Palabra.ai logo
Translation Language · Voice Synthesis

Palabra.ai

Palabra.ai is a real-time speech engine for voice-to-voice translation, speech synthesis and transcription across 60+ languages, with sub-second latency. It ships as a developer API and as ready-to-use apps for meetings, events, webinars and live streams.

Active GDPR compliant Free trial Subscription API available 18+ Verified by Guidaio
Overview

What is Palabra.ai?

Palabra.ai is a real-time speech platform built on three engines: speech-to-speech translation, text-to-speech synthesis and speech recognition. The company says it trains its own translation LLM and controls the whole pipeline, from ASR through to TTS, rather than stitching together third-party models.

The headline numbers are about speed. End-to-end voice translation is announced at under one second, while text-to-speech reaches 35 ms time-to-first-audio (P90, excluding network latency), a figure attributed to a Coval benchmark published in July 2026. Synthesis starts streaming after two or three words instead of waiting for a full sentence, a capability the site presents as exclusive. Voice cloning needs about three seconds of audio and no fine-tuning, and a deaccenting layer lets a voice cloned in English speak German or Spanish without a foreign accent, across 21 TTS languages.

The catalogue is announced at 60+ languages: the dedicated page lists 48 speaker languages plus 5 additional listener languages, and speech-to-text was in early access at the time of collection, covering 13 transcription languages.

There are two ways in. The API route runs on platform.palabra.ai, with WebRTC for client-side apps, WebSocket for server integrations and official Python and TypeScript clients on GitHub. The no-code route runs on app.palabra.ai and covers meeting translation for Zoom, Google Meet and Microsoft Teams, in-person events, webinars and live streams. Around both sit automatic language detection including mid-sentence code-switching, speaker diarization and custom glossaries.

Deployment can be cloud, self-hosted or on-premises, with regional servers offered on Business accounts. On data, the stance is zero retention by default and no training on customer data. The About page claims 500,000+ minutes translated, 60+ languages, sub-second latency and 99% accuracy.

Palabra.ai Ltd is a London company, which announced the acquisition of Talo in November 2025. The site names its own competition, citing ElevenLabs in a TTS comparison table and Deepgram in a customer quote, and announces compatibility with the Vapi, Retell, LiveKit, Pipecat and Vocode voice agent frameworks. Its recurring commercial argument on the meetings page is being four times cheaper than a human interpreter.

What it does

  • Translate live speech into more than 60 languages while preserving the speaker's own voice
  • Generate speech from text with 35 ms time-to-first-audio (P90, network latency excluded)
  • Transcribe audio in real time with precise timestamps and end-of-turn detection
  • Send a translation agent into a Zoom, Google Meet or Microsoft Teams call
  • Clone a voice from three seconds of audio and strip the foreign accent (deaccenting)
  • Broadcast an RTMP, SRT or HLS stream with a translated audio track and subtitles
  • Give event attendees access through a QR code, on their own phone
Audience

When to use Palabra.ai / When not to

A quick filter to help you decide if Palabra.ai is the right fit.

When to use Palabra.ai

  • Product and engineering teams embedding real-time translation, speech synthesis and transcription into their own software through a single streaming API
  • Organisers of multilingual in-person events who want listener access by QR code, with no booth, no headset and no app to install
  • Distributed teams that run Zoom, Google Meet or Microsoft Teams meetings across mixed languages
  • Broadcasters, media and streamers with an RTMP/SRT/HLS chain that needs translated audio and subtitles on output
  • Privacy-constrained organisations and voice agent builders: no retention by default, self-hosted or on-premises deployment, and announced compatibility with Vapi, Retell, LiveKit, Pipecat and Vocode

When not to use Palabra.ai

  • Anyone who needs written document translation: this is a speech engine, not a text or document translator
  • Consumers looking for a free tool: there is no permanent free tier, only a 7-day trial and $50 in credits at sign-up on the API platform
  • Tight budgets on the events side, where the Starter plan is $500 per month for five hours and $100 per hour beyond it
  • Under-18s, and contexts that require a sworn human interpreter (legal, regulated medical): the site positions the tool as a cheaper alternative, without any interpreting certification
  • Mobile-first or offline use: no iOS or Android app is referenced, and everything runs through a WebRTC or WebSocket stream
Get started

How to use Palabra.ai

A typical end-to-end flow, from setup to results.

