# How to migrate from Hume Octave to Bland Speech (2026): What Hume's site shows, and a plan for the move

By Samuel Oram-Jones

A developer who goes looking for Hume's text to speech documentation in November 2026 ends up on a page about audio decisions. Hume's old text to speech docs address answers with an HTTP 308 redirect to the Audio Decisions overview, the old EVI docs address ends up on the same page, and the product pages for Octave and EVI return a 404.

A team with Octave in production has no notice to plan against, and nothing on Hume's site says what happens next. Octave's features still map onto Bland Speech's, and a cutover can start without a date from Hume.

## Is Hume Octave still available for new text to speech work?

As of November 2026, Hume's site, developer docs, and status page show no text to speech product, and none of them gives an end date for Octave.

Hume's homepage leads with the Audio Decisions API, and its product list names the Audio Decision API, Expression APIs, Human Feedback API, User Testing API, and Speech Data. Hume's docs sitemap holds 61 addresses and none of them is about text to speech, Octave, EVI, or speech, and Hume's status page lists two APIs, Decisions and Expression Measurement. Hume's pricing address redirects to its homepage. Pipecat marks its Hume service as deprecated.

Two older pages are still online: the Octave 2 launch post, dated 1 October 2025, and the EVI 3 post, dated 17 July 2025. Neither carries a notice that the product is ending. A team that reads those posts as a roadmap is reading something Hume's current navigation has moved past, so the safer assumption is that Octave could stop answering requests without warning, and a cutover plan written now costs little.

## What maps to what when moving from Hume Octave to Bland Speech?

Octave's voice cloning maps to Bland's instant and professional clones, its low-latency generation maps to Bland's WebSocket and HTTP endpoints, and its delivery inference maps to written tags and controls.

| What a team touches | Bland Speech | Hume Octave 2 (launch post, 1 October 2025) |
| --- | --- | --- |
| Voice from a recording | Instant clone from one clean sample of about 10 seconds | Sample voices made from 15-second recordings |
| Speed | Bland publishes 199ms to first audio | Hume says audio in under 200ms |
| Delivery | Bracket tags, pause markers, and two controls from 0.0 to 1.0 | A speech-language model that infers emotional tone from the script |
| Pronunciation edits | Wording changes, such as spelling a number as words, and pause markers | Phoneme editing, described as in exploration |
| Price | $0.04 per 1,000 characters | Half the price of Octave 1, and under a cent per minute on dedicated deployments |

<details><summary>What these rows mean</summary>

- **Voice from a recording:** Bland says an instant clone needs one clean sample of about 10 seconds. Hume said its two sample voices were made with instant cloning from 15-second recordings.
- **Speed:** each figure is the company's own claim. Bland says its 199ms runs to the first audio chunk on a streamed request, and Hume's wording is "under 200ms".
- **Delivery:** Hume describes Octave as a model that understands how a script informs tune, rhythm, and timbre. Bland's model performs what the text asks for, and delivery is written into the text.
- **Pronunciation edits:** Hume presented direct phoneme editing as something it was working on at the Octave 2 launch. On Bland, a pronunciation is fixed in the wording.
- **Price:** Hume stated Octave 2's price as ratios rather than a per-character rate, so the two cells can't be set side by side.

</details>

Delivery is where the work lands. Hume said Octave infers "when to whisper secrets, shout triumphantly, or calmly explain a fact" from a script. Bland publishes a verbatim tag set, such as `[say in a whisper]`, `[say loudly]`, and `[laughs]`, placed right before the words they shape, and says invented variants are ignored or garbled. Two controls, `expressiveness` and `stability`, each run from 0.0 to 1.0, and a pause is written `<|0.8|>`, with a length from 0.1 to 10.0 seconds. A team that relied on Octave to read a flat line with feeling now writes that feeling in.

## What are the steps to switch from Hume Octave to Bland Speech?

A switch takes six steps: inventory every call site, collect original recordings, build a voice table, pick endpoints, rewrite delivery, and run both providers side by side while Hume still answers.

