Hume has turned from a voice generator into an audio judgment API. If you need to score thousands of calls or voice-model outputs, the Audio Decisions API is cheap to try and easy to call.
Home / Tools / Audio / Voice
Audio / Voice · Tool review
Hume AI Review: Voice API Pricing 2026
The AI news that actually mattered this week.
One short email every Friday: the biggest AI stories, what they mean for you, and the tools worth trying. A 3-minute read, no hype.
Free. Unsubscribe anytime.
You're in. Watch for the brief on Friday. Free. Unsubscribe anytime.
Summary
Pros and cons
What we like
- Asks audio questions in plain language.
- Cheap per-token billing, no commitment.
- 1M-token trial credit.
- Grounded in large human-rating datasets.
- Python and TypeScript SDKs.
What could be better
- Octave TTS and EVI end Nov 13, 2026.
- Developer API only, no app.
- 5-minute limit per request.
- Answers need your own thresholds.
Key features
- Audio Decisions: Ask questions, get probabilities.
- Expression Measurement: Emotion scores for voice and face.
- VoiceEQ: Benchmark for voice AI systems.
- Voice APIs: Octave and EVI end Nov 13.
- Limits: 1 s to 5 min audio, 600 req/min.
- Pricing: $0.47 per 1M input tokens.
Screenshots
Full review
Independent and reader-supported. Some links may earn us a commission; it never changes our scores. Editorial policy
Hume AI now sells the Audio Decisions API, which answers plain-language questions about audio for $0.47 per 1M tokens. Octave TTS and EVI end November 13.
Quick answer: Hume AI is best known for Octave, its expressive text-to-speech model, and EVI, its Empathic Voice Interface for voice agents. Hume is now sunsetting both: access ends at 12:01 a.m. EST on November 13, 2026, and account data is deleted after that. The product Hume sells today is the Audio Decisions API, launched on November 2, 2026, which listens to a recording and returns probabilities for questions you write, such as “Does the caller sound frustrated?” It costs $0.47 per 1M input tokens, with a 1M-token trial credit. Prices checked in Hume’s billing docs on November 10, 2026.

What Hume AI does in 2026
- Audio Decisions API: send an audio file (1 second to 5 minutes) with up to 25 plain-language questions. Answers come back as yes/no, choice, score or time-span probabilities, with no transcript.
- Expression Measurement API: scores emotional expression in speech and faces, with separate audio and video endpoints.
- Real-World VoiceEQ: Hume’s benchmark and leaderboard for comparing voice AI systems in real conditions.
- Octave TTS and EVI (ending): the expressive voice generator and speech-to-speech voice agent API. Both stay supported until November 13, 2026, then shut down.
Hume AI voice products: what changes on November 13
Hume posted the sunset notice in its developer changelog on October 2, 2026. Octave 1 and 2, voice cloning, voice design and every EVI configuration go away with it, and Hume says any data left in your account is permanently deleted after the cutoff. If you built on Hume voices, export generated audio, prompts and EVI configs now. The notice points users to Real-World VoiceEQ to compare replacement voice models.
For new voice projects, look at ElevenLabs for realistic voices and cloning, or our list of free unlimited text to speech tools if you just need narration. Luvvoice is a simple free option with 200+ voices.
Hume AI Pricing (2026)
| Product | Price | Notes |
|---|---|---|
| Audio Decisions API | $0.47 per 1M input tokens | No commitment. Answers are not billed, only audio plus question text |
| Trial credit | 1M input tokens | Granted once your organization gets access; expiry shown on the Plan page |
| Enterprise | Custom | Volume pricing, deployment in your own VPC, on premises or in country |
| Octave TTS and EVI | No longer sold | Access ends November 13, 2026; hume.ai/pricing now redirects to the homepage |


What does that cost in practice? Hume’s own examples put a 5-minute support call with six questions at about 26,400 input tokens, roughly $0.012. Rating a 10-second text-to-speech clip is about 1,000 tokens. Audio length drives most of the bill, so ten questions about a minute of audio cost little more than one.
Who Hume AI is for
Teams that review calls or test voice models: contact centers scoring calls for frustration or interruptions, voice-agent builders checking if a reply sounds natural, and labs grading text-to-speech output without a listening panel. It is a developer API, not an app. If you wanted Hume to read text aloud, that part of Hume ends this week.
Limitations
- No more voice generation: Octave and EVI shut down on November 13, 2026.
- Developer-only: there is no consumer app; you need an API key and some code.
- Five-minute cap per request: longer recordings must be split into files.
- Perceptual answers are probabilities: Hume reports 70.7% agreement with a human panel’s average, so set thresholds on your own labeled audio.
TechPilot.ai’s Verdict on Hume AI
Hume has turned from a voice generator into an audio judgment API. If you need to score thousands of calls or voice-model outputs, the Audio Decisions API is cheap to try and easy to call.
If you came for Octave voices or EVI agents, plan your move now: both stop working on November 13, 2026.
Hume AI FAQ
Is Hume AI free?
Not fully. New Audio Decisions organizations get a one-time trial credit of 1M input tokens. After that the API is pay as you go, and a payment method is required.
•
How much does Hume AI cost?
The Audio Decisions API costs $0.47 per 1M input tokens with no commitment. Enterprise pricing is custom. Octave TTS and EVI plans are no longer sold (checked November 10, 2026).
•
Is Hume AI shutting down?
No, but its voice APIs are. Hume is sunsetting Octave TTS and EVI, with access ending November 13, 2026 at 12:01 a.m. EST. Account data is deleted after that.
•
What is the Hume Audio Decisions API?
It is an audio-native model launched on November 2, 2026. You send audio and plain-language questions, and it returns probabilities, such as how frustrated a caller sounds.
•
What can replace Hume Octave text to speech?
Voice generators such as ElevenLabs, or free tools like Luvvoice and TTSMaker for simple narration. Export your Hume audio and configs before November 13.