Hume Octave and EVI shut down Nov 13, 2026: what to export

Hume's text-to-speech (Octave) and Empathic Voice Interface (EVI) APIs stop working on November 13, 2026 at 12:01 a.m. EST, and both stay fully supported until then. Everything you leave in the account after that moment is permanently deleted, so the export has to happen before the cutoff, not after.

This page is the checklist: what you can download, what cannot leave Hume at all, and how the replacement call looks if you move batch narration to Sume TTS. Hume itself continues, according to a competitor's write-up, with voice evaluation and expression measurement; the two voice-generation APIs are what end.

What Hume says ends, and what it deletes

Hume's own TTS overview states the sunset and the exact time. A Cartesia guide for Hume users adds the practical detail about data and voices. Treat the Cartesia page as a secondary read, since Cartesia sells a competing product.

Export list before the cutoff

Work from the account outward, and keep everything in your own storage.

• Download every generated audio file you still need. Re-generating later is impossible once the API is gone.
• Download EVI chat transcripts and logs if you keep them for support or compliance.
• Write down each voice: its name, the prompt that made it, and the clip you used to clone it, if any.
• Keep the original recordings behind any clone. They are the only way to recreate the voice elsewhere.
• Save the exact scripts and descriptions that produced your approved takes, so you can audition replacements on the same text.

Voices that cannot move

A voice designed from a text prompt has no source audio. The practical path is to choose the closest match in the new provider's library and test it on your real scripts, not on a demo sentence. A cloned voice is different: if you hold the original recording and the speaker's consent, you can recreate it with a provider that offers cloning.

Sume's TTS route takes a voice id from its catalog and has no voice-cloning or voice-upload widget on the tool surface, so a Hume clone does not transfer there. Pick a catalog voice and audition it. If a cloned voice is a hard requirement, use a provider whose terms cover your recording and keep the consent record.

Replacing batch narration with Sume TTS

If your Octave use was offline narration rather than live conversation, Sume's primary route is POST /v1/tts-1.0/generate with the public model id sume/tts-1.0. It is an async job: you submit, then poll or receive a webhook, and the result exposes mirrored audio artifacts. The tool description accepts a transcript of 1 to 20,000 characters per request, so long chapters go in as several jobs.

EVI is a live speech-to-speech product. Sume TTS is not a drop-in for it, because there is no streaming, turn-taking or interruption handling in the job model. Plan a different vendor for the live agent and keep Sume for rendered audio that goes onto video.

Budget a short audition: take five approved Hume lines, render them against two or three Sume voices, and compare before you commit your catalog. Details of the routes are in the Sume API reference. For comparison math on Hume plans, see the plan ladder post, and for language coverage, the Octave 2 language list.

Retest checklist after you switch

Voice IDs and EVI configurations do not carry over, so nothing works by renaming a field.

• Map each Hume voice to a new voice id and record the pairing.
• Rerun your approved scripts and listen for pronunciation of brand names and numbers.
• Check job failure paths: your code should handle an async job that finishes later, not an immediate audio response.
• For live agents, retest turn-taking and interruptions on the new vendor before cutover.
• Keep consent records for any cloned voice you recreate elsewhere; see what to keep before uploading reference audio.