6-second ad legal line: Sume TTS speed 1.5 and checking word times

The fastest a Sume TTS voice will speak is generation_config.speed 1.5, and the API rejects higher values. If a legal or disclaimer line does not fit its 6-second slot at 1.5, the fix is fewer words, not a faster setting. Ask for timestamps.words and read the end time of the last word to find out before you build the cut.

The speed range and the timings option are documented in the Sume API reference (read 2026-10-06).

How do I measure the line?

Send the line with generation_config: { "speed": 1.5 } and timestamps: { "words": true }. When the job is complete, GET /v1/jobs/{id}/result returns words[] with a start and an end in seconds for each word. The end of the last word is the spoken length. Compare it with the slot, leaving a little room for the cut.

• Speed 1.5 is the ceiling, the floor is 0.6.
• Words are monotonic, so the last word's end is the length.
• Spaces and punctuation count toward usage, so a tighter line is also cheaper.

What if it still runs long?

Shorten the wording. A fast read is also harder to follow, so keep the line to the words the rule needs and put anything else in the on-screen text. If you burn the line as captions, the Video captions docs describe script_text, which keeps your wording while the speech timings stay the source of time.

Is a faster read always compliant?

That is a question for your legal reviewer. Sume's docs say what the speed field does, not what a regulator accepts as readable. Test the line at 1.2, 1.35 and 1.5 and have a person listen to each one.