  1. Pick your route: no-code apps on app.palabra.ai, or the developer API on platform.palabra.ai
  2. No-code route: create an account and start the 7-day free trial
  3. For a meeting, paste the Zoom, Google Meet or Teams link, choose the languages and the voice, then click Join Call so the agent enters the conversation
  4. For an in-person event, share the QR code: attendees scan it, pick their language, then listen or read the subtitles on their own phone, with nothing to download
  5. For a stream, connect OBS, vMix, Castr, YouTube Live or any RTMP/SRT chain and take the translated audio and subtitles on output
  6. Define your glossaries beforehand to lock down the business terms you want translated consistently
  7. API route: sign up on platform.palabra.ai, collect the $50 in credits and create an API key (documented under /docs/auth/api-keys)
  8. Choose your transport: WebRTC for a client-side application, WebSocket for a server-side integration
  9. Follow the four documented steps: import the official Python or TypeScript client, set your API keys, choose source and target languages, then wire it into your interface
  10. Try the browser playgrounds for TTS, S2S and STT without signing up first, and watch the public status page for availability and incidents
Quick read

Pros & Cons

Pros

  • Latency among the lowest published on the market: 35 ms TTFA for TTS and under one second end-to-end for translation
  • Full pipeline owned in-house (ASR, translation, TTS) rather than assembled from third-party models
  • Zero retention by default and no training on customer data, written into the terms and the privacy policy
  • Public, itemised unit pricing: $0.002 per minute for STT, $0.03 per 1,000 characters for TTS, $0.04 per minute for S2S
  • Two ways in on the same infrastructure: no-code apps for business teams, API for developers
  • Listener access with nothing to install (QR code in a browser), and self-hosted or on-premises deployment with regional servers
  • Public documentation with official Python and TypeScript clients, a public status page, a named Article 27 GDPR representative and a published ICO registration

Cons

  • No permanent free tier: the 7-day trial converts automatically into a paid subscription unless it is cancelled
  • High entry ticket outside meetings: $500 per month for the events Starter plan (then $100 per hour) and $300 per month for streaming
  • The language catalogue varies from page to page: 48 speaker languages on the dedicated page, other names on the pricing page, 28 on the homepage, and the 60+ claim is never reconciled in a single place
  • Performance figures are published by the vendor itself, apart from the Coval benchmark cited for TTS
  • The ISO 27001 and SOC 2 Type II certifications quoted in the privacy policy are the hosting providers', the Vanta Trust Centre is JavaScript-rendered and unreadable without an account, and no named list of subprocessors is published
  • Speech-to-text was still in early access at the time of collection with only 13 languages, there is no iOS or Android app, and emotion transfer is announced but not available
  • Narrow refund window (7 days, no more than 20% of credits consumed), unused credits forfeited on cancellation, and pay-as-you-go concurrency capped at 10 sessions per account
Pricing

Pricing & Plans

Palabra.ai offers no permanent free plan. A 7-day free trial is available (terms 6.6), and signing up on the API platform grants $50 in credits. Subscriptions for meetings and presentations start at $60.00 per month for three hours, or $45 per month on annual billing, with additional hours at $20. In-person events start at $500 per month for five hours and live streams at $300 per month for five hours. Usage-based API pricing is $0.03 per 1,000 characters for text-to-speech, $0.002 per minute of audio for speech recognition, and $0.04 per minute for speech-to-speech translation. Annual commitment is advertised as saving up to 62%. A credit system applies to the subscription tiers, with 150 credits per month on Pro, 900 on Scale and 3,500 on Business, unused credits rolling over while the subscription stays active. Event support is billed at $1,000 per 10-hour block, and offline dubbing and subtitling are quoted on request. Refunds are possible within 7 days provided less than 20% of the plan's credits have been consumed (terms 6.9); EU and UK consumers hold a 14-day statutory withdrawal right, lost once the service starts. All displayed prices are in US dollars.

