tts_text_source_conflict 400: transcript or transcript_source

A Sume TTS generate request needs exactly one text input: transcript (literal text, 1 to 20,000 characters) or transcript_source (a reference to an accepted script: script_revision_id plus sentence_ids). Send both, neither, or a malformed source, and you get a 400. Two codes are involved, and they are easy to mix up. These come back in Sume's standard error envelope with an HTTP status and a stable code.

The request schema itself says "Provide exactly one of transcript_source or transcript." The source service then answers with the code that tells you which way you erred.

Which code means what

The shape that works

transcript_source accepts only the keys script_revision_id and sentence_ids. Ids match ^[A-Za-z0-9_-]{1,128}$. Anything else, including a stray thread_id inside the object, is treated as invalid. Source resolution is thread-scoped, so a bare key with no thread behind it cannot use transcript_source.

Use the literal form when you do not need provenance. The two forms cost the same, $0.0475 per 1,000 characters with a one-cent minimum per job; the source form adds a server-owned receipt to the finished job.

curl -s https://api.sume.com/v1/tts-1.0/generate \
  -H "x-api-key: $SUME_API_KEY" -H "idempotency-key: line-001" \
  -H "content-type: application/json" \
  -d "{\"transcript\": \"Hello from Sume.\", \"voice\": {\"mode\": \"id\", \"id\": \"$VOICE_ID\"}}"

Fix by symptom

• Conflict: delete one of the two fields. Do not send an empty string for the one you are not using; an empty transcript is its own error.
• Required: you sent neither; add the one you meant.
• Invalid transcript_source: print the object and check for unknown keys and duplicate ids.