Can Conversation to Text handle accents?

Overview

This article explains how the Conversation to Text module handles speakers with regional accents or non-native speech patterns.

Applies to

All Users

Accent handling

The Conversation to Text engine powered by Microsoft Azure Cognitive Speech Services is trained on large and diverse datasets of speech from many different accents, dialects, and regional varieties of supported languages. As a result, it handles a wide range of accents reasonably well, including regional British accents (Scottish, Welsh, Northern Irish, regional English), Irish accents, South Asian accents, and many others.

Factors that affect accuracy with accents

Accuracy may be reduced for very strong regional accents, non-native speaker accents that differ significantly from the training data, speakers who mix languages or use a high proportion of dialect specific vocabulary, and recordings where accent combined with background noise creates additional ambiguity.