Transcribe and translate · 96 languages

Sovereign speech AI in 96 languages.

Transcribe and translate audio. Use our cloud, or run it inside your own country.

Start with 1 free hourSovereign deployment
The listen-along view of a 3 minute 41 second ENT clinic recording, playing at 0:49. The waveform is two-thirds grey with the played part in green, and the transcript line starting at 0:46, in Kiswahili, is highlighted.

Transcribing

  • 00:00 Nice to see you today.
  • 00:23 Sawa, pole sana.
  • 00:46 Haya yanawasha tu hayaumi, hayaumi hakuna kitu.
  • 01:09 Mara nyingi ni dalili ya kwamba pia una allergy.

Powering ADB's own products

Use cases

Hear what it does for your team.

Pick an industry and watch a recording become text, then data.

Telecom and broadcast regulators

Media monitoring for broadcast regulators, in the languages stations actually air.

  • Search all stations by word, name or topic
  • Flag hate speech and misinformation for review
  • Evidence with timestamps and audio
TranscriptionEntitiesSentimentTranslation
How it works for telecom and broadcast regulators

Transcribing· Radio bulletin · Kiswahili

00:15
  • 00:04 Habari za asubuhi, wasikilizaji wa Redio Jamii.
  • 00:09 Leo tunazungumzia bei ya mafuta jijini Mombasa.
  • 00:15 Mgeni wetu ni Waziri wa Nishati, live from the studio.
Example outputSearchable · flagged for review

Capabilities

One recording, everything in it.

Transcribe once. Translate it, summarise it, find the names in it, mask the numbers. Ask for all of it in the same call.

  1. Input

    Audio

    Any recording, any length

  2. Result

    Transcript

    00:46Haya yanawasha tu hayaumi.

    Timestamped, line by line

  3. Add-on

    Translation

    ENThey just itch, they don't hurt.

    Any language pair · $0.001/min

  4. Add-on

    Insights

    Ear itching, no painSentiment · neutral

    Summary, entities, sentiment, redaction, speakers

See every feature

Integrations

Works with what you already use.

Recordings arrive from the tools your team already uses, and results go back to them. AI assistants can use it too.

Meetings

  • ZoomCloud recordings, as they finish
  • Google MeetThe Meet Recordings folder
  • Microsoft TeamsRecordings from OneDrive
  • Meeting botJoins any call and records it

Files

  • Google DriveWatches a folder
  • DropboxWatches a folder
  • OneDriveWatches a folder
  • Amazon S3Reads a bucket, writes results back

Calls and results

  • TwilioCall recordings, as they finish
  • SlackPosts each summary to a channel
  • EmailSends each transcript to a list
  • WebhooksSigned events to your own app

AI assistants

An MCP server: search, transcribe, translate and summarise from your assistant, with an API key.

  • Claude
  • Cursor
  • Any MCP client

Inside the platform

Find what was said, fast.

Real recordings, real console.

Click any line to hear it. Slow down, loop, copy. The cost of the job is shown beside it.

A finished job: 3:41 of Kiswahili audio billed as 3:42 for $0.0185 at $0.005 per minute, above the waveform and the first transcript lines.

The whole console also works on a phone.

Languages

Your language is probably here.

From Kiswahili, Amharic, Hausa, Yoruba and Zulu to the world's major languages.

  • HabariKiswahili
  • HelloEnglish
  • BonjourFrench
  • SannuHausa
  • HolaSpanish
  • ሰላምAmharic
  • OláPortuguese
  • SawubonaZulu
  • مرحباArabic
  • नमस्तेHindi
  • HalloGerman
  • 你好Chinese
  • CiaoItalian
  • MerhabaTurkish
Browse all 96 languages

Africa

18
  • Kiswahili
  • Afrikaans
  • Amharic
  • Chichewa
  • Fulah
  • Hausa
  • Igbo
  • Lingala
  • Luganda
  • Luo
  • Northern Sotho
  • Oromo
  • Shona
  • Somali
  • Wolof
  • Xhosa
  • Yoruba
  • Zulu

Rest of the world

78
  • English
  • Arabic
  • Armenian
  • Assamese
  • Azerbaijani
  • Belarusian
  • Bengali
  • Bosnian
  • Bulgarian
  • Burmese
  • Catalan
  • Cebuano
  • Chinese
  • Croatian
  • Czech
  • Danish
  • Dutch
  • Estonian
  • Filipino
  • Finnish
  • French
  • Galician
  • Georgian
  • German
  • Greek
  • Gujarati
  • Hebrew
  • Hindi
  • Hungarian
  • Icelandic
  • Indonesian
  • Irish
  • Italian
  • Japanese
  • Javanese
  • Kannada
  • Kazakh
  • Khmer
  • Korean
  • Kyrgyz
  • Lao
  • Latvian
  • Lithuanian
  • Luxembourgish
  • Macedonian
  • Malay
  • Malayalam
  • Maltese
  • Maori
  • Marathi
  • Mongolian
  • Nepali
  • Norwegian
  • Occitan
  • Odia
  • Pashto
  • Persian
  • Polish
  • Portuguese
  • Punjabi
  • Romanian
  • Russian
  • Serbian
  • Sindhi
  • Slovak
  • Slovenian
  • Spanish
  • Swedish
  • Tajik
  • Tamil
  • Telugu
  • Thai
  • Turkish
  • Ukrainian
  • Urdu
  • Uzbek
  • Vietnamese
  • Welsh

Flagship: tuned on our own recordings, including Kiswahili–English in the same sentenceTranscribe in any of them, translate into any of them.

