Transcribing with Computer Aided Transcription (CAT) - a curse and a blessing
The development of artificial intelligence (AI)-based computer tools for speech recognition is flourishing. It looks like the universal solution to all language barriers. But is that really true?
Artificial intelligence can reproduce a short, perfectly pronounced sentence just as perfectly. But truthfully, how often do you hear such sentences at a conference or meeting? In 20-25% of cases, the verb or subject is missing, or the sentence is incomplete in some other way.
Our brain, on the other hand, is a master at piecing things together, interpreting them and ignoring mistakes. So, it fills in the gaps and completes sentences without us even having to think about it. And more importantly, the meaning is preserved.
IT tools, on the other hand, face an almost insoluble challenge. Without context, background knowledge and decision-making skills, they have no way of compensating for the linguistic deficits and so produce gibberish that cannot be deciphered without human intervention.
Hence transcription software still relies on input from people. You can find out how you can help to maximise accuracy in transcription in our Practical guide on Subtitling. You will be pleased to discover that the first steps for voiceover and subtitling are the same!
Poor sound recordings
Concentrating hard when a speaker strays too far from the microphone is not enough. It takes just one or two missed words to lose the meaning of a sentence. For AI, unable to recognise what has happened, it is as if there had been an interruption. But interruption means the end of the sentence, so let’s insert a full stop! The machine transcription can lose a further 20% of intelligibility if the sound recording is poor.
Fortunately, there are ways to overcome these obstacles. With the following tips, you can achieve over 95% accuracy with modern AI (CAT). This means lower costs and shorter delivery times.
For example, if the transcription quality is above 95%, we offer a discount of 25%. If the quality is below 90%, we add 50% to the flat rate.
- Make sure that the recording quality is perfect. Every speaker should use a professional headset (photo)
- Run a rehearsal and ask the speaker to speak in short and precise sentences. Improve the phrasing if needed
- To perfect the phrasing, use a teleprompter when recording
- Provide us with the original script with a list of your organisation's abbreviations, acronyms and keywords
And last but not least, we can provide all the recording equipment, including the technicians or even arrange the audio-video recording for you in a studio.