Light years away from perfection!
Speech to text applications work fine as long as you don’t go beyond its comfort zone. Simple words that are clearly dictated are transcribed almost accurately by speech to text apps. Go a little further and all you get is a mishmash of nonsensical words. The web is filled with bloopers that a speech to text app makes, when a human being makes the mistake of, well, speaking like one.
To wreck a nice speech!
Ironically this is what an automated voice to text application translates when you say “to recognize speech”. After years of research and technological advances the highest accuracy a speech to text app can achieve is 80%. Though the latest applications are much better, they still leave a lot to be desired.
One of the biggest drawbacks of these applications is that they do not recognize even simple foreign phrases. Commonly used expressions such as “bon voyage” “gracias” can drive voice recognition applications into a tizzy. Most users of this technology change their vocabulary and accent so the poor picks up what’s spoken. For those who have no patience with slow learning apps, working with one can be a disaster.
Weird video captions!
Have you watched a YouTube video with the automatic captions on? It can be a nightmarish experience. Ridden with mistakes that would make a two year old cringe, it is certainly a wiser choice to turn them off.
So what is the average accuracy score of a human transcriber?
No points for guessing that it is way higher that of a speech to text application. The range of human error in transcription is also considerably lower. Between 2 and 4 % it is much better than an automated voice recognition system.
Is it the death knell for speech to text applications?
Definitely not! It would be unfair to pronounce the technology dead as hundreds of speech scientists and linguists are working on making it better. It may not have an amazing level of comprehension but it is still a work in progress. A work that would hopefully get better with time.