Sume TTS pace test: 1,000 characters a minute decides the cap

On Sume TTS the character cap binds first if your voice speaks faster than 1,000 characters a minute, and the duration cap binds first if it speaks slower. A request may carry at most 20,000 characters (priced at $0.0475 per 1,000, so $0.95 at the cap), and synthesized audio longer than 1,200 seconds fails with tts_duration_exceeded and no credit capture.

The break-even is plain division: 1,200 seconds / 20,000 characters = 0.06 seconds per character, or 16.67 characters per second, or 1,000 characters a minute. Both limits come from the Sume API catalog and the TTS request schema in the repository (read 2026-10-09).

Which limit hits for which script

A 20,000-character transcript read at 1,000 characters a minute runs exactly 20 minutes, which is 1,200 seconds. At 1,100 characters a minute it runs about 18.2 minutes and the request is fine. At 800 characters a minute it would run 25 minutes, so the duration cap fails it. These speeds are examples, not measurements of any voice; measure your own voice with a short test line.

Spaces and punctuation count toward the character total. A script that is 19,800 characters in your editor may be a different number once you add stage directions or pronunciation hints, so count what you actually send.

Planning around the cap

Split at natural breaks, usually chapters or scene changes, and keep each request under both limits with margin. A safe working rule for a slow voice is to cut at 15,000 characters, which at 800 characters a minute is 18.75 minutes. Each job is billed on its characters, so splitting does not change the total: 100,000 characters is $4.75 whether it is five jobs or more.

Billing for a failed duration check matters here. The schema says a request over 1,200 seconds of audio fails with tts_duration_exceeded and no credit capture, so you are not charged for that attempt, but you still lose the time. Test with a short slice if your voice is slow.

Putting the parts back together

After you synthesize the parts, timeline audio concatenates up to 20 Sume-hosted audio parts into one gapless file for $0.01 a job. Request wav for parts you plan to join, since mp3 adds padding at every edge. The 20-part limit means 20 requests of 20,000 characters each, which is 400,000 characters, joins in one call.

See the related note on counting characters before you pay for how spaces and punctuation are counted.

Measure your own pace

Send a 500-character test line, divide the characters by the audio seconds you get back, and multiply by 60. That is your characters-per-minute figure, and the table above tells you which limit you will hit. Different voices and languages read at different paces, and a voice speaking Korean or Japanese will not match an English pace, so measure per voice and per language.

A test line costs 500 / 1,000 x $0.0475 = $0.02375, which the catalog rounds up to a whole cent when it reserves the job.