Pay-as-you-go API
  • $0.03 per 1
  • 000 characters for TTS
  • $0.002 per minute for STT
  • $0.04 per minute for S2S
  • with $50 in credits at sign-up
  • up to 10 concurrent sessions and zero retention
Pro (meetings and presentations)
  • $150 per month
  • $113 per month billed annually
  • 10 hours included
  • then $15 per hour. Everything in Starter with more hours and a better hourly rate
Team (meetings and presentations)
  • $500 per month
  • $375 per month billed annually
  • 50 hours included
  • then $10 per hour. Everything in Pro plus a multi-seat workspace with roles and permissions
  • SSO and audit logs
  • a dedicated account manager and setup assistance
Business (meetings and presentations)
  • quote on request. Everything in Team plus custom development and integrations
  • SLA and priority support
  • security and procurement support
  • enterprise onboarding and regional server deployment
Starter (in-person events)
  • $500 per month
  • $375 per month billed annually
  • 5 hours
  • then $100 per hour. Unlimited listeners
  • single-stage event
  • 2 cloned voices with accent control
  • QR-code access
  • glossaries
Pro (in-person events)
  • $1
  • 600 per month
  • $1
  • 200 per month billed annually
  • 20 hours
  • then $80 per hour. Up to 3 stages per event
  • 10 cloned voices
  • speaker diarization and setup assistance
Team (in-person events)
  • $3
  • 000 per month
  • $2
  • 250 per month billed annually
  • 50 hours
  • then $60 per hour. Unlimited stages and cloned voices
  • multi-seat workspace
  • dedicated account manager and enterprise onboarding
Starter (streams and broadcast)
  • $300 per month
  • $225 per month billed annually
  • 5 hours
  • then $60 per hour. Unlimited listeners
  • 3 simultaneous output languages
  • 2 cloned voices
  • translated audio and subtitles
  • RTMP/SRT and HLS
Pro (streams and broadcast)
  • $800 per month
  • $600 per month billed annually
  • 20 hours
  • then $40 per hour. 10 simultaneous output languages
  • 10 cloned voices and speaker diarization
Team (streams and broadcast)
  • $1
  • 500 per month
  • $1
  • 125 per month billed annually
  • 50 hours
  • then $30 per hour. Unlimited output languages and cloned voices
  • multi-seat workspace and dedicated account manager
Business (events and streams)
  • quote on request
  • with custom development
  • SLA and priority support
  • and regional server deployment
Plan 13
  • All plans include sub-second latency
  • 60+ languages and no data retention
Special offers — $50 in credits granted on sign-up to the API platform · 7-day free trial on the apps, with cancellation presented as straightforward · Annual commitment advertised as saving up to 62% compared with monthly billing · An affiliate programme is referenced in the site navigation and footer, but its terms were not collected · No public promo code and no time-limited offer were displayed at the time of collection
Prices and plans listed above may evolve. Always check the official pricing page before subscribing.
Trust & Privacy

Data, GDPR & hosting

A consolidated view of how Palabra.ai handles your data.

GDPR overview

GDPR implementation is concrete and documented. Palabra.ai Ltd is registered in England and Wales under number 15047379, at 86-90 Paul Street, London EC2A 4NE, and is registered with the UK regulator (the ICO) under number ZC087866. An EU representative has been appointed under Article 27: Instant EU GDPR Representative Ltd., Adam Brogden, contact@gdprlocal.com, +35315549700, based in Lucan, Co. Dublin, Ireland. Palabra acts as controller for its site and services, and as processor for business customers, under a DPA incorporated by reference into the terms (1.2 and 4.2), applying automatically without separate signature and prevailing in case of conflict. Listed rights cover access, rectification, erasure, restriction, notification, portability, objection, withdrawal of consent and automated decision-making. Requests are handled within 30 days, extendable by two months. Transfers outside the UK/EEA rely on standard contractual clauses.

