Using machine learning to rapidly build and adapt languages for speech-to-text transcription

We’re dedicated to providing the best speech recognition services possible to our global customers and have developed an innovative machine learning framework called The Automatic Linguist (AL). The framework allows us to build and adapt languages quicker than ever before, proving this when it built Hindi in just one week.

Traditionally, building and adapting new language packs has been a lengthy, laborious affair, involving gathering vast amounts of data, building a one-off system and continually refining it with input from experts in that language. It’s time-consuming, expensive and difficult, which is why Speechmatics developed an entirely different solution for training languages.

Speechmatics used its expertize and knowledge in machine learning and neural networks to create a capability for building and adapting languages for use in Speechmatics’ any-context speech recognition engine.

Read the Automatic Linguist whitepaper

Download our Automatic Linguist whitepaper

Building language packs for the languages of the world

“AL is a flexible, evolving framework enabling us to provide high-quality languages for our customers and partners to use in our any-context speech recognition engine.”

Inside the whitepaper you’ll find:

  • The traditional approach to building and adapting speech recognition languages
  • Speechmatics’ approach to building new languages using machine learning and neural networks
  • An overview of Speechmatics’ language portfolio
  • Speech recognition accuracy and competitor benchmarking figures
  • The real-world benefits of AL
  • An introduction to the challenges ahead

Read the whitepaper to learn how Speechmatics builds and adapts speech recognition languages using a unique machine learning framework, The Automatic Linguist.

Speechmatics has developed an innovative machine learning framework called The Automatic Linguist (AL) to rapidly build languages for speech recognition.

Key stats from the whitepaper:


speech recognition languages


years of R&D experience


weeks to build 46 working research languages

Ready to try Speechmatics?

Try it now