Kejue
Product · 20 October 2026

Speech-to-text V2: faster, more accurate, and better with everyday speech

The second generation of our own speech recognition is live. It is faster, it is more accurate, and it understands colloquial speech and slang across the dialects it was trained on.

Speech-to-text is the stage of a call that decides everything after it. If the agent heard the caller wrong, the best reply in the world is a reply to the wrong question. The second generation of our own speech recognition is live today, selectable per agent on the Config tab, and it improves the three things that matter on a phone line.

Faster

V2 keeps up with the caller instead of waiting for them to stop, so the agent's reply starts sooner after the last word.

More accurate

Fewer misheard words means fewer repeated questions and fewer wrong lookups. The gains are largest where the first generation struggled: noisy lines, names, numbers and the places where people speak quickly.

Better with the way people actually talk

People do not speak to a phone line the way a textbook is written. V2 is trained on colloquial speech and slang across its dialects, so a caller who says it the way they would say it to a friend is understood the first time.

Your vocabulary

Every business has words a general model has never heard: product names, branch names, plan names, the way your customers say your brand. V2 supports custom vocabulary. Give the agent the terms that matter to you and they are recognised as what they are, rather than as the nearest common word.

The agent's Config tab with speech-to-text V2 selected and custom terms added

Custom terms on the agent's Config tab. Type a term, press Enter, and the transcriber favours it from the next call.