Speechmatics logo

Speechmatics

Speechmatics is a speech-to-text and voice AI API platform covering 55+ languages, with pre-recorded batch transcription, sub-second real-time streaming, dedicated medical and voice-agent models, and text to speech. It targets regulated industries and deploys in the cloud, on-prem or on-device. Usage is metered per hour of audio, and new accounts start with $100 in free credit.

Visit Website

What is Speechmatics?

Speechmatics is a UK speech recognition company that sells speech-to-text and voice AI as an API rather than as a consumer app. Its core product converts audio and video into text in two modes: pre-recorded transcription for files and archives, and real-time streaming for live audio, both reached through the same account, API key and credit balance. Coverage spans 55+ languages, and the newer Melia 1 model handles 56 languages without asking you to choose one up front, switching language mid-sentence when speakers code-switch. Real-time latency sits under one second, and the vendor cites a 1.07% pooled word error rate in Pipecat's August 2026 benchmark of twelve services. Clients ship as Python, JavaScript and .NET SDKs over REST and WebSocket, with documented integrations for LiveKit, Pipecat, Vapi and Zapier.

Pricing is split by model and by mode. On the pre-recorded side Melia 1 costs $0.12 per hour, Enhanced $0.38 per hour, the medical Oak 1 $0.15 per hour, Linden 1 $0.15 per hour and Standard $0.23 per hour. Real-time carries a premium: Melia 1 $0.20, Enhanced $0.40, Oak 1 $0.23, Linden 1 $0.15 and Standard $0.23 per hour. Enhanced is the highest-accuracy single-language model and supports a custom dictionary plus advanced punctuation; Oak 1 is tuned for clinical vocabulary such as drug names, dosages, abbreviations and procedure terms; Linden 1 is the voice-agent model with built-in turn detection, voice activity detection and segment-level timings; Standard reuses Enhanced features at a lower rate with lower accuracy.

Feature coverage includes speaker diarization, word-level timings, smart formatting, partial transcripts and language hints and labelling, with availability varying by model. Optional add-ons are billed per hour on top of transcription: translation at $0.49, chapters at $0.30, topics at $0.15, summaries and sentiment at $0.09 each, while PII redaction, entity detection and audio alignment are quoted on request.

Billing runs on credits, where one credit equals one dollar before discounts, consumed to the second at the hourly rate. Buyers choose pay as you go, a credit pack from $25, or a monthly or annual subscription, with volume discounts up to 25% off every credit. Opting into the model-training program takes a further 33% off speech-to-text rates in exchange for permission to use audio and transcripts for model improvement. New accounts receive $100 in credit without a card.

Compliance is a central selling point: SOC 2 Type II, ISO/IEC 27001:2022, GDPR and HIPAA, with data processing in the EU, US or Australia, no logging for training by default, real-time audio never stored, and pre-recorded data auto-deleted after seven days. Enterprise adds on-prem containers, virtual appliances, hybrid and on-device deployment, unlimited files per second and elastic concurrency; self-serve is capped at 10 files per second and 50 concurrent real-time sessions, or 2 on the free tier. Text to speech is sold separately at $0.011 per 1,000 characters with the first million characters free.

Speechmatics AI Speech-to-Text product interface screenshot

Speechmatics Core Features

✦

Real-time streaming

Sub-second WebSocket transcription for live audio, used for voice agents, live captioning and call analytics, with partial results and word-level timings.

✦

Pre-recorded batch

Asynchronous transcription of audio and video files at up to 10 files per second on self-serve, with diarization and smart formatting in the output.

✦

Multilingual Melia 1

A single model covering 56 languages that does not require a language choice up front and handles mid-sentence code-switching.

✦

Medical model Oak 1

Clinical vocabulary for drug names, dosages, abbreviations and procedure terms, with full multilingual support for consultations and dictation.

✦

Voice agent model Linden 1

Turn detection and voice activity detection built into the recogniser, priced at $0.15 per hour in both real-time and pre-recorded modes.

✦

Compliance and residency

SOC 2 Type II, ISO/IEC 27001:2022, GDPR and HIPAA, with EU, US or Australian data processing and no logging for model training by default.

✦

Flexible deployment

SaaS, multi-region cloud, on-prem containers, virtual appliances, hybrid and on-device inference on Enterprise plans.

✦

Text to speech

Low-latency speech synthesis at $0.011 per 1,000 characters, with the first million characters free for voice assistants, IVR and accessibility.

Who is Speechmatics for?