Sovereign

Your audio. Your country. Your keys.

License ADB Voices to run in your data centre or national cloud. Recordings never leave the country.

Annual licence

In-country licence

Your servers or national cloud, behind your identity provider and keys.

Annual licence

Air-gapped

For classified and regulated work. No internet needed.

Live today

Managed cloud

We run it. Pay per minute. Start today.

  • Same API and console
  • Your identity provider
  • Your encryption keys
  • Every access audited
How sovereign deployment works Talk to us about a licence

Security and compliance

  • ISO/IEC 27001

    Certified information security management.

  • GDPR

    Compliant processing, with data-processing agreements.

  • Encrypted

    In transit and at rest, with your keys on a licence.

  • Audited

    Every access and deletion is logged.

Why ADB Voices

Built for where your audio lives.

FeatureADB VoicesGlobal speech APIs
Runs inside your countryYes, on your serversUsually US or EU cloud
Works fully offlineYes, air-gappedRarely
Licensed to regulators and governmentsYesRarely
Tuned on Kiswahili clinic and broadcast audioYesGeneral-purpose models
Two languages in one sentenceYesVaries
African languages18Varies
Billed by the second, failed jobs freeYesVaries

Pricing

Simple, usage-based pricing.

No subscription. Start with 1 free hour. Pay less as you grow.

Pay as you go

For teams, researchers and developers.

Starts at

$0.30/hour

Start with 1 free hour

  • 96 languages
  • Billed by the second, failed jobs free
  • Rate drops as you grow
  • Minutes never expire
Create free account

Sovereign

For regulators, governments and enterprises.

Annual licence

Custom

Priced per deployment

  • In-country or air-gapped
  • Your identity provider and keys
  • Uptime SLA and dedicated support
  • Custom vocabulary and fine-tuning
Talk to us

  • Volume discounts

    Down to $0.24/h

    Automatic as your monthly audio grows

  • Top-up bonus

    Up to +20%

    Free minutes on bigger top-ups

  • Add-ons

    From $0.018/h

    Translation, summary, entities and more

See the full price list and estimate your cost

Estimate your monthly cost

$28.80/month

100 h at $0.288/h

Volume discounts

Automatic. Your rate follows the audio you transcribe each month.

Monthly volumeRate
0 – 100 h$0.30/h
100 – 500 h$0.288/h
500 – 2,000 h$0.276/h
2,000 – 5,000 h$0.264/h
5,000 – 10,000 h$0.252/h
10,000+ h$0.24/h

Top-up bonuses

Bigger top-ups earn free minutes.

Top upBonusYou get
$100+10%367 h
$500+15%1,917 h
$1,000+20%4,000 h

Add-ons

Billed only when you turn them on, per hour of audio.

  • Translation (96 languages)$0.06/h
  • Summary$0.03/h
  • Sentiment$0.018/h
  • Named entities$0.06/h
  • PII redaction$0.06/h

Audio storage

You choose how long audio is kept. Transcripts are not charged.

First 30 days includedthen $0.025 per GB a month

How it works

One call to a transcript.

Send the recording with what you want back. See the price first, and the bill after.

  1. 1.Send it

    A file, a link or an upload, with the language and any translations or features.

    POST /v1/transcriptions

  2. 2.See the price

    Add dry_run to get the cost of every item before anything runs.

    ?dry_run=true

  3. 3.Get it back

    A signed webhook, or poll. The transcript, translations and results together.

    GET /v1/transcriptions/{id}

  4. +The bill

    Every job carries its cost, line by line. Failed jobs are free.

    "cost": { "total": … }

FAQ

Straight answers

What does sovereign mean here?

You can license ADB Voices to run in your own data centre or national cloud, with your keys, so recordings never leave the country.

Which languages does it handle?

96. Kiswahili and English are tuned on our own recordings, including both in one sentence. For other languages, try a sample of your audio first.

Are timestamps word-level? Are speakers labelled?

Lines are timestamped every few seconds; word-level timestamps are not available yet. Speaker labels are in beta: turn them on for a job and each line says Speaker A, B and so on.

Is it safe for clinical use?

It produces drafts for a person to review, not unattended clinical decisions.

How does translation work?

Pick a language on any finished transcript. Each line keeps its timestamp, so subtitles work too.

How is a job priced?

By audio length, per second. Failed jobs are free. Pay by card or invoice.

Where does our audio go, and for how long?

Encrypted storage, for as long as you choose. The first 30 days of each file are included; keeping audio longer costs $0.025 per GB a month. Licensed deployments keep it on your premises.

How do we get access?

Create an account at /signup. Your workspace starts with 1 free hour, and paid minutes never expire.

Start today

Hear your own audio turn into text.

1 free hour, no card. Minutes you buy never expire.

Create your free accountTalk to us about a licence