1. A team lists every place that requests Octave audio, with the voice, language, and output format each one uses, working from its own code because Hume's docs addresses redirect.
2. A team collects the original recordings behind each cloned voice. Bland says voice IDs are provider-specific and can't be reused, so a Bland clone starts from audio and the speaker's permission.
3. A team builds a voice table with each Hume voice in one column and a Bland voice ID beside it. Three built-in voices, Karen, Matthew, and River, are open to every organization, and a clone stays private to the organization that made it.
4. A team picks an endpoint per use. A voice agent that streams language model tokens opens one WebSocket per conversation at `/v2/tts/ws`, and complete text goes to `POST /v2/tts`. Phone audio is `mulaw` at 8,000 Hz, requested directly:

```bash
curl -X POST "https://api.bland.ai/v2/tts" \
  -H "Authorization: Bearer $BLAND_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Your appointment is confirmed for Tuesday at 3 PM.",
    "voice": "29158307-9893-4149-8a75-bc9ce313d64e",
    "audio": { "encoding": "mulaw", "sample_rate": 8000 }
  }' \
  --output prompt.ulaw
```

5. A team rewrites delivery as tags, pause markers, and the two controls, and renders each line in Speech Studio, which uses the same models as the API, before putting it in code.
6. A team sends a small share of calls to Bland, compares recordings against Octave output for the same lines, and moves the rest once the numbers hold. Because Hume has published no end date, the old path stays wired until the new one has carried traffic for long enough to trust.

Bland's WebSocket takes text exactly as a language model writes it, with leading spaces and punctuation intact, and picks the synthesis boundaries itself. A new `context_id` cuts off the turn that is playing, which is how a caller's interruption is handled, and the client drops any audio it hadn't played yet.

## Which text to speech API fits a former Octave workload?

Bland Speech fits a workload that ends on a phone line, because it outputs mu-law at 8 kHz directly and Bland Speech v3 ranked first of 13 AI voices in Design Arena's [Audio Realism Bench results of 4 August 2026](https://notes.designarena.ai/audio-realism-benchmark-measuring-realism-in-audio-models/), at 1365, with the human recording at 1500.

Hume wasn't in those results, so the benchmark can't say how Octave sounds beside Bland. The benchmark has native speakers pick, blind, which of two clips of the same text sounds more like a real person, and it has a Phone Agents category. A team deciding between providers renders a call script's most repeated lines through each and lets a listener who can't see the labels choose.

Bland owns its speech model, and Bland's trust page says speech synthesis runs inside its FedRAMP boundary. Bland is SOC 2 Type II audited, HIPAA compliant with a signed BAA, and FedRAMP 20x Class C certified at Moderate, granted 17 November 2026.

## What does the move cost, and what can't come along?

Bland Speech costs $0.04 per 1,000 characters ($40 per million), pay as you go with no monthly plan, and 50,000 characters are free once for a new account. A 10-minute narration comes to $0.36, and each request carries a $0.001 minimum. Enterprise pricing is custom, set with Bland's sales team.

Hume's pricing address now redirects to its homepage, and the Octave 2 launch stated prices as ratios, so a team that knows its own Octave invoice divides it by the characters it sent and sets the result beside $0.04.

Two things stay behind. A cloned voice doesn't move, because Bland voices are made on Bland from audio, so each one is cloned again on Bland from its original recording: one clean sample of about 10 seconds for an instant clone, or 30 minutes or more for a professional clone that goes through Bland's consent and rights review. EVI, Hume's speech-to-speech model, is a separate product, and Bland Speech takes text in and returns audio, so a team that used EVI for whole conversations pairs Bland Speech with its own speech recognition and language model.

For other routes, the list of [Hume Octave alternatives](/blog/hume-tts-alternatives) compares the field, and the same plan applies to [moving off PlayHT](/blog/migrate-from-playht). Every comparison sits in the [text to speech comparison hub](/compare/text-to-speech), the other [text to speech migration guides](/compare/text-to-speech/guides) follow the same shape, the [text to speech API](/text-to-speech-api) page lists each endpoint, and [Bland Speech](/speech) has the studio.

## Frequently asked questions

### Has Hume shut down Octave?

No Hume page says so. Hume's site, docs, and status page show no text to speech product, its old text to speech docs address redirects to the Audio Decisions docs, and none of those pages gives an end date.

### Can a Hume voice be exported to Bland Speech?

No. Bland says voice IDs are provider-specific and can't be reused, so a team clones the voice again on Bland from its original recording. An instant clone needs one clean sample of about 10 seconds, and every clone is private to the customer's organization.

### How is pronunciation controlled on Bland Speech?

Through the text: a number spelled as words is read exactly as written, a long digit run is split with a pause marker, and every sentence ends with terminal punctuation.

### Do Pipecat and LiveKit apps need changes to move off Hume?

A small change in each. Pipecat marks Hume as deprecated and supports Bland as a service the Pipecat team maintains, installed with `uv add "pipecat-ai[bland]"`. LiveKit Agents has Hume and Bland plugins for Python, and Bland's installs with `uv add "livekit-agents[bland]~=1.8"`.

### How does Bland Speech pricing compare with Hume's?

Bland Speech is $0.04 per 1,000 characters with no monthly plan. Hume stated Octave 2's price as a ratio to Octave 1, so the comparison runs through a team's own Octave invoice divided by the characters it sent.
