Hume AI Octave Is Shutting Down: What to Use Next
Hume AI Octave is shutting down: the TTS and EVI APIs end November 13, 2026. See what Octave did, what changed at Hume and what to use next, as of October 2026.
Hume AI Octave is Hume's text to speech model, and Hume is sunsetting its TTS and EVI APIs. Access ends November 13, 2026 at 12:01 a.m. EST, and account data is permanently deleted after that. Do not sign up for Octave now. Download your audio, keep your original recordings, and move to another tool before the deadline.
Is Hume AI Octave shutting down?
Yes. Hume's developer documentation says: "Hume is sunsetting the TTS and EVI APIs." It adds that "access ends November 13, 2026 at 12:01 a.m. EST, and both APIs remain fully supported until then." The same notice says that "any data in your account is permanently deleted after November 13, 2026."
The notice does not name Octave. The TTS API is the Octave product, so the shutdown covers it. A roundup of AI news, citing Hume's changelog, dates the announcement to October 2, 2026. Hume gave no reason in that changelog, and its notice contains no export or migration guidance. It points customers to its Real World VoiceEQ evaluation and to assist@hume.ai.
Hume is not closing as a company. It now sells a voice AI evaluation platform called VoiceEQ, along with expression measurement APIs. The TTS and EVI APIs are the parts that end.
What is Hume AI Octave?
Octave is Hume's speech-language model for text to speech. At its launch on February 26, 2025, Hume called it "the first text-to-speech system powered by a large language model," according to VentureBeat's launch coverage. It supported English and Spanish at launch.
Octave's distinctive feature was voice design from a text description, such as "a sarcastic medieval peasant." You could then steer the result with instructions such as "happier, sadder, more frustrated." Hume's own study, with 180 human raters and 120 prompts, found that listeners preferred Octave to ElevenLabs for audio quality in 71.6 percent of trials. That study compared against an older ElevenLabs model, and Hume ran it, so treat it as a vendor result.
Is Octave 1 or Octave 2 the one you used?
Octave 2 launched as a preview on October 1, 2025. Hume's Octave 2 launch post and its TTS documentation describe the differences below.
| Feature | Octave 1 | Octave 2 (preview) | |---|---|---| | Languages | English and Spanish | 11: English, Japanese, Korean, Spanish, French, Portuguese, Italian, German, Russian, Hindi, Arabic | | Latency | About 200 ms | "As low as ~100 ms, not including network transit" in the docs; "under 200 ms" in the launch post | | Voice design from text | Yes, at launch | English only | | Instant cloning | Not stated in the pages we read | From a 15-second recording | | Acting instructions | Yes | Marked "Coming soon" in the docs | | Status | Listed in the docs | Still labeled preview in the docs we read |
Hume's launch post also promised "at least 20 languages in the coming months." We found no evidence that this happened, and voice conversion and phoneme editing were not available at launch.
Hume AI Octave timeline
The table below separates Hume's own announcements from reports by other outlets.
| Date | Event | Source type | |---|---|---| | 2025-02-26 | Octave launches, English and Spanish | Hume, reported by VentureBeat | | 2025-10-01 | Octave 2 preview launches with 11 languages | Hume blog | | 2026-01-22 | Google DeepMind reportedly signs a licensing agreement with Hume and hires CEO Alan Cowen | Reported by PYMNTS | | 2026-03 | TADA, Hume's open-source TTS model, is last updated in the README | GitHub | | 2026-06-14 | Hume's Expression Measurement API shuts down | Reported by a competitor | | 2026-10-02 | Sunset of the TTS and EVI APIs announced | Roundup citing Hume's changelog | | 2026-11-13 | Access ends at 12:01 a.m. EST, and data is deleted afterward | Hume documentation |
The DeepMind deal is reported as a licensing agreement and a set of hires. Terms were not disclosed, and it is not an acquisition. Hume's CEO role passed to Andrew Ettinger. Nothing in Hume's shutdown notice connects the two events, and we do not claim a link.
What to do before November 13
If you still have Octave voices or audio, work through this list while the APIs are running:
- Download every audio file you need. Hume says account data is permanently deleted after November 13, 2026.
