Every European Speech-to-Text API reviewed
Cambridge, United Kingdom
Free tier with $100 credit, no card required / Pro pay-as-you-go from $0.129 / Enterprise custom pricing with volume discounts
Demo on request
Best for: Developers who need live and batch recognition at scale and may have to run the engine offline or on a device
Speechmatics is the Cambridge speech recognition company that sells a developer API for turning audio into text. The service covers 55+ languages and accents, attaches speaker labels to each turn, and returns streaming transcripts in under a second. A Medical Model is offered for clinical vocabulary, and the vendor says it halves errors on such terms at best.
Speechmatics operates as Cantab Research Limited and is based in Cambridge, which places it under UK law and links it to the EU through an adequacy decision. Its pages list ISO 27001, GDPR, HIPAA and SOC 2 Type II. Pro pay-as-you-go starts at $0.129, volume discounts apply above 500 hours a month, and the 1.07% pooled word error rate is a benchmark result the company reports.
What Speechmatics does well
- Reports the lowest pooled error rate among twelve independently benchmarked services
- Handles 55+ languages and separates speakers in the same call
- Returns streaming results in under a second
- Runs in a hosted cloud, on-premise or on-device
- Includes a free tier with $100 of credit and no card
Where Speechmatics falls short
- Covers recognition only, with no speech generation or cloning
- States a headline Pro rate without itemising the billing unit
- Sits in the UK, so the EU link is an adequacy decision
- Targets developers who call an API, not people who want an editor
Standout feature. Speechmatics is the only API here that pairs a published independent accuracy benchmark with a choice of cloud, on-premise or on-device deployment.
Roubaix, France
Founded 2022
Starter pay-as-you-go from $0.61/hr async, $0.75/hr real-time (€50 free credits) / Growth from $0.20/hr with commitment / Enterprise custom
Demo on request
Best for: Product teams that want a hosted EU API billed by usage, with automatic language detection across 100+ languages
Gladia is a French speech-to-text API that offers asynchronous and real-time transcription with automatic language detection. Starter pricing is $0.61 per hour for async and $0.75 for real-time, with €50 of credit. Growth begins at $0.20 per hour on a volume commitment, and the Enterprise tier adds zero data retention, dedicated infrastructure and an SLA.
Gladia SAS was registered in January 2022 and sits at the same Roubaix address as OVHcloud, whose entity appears as president in the French register. Its pages list GDPR, HIPAA and SOC 2 Type II, and the vendor states it is headquartered in the EU. Gladia Inc. in Delaware also publishes terms, so check which entity signs your contract.
What Gladia does well
- Bills by usage from $0.61 per hour with no quote needed
- Detects the spoken language automatically in 100+ languages
- Includes speaker separation in the standard offer
- Gives €50 of credit that does not expire
- Belongs to OVHcloud, a European listed cloud group
Where Gladia falls short
- Offers a hosted API only, with no on-premise or device edition
- Takes its founding year from the company register, not an about page
- Entered the category later than Speechmatics
- Handles transcription only, with no speech output
Standout feature. Gladia is the only hosted speech API here with a European cloud group, OVHcloud, named as its parent in the company register.
Paris, France
API: Voxtral Mini Transcribe 3 $0.0035 per minute, Voxtral Small Transcribe 3 $0.004 per minute, Voxtral Mini Transcribe Realtime 3 $0.008 per minute (Mistral pricing page, USD or EUR toggle)
Demo on request
Best for: Developers who want a cheap per-minute API now and the option to run open weights themselves later
Voxtral is the speech model family from Mistral AI, available as a metered API and, for the original models, as open weights. The API lists Mini Transcribe 3 at $0.0035 per minute, Small Transcribe 3 at $0.004 and a Realtime model at $0.008. Documented functions include live streams, speaker labels, timestamps, custom vocabulary, language detection and audio question answering.
Mistral is a simplified joint stock company in Paris, registered under RCS 952 418 325. The original 24B and 3B models carry an Apache 2.0 licence, so a team can host them itself. The pages we read do not state where the hosted API processes audio, and the benchmark claims come from Mistral, so verify both before sending sensitive recordings.