Who owns the data?

Under section 3.3 of the terms, you keep all ownership rights in your User Content, meaning both the Input you provide and the Output the service generates. Section 3.4 grants Palabra a worldwide, non-exclusive, royalty-free licence, but strictly to provide, operate, maintain and support the services and to comply with the law. The terms state explicitly that this licence transfers no ownership and permits no other use unless expressly agreed otherwise in writing. For voice recordings, it covers generating Output and delivering the service, nothing more. Palabra retains ownership of its services, models, algorithms and interfaces (5.1), and cloned voices require you to hold the rights and consents yourself (3.2).

Reuse rights

Because you keep ownership of both Input and Output, you can reuse what the service produces without asking permission: the licence granted to Palabra runs one way and adds no restriction on your side. On the vendor's side, user content is processed ephemerally by default, cached for at most one minute to enable processing, then continuously overwritten, so full conversations are never retained and cannot be recovered. Collected categories are Account Data (name, email), User Content, Voice Data, Technical Data, Usage Data and Analytics Data, under performance of the contract, legitimate interest, legal obligation or consent for marketing. Sharing is limited to contractually bound subprocessors (cloud hosting, payment, analytics, communication, security) with no use of their own, and no personal data is sold. Customer data is never used to train the models. Two exceptions to the no-storage rule exist, both switched on explicitly by the user: the Recording Feature (off by default, enabled through a checkbox with a legal notice, terms 3.8) and voice cloning, which keeps a voice sample.

Data retention & training

Retention summary
The default is simple: user content is not stored. Audio and text are processed in real time, cached for a maximum of one minute solely to enable processing, then continuously overwritten, so full conversations or sessions are never retained and cannot be recovered. Two optional features change this. The Recording Feature, off by default, keeps transcripts and recordings for as long as it stays enabled and the account remains open, or until you delete them, whichever comes first. Voice cloning keeps a voice sample used only to deliver that feature, with no biometric identification and no training. Backup and disaster-recovery copies may persist for a limited period after deletion. Deletion requests are actioned within 30 days, extendable by two months for complex or high-volume requests. Personal data is otherwise kept only as long as needed for the stated purposes and legal obligations.
Trains on customer data
No
DPA available
Yes
GDPR contact

Hosting summary

Personal data is primarily stored and processed in the United Kingdom and the European Economic Area, which the privacy policy presents as a way to ensure a high level of data protection. Processing or transfers outside the UK/EEA remain possible where third-party subprocessors support the operation, scaling or performance of the service. For those transfers, the announced safeguards are compliance verification, standard contractual clauses, technical and organisational measures, internal policies and role-based access controls. All data is encrypted in transit using standard protocols, with secure key management, and logging is limited to operational metadata stored separately from processing systems, with user content excluded from logs. Product pages carry a "GDPR - EU data processing, DPA ready" badge. The hosting providers maintain industry-standard security certifications, cited as ISO 27001 and SOC 2 Type II; these belong to the hosting providers rather than to Palabra.ai Ltd itself. Self-hosted and on-premises deployment is offered to customers with advanced security requirements, and regional servers are included in the Business tiers.

Hosting countries
🇬🇧 United Kingdom
Hosting regions
UKEEA
Availability

Where Palabra.ai works

Country-level availability.

Not available in

No country or region restriction is published. The site lists no excluded territories and sets no residency condition; the terms only require a minimum age of 18 and the legal capacity to enter into a contract. The 14 Day statutory withdrawal right granted to EU and UK consumers implies availability in those areas.
Watch-outs

Things to keep in mind

Risks and trade-offs to weigh before adopting Palabra.ai.

