Speechmatics
Speechmatics develops speech models in Cambridge, England. Tony Robinson founded the business in 2006 after earlier neural-network research at Cambridge University. Its work now includes recognition models for recorded audio, live conversations, clinical speech and local device processing.
Company & leadership
- CEO, November 2026
- Katy Wigdahl
- Legal entity
- Cantab Research Limited, trading as Speechmatics; England and Wales company 05697423
- Registered office
- One Cambridge Square, Milton Avenue, Cambridge, CB4 0AE, United Kingdom
Founder
- Dr Tony Robinson
Research roots and company formation
Robinson began applying neural networks to speech recognition during his Cambridge University research in the 1980s. He later founded Speechmatics in 2006. The company’s early research history helps explain its focus on speech recognition rather than a general-purpose language assistant.
Katy Wigdahl leads the business. The current board also lists Robinson and representatives of Susquehanna Growth Equity, AlbionVC and IQ Capital. IQ Capital and Amadeus invested in late 2016.
From recognition to speech models
Autonomous Speech Recognition introduced self-supervised models in October 2021. These learned from much more unlabelled audio than the earlier system.
Speechmatics also builds speech synthesis. Its 2025 text-to-speech preview uses its own models.
Its conversational products can connect separate recognition, language and synthesis components. A combined agent product is not one newly trained Speechmatics foundation model.
Adobe has used Speechmatics technology in Premiere since 2021. A new local recognition model was announced in April 2026. The partnership shows a second distribution route for the company’s models: software installed on the user’s device.
Financing and software access
The company’s July 2022 funding account records a completed $62 million Series B, led by Susquehanna Growth Equity with AlbionVC and IQ Capital participating. That amount was investment capital, not a disclosed company valuation.
The service terms provide a limited software and API licence.
Container deployments require repository credentials and a customer licence token. These are commercial access routes; downloading a container does not provide an open-weights licence.
Model families and specialist releases
Standard and Enhanced are maintained choices. Batch and Realtime share models. Linden uses Agent STT.
The medical range contains separately developed language releases selected through Enhanced Medical. These articles identify each documented release. Ordinary language packs, container tags and synthesis voice names remain within the relevant model family.
The current catalog does not confirm access to exact historical checkpoints.