What Voxtral does well
- Per-minute rates published, starting at $0.0035
- Open Apache 2.0 weights for the original models
- Live streams, speaker labels and custom vocabulary documented
- French company registered in Paris
- Transcription, text-to-speech and chat on one API
Where Voxtral falls short
- API data location not stated on the pages we read
- Benchmark claims come from Mistral itself
- Prices are shown in USD by default
- No contact-centre specific tooling mentioned
Standout feature. Voxtral is the only entry here that releases model weights under Apache 2.0, so a team can leave the hosted API without leaving the model.
Amsterdam, Netherlands
Prerecorded transcription in 40+ languages 'from EUR 0.01 per hour' (home page banner for Rumble-2-Flash); full rate card not read
Demo on request
Best for: Teams transcribing Dutch, German, French and other European audio with heavy domain terms such as drug names or case numbers
Reson8 is an Amsterdam speech API built around vocabulary adaptation: you upload a plain list of terms such as drug names or case numbers and the recogniser favours them, without a fine-tuning project. Prerecorded transcription covers 40+ languages with a European focus that includes Dutch, German, French and Frisian. A banner on the home page quotes prerecorded audio from €0.01 per hour.
Resonate Labs B.V. operates Reson8 under Chamber of Commerce number 98891340. The company states that all GPU processing happens in the EU and that no audio is retained. The full rate card could not be read as text, and no hosting provider, certification or self-hosted edition appears on the pages we read. Terms and privacy policy were updated in October 2026.
What Reson8 does well
- Dutch B.V. with chamber of commerce number published
- Custom vocabulary from a text list, no fine-tuning
- EU GPU processing and zero audio retention stated
- Banner price from EUR 0.01 per hour for prerecorded audio
- European language focus including Dutch and Frisian
Where Reson8 falls short
- Full pricing page not readable as text
- Hosting provider and sub-processors not named on the pages we read
- No self-hosted option mentioned
- No text-to-speech or voice agent product mentioned
Standout feature. Reson8 is the only API here that takes a pasted word list as recognition vocabulary and states zero audio retention at the same time.
Paris, France
Free: EUR 0 a month with 5 hours of audio; Pay-as-you-go: EUR 0.006 per minute billed per second; Enterprise: custom
Demo on request
Best for: French and European teams that want one EU-hosted API for transcription, translation and synthesis
Voxist is a Paris API company that sells streaming and batch recognition in 40+ languages, translation, neural text-to-speech and a voice-to-voice endpoint. Calls go over REST or gRPC, and SDKs exist for Python, Node.js, Go, Rust, Java and .NET. The free plan gives 5 hours of audio a month, and pay-as-you-go costs €0.006 per minute, billed per second.
Voxist SAS states that it hosts in EU data centres operated by OVHcloud and Scaleway. The Enterprise plan adds on-premise and air-gapped options. SOC 2 Type II and ISO 27001 are listed as in progress, accuracy figures for French are the vendor's own, and the homepage names English, French, German, Spanish and Italian, so Dutch or Nordic audio needs a test. The free plan allows 5 requests per second.
What Voxist does well
- Euro pricing with a free tier of 5 hours a month
- EU hosting on OVHcloud and Scaleway
- ASR, translation and TTS in one API
- On-premise and air-gapped option on Enterprise
- Six SDKs and an OpenAPI spec
Where Voxist falls short
- SOC 2 and ISO 27001 still in progress
- Accuracy figures are the vendor's own
- Smaller language list on the homepage than some rivals
- Free plan rate-limited to 5 requests a second
Standout feature. Voxist is the only entry here that bills recognition, translation and speech output in euros from stated EU data centres.
Mannheim, Germany
Pay per use: Standard $0.20 per hour, Premium (Topaz) $0.30 per hour, Premium (Jade) $0.32 per hour; Pro $22.00 per month; 3 free transcriptions
Demo on request
Best for: German-speaking teams that want hosted transcription in Germany with a simple per-hour price and an API
DeepScript is a transcription service from Mannheim that offers a web interface and an API. It lists 99+ languages with DACH dialect recognition, diarization, custom vocabulary and export to TXT, SRT, VTT and JSON. Developers get a REST API, webhooks and an MCP endpoint, which lets AI assistants submit and fetch transcripts.
Aliru GmbH runs the brand and is registered at Amtsgericht Mannheim as HRB 739079. Audio is processed on Hetzner servers in Nuremberg. Standard costs $0.20 per hour, Premium Topaz $0.30 and Premium Jade $0.32, Pro is $22.00 a month, and three transcriptions are free. No real-time streaming appears in the feature list, and transcripts vanish after 30 days without Pro.
