Azure word boundaries in ms vs Sume TTS timestamps in seconds

With Azure batch synthesis you turn on wordBoundaryEnabled or sentenceBoundaryEnabled and get a [nnnn].word.json or .sentence.json file in the ZIP, where each word has Text, AudioOffset and Duration in milliseconds. On Sume TTS 1.0 you send timestamps.words: true and the completed job result includes words[] with start and end in seconds. Convert units before you share code.

Azure's format is from its batch synthesis page, read 2026-10-01; Sume's from the API reference.

How do the two timing shapes compare?

Different file layout, different unit, same purpose.

What does Sume do for sentences?

Add segmentation: { "mode": "sentence" } with timestamps.words: true. The result has gapless segments[], where each segment ends exactly where the next starts, using a 70 ms post-word boundary by default that you can set from 0 to 500. With a wav or raw container each segment also gets its own audio URL; with mp3 you get timings only.

How do I convert an Azure word file?

Divide by 1,000. An Azure word with AudioOffset 200 and Duration 350 starts at 0.2 seconds and, if you want an end, ends at 0.55. Azure gives a duration, Sume gives an end, so compute end = (offset + duration) / 1000.

What does the request look like?

Ask for wav so the sentence slices come back as audio.

curl -X POST https://api.sume.com/v1/tts-1.0/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "transcript": "The rainbow has seven colors.",
    "avatar_handle": "@narrator",
    "output_format": { "container": "wav", "sample_rate": 44100, "encoding": "pcm_s16le" },
    "timestamps": { "words": true },
    "segmentation": { "mode": "sentence" }
  }'