Speechmatics
UK speech-to-text API benchmarked to the lowest error rate among 12 rivals, deployable on-premise
Quick Overview
| Company | Cantab Research Limited, trading as Speechmatics |
|---|---|
| Category | AI Voice & Speech |
| Headquarters | Cambridge, United Kingdom |
| EU Presence | United Kingdom (adequacy decision, outside the EEA) |
| Data Location | Multi-region cloud, on-premise and on-device deployment available |
| Open Source | No |
| Compliance | ISO 27001, GDPR, HIPAA, SOC 2 Type II |
| Pricing | Free tier with $100 credit, no card required / Pro pay-as-you-go from $0.129 / Enterprise custom pricing with volume discounts |
| Free Option | Free tier, $100 credit, no card required |
| Replaces | Amazon Transcribe, Google Speech-to-Text, Deepgram |
Detailed Review
Every other tool in this category turns text or a voice into speech; Speechmatics runs the tape the other way, turning speech into text.
It grew out of research applying neural networks to speech recognition at Cambridge University, and it competes on accuracy rather than voice quality, reporting a 1.07% pooled word error rate — the lowest of twelve services benchmarked independently. That distinction matters for anyone building a product on top of transcription rather than reading text aloud: a voice agent, a captioning pipeline, a call-centre analytics tool, where every recognition mistake compounds downstream.
The API covers 55+ languages and accents with built-in speaker diarization, separating individual speakers in meetings, calls and legal proceedings, and returns real-time results in under a second.
A specialised Medical Model cuts errors on clinical terminology by up to half, and deployment is not fixed to Speechmatics' own cloud: customers can run it on-premise or on-device where privacy rules require it, alongside the standard multi-region hosted option. That range, from a hosted API to fully offline, is unusual for a speech company of this size.
Speechmatics trades under Cantab Research Limited, based in Cambridge, and carries ISO 27001, GDPR, HIPAA and SOC 2 Type II certifications.
Pricing starts with a free tier and $100 in credit with no card required, moves to a pay-as-you-go Pro tier from $0.129, and volume discounts apply automatically above 500 hours a month; Enterprise pricing is quoted separately for no-rate-limit and privacy-first deployments. As with every UK company in this directory, that is an adequacy-decision relationship with the EU rather than establishment inside it.
What Speechmatics does well
- Independently benchmarked as the most accurate of 12 STT services
- 55+ languages with built-in speaker diarization
- Sub-second real-time transcription
- Cloud, on-premise or on-device deployment
- Free tier with $100 credit, no card required
Where Speechmatics falls short
- Speech-to-text only — no voice synthesis or cloning
- Pro tier pricing unit not itemised beyond the headline rate
- UK company, so adequacy rather than EU establishment
- Built for developers integrating an API, not end users
Standout feature. Speech recognition rather than speech generation — independently benchmarked as the most accurate of twelve services tested.
Pros and Cons
Pros
- Independently benchmarked as the most accurate of 12 STT services
- 55+ languages with built-in speaker diarization
- Sub-second real-time transcription
- Cloud, on-premise or on-device deployment
- Free tier with $100 credit, no card required
Cons
- Speech-to-text only — no voice synthesis or cloning
- Pro tier pricing unit not itemised beyond the headline rate
- UK company, so adequacy rather than EU establishment
- Built for developers integrating an API, not end users
Alternatives to Speechmatics
Frequently Asked Questions
What is Speechmatics?
Every other tool in this category turns text or a voice into speech; Speechmatics runs the tape the other way, turning speech into text.
It grew out of research applying neural networks to speech recognition at Cambridge University, and it competes on accuracy rather than voice quality, reporting a 1.07% pooled word error rate — the lowest of twelve services benchmarked independently. That distinction matters for anyone building a product on top of transcription rather than reading text aloud: a voice agent, a captioning pipeline, a call-centre analytics tool, where every recognition mistake compounds downstream.
Where is Speechmatics based?
Speechmatics operates from Cambridge, United Kingdom, which places it under United Kingdom (adequacy decision, outside the EEA). Compliance: ISO 27001, GDPR, HIPAA, SOC 2 Type II.
What does Speechmatics cost?
Free tier with $100 credit, no card required / Pro pay-as-you-go from $0.129 / Enterprise custom pricing with volume discounts. Free tier, $100 credit, no card required.
Who is Speechmatics best for?
Developers who need accurate speech-to-text at scale. Speech recognition rather than speech generation — independently benchmarked as the most accurate of twelve services tested.
What are the drawbacks of Speechmatics?
Speech-to-text only — no voice synthesis or cloning. Pro tier pricing unit not itemised beyond the headline rate. UK company, so adequacy rather than EU establishment. Built for developers integrating an API, not end users.