160 ads a month at 450 characters: the TTS bill on Sume, MAI and Flash

One hundred sixty ads of 450 characters each cost $4.80 a month on Sume, $1.584 on MAI-Voice-2.1 and $1.08 on MAI-Voice-2.1-Flash. All three are small next to media spend, so pick on workflow and language fit.

The sums

On Sume 450 characters is 450 x $0.0475 / 1,000 = $0.021375, which rounds up to 3 cents per job; 160 x $0.03 = $4.80. On Microsoft's list rates: 450 x $22 / 1,000,000 = $0.0099 and 450 x $15 / 1,000,000 = $0.00675; times 160 that is $1.584 and $1.08.

What the difference is

The difference between Sume and Flash is $3.72 a month. The difference buys a hosted job API: POST /v1/tts-router/generate, a status and result endpoint, idempotency, a voice library with language tags, word timestamps with timestamps.words, and one account for captions and rendering. Microsoft's page positions Flash for call-center agents and voice assistants, and the standard model for audiobooks, content creation and voice-over, which is the one this scenario would use.

Retries do not double the bill

Rework is the real cost driver: if a creative director asks for three takes per ad, the Sume bill triples to $14.40. Use generation_config.speed and emotion to steer takes, and send an Idempotency-Key per ad-and-take so repeated submits cannot double-charge.

ads = 160
chars = 450
sume = ads * 0.03
mai = ads * chars * 22 / 1e6
flash = ads * chars * 15 / 1e6
print(round(sume, 3), round(mai, 3), round(flash, 3))
print("three takes on sume", round(sume * 3, 2))

Limits

I did not test any of the three for audio quality or for Microsoft's real invoice, only the published rates. Audition one ad per option before you decide.