What DeepScript does well
- German GmbH with commercial register number published
- Servers at Hetzner in Germany, no US cloud
- Simple per-hour pricing from $0.20
- REST API, webhooks and MCP endpoint
- DACH dialect recognition listed
Where DeepScript falls short
- No real-time streaming in the feature list we read
- Prices shown in USD by default
- Transcripts deleted after 30 days without Pro
- No self-hosted edition mentioned
Standout feature. DeepScript is the only vendor here that names both a German city and a German data centre operator for its servers.
Vădăstrița, Olt, Romania
Free 10 minutes; Basic €10/month (10 h), Standard €27/month (30 h), Pro €89/month (100 h); extra hours €5; API from €0.22/hour
Demo on request
Best for: Small companies that want plan and API prices in euros and a route to on-premise deployment later
Vatis Tech is a Romanian transcription vendor that pairs an editor-based web app with an asynchronous speech API. The API costs €0.22 per hour with diarization and word timestamps, and audio intelligence is billed per token on top. Monthly plans run from Basic at €10 for 10 hours to Pro at €89 for 100 hours, with extra hours at €5.
VATIS TECH S.R.L. is entered in the Romanian Trade Registry under J28/696/2021 and states that GDPR and ISO 27001 apply, while SOC 2 Type II is still in progress. The company offers on-premise installation or an isolated cloud for regulated audio. The accuracy comparisons on its site are its own, and live Romanian recognition is announced rather than released.
What Vatis Tech does well
- Lists every plan and API rate in euros
- Offers on-premise or isolated-cloud deployment
- Covers 50+ languages in the app
- States ISO 27001 certification
- Combines a transcript editor with an API
Where Vatis Tech falls short
- Has not yet completed SOC 2 Type II
- Does not say where hosted data is stored
- Backs accuracy claims with its own benchmarks only
- Still lists real-time Romanian as upcoming
Standout feature. Vatis Tech is the only vendor here that sells both a euro-priced subscription editor and a per-hour API rate on one price list.
Copenhagen, Denmark
Speech to Text $0.0065 per minute; free credits across capabilities; enterprise plan on request
Demo on request
Best for: Healthcare software vendors that need clinical transcription and note generation from one Danish supplier
Corti is a Copenhagen company whose speech API is built for clinical conversations, in real time and in batch. Speech to Text costs $0.0065 per minute, and the same platform adds a medical coding API, text generation and an agentic framework. The customer chooses whether data is hosted in the EU or the US, and free credits cover each capability.
Corti ApS carries VAT number 35669825. Its privacy policy names two US affiliates, Corti Inc. and Corti America Inc., and the pages we read do not state who owns them. The speech product is clinical, so general audio and non-medical languages are undocumented, prices appear in dollars, and the API master terms were not read for this review.
What Corti does well
- Danish ApS with VAT number published
- Real-time and batch clinical transcription
- Per-minute price of $0.0065
- EU or US hosting chosen by the customer
- Medical coding and text generation on the same platform
Where Corti falls short
- US affiliates named in the privacy policy
- Clinical focus, less suited to general audio
- Prices in USD
- API master terms not read
Standout feature. Corti is the only vendor here that puts medical coding and note generation on the same API as transcription.
Brno, Czech Republic
Pricing on request
Demo on request
Best for: Security agencies, courts and call centres that must process recordings inside their own network
Phonexia is a Brno company whose speech platform combines transcription with speaker identification, voice biometrics and deepfake detection. The vendor describes 14 speech technologies behind one API and says the software is designed for on-premise and private-cloud installation. It names national security, law enforcement and call centres among its operators.
Phonexia a.s. is registered in the Czech Republic and its privacy policy is governed by Czech law. The company advertises 87+ languages, which is its own figure. Prices are not published and an online demo replaces self-service sign-up, so evaluation starts with a sales conversation rather than an API key.
What Phonexia does well
- Deploys on-premise or in a private cloud
- Operates as a Czech company under Czech law
- Adds voice biometrics to speech recognition
- Claims 87+ languages, a vendor figure
- Targets recordings that cannot leave an organisation
Where Phonexia falls short
- Publishes no prices
- Sells to organisations, not individual users
- Advertises no editor or subtitle export
- Gives its language count without independent checks
Standout feature. Phonexia is the only entry here that treats voice biometrics and deepfake detection as part of the same on-premise platform as transcription.
