Private preview — sign-ups opening soon

Blab is the AI voice platform built on its own infrastructure — streaming speech that starts in under a second, dubbing that preserves every note of the soundtrack, and cloning that keeps the soul in the voice.

tts-studio
AAsel · Studio voiceEN
Generate
···
languages, every voice
0
languages, every voice
studio voices
0
studio voices
to first audio
<0s
to first audio
clips per batch
0
clips per batch
TürkçeEnglishالعربيةБългарскиČeštinaDanskDeutschΕλληνικάEspañolEestiSuomiFrançaisहिन्दीHrvatskiMagyar
Bahasa IndonesiaItaliano日本語한국어LatviešuNederlandsPolskiPortuguêsRomânăРусскийSlovenčinaSlovenščinaSvenskaУкраїнськаTiếng Việt

Every voice ships fluent in all thirty. No extra voices to buy. No accents to apologize for.

Text to speech

It starts speaking before you finish blinking.

Streaming synthesis pushes the first audio frame in under a second — the take is playing while it's still being generated. Pick from 64 studio voices, direct the pace and stability, and every take lands in your history with its own player.

  • Sub-second first audio over streaming SSE
  • Pace, stability and style direction per take
  • Takes feed with replay, download and regenerate
tts-studio
AAsel · Studio voiceEN
Generate
···

Dubbing studio

We translate the voice. The soundtrack never finds out.

Blab separates the vocals from the world around them, re-voices every line in the target language at millisecond-accurate timestamps, and lays the new performance back over the untouched music and ambience. The result doesn't sound dubbed. It sounds shot that way.

dubbing-studio
SeparateTranscribeTranslateSpeakMix
VocalsEN
Music + ambienceuntouched
00:03.214

The city never sleeps, it just changes shifts.

Şehir hiç uyumaz, sadece vardiya değiştirir.

00:07.842

Somewhere a saxophone argues with the rain.

Bir yerlerde bir saksofon yağmurla tartışıyor.

00:12.090

And the music plays on, untouched.

Ve müzik, hiç dokunulmadan çalmaya devam ediyor.

Music preserved

Source separation lifts the speech out and leaves the score, the rain and the room exactly where they were.

Millisecond timing

Every line lands on the original utterance window — down to the silence between sentences.

Transcripts, in sync

Original and translation side by side, highlighted live, click any line to jump the player there.

Voice cloning

Your voice, twice.

Two engines, one voice. Instant cloning is ready in seconds. The studio-grade engine runs a deep optimization pass that chases the original until the difference stops mattering. Both speak all 30 languages — and neither starts without recorded consent.

  • Instant engine: cloned in seconds
  • Studio engine: a ~10-minute deep match
  • Consent is mandatory, by design — not by checkbox
voice-cloning
Drop a 30-second clip
Consent recorded

Batch generation

Two hundred clips. One click. Go get coffee.

Queue up to 200 clips in a single job and watch them stream in live — every row gets its own player, download and retry. Progress arrives over SSE, so the page is just a window: refresh it, close it, come back. The job doesn't care.

  • Up to 200 clips per job
  • Live per-clip progress over SSE
  • Download everything as one zip
batch · 200 clipsSSE live
184 / 200
clip-188.mp3
clip-189.mp3
clip-190.mp3
clip-191.mp3
clip-192.mp3
clip-193.mp3
clip-194.mp3

Voice agents

Give your agent a mouth. And ears.

Speech in, reasoning in the middle, speech out — one API call per turn. Your agents listen, think with full conversation history, and answer in any of the 64 voices. The same infrastructure that powers the studio powers them.

  • Speech → reasoning → speech in one turn
  • Conversation memory built in
  • Any studio or cloned voice
voice-agent
listening…

Languages

Hire one voice. Get thirty native speakers.

Every Blab voice — curated or cloned — speaks all thirty languages. Same timbre, same character, new language.

Türkçe
English
العربية
Български
Čeština
Dansk
Deutsch
Ελληνικά
Español
Eesti
Suomi
Français
हिन्दी
Hrvatski
Magyar
Bahasa Indonesia
Italiano
日本語
한국어
Latviešu
Nederlands
Polski
Português
Română
Русский
Slovenčina
Slovenščina
Svenska
Українська
Tiếng Việt

64 voices × 30 languages = 1,920 ways to say it.

Developer platform

The voice API you wish you'd built.

An OpenAPI-documented REST API with streaming first-class, a typed TypeScript SDK, an MCP server so AI agents can call Blab natively, signed webhooks and scoped keys. Four lines of code to a speaking product.

REST API

OpenAPI + live Swagger. Streaming TTS over SSE.

TypeScript SDK

Full coverage: tts, dubbing, voices, jobs, usage.

MCP server

Your AI agents get Blab as native tools.

Signed webhooks

HMAC-signed events with retries and delivery logs.

Scoped API keys

Hashed at rest, least-privilege scopes.

Usage analytics

Per-model, per-day breakdowns via API.

blab-api
curlTypeScriptMCP

Infrastructure

Our models. Our gateway. Our problem — not yours.

The entire pipeline — synthesis, separation, translation, cloning — runs on Blab's own infrastructure behind the Qevron gateway. No third-party voice APIs in the loop, no surprise deprecations, no per-vendor privacy story.

Own infrastructure

GPU workers behind our own gateway. The stack is the product.

Consent-gated cloning

Voice cloning will not start without explicit recorded consent.

Audio as personal data

Uploads served only through signed, expiring URLs.

Jobs that survive

Queued pipeline with checkpoints — retries resume, never redo.

Early access

The studio is warming up.

Blab is in private preview with early teams. Sign-ups open soon — the studio is already live, already dubbing, already speaking thirty languages.

Already invited? Sign in