AI vs Human Transcriptions, Which Is Better?

  • Facebook
  • Twitter
  • Pinterest
  • LinkedIn

On the surface, audio transcription seems like a pretty easy task, you just write down exactly what you hear in an audio file. However, there are so many intricacies that make it far more complicated than you may initially think. Transcribers need to be able to understand accents, dialects, industry-specific jargon, facts and a whole lot more. For these reasons accuracy is hugely important, in specific industries the inaccuracies of just a few words can truly be catastrophic. 

Automatic speech recognition (ASR) is software that converts audio files into text. It’s often fast, cheap but highly inaccurate. Transcriptions completed by ASR are produced with around 60-70% accuracy (according to The University of Sussex). To put that into perspective, if you had a 1500 word document, between 450 and 600 words would be inaccurate. This level of inaccuracy is very poor and likely to negatively affect a customer no matter what field they’re in. 

In stark contrast to this, the human transcribers at McGowan Transcriptions boast a 99.5% accuracy, that’s only 7.5 potential inaccuracies on a 1500 word document. Humans have the unique ability to pick up on the intricacies of dialect. This is something that no AI software has the intelligence to do.

With human transcribers, you are guaranteed the highest quality work possible. You’re guaranteed almost no inaccuracies and can rest assured that your work is in the hands of experienced professionals.

Industries That Need Accurate Transcriptions

Some industries demand accuracy and can face huge repercussions if these accuracy levels are not met. A brilliant example of this is in the legal industry. Poor quality transcriptions can cause judges a lot of problems within the court system and make it far more difficult for them to come to a resounding conclusion. 

In 2011 66-year-old Carlos Ortega was arrested for a crime he did not commit. He had to spend $300,000 and a year in jail until it came to light that US authorities had ordered the extradition of the wrong Carlos due to a simple transcription error. 

In addition to this, the accuracy of transcription can actually be used as grounds for an appeal in the legal industry. In countries like Australia, an inaccurate phrase could result in a complete mistrial in a criminal case, costing the courts thousands of dollars and countless hours of their time. 

Other industries needing accurate transcriptions include:

Medical – Accurate transcriptions in the medical industry can potentially play a part in saving the life of an individual. Accurate patient histories make sure that doctors know what exact steps to take next and the proper course of action to follow. Inaccurate transcriptions could end up in a patient receiving the wrong treatment which could, in turn, be disastrous.

Educational – Education is hugely important to us all, however, it is especially important to students in higher education such as college or university. Inaccuracies in this industry can be the difference between grades. This could be catastrophic to a customer paying for your service expecting the highest quality work.
Corporate – Transcribing corporate meetings can be difficult. There are normally numerous speakers and a wide variety of topics discussed. An inaccuracy in something as small as a figure can be catastrophic to a company when trying to finalise a deal or new venture.

ASR vs Human Transcriptions

Understanding Accents and Dialects –  While technology is advancing phenomenally, ASR will struggle to ever be intelligent enough to fully understand the unique differences in accents and dialects. With over 40 different dialects in the UK alone, understanding each one is vital for complete transcription accuracy. Human transcribers have years of experience with a huge variety of different accent variations. 

Knowledge of Industry Specific Jargon – Full understanding of industry jargon comes with experience. In many sectors, jargon can be challenging to understand. A human transcriber’s job is to understand the jargon of all industries. In contrast to that, ASR cannot understand such specific language. There are thousands of jargon terms in each industry, how could ASR ever understand so many terms? This again could be detrimental in more serious and formal industries.

Identifying Number of Speakers – In many transcriptions, there is commonly more than one speaker, sometimes several. ASR can struggle to fully identify and differentiate each individual speaker. This is hugely problematic for situations such as court cases where an inaccuracy on who said what could truly be catastrophic. Meanwhile, human transcribers are easily able to identify the number of speakers and differentiate them from each other. This is vital for the all-important accuracy