It is a very complex processes for a computer to be able to recognize speech, essentially turning minimal changes in air pressure into language. Sometimes computers get it wrong, sometimes they get it right. This is the main reason why at Net Transcripts, all of our transcripts are typed by actual humans based domestically here in the United States.
Aside from our previous concerns of privacy and confidentiality, voice recognition still does not have a place in Law Enforcement and Criminal Justice Transcription. Especially in some cases when that transcript can be someone’s only voice in a trial. We take our quality and accuracy very seriously having one of the best in the industry (98.8%). We realize and understand the importance of an accurate transcript when going to trial.
Following is a very interesting read from Mental Floss on all the processes involved in for a computer to understand speech. It is no wonder with all these steps how there is potential for error.