Hume AI Alternative: Octave TTS and EVI End Nov 13
Hume AI alternative for Octave users: Hume shuts its TTS and EVI APIs on Nov 13, 2026. What to export first, what replaces what, and where we fall short.
Last updated: 2026-10-26
If you are looking for a Hume AI alternative, there is a deadline: Hume is sunsetting its text-to-speech (Octave) and EVI APIs, and according to Hume's documentation access ends on November 13, 2026 at 12:01 a.m. EST, after which any data left in your account is permanently deleted. VoiceSmiths can replace the content side of Octave: expressive voiceovers, voices designed from a description, character dialogue and audiobooks, all in a web studio. It cannot replace the API side: there is no public API, no streaming endpoint and no self-serve real-time voice agent. Below is what maps to what, what to export first, and a migration checklist.
Hear Eleven v4
Tap a voice to hear it. To voice your own text, paste it into the workbench: 7 credits per 100 characters, starting with the $3.90 trial pack.
What Hume is shutting down
| Hume product | What it did | Status |
|---|---|---|
| Octave (TTS API) | Expressive text to speech with acting instructions, voice design from a description, voice cloning from as little as 15 seconds, streaming over HTTP and WebSocket at about 100–200 ms | Access ends November 13, 2026, 12:01 a.m. EST |
| EVI (speech-to-speech API) | A real-time voice AI that analyzes the vocal qualities of the user's speech and answers in a way that fits their expressed emotion | Access ends November 13, 2026, 12:01 a.m. EST |
Both notices say the same thing (TTS overview, EVI overview): the APIs stay fully supported until the cutoff, and account data is deleted after it. Neither describes a migration tool or an export feature; Hume points customers to Real World VoiceEQ for comparing other voice models, and to its support team for questions.
Hume is not the only voice platform to close its doors recently. Meta acquired PlayAI, the company behind PlayHT, in July 2025 (Verdict, citing Bloomberg), and PlayHT has since been shut down; we found no official closing notice, only third-party reports dating it to the end of 2025. Resemble AI's CEO wrote on August 13, 2026: "We're not selling voice AI to new customers", while it keeps supporting existing ones (Resemble AI).
What replaces what on VoiceSmiths
| What you used on Hume | On VoiceSmiths | Fit |
|---|---|---|
| Octave expressive TTS | Text to speech on Eleven v4: 54 curated voices plus 5,000+ community voices, MP3 with character-timed SRT | Yes |
Acting instructions (description) | Inline audio tags such as [whispers], [nervous], [sarcastic], written before the words they color | Mostly (see the table below) |
| Voice design from a prompt | Voice design in the workbench, 20 credits per voice; hear designed voices on prompt to voice | Yes |
| Voice cloning | Instant voice cloning from 1–3 minutes of clean audio, 30 credits; your own voice or one you have written permission for | Yes, but needs more audio than Octave's 15 seconds |
| Languages | 80+ on Eleven v4, including Mandarin with 15 native Mandarin voices; Octave 2 listed 11 languages, without Chinese | Yes |
| Several characters in one scene | Multi-speaker dialogue, or the audio drama studio, which casts and directs a whole chapter | Yes |
| Long scripts | 5,000 characters per generation, 50,000 in one long-form job | Yes |
| Streaming TTS API, WebSocket, low latency | None: no public API, no streaming | No |
| EVI real-time voice agent | A demo agent in the workbench (15 credits per call, up to 60 seconds); custom agents built on request via ticket; no agent API | No self-serve option |
Hear the replacement
Delivery directed with tags instead of an acting-instruction field. Sweet Girl, one of our designed voices, on Eleven v4:
[sighs] I know today was a lot. Let's slow down for a minute. [curious] What's the one thing still on your mind? [laughs] No, not the inbox. The other one.Two designed voices in one scene, the way you would script a podcast intro or an in-app character exchange:
[curious] You moved forty voices off Hume in one afternoon?
[laughs] Thirty-eight. Two of them were my cat.
[sarcastic] Naturally. And the deadline?
November thirteenth. [whispers] Which is why we're recording today.At length: British Butler reading narration, and a 7-voice audio drama rendered on v4.
Turning Octave acting instructions into audio tags
Octave took a natural-language description per utterance, such as "calm, pedagogical". Eleven v4 takes short tags inside the text instead. Most instructions translate directly; a few become a voice choice or a setting.
| Octave acting instruction | On Eleven v4 |
|---|---|
| "whispering", "secretive" | [whispers] before the line |
| "nervous", "hesitant" | [nervous] |
| "sarcastic", "dry" | [sarcastic] |
| "excited", "energetic" | [excited] |
| "sad", "tearful" | [sad] or [crying] |
| "calm, pedagogical" | No tag: pick a calm voice (Composed Lady, British Butler) and keep the line plain |
| Speaking rate | The speed slider in the voice settings, 0.7× to 1.2× |
| Trailing silence | No setting: use line breaks and ... for pauses, add silence at the end in your editor |
Tags are performed, never read aloud, and never reach the subtitles. Short free-form directions in brackets are sometimes acted too: in our test of Fish Audio Drama 3 scripts, v4 never spoke a direction and acted some of them. The full tag list is in the emotion tags cheat sheet.