  • The 7-day free trial converts automatically into a paid subscription if you do not cancel before it ends (terms 6.6)
  • Unused credits are forfeited on cancellation or non-renewal (terms 6.3 and 7.3), and the refund window is narrow: 7 days, with no more than 20% of credits consumed
  • Voice cloning puts the burden on you: you must hold the rights and the explicit consent of the person whose voice is cloned (terms 3.2)
  • Turning on the Recording Feature transfers to you the responsibility for obtaining recording consent from every participant, including under local wiretapping laws (terms 3.8)
  • AI output can contain inaccuracies, errors or omissions: terms 8.2 forbid relying on it as a single source of truth or as a substitute for professional advice, which matters in legal, medical or financial settings
  • The ISO 27001 and SOC 2 Type II certifications mentioned in the privacy policy belong to the hosting providers, not to the publisher, and no named list of subprocessors is published
  • Liability is capped at the amounts paid over the six months preceding the claim, governing law is England and Wales, and the privacy contact reads support@palabra.ai while the mailto link on the same line points to legal@palabra.ai
Setup

Setup & Integrations

Technical difficulty

Low on the no-code route: nothing to install, no plugins. You paste a meeting link, pick the languages and a voice, and join the call. Event listeners simply scan a QR code and listen on their own phone. The API route needs developer skills but stays light: an API key, a transport choice (WebRTC client-side, WebSocket server-side) and an official Python or TypeScript client, with a documented four-step path; one customer quote mentions a 40-minute integration. Streaming assumes an existing RTMP/SRT chain and video production know-how. Setup assistance comes with the higher tiers, and self-hosted deployment goes through sales.

Deployment

Web appAPI

Integrations

Zoom Google Meet Microsoft Teams OBS VMix YouTube Vimeo Castr Twitch Cloudflare Vapi Retell LiveKit Pipecat Vocode

Supported languages

ArabicMoroccan ArabicBelarusianBengaliBulgarianCantoneseChinese (Simplified)Chinese (Traditional)CroatianCzechDanishDutchEnglishFilipinoFinnishFrenchGermanGreekHaitian CreoleHebrewHindiHungarianIndonesianIrishItalianJapaneseKazakhKoreanLatvianLithuanianMacedonianMalayMarathiMongolianNorwegianPersianPolishPortugueseEuropean PortuguesePunjabiRomanianRussianSerbianSlovakSlovenianSpanishMexican SpanishSwahiliSwedishTagalogTamilThaiTurkishTurkmenUkrainianUrduUzbekVietnameseWelshZulu
Company

Behind Palabra.ai

Company name
Palabra.ai Ltd
Founded
20/12/2021
Country of origin
🇬🇧 United Kingdom
Headquarters
86-90 Paul Street, London EC2A 4NE, United Kingdom
UBO
INFORMATION_NOT_FOUND
UBO country
INFORMATION_NOT_FOUND
Domain registrar country
🇺🇸 United States
Legal contact
Support contact

Fundraising

Pre-seed round of $8.4 million announced in August 2025
Lead investor: Seven Seven Six, the fund founded by Alexis Ohanian, co-founder of Reddit
Other participants: Creator Ventures, Max Mullen, Anne Lee Skates, Mehdi Ghissassi and Namat Bahram
Press coverage: TechCrunch on 14 August 2025 (linked from the homepage and the About page), plus Slator, UKTechNews, The SaaS News and Finsmes
Acquisition of Talo announced in November 2025 through a BusinessWire release linked from the site, alongside the launch of a consumer product line; Talo AI is also an API customer, quoted on the API page

Social

Official links

Resources

All the official URLs gathered for verification and reference.

Compare

Alternatives

Tools that compete with or complement Palabra.ai.

