42 museum audio-guide stops of 380 characters: MAI vs Sume TTS bill

A 42-stop museum audio guide with 380 characters of script per stop is 15,960 characters. At Microsoft's list prices that is about $0.35 on MAI-Voice-2.1 and $0.24 on MAI-Voice-2.1-Flash. On Sume TTS the same text is about $0.76. MAI is cheaper per character. Sume does not list MAI voices, so this is a price comparison, not a switch you can make inside Sume.

The arithmetic

Take 42 stops x 380 characters = 15,960 characters. Microsoft lists MAI-Voice-2.1 at $22 per 1M characters and MAI-Voice-2.1-Flash at $15 per 1M characters. Sume TTS 1.0 lists $0.0475 per 1,000 characters, which is the provider list price multiplied by 1.25. Spaces and punctuation count on Sume.

One job per stop on Sume

An audio guide wants one file per stop, because visitors jump between stops. On Sume that means 42 TTS jobs. Each stop is 380 characters, far under the 20,000-character cap for one request, and costs about 380 x 0.0475 / 1,000 = $0.01805 at the per-character rate.

The catalog says Sume reserves the estimated amount at submit, rounds the reservation up to whole cents, and captures on completion. A 380-character stop therefore reserves about 2 cents. Across 42 jobs the reservation can reach roughly $0.84 before capture settles. Read the captured amount on a finished job instead of assuming the reservation is the bill.

What the price gap does and does not tell you

Four points decide whether the 41-cent difference matters to you.

• Microsoft's page lists 23 languages for both MAI voices, and positions the standard model for audiobooks, content creation and voice-over, and Flash for call-center, assistant and IVR use. A museum guide is the first kind of use.
• The gap on this job is about 41 cents. For one exhibit the engine price is not the cost that matters; recording rooms, translation and QA are.
• If you add a second and third language, the Sume bill scales the same way: 15,960 characters per language, so three languages are about $2.27 on Sume.
• Sume exposes voices through voice.id or an avatar voice, not through MAI voice names. Do not write a Sume request with an MAI voice name; the API rejects names that are not voice ids.

When Sume is still the practical choice

The guide is only one step. Stops usually need a video loop or a looped still, and a caption track for accessibility. On Sume those are the same API key: the Timeline render costs $0.10 per output minute, and a standalone caption job is $0.20 for a clip up to 60 seconds. If you only need audio files and nothing else, price the MAI engine directly. If the audio feeds a render, one vendor and one balance is simpler.

Before you commit either way, run three representative stops through the engine and listen. Price per character does not tell you whether the pronunciation of artist names is right.