The Hague, Netherlands
Bronze EUR 119 a month, Silver EUR 479 a month, Gold EUR 1,199 a month (annual plans paid upfront); on-premise from EUR 1,999 per instance
Demo on request
Best for: Public bodies, courts and broadcasters that need transcription and subtitles with EU or air-gapped deployment
Scriptix is a Hague transcription platform that combines batch and realtime recognition with transcript and subtitle editors, machine translation in 100+ languages and meeting minutes. Custom model fine-tuning and an API are included. Annual plans are Bronze at €119 a month, Silver at €479 and Gold at €1,199, paid upfront, with on-premise installation from €1,999 per instance.
Scriptix B.V. is registered at the Chamber of Commerce under 75668475 and runs on EU infrastructure by default, or inside your own tenant when installed on-premise. The model is a subscription with fair-use allowances rather than metering per minute, the language count is 50, and the company is a team of four. Courts, public bodies and broadcasters are its stated audience.
What Scriptix does well
- Dutch B.V. with chamber of commerce number published
- Public monthly plans in euros
- Air-gapped on-premise editions with starting prices
- Editors, translation and meeting minutes included
- EU data residency by default
Where Scriptix falls short
- Subscription model, not metered per minute
- Allowances described only as fair use
- 50 languages
- Small team of four
Standout feature. Scriptix is the only vendor here that publishes a starting price for air-gapped on-premise transcription.
Orsay, France
On the order of €0.01 per minute for generic systems in large quantities (vendor statement); no pay-as-you-go plan
Demo on request
Best for: Broadcasters, archives and developers who batch-process large audio collections
Vocapia is a French research company whose VoxSigma suite handles speech-to-text, language identification, speech-text alignment, translation and summary generation. The web service is a REST API over HTTPS that accepts files or streams and returns XML or JSON with diarization, punctuation and confidence scores. It lists 28 languages, among them Dutch, French, German, Italian, Spanish and Polish.
Vocapia Research SAS is registered in Evry (number 432.162.063) with its seat at Parc Orsay Université. The vendor says generic systems cost on the order of €0.01 per minute in large quantities, billed on speech duration. It no longer offers pay-as-you-go and points small or irregular users to a partner service, and hosting regions are not described.
What Vocapia does well
- Registers the company in Evry, France
- Ships as standalone software or as a web service
- Provides diarization and speech-text alignment
- Processes whole archives on request
- States a low per-minute price for volume
Where Vocapia falls short
- Offers no pay-as-you-go plan
- Quotes only an order-of-magnitude price
- Leaves the hosting region undescribed
- Provides little editing tooling for non-developers
Standout feature. Vocapia is the only vendor here that offers its recognition engine as standalone software while also stating a bulk price near €0.01 per minute.
Neuilly-sur-Seine, France
Developer EUR 19 a month with EUR 19 usage credit; Starter EUR 99 a month with EUR 99 credit; Enterprise custom; 30-day trial
Demo on request
Best for: Teams that already have a transcription engine and need reliable speaker labels in meetings and calls
pyannoteAI is a French specialist in speaker diarization, the task of working out who spoke when. The API offers streaming and batch diarization, voice activity detection, speaker identification and separation, and speaker-attributed transcripts. Developer costs €19 a month with €19 of usage credit, Starter €99 with €99 of credit, and a 30-day trial is available.
pyannote SAS is registered at Nanterre under 924 837 636 and works from Neuilly-sur-Seine. It can be used as open source, through the API or on your own infrastructure. It does not replace a general speech recogniser, and the engine behind its transcripts, the data location and the open source licence are not stated on the pages we read. Starter includes three users.
What pyannoteAI does well
- French SAS with register number published
- Streaming and batch diarization
- Plans in euros from EUR 19 a month
- Open source, API or own infrastructure
- 30-day trial without a card
Where pyannoteAI falls short
- Not a standalone speech recogniser
- Underlying speech engine not named
- Data location not stated
- Starter plan includes only three users
Standout feature. pyannoteAI is the only entry here devoted entirely to speaker labels, offered as open source, API or self-hosted software.