What VoiceSmiths does not do (yet)
- No public API. You can create API keys in settings, but no endpoint accepts them today. For bulk work or an integration, email voicesmiths@126.com.
- No streaming or low-latency endpoint for your app. You generate in the browser and download a finished MP3.
- No self-serve real-time agent. The voice agent page describes the demo agent and custom builds on request; there is no EVI-style API to build your own.
- No reading of the listener's emotion. EVI analyzed the user's voice; nothing here does.
- Cloning needs 1–3 minutes of audio, not 15 seconds.
If any of these is the reason you used Hume, choose a provider that sells its models by API. ElevenLabs, whose models VoiceSmiths runs on, sells TTS and agents by API directly; the text-to-speech API guide covers what to compare.
Migration checklist
- Export before November 13, 2026, 12:01 a.m. EST. Download every generated file you want to keep. Hume says account data is deleted after the cutoff.
- Save your voice recipes. Copy the text description of each designed voice, and keep the source audio for each clone (only audio you own or have permission to use).
- Save the rest of your setup. Acting instructions you reuse, EVI system prompts, tool definitions and any transcripts you need.
- Decide which side you need. Content you publish (voiceovers, audiobooks, characters) can move to a studio. Speech generated live inside your product needs an API provider.
- Rebuild the voices. Paste each description into voice design (20 credits), or clone from your audio (30 credits). A new design will be close to your old voice, not identical, so run the same script through two or three candidates.
- Translate the acting instructions into tags with the table above, and keep each tag right before the words it colors.
- Re-render a reference script of three or four lines and compare it with your old Hume files before converting a whole catalog.
- Update your credits and disclosures wherever you name the voice provider.
What it costs
| Characters | Credits | At list price (new-user first month) | |
|---|---|---|---|
| One minute of English | ~850 | 60 | $1.25–2.00 ($0.50–0.79) |
| One designed voice | 20 | $0.42–0.66 ($0.17–0.26) | |
| One cloned voice | 30 | $0.62–1.00 ($0.25–0.40) |
Rate: 7 credits per 100 characters (4 on the faster Flash model); price ranges are the Creator and Starter monthly plans. The one-time trial pack is 150 credits, enough to design a voice and hear it read about 1,800 characters, for $3.90 in the first 24 hours after sign-up (list $9.90), and its credits never expire. A first order is fully refunded within 7 days if you have used 150 credits or fewer. The audio you generate is yours to use commercially. See pricing, or compare more options in ElevenLabs alternatives.
FAQ
Is Hume AI shutting down? Hume is sunsetting its text-to-speech (Octave) and EVI APIs. According to Hume's documentation, access ends on November 13, 2026 at 12:01 a.m. EST, both APIs stay fully supported until then, and any data in your account is permanently deleted after November 13, 2026.
What is the best Hume Octave alternative? It depends on how you used Octave. For content (voiceovers, audiobooks, characters, ads), a studio like VoiceSmiths covers it: Eleven v4 with inline audio tags, voice design from a description, voice cloning and full-cast audio drama. For speech generated inside your own app in real time, you need a provider with a streaming API, and VoiceSmiths has no public API.
Can I move my Hume voices to another platform? Not the voice files themselves. Save each designed voice's text description and design a new voice from it (20 credits per voice on VoiceSmiths; expect a close match, not an identical voice), or clone it again from 1–3 minutes of audio you have the rights to (30 credits). Download everything before November 13, 2026, because Hume deletes account data after that.
Does VoiceSmiths have an EVI-style voice agent? Not a self-serve one. The workbench has a demo voice agent you can talk to (15 credits per call, up to 60 seconds), and custom agents for a website or phone line are built on request through a support ticket. There is no API for building your own real-time agent on VoiceSmiths.
Does VoiceSmiths have an API? Not yet. VoiceSmiths is a web workbench: you can create API keys in settings, but no endpoint accepts them today. For bulk work or an integration, email voicesmiths@126.com with your volume and use case.
How much does VoiceSmiths cost? A minute of English is about 850 characters, which is about 60 credits: roughly $1.25–2.00 on monthly plans at list price, or $0.50–0.79 in a new account's first month. The one-time trial pack (150 credits, never expires) is $3.90 in the first 24 hours after sign-up (list $9.90), and a first order is fully refunded within 7 days if you have used 150 credits or fewer.