Blog chevron right Transcription

AI-based Transcriptions Fall Short Compared to the Human-Made Ones, Part 2

Michael Gallagher
Michael Gallagher
Posted in Zoom Oct 6 · 9 Oct, 2021
AI-based Transcriptions Fall Short Compared to the Human-Made Ones, Part 2

In part 1, we have mentioned that AI sometimes makes some silly or big mistakes and in some instances it takes the media like a storm. Often it is funny to read these news, but not when oneself is at the butt of the joke. Therefore, at GoTranscript we put an absolute effort with our human-driven transcriptions to deliver top quality that supersedes any AI application. 

On one hand, we do that by employing native speakers for transcription jobs and for editing to ensure top quality. On the other hand, there are some more reasons, where AI is being beaten by a person. 

Let’s look at a few more such occasions, where AI is not coping too well. 

  1. Interlocution

AI transcription of some clear monologues can be delivered rather well. But what if the audio track has interchange of speech, a.k.a. interlocution. In discussions and dialogues there are multiple persons speaking, sometimes at the same time.

Transcriber has to catch who’s talking, and who’s saying what. Can software tell in between the voices? Can AI match each sentence to a distinct person? Most likely it is not worth taking the risk if you are looking for a superb transcription that would ease you of work instead of making you edit while listening to the audio track over and over again. 

On top of that, individuals have varying levels of articulation and in some occasions different ethnic accents. This makes the interlocution even more complicated. 

  1. Accents

English is the most adopted language for international use. It's being spoken in a business context, lots of media and entertainment is produced in English to enter international markets. The world is global, and lots of economic, political, cultural or academic discussions are being held in English. 

⅓ of all English speakers are not native! Most of them definitely speak with a stronger or weaker hint of their nativr language. No matter how attractive some accents are, or maybe makes us unintentionally smile, they can be a tough cookie for AI to catch the right words, more so, the right meaning. 

  1. Proofreading

Some of the very best AI transcription services claim 99% accuracy. This is 1 mistake in a 100 words, or 10 mistakes in 1000 words. Way too much if you take the content seriously as it may deteriorate the perceived value of the text. 

At GoTranscript, the transcriptions are thoroughly checked by a team of professional editors. The risk of delivering obvious mistakes is virtually zero. That makes a big difference in various transcription applications, such as legal or medical. Naturally, for online use it is also important, because there could be a risk of missing a few valuable pieces for a killer SEO ranking.