How accurate is Conversation to Text?

Overview

This article explains the accuracy levels achievable with Conversation to Text and the factors that influence it.

Applies to

All Users

Expected accuracy

Under optimal conditions clear audio, minimal background noise, distinct speakers, standard vocabulary, and a supported language. Conversation to Text can achieve high levels of accuracy. Dictalogic’s AI engine, powered by Microsoft Azure Cognitive Speech Services, is capable of accuracy levels comparable to those of experienced human transcriptionists for high quality recordings. Dictalogic’s platform is designed to produce up to 99.8% accuracy under ideal conditions for single-speaker dictation. Multi speaker conversation transcription introduces additional complexity, and accuracy in practice will vary depending on the factors below.

Factors that affect accuracy

Audio quality is the most significant factor. Poor microphone placement, background noise, telephone quality audio, and compression artefacts all reduce accuracy. Speaker clarity speaking pace, diction, and accent also plays a role. The number of speakers and how distinct their voices are affects diarization quality. Use of specialist or domain specific vocabulary that the AI has not been trained on can lead to misrecognition, which can be mitigated with Speech to Text Replacements.