Speechmatics is a company that builds its own speech recognition models and sells them through APIs, interfaces that other programs call over the internet. Its main product is speech-to-text; it also offers text-to-speech.
The Batch API transcribes recorded audio or video files. The Realtime API receives live audio over a WebSocket connection, a connection that stays open in both directions, and returns transcript text continuously, within milliseconds of the spoken words. The models cover more than 55 languages, including speech that switches language in the middle of a sentence, and speaker diarization marks which speaker said each part. The models include Melia 1 (multilingual), Enhanced (highest accuracy for one chosen language), Oak 1 (medical vocabulary) and Linden 1 (for voice agents, with turn detection). Add-ons translate, summarize or find topics. Client libraries exist for Python and JavaScript, and the models can run in Speechmatics' cloud, on the customer's own servers or on a device.
Use is billed per hour of audio, with a different rate for each model; new accounts receive US$100 in free credit.