- Keep the original recordings you used for each clone. Cartesia's guide, a vendor source, says designed voices cannot be exported because they have no source recording. Hume's notice says nothing about export, so treat that as one vendor's reading.
- Assume voice IDs and configurations will not move. Cartesia's guide says Hume voice IDs and EVI configurations do not transfer to other providers. Pick the closest match elsewhere and test it on your own sample text.
- Email assist@hume.ai if you need help with your account before the deadline. The notice lists this address.
- Plan around the time zone. The deadline is 12:01 a.m. EST, which may be the evening of November 12 where you are.
How Octave compared with independent arena scores
The Artificial Analysis TTS leaderboard lists Octave 2 at 1051 Elo and the original Octave TTS at 1039 Elo. The page shows rank ranges rather than exact ranks for these models, so we quote Elo only. The top of the same leaderboard has Eleven v4 Turbo at 1330 and Cartesia Sonic 3.6 at 1285.
Hume's own Real World VoiceEQ leaderboard lists 39 TTS models with scores from 0 to 1. The version we read did not show Octave, so we cannot say how Octave ranks there. Elo and benchmark scores measure listening preference or test results. They do not measure price, latency or fit for your project.
What to use instead
The right replacement depends on what Octave did for you. If you used it to steer delivery with words, look for a tool with explicit delivery controls, and test them on your own script.
- Hosted TTS with delivery controls. Cartesia Sonic has speed, volume, emotion and laughter controls. Our Cartesia Sonic explainer covers its models and pricing. ElevenLabs and Inworld also publish API prices.
- Open models you can run yourself. TADA is Hume's own open-source TTS model. Its code is MIT licensed, and its weights use the Llama 3.2 Community License, so read that license before you use it. Chatterbox, from Resemble, is also MIT licensed.
- Other voice clone sources. If you need to move voices from a platform that is closing, our PlayHT alternatives guide explains what to check before you switch.
Choose on your own tests. A short script read by two candidate voices will tell you more than any leaderboard.
Where OndaLabs fits
Octave's pitch was steering delivery with words. OndaLabs also controls delivery by describing it in plain words and with supported expressive tags in the script. It can design a custom voice from a text description, which costs 1,500 credits, and Creator at $22 a month includes 3 custom voices. Think of it as a similar idea, not an equivalent one, and do not treat it as a better version of Octave.
Two-speaker dialogue is available, and the free plan includes dialogue and 5 saved clips. OndaLabs does not offer emotion detection, which is part of Hume's expression measurement work. We make no latency or quality claims here.
Because Hume deletes account data after November 13, keep your original recordings and downloaded audio. With OndaLabs, a clone is made from a recording you have permission to use. You can manage custom voices, try text to speech, or build a script with dialogue.
Frequently asked questions
What is Hume AI Octave?
Octave is Hume's text to speech model, launched in February 2025 and described by Hume as the first TTS system powered by a large language model. Its distinctive feature was voice design from a text description. Octave 2 followed as a preview in October 2025.
Is Hume AI shutting down its text to speech API?
Yes. Hume says access to its TTS and EVI APIs ends on November 13, 2026 at 12:01 a.m. EST. Account data is permanently deleted after that date. Hume's notice does not give a reason or export guidance.
What happens to my Hume voices and data?
Hume says account data is permanently deleted after November 13, 2026. Voice IDs and EVI configurations do not transfer to other providers, according to one vendor's guide. Keep your original recordings and download your audio before the deadline.
Did Google buy Hume AI?
Not according to the reports we read. The January 2026 reports describe a licensing agreement with Google DeepMind and the hiring of CEO Alan Cowen and several engineers. Terms were not disclosed, so do not treat the deal as an acquisition.
What is the difference between Octave 1 and Octave 2?
Octave 1 supports English and Spanish, with about 200 ms latency. Octave 2, which was in preview, supports 11 languages and was described as having latency as low as about 100 ms, excluding network transit. Octave 2's voice design is English only, and acting instructions were marked as coming soon.
What are good alternatives to Hume Octave?
Start with the controls you used in Octave, such as voice design and delivery instructions. Then test hosted options such as Cartesia Sonic, Inworld and ElevenLabs on your own script, and consider open models such as TADA or Chatterbox if you can run them. Check each vendor's current pricing and license terms, because they change.