ElevenLabs speech to speech API: what Sume offers instead

Sume has no speech-to-speech endpoint. Its text-to-speech request takes a transcript, not source audio, so the nearest route is three steps: transcribe with STT 1.0, edit the text, then synthesize it with TTS 1.0. The new voice follows the words, not the original performance.

The ElevenLabs models page (read 2026-10-01) lists eleven_multilingual_sts_v2 as a multilingual voice changer (Speech to Speech) and eleven_english_sts_v2 as an English-only one. Sume facts are from its OpenAPI file and Timeline audio docs.

What does a speech-to-speech model do?

On the vendor's page these are voice changer models: audio in, audio out in a different voice. Sume does not list an equivalent, so nothing in the Sume API takes a recorded take and re-voices it.

What does Sume TTS take as input?

The TTS 1.0 schema says "Exactly one of transcript_source or transcript is required." Text goes in; audio comes out. The voice is picked by the top-level avatar_id / avatar_handle, or voice.id.

STT 1.0 is the route that supplies the words: the schema describes POST /v1/stt-1.0/transcribe as the primary Sume STT 1.0 route for speech-to-text.

What is lost compared with voice conversion?

Pacing, emphasis and breath from the original take are not carried into TTS; only the transcript is. If the original delivery matters, keep it and do not use this route.

How do I put the new audio back on a video?

Timeline audio returns one hosted file, and the docs say to use that file as audio.url on the render. For lip-sync limits when the picture is a talking face, read replace video audio with an AI voice. For the full transcribe, translate, speak chain see build an AI dubbing pipeline.