Streaming TTS API for Voice Agents

Built for production. Certified for compliance.

Cloud, self-hosted, or on-premises.

ISO 27001

Certified infrastructure

GDPR

EU data processing, DPA ready

99.9% SLA

Status page & incident response

Zero Training

Your data never trains our models

Other TTS APIs wait for a full sentence. We don't.

Stream text into our WebSocket as it arrives: straight from an LLM's output, a live transcript, or any other incremental source.

Send it word by word, with no need to wait for complete sentences. Synthesis starts as soon as our model has enough context, often just 2–3 words.

This is the single biggest improvement you can make to perceived voice agent speed — and only Palabra offers it.

Palabra starts synthesis after 2 words.  Others wait for the full sentence.

LLM

Tokens generating…

Others

Waiting… then audio

~500ms

Palabra

Audio from word 2

~35ms

One cloned voice. Eight languages. Zero foreign accent.

Same voice identity in German, French, Spanish — each sounds like a native speaker. 6 seconds of reference audio. No fine-tuning.

Purpose-built for real-time voice

Not a general-purpose TTS. Built for where latency, naturalness, and unit economics all matter at once.

VAPI • RETELL • LIVEKIT

Voice Agents

Streaming input eliminates text buffering — the #1 source of perceived lag. Clean barge-in handling. Your agent responds in real-time.

Μ-LAW • PCM • G.711

Telephony & IVR

Native G.711 μ-law for direct SIP trunk integration. Sub-100ms latency keeps calls flowing. Clone your brand voice once, use across every region.

DEACCENTING • 21 LANGUAGES

Multilingual Content

One speaker, 21 languages, zero foreign accent. Your German content sounds German. That used to require 8 voice actors.

CAPABILITIES

Next-Gen AI TTS Engine: What You Can't Get Anywhere Else

Two exclusive features plus the fastest, cheapest voice cloning in the market.

Real-Time Deaccenting

Clone a voice in English, speak in German — it sounds like a native German speaker. The voice identity stays. The foreign accent disappears. Adjust the accent with real-time settings.

Blue and black fuzzy sphere with speech bubbles saying 'Hi!' and '¡Hola!' on a black background.

Abstract sphere in the center with sound wave patterns and arrows pointing toward and away from it.

True Streaming Input — From 2 Words

Stream text to our WebSocket as it becomes available—directly from an LLM, a live transcript, or any other incremental source. Send text word by word without waiting for complete sentences.

Voice Cloning

Clone a voice from just three seconds of audio and adjust accent strength in real time.

$0.03 per 1,000 Characters

Competitive, transparent pricing. No concurrency caps. Voice cloning included.

DEVELOPER-FIRST

Ship in an afternoon

Python SDK that is compatible with the platforms you already use and love.

VAPI

RETELL

LIVEKIT

PIPECAT

Python

async with palabra.tts(language="en", voice_id="default_low") as tts:
    await tts.send_text("The sun was setting over the mountains,")
    await tts.send_text(" casting long golden shadows.", eos=True)
    async for chunk in tts: # TtsChunk: audio, generation_id, last_chunk, audio_len
        play(chunk.audio)
        if chunk.last_chunk:
            break
    await tts.cancel() # stop current synthesis, session stays open

Transparent TTS Pricing: How Palabra Stacks Up

We show where we win and where others lead. You decide what matters.

Streaming Input

★ From 2 words of sentence

Full sentence

Partial

Realtime Deaccenting

★ Yes

Voice Cloning

3 seconds

10s-60s

Yes

TTFA (P90)

~35ms

~150ms

90ms

~200ms

Languages

21

70+

42

7

Perceptual Quality

Optimized for speed & business fit

Optimized for natural-sounding voice

Good

Streaming Input

★ From 2 words of sentence

Full sentence

Partial

Realtime Deaccenting

★ Yes

Voice Cloning

2 seconds

10s-60s

Yes

Price / 1K chars

~$0.015

$0.05–$0.10

~$0.030

$0.030

TTFA (P90)

~35ms

~150ms

90ms

~200ms

Languages

21

70+

42

7

Perceptual Quality

Optimized for speed & business fit

Optimized for natural-sounding voice

Good

FROM OUR USERS

Developers ship faster with Palabra

"Switched from ElevenLabs Multilingual. TTS bill dropped 80%, latency is comparable, and streaming input eliminated our text buffering step entirely."

M

Voice Agent Startup

Series A • Bay Area

"Deaccenting is magic. We serve 6 EU markets with one cloned brand voice and each market thinks it's local. Replaced 6 voice actors."

S

Enterprise SaaS

EU • Multilingual support

"Integration was 40 minutes. Swapped out Deepgram, TTFA dropped from 210ms to 98ms. Vapi plugin made it trivial."

J

Contact Center Platform

Vapi integration

Twenty one languages. Each trained to near-native.

Including some country-specific dialects

English

Spanish

Spanish (LA)

Russian

German

Dutch

Italian

Hindi

Korean

Czech

Finnish

Indonesian

Polish

Romanian

Slovak

Swedish

Turkish

French

French (Canada)

Portuguese (Europe)

Portuguese (Brazil)

FAQ About Our Neural TTS

How does voice quality compare to ElevenLabs?

ElevenLabs optimizes for studio-quality English voice. Palabra is optimized for production voice agents — lowest latency, multilingual cloning with native accent, and transparent pricing.

Can I self-host or deploy on-premises?

Yes. Palabra TTS is available as a cloud API, self-hosted deployment, or on-premises installation. All options are ISO 27001-certified and GDPR-compliant. Your data never trains our models. Contact sales for self-hosted and on-premises options.

What does "streaming input" actually mean?

Every other TTS provider requires a complete sentence before synthesis starts — that buffer adds 300–800ms of dead air your users hear as lag. Palabra accepts LLM tokens as they stream and begins audio synthesis after just 2 words, while the LLM is still generating. At 35ms time-to-first-audio, Palabra is the fastest TTS in the world.

How does voice cloning work, and how much audio do I need?

Just 3 seconds of reference audio. We use dual conditioning — discrete codes for the LLM and continuous embeddings for the decoder — to capture both voice identity and natural prosody without fine-tuning.

What is deaccenting and why does it matter?

Most TTS engines carry the phonetic habits of a cloned voice into every language it speaks — so an English-cloned voice sounds foreign in German, even if the words are correct. Deaccenting separates voice identity (timbre, resonance, prosodic character) from accent, so the output sounds like a native speaker of the target language while still sounding like the original person.