Speechmatics

Speechmatics is a speech recognition API that turns live and recorded audio into text for developers, enterprises and voice AI builders. It is one of the older independent players in automatic speech recognition, and its pitch rests on accuracy with accents, noise and several speakers, plus a choice of where the models run. Few rivals offer the same API in the cloud, inside your own infrastructure and on a device.

What is Speechmatics used for?

Teams use Speechmatics as the transcription layer inside their own products rather than as an app for end users. You send audio over a REST call for files or a WebSocket for streams, and get back timestamped text with speaker labels.

• Live captioning for broadcast, events and media workflows
• Ambient scribing and dictation in healthcare through its medical model
• The listening half of a phone or web voice agent, with turn detection on the agent model
• Contact center analytics with summaries, topics and sentiment add-ons
• Court reporting and legal transcription where accuracy on names and terms matters
• Embedded transcription inside desktop software through on-device SDKs

Who Speechmatics is best for

Speechmatics suits engineering teams at media companies, healthcare software vendors, contact center platforms and meeting tool makers, especially when audio is multilingual or messy. It is also a fit for regulated organizations that cannot send recordings to a public cloud, since containers and on-premises installs are offered. If you only need to transcribe a few interviews a month, a finished app is a better use of your time than an API.

Speechmatics pricing

As of October 2026, new accounts get $100 in free credits without a card, with limits of two concurrent real-time sessions. After that, pay-as-you-go rates depend on the model: the multilingual Melia 1 model lists at $0.24 per hour for recorded audio and $0.40 per hour in real time, while the Enhanced model costs $0.75 per hour for files. Monthly subscriptions start at $10, annual plans at $120 per year, and both add small discounts, as do prepaid credit packs. Opting in to model training cuts speech-to-text rates by a third. Add-ons such as translation and summaries are billed per hour on top. Enterprise contracts add on-premises deployment, unlimited throughput, a dedicated success team and invoicing. Check the pricing page for current rates.

Limitations to keep in mind

• Pricing mixes credits, models and add-ons, so estimating a monthly bill takes some arithmetic
• PII redaction and audio alignment are listed as enterprise features
• It is an API, so non-developers will need another product built on top of it
• Single sign-on is a paid add-on rather than part of the base enterprise package

Speechmatics vs alternatives

Deepgram is the closest rival for streaming use and is popular with voice agent builders who want a single vendor for speech-to-text and text-to-speech. AssemblyAI is often chosen for its audio intelligence features and developer documentation. OpenAI Whisper is free to self-host and a good choice when you have GPUs and engineers to run it. Rev pairs an API with human transcription when you need certified accuracy. Pick Speechmatics when languages are mixed, when you need an on-premises or on-device install, or when you want a medical model from the same vendor.

Tips for getting started with Speechmatics

Spend the free credit on your own worst recordings, such as noisy calls or accented speakers, rather than clean samples. Compare the multilingual and Enhanced models on the same files before committing, since the price gap is large. Add domain terms through custom vocabulary early, and only switch on add-ons you will actually use, because each is billed per hour of audio.

Speechmatics FAQ

Is Speechmatics free?

Speechmatics gives new accounts $100 in free credits with no card required. After the credit runs out, usage is billed per hour of audio or through a subscription.

How many languages does Speechmatics support?

Speechmatics supports more than 55 languages for speech-to-text, and its multilingual model can follow speakers who switch languages mid-sentence.

Can Speechmatics run on-premises?

Yes. Enterprise customers can deploy Speechmatics in containers, on a virtual appliance, on-device or in a hybrid setup inside their own infrastructure.

Speechmatics vs Deepgram: which is better?

Deepgram is a strong pick for low-latency voice agents in the cloud. Speechmatics is the better fit when you need mixed-language accuracy or a self-hosted deployment.