# How to migrate from ElevenLabs text to speech to Bland Speech (2026): Streaming, voices, and billing

By Samuel Oram-Jones

A voice agent built on ElevenLabs' standard WebSocket can't use ElevenLabs' newest models without a rewrite. ElevenLabs says that endpoint does not support the eleven_v3 or eleven_v4 models, and routes Eleven v4 through its Text to Dialogue API and Eleven v4 Turbo through a Text to Dialogue WebSocket.

A team in that spot is rewriting its streaming layer either way, and the rewrite can point at ElevenLabs' newer endpoints or at Bland Speech's. Pointing it at Bland means mapping each ElevenLabs call, audio format, and tag onto its Bland counterpart, and weighing what costs more or less afterward.

## Why would a team move from ElevenLabs text to speech to Bland Speech?

A team moves from ElevenLabs to Bland Speech when it wants a phone-first voice with one flat rate and no monthly plan to manage, and when its streaming layer is being rewritten anyway.

As of November 2026, ElevenLabs publishes lower latency numbers than Bland does: a median of about 100ms for Eleven v4 Turbo, excluding application and network latency, and about 150ms median time to first speech in its launch post, against Bland's published 199ms to first audio. Each is a company's own claim, not a race either has run. ElevenLabs also sells plans from Free to Business, and its Business plan advertises low-latency text to speech "as low as 5c/minute".

Bland Speech differs in two ways that matter on a phone line. Bland bills each character at $0.04 per 1,000 with no plan tiers, and Bland Speech v3 ranked first of 13 AI voices in Design Arena's blind realism results of 4 August 2026. The trade is a higher published latency, which a team can check on its own network before it moves traffic.

## What maps to what when moving from ElevenLabs text to speech to Bland Speech?

ElevenLabs' `stream-input` WebSocket maps to Bland's `WSS /v2/tts/ws`, its `voice_id` maps to a Bland voice UUID, `ulaw_8000` maps to `mulaw` at 8,000 Hz, and audio tags map to Bland's bracket tags.

| What a team touches | Bland Speech | ElevenLabs text to speech |
| --- | --- | --- |
| Streaming text in | WebSocket at /v2/tts/ws, one socket per conversation | /v1/text-to-speech/{voice_id}/stream-input, which takes no v3 or v4 model |
| Complete text | POST /v2/tts, up to 5,000 characters | HTTP streaming with chunked transfer encoding |
| Voice | Voice UUID in the request body | voice_id in the request path |
| Phone audio | `mulaw` at 8,000 Hz, or `pcm_s16le` at 8,000 to 48,000 Hz | `ulaw_8000`, `alaw_8000`, `pcm_8000`, and `wav_8000`, among others |
| Delivery | Bracket tags, pause markers, and two controls from 0.0 to 1.0 | Audio tags on Eleven v4 and v3, SSML breaks on older models |
| Instant clone | One clean sample of about 10 seconds | 10 seconds on v4, Starter plan or above |
| Professional clone | 30 minutes or more, plus Bland's consent and rights review | About 30 minutes, Creator plan or above, plus a voice-captcha check |
| Price | $0.04 per 1,000 characters | v4 $0.08, v4 Turbo and Flash $0.04 per 1,000 characters |

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

- **Streaming text in:** Bland describes the ElevenLabs job as replacing "stream-input messages, base64 audio, and manual flush behavior". On Bland, one `init` message opens the session, text arrives as `speak` deltas under a `context_id` for each turn, `end_of_turn` takes the place of a flush, and audio comes back as binary frames.
- **Phone audio:** Bland offers two encodings, `pcm_s16le` and `mulaw`, and mu-law is fixed at 8,000 Hz. ElevenLabs gates 44.1 kHz PCM and WAV to Pro or above.
- **Delivery:** ElevenLabs says v4 and v3 take no SSML break tags and that SSML phoneme tags work only on `eleven_flash_v2`.
- **Price:** each is the published list rate. ElevenLabs bills API usage in US dollars per character, and a plan bundles characters at those rates.

</details>

Delivery is the part that takes thought. ElevenLabs steers Eleven v4 and v3 with audio tags such as `[whispering]`, `[shouting]`, and `[laughing]`, and Bland publishes its own verbatim set, so a team maps by meaning and checks by ear. `[whispering]` becomes `[say in a whisper]`, `[laughing]` becomes `[laughs]`, and `[shouting]` is closest to `[say loudly]` or `[very loud]`. Bland says invented tag variants are ignored or garbled and that tags work on current-generation voices only.

Pauses change spelling. ElevenLabs' `<break time="x.xs" />` on older models, which takes up to 3 seconds, becomes `<|x|>` on Bland, with a length from 0.1 to 10.0 seconds, and each marker costs roughly a second of extra breath. For exact pronunciation, Bland recommends wording, such as spelling a number out as words.

## What are the steps to switch from ElevenLabs text to speech to Bland Speech?

A switch takes six steps: list models and voices, build a voice table, replace the streaming layer, request phone audio directly, rewrite tags, and run both providers on a share of traffic.

