Slow, clear learner audio: Sume TTS speed 0.6 and sentence slices

Learners want audio that is slower than native speech but not robotic. Sume TTS exposes generation_config.speed from 0.6 to 1.5, so you can render slower audio without editing a file.

Pick the speed

Slice per sentence

Ask for segmentation: {mode: "sentence"} with timestamps.words true and a wav container. You get one clip per sentence and the timings, so the app can play a single sentence on tap. mp3 returns timings only, no per-slice audio URLs. Lines are gapless, so add a short pause in the player if you want repeat-after-me gaps.

Set the language

Always send language for a non-English lesson, and a voice tagged for that language. Sume answers a known mismatch with a 409 (tts_voice_language_mismatch) before any job or charge, and you retry with confirm_language_mismatch: true only if you meant it. Speed 0.6 on a wrong-language voice sounds worse, not clearer.

Cost

A 600-character lesson is $0.0285 raw, so 3 cents billed. Render it at 0.8 and 1.0 and the pair is 6 cents. A 30-lesson course with both versions is under $2.