E ElevenLabsD Deepgram
FAQ

Frequently asked questions

How many languages does Palabra.ai support?
The headline claim is 60+ languages. The dedicated page lists 48 speaker languages plus 5 additional listener languages, text-to-speech covers 21 languages, and speech-to-text covers 13. The counts differ from one page to another, so it is worth confirming your specific language before committing. Missing languages can be added on commercial request.
How fast is it really?
End-to-end voice translation is announced at under one second. Text-to-speech is quoted at 35 ms time-to-first-audio (P90, network latency excluded), a figure attributed to the Coval benchmark of July 2026. Synthesis starts streaming as soon as two or three words have arrived, without waiting for a complete sentence.
Are my conversations stored?
Not by default. Content is processed in real time, cached for at most one minute to enable processing, then continuously overwritten, so full conversations or sessions are never retained and cannot be recovered. Two exceptions exist and both are switched on explicitly by the user: the Recording Feature, which keeps transcripts and recordings, and voice cloning, which keeps a voice sample.
Is my data used to train the models?
No. Terms 3.6 and the privacy policy both state that customer data is not used for training. The only stated exception concerns glossaries and terminology lists you provide voluntarily, which may feed terminology improvements.
Is there a free plan?
There is no permanent free tier. You get a 7-day free trial on the apps, and $50 in credits when you sign up on the API platform. The trial converts automatically into a paid subscription if you do not cancel before it ends, so set a reminder.
Which meeting and streaming tools does it work with?
For meetings, Zoom, Google Meet and Microsoft Teams, with no extra software to install. For streaming, OBS, vMix, YouTube, Vimeo, Castr and Twitch, through the SRT, RTMP and HLS protocols.
Is there an API?
Yes. The API supports WebRTC for client-side applications and WebSocket for server-side integrations, with official Python and TypeScript clients. Documentation lives on platform.palabra.ai/docs, and a public status page tracks availability and incidents.
Can I host it myself, and where is the data hosted?
Cloud, self-hosted and on-premises deployments are all offered, with regional servers on Business accounts, arranged through sales rather than self-service. Data is primarily stored and processed in the United Kingdom and the European Economic Area, with transfers outside that area covered by standard contractual clauses.
Can I clone a voice, and can I get a refund?
Voice cloning works from three seconds of audio, with accent control through deaccenting, but you must hold the necessary rights and consents for the voice you clone. On refunds, the window is 7 days and requires that less than 20% of the plan's credits have been consumed; EU and UK consumers also hold a 14-day statutory withdrawal right, which is lost once the service starts. The minimum age is 18.
Conclusion

Should you pick Palabra.ai?

Palabra.ai is technically mature for a company this young: proprietary models, precise latency figures, public documentation, a status page and an announced SLA. Its clearest strength is the deliberate double audience. Business teams get no-code apps for meetings, events and streams; developers get a single streaming API covering ASR, translation and TTS. Both run on the same infrastructure.

The data posture is unusually clear for this sector. No retention by default and no training on customer data are not marketing lines here: they are written into the terms and the privacy policy, and backed by a named Article 27 EU representative, a published ICO registration and a DPA that applies automatically. Self-hosted and on-premises deployment are available for organisations that need them.

Two reservations deserve attention before you commit. First, the language catalogue is not reconciled across the site: the dedicated page lists 48 speaker languages, the pricing page names others, the homepage shows 28, and the 60+ claim cannot be verified in one place. If a specific language matters to your project, ask for confirmation. Second, the ISO 27001 and SOC 2 Type II certifications on display belong to the hosting providers rather than to Palabra.ai Ltd, and the Trust Centre is not readable without an account.

Pricing is legible on the API side, where per-minute and per-character rates are published. It gets heavier on events and streaming, where the entry tier alone runs to several hundred dollars a month. There is no permanent free plan: a 7-day trial and $50 in API credits are the way in, and the trial converts automatically if you do not cancel.

A young British company, pre-seed funded in August 2025 and already growing by acquisition. Worth a trial for anyone who needs live multilingual speech and cares where the audio goes.