Speechmatics is aimed at teams that treat transcription as infrastructure rather than a desktop utility. The primary buyer is a developer or platform engineer embedding speech recognition into a product: a voice agent that needs sub-second turns, a meeting recorder that must label who spoke, a media pipeline captioning live broadcasts, or a contact center analysing thousands of calls a day. These teams care about API ergonomics, WebSocket streaming, SDKs in Python, JavaScript and .NET, and predictable per-hour economics, because they are shipping features on top of the transcript rather than reading it themselves. The second group is regulated organisations that cannot send audio to a generic consumer service. Healthcare providers building ambient scribes and clinical dictation need HIPAA alignment and a model that recognises drug names, dosages and procedure terms. Hospitals and health systems working across languages need mixed-language consultations transcribed without picking a language first. Legal teams and court reporters need accuracy across accents and speaker separation, sometimes on a machine with no internet connection. Financial services and public sector buyers ask about data residency in the EU, US or Australia, and about on-prem containers or virtual appliances that keep audio inside their own network. The third group is product and operations leaders at contact centers, broadcasters and media monitoring firms who buy transcription at volume and negotiate on rate. They compare pooled word error rate benchmarks, concurrency ceilings and volume discounts, and they want a vendor that will sign an MSA rather than sell them a monthly seat. Speechmatics is a weaker fit for individuals who want a free unlimited transcription tool for occasional personal use, for teams that only need English and are happy with a bundled cloud default, and for buyers who want a fully managed note-taking application with a UI, since Speechmatics sells the engine and the API, not the end-user app.

Speechmatics Use Cases

Power AI voice agents with sub-second speech recognition, turn detection and speaker-aware transcripts across 55+ languages.

Generate live captions for broadcasts, sports and news with accuracy that holds up under accents and background noise.

Transcribe clinical consultations and dictation with a medical model tuned for drug names, dosages and procedure terms.

Analyse contact center calls at scale to measure sentiment, reduce wait times and coach agents on real conversations.

Produce court reporting and legal transcripts in real time, including on-device deployments where audio cannot leave the room.

Add automated meeting notes to a collaboration platform with diarization, custom vocabulary and multilingual support.

Caption and index media archives in bulk with pre-recorded transcription, translation and chapter generation add-ons.

Build accessible products with read-along captions and synthesized speech for screen readers and assistive technology.

Speechmatics Pros and Cons

Pros

  • Transparent metered pricing with published per-hour rates for every model and mode, plus a $100 credit that requires no card.
  • Strong regulated-industry posture: SOC 2 Type II, ISO 27001, GDPR and HIPAA, regional data processing and optional on-prem or on-device deployment.
  • Genuine multilingual handling, including a model that switches language mid-sentence instead of forcing an upfront choice.
  • Low latency streaming suitable for production voice agents, backed by integrations with LiveKit, Pipecat, Vapi and Zapier.
  • No data logging for training by default, with real-time audio never stored and pre-recorded files deleted after seven days.

Cons

  • Real-time rates are roughly double pre-recorded rates for the same model, which raises costs for always-on streaming workloads.
  • Optional analytics such as translation, chapters, topics, summaries and sentiment are billed per hour on top of transcription.
  • The free tier is a $100 credit rather than a permanent allowance, and it caps real-time concurrency at two sessions.

FAQ About Speechmatics

Speechmatics Pricing

FreemiumFrom USD 0.00

Speechmatics uses usage-based, metered pricing in credits where 1 credit equals $1 before discounts, consumed to the second at a published hourly rate. Pre-recorded transcription costs $0.12/hr for Melia 1 (multilingual), $0.38/hr for Enhanced, $0.15/hr for Oak 1 (medical) and Linden 1 (voice agents), and $0.23/hr for Standard. Real-time is priced higher: Melia 1 $0.20/hr, Enhanced $0.40/hr, Oak 1 $0.23/hr, Linden 1 $0.15/hr (discounted from $0.20) and Standard $0.23/hr. Add-ons are billed per hour on top: translation $0.49, chapters $0.30, topics $0.15, summaries $0.09, sentiment $0.09, with PII redaction, entity detection and audio alignment on request. New accounts get $100 in credit with no card, and the free tier allows 2 concurrent real-time sessions. Paid self-serve offers pay as you go, credit packs from $25, or monthly and annual subscriptions, with volume discounts up to 25% off every credit and an optional 33% model-training discount on speech-to-text. Text to speech costs $0.011 per 1,000 characters with the first million characters free. Enterprise pricing is custom and adds on-prem, hybrid and on-device deployment, unlimited files per second and SSO at $280/mo.

Check official pricing

Free

$0one-time $100 credit

$100 in credit with no card required to test transcription quality and languages; 2 concurrent real-time sessions; community support; API access pauses when credit runs out.

Pay as you go (Pro)

$0.12per hour of pre-recorded audio (Melia 1)

Standard per-hour rates with no commitment, billed to a card. Melia 1 $0.12/hr, Oak 1 and Linden 1 $0.15/hr, Standard $0.23/hr, Enhanced $0.38/hr pre-recorded; real-time from $0.15-$0.40/hr.

Credit packs

from $25per pack

One-off credit purchase discounted by pack size, no recurring commitment. Pack credits last 12 months and are spent down to the second at the hourly rate.

Subscription

Customper month or year

Monthly, annual or committed subscription for the best per-credit price, with volume discounts up to 25% off every credit and overages billed at the discounted rate.

Enterprise

Customper contract

Custom pricing with unlimited files per second, elastic real-time concurrency, on-prem containers, virtual appliances, hybrid and on-device deployment, SSO at $280/mo and dedicated support.

Text to Speech

$0.011per 1,000 characters

Low-latency speech synthesis for voice AI, translation, accessibility and content creation, with the first 1 million characters free (about 20 hours of speech).

Speechmatics Alternatives