1. A team lists which ElevenLabs model each flow uses (Eleven v4, v4 Turbo, v3, Flash, or Multilingual v2), because the answer decides which endpoint it is leaving.
2. A team builds a voice table with each `voice_id` in one column and a Bland voice UUID beside it. A GET to `/v1/voices` returns every voice available to the organization, and Karen, Matthew, and River are open to every organization.
3. A team replaces the `stream-input` loop with a Bland session, and sends text as a language model writes it:

```json
{ "type": "init", "voice": "29158307-9893-4149-8a75-bc9ce313d64e", "audio": { "encoding": "mulaw", "sample_rate": 8000 } }
{ "type": "speak", "context_id": "turn-42", "text": "Hello" }
{ "type": "speak", "context_id": "turn-42", "text": ", how can I help?" }
{ "type": "end_of_turn", "context_id": "turn-42" }
```

4. A team that requests `ulaw_8000` for a phone line asks Bland for `{"audio": {"encoding": "mulaw", "sample_rate": 8000}}`, which skips a conversion step. A team on `pcm_48000` asks for `pcm_s16le` at 48,000 Hz, the rate `BTTS_V3` renders at natively.
5. A team rewrites audio tags and pauses into Bland's tag set and pause markers, and renders each line in Speech Studio, which uses the same models as the API, before putting it in code.
6. A team routes a small share of calls to Bland, compares recordings against ElevenLabs output for the same lines, and moves the rest once the numbers hold.

A voice agent that sends one large `speak` at the end of a turn gets no audio until `end_of_turn`, because release is driven by text continuing to arrive, so text goes out as deltas with its leading spaces and punctuation intact. A new `context_id` cuts off the turn that is playing, which is how a caller's interruption is handled, and the client discards audio it hadn't played yet.

## Will callers hear a difference after moving from ElevenLabs to Bland Speech?

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, while ElevenLabs' Eleven v3 ranked 12th at 938.

Eleven v4 and Eleven v4 Turbo launched on 28 September 2026, after those results, and weren't scored in them. In the benchmark, native speakers pick blind which of two clips of the same text sounds more like a real person, one voice per model per gender, and the benchmark has a Phone Agents category. A score across hundreds of prompts can't say how one team's lines sound, so the check that settles it is a call script's most repeated lines rendered by Eleven v4 Turbo and by Bland, played to a listener who can't see the labels.

## 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. Against ElevenLabs' list rates, Bland is half the price of v4 and v3 and the same as v4 Turbo and Flash.

At 5 million characters a month, the list prices come to $200 on Bland, $400 on ElevenLabs v4 at $0.08 per 1,000, and $200 on v4 Turbo or Flash at $0.04. ElevenLabs sells those characters inside monthly plans from Starter at $6 to Business at $990, and each paid plan's price divided by its included v4 characters comes to the same $0.08 per 1,000. Bland has one standard rate for studio and API, a $0.001 minimum per request, and billing that follows delivery, so a synthesis that fails before any bytes are written isn't charged. Enterprise pricing is custom, set with Bland's sales team.

ElevenLabs voices don't come along. Bland says voice IDs are provider-specific and can't be reused, so each voice is cloned again on Bland from its original recording. An instant clone needs one clean sample of about 10 seconds, a professional clone is fine-tuned on 30 minutes or more of audio and goes through Bland's consent and rights review, every clone is private to the customer's organization, and the customer confirms the voice is theirs to use. ElevenLabs verifies a professional clone by having the voice owner read lines from a captcha image aloud, and Bland's check is a review of consent and rights, so a team plans for a different process.

A head-to-head sits at [Bland Speech vs ElevenLabs](/compare/bland-speech-vs-elevenlabs), and the [ElevenLabs voice cloning alternatives](/blog/elevenlabs-voice-cloning-alternatives) list covers clones. Moves from other providers have their own [text to speech migration guides](/compare/text-to-speech/guides), the [text to speech comparison hub](/compare/text-to-speech) holds every comparison, the [text to speech API](/text-to-speech-api) page lists each endpoint, and [Bland Speech](/speech) has the studio.

## Frequently asked questions

### Can ElevenLabs voices be exported to Bland Speech?

No. Bland says voice IDs are provider-specific and can't be reused, so a team clones each voice again on Bland from its original recording. An instant clone needs one clean sample of about 10 seconds, and a professional clone needs 30 minutes or more and passes Bland's consent and rights review.

### How do ElevenLabs audio tags translate to Bland Speech?

By meaning, checked by ear. `[whispering]` becomes `[say in a whisper]`, `[laughing]` becomes `[laughs]`, and `[shouting]` is closest to `[say loudly]` or `[very loud]`. Bland says invented tag variants are ignored or garbled, and tags work on current-generation voices only.

### Does ElevenLabs' 8 kHz mu-law output carry over?

Yes. `ulaw_8000` becomes `mulaw` at 8,000 Hz on Bland's `/v2/tts` and `/v2/tts/ws`, so a phone line gets the format it expects without a conversion step.

### Is Bland Speech cheaper than ElevenLabs?

Against ElevenLabs v4 and v3, yes: half the price at list rates. Against v4 Turbo and Flash, no: the same price. Bland Speech is $0.04 per 1,000 characters with no monthly plan, and ElevenLabs charges $0.08 for v4 and $0.04 for v4 Turbo and Flash.

### Is Bland Speech as fast as Eleven v4 Turbo?

ElevenLabs publishes a median of about 100ms for Eleven v4 Turbo, excluding application and network latency, and Bland publishes 199ms to first audio on a streamed request. Those are two companies' own figures, so a team that cares tests both on its own network.

### Do Pipecat and LiveKit agents need code changes to move?

A small change in each. Pipecat supports both ElevenLabs and Bland as text to speech services, and the Pipecat team maintains Bland's, installed with `uv add "pipecat-ai[bland]"`. LiveKit Agents has ElevenLabs plugins for Python and Node.js and a Bland plugin for Python, installed with `uv add "livekit-agents[bland]~=1.8"`.
