This article explains how you can maximise the accuracy of the automatic transcription using context-specific lexicons

Machine transcription is an automated process. Even when using the best possible audio quality, it will be prone to errors. In particular, proper names and specialised words are challenging for the Automatic Speech Recognition (ASR) engine. 

It is simply not possible to train an ASR engine to be equally expert in all possible domains, because you would see the number of false positives increasing. For proper names there is an additional complexity, given a sometimes wide range of possible spelling variations of such names. 

So in cases where you need a 100% accurate transcription, we allow you to post-edit the machine generated result using the EDIT mode (see: Editing the Transcript).

However, if you regularly transcribe the same type of material, you might notice that you are often correcting the same recognition errors. To minimise such repeated corrections, you can send us a list of words that should be recognised by preference. We will use it to create a specific vocabulary for your transcriptions. By doing so, you can prevent having to make the same corrections over and over again.

Note: if you provide us with a specific dictionary or lexicon, the list of words remains your exclusive intellectual property. We consider it as confidential information and we will not share this with third parties, in accordance with our terms of service.

For more information on the use and the implementation of custom dictionaries, please contact the service desk.