip under a minute, and the catalog allows up to ten cents for longer audio. Probing the file first is free.

If the check shows a mismatch, regenerate only that voiceover. At roughly two cents it is the cheapest retake in the whole pipeline, which is a good reason to check before you render the video around it.

VOICE_PER_1K = 0.0475
MUSIC = 0.125
CHECK = 0.01

variants, chars, tracks = 20, 450, 4
total = (
    variants * chars / 1000 * VOICE_PER_1K
    + tracks * MUSIC
    + variants * CHECK
)
print(f"${total:.2f}")

What would change this total?

Treat the total as a floor, then add your expected retake rate on top.

• Longer scripts. Voiceover is linear in characters, so 900 characters doubles that line.
• Retakes. Each regenerated voiceover or track is a new job at the full rate.
• Music length. The catalog lists one price per audio, so check the route's duration options before assuming a single track covers a long ad.

How do I read what I actually spent?

Use the usage ledger with the run, thread or job id and quote summary.debited_usd. The ledger separates reserved, captured and refunded money, and a refunded row keeps its hold amount, so summing rows by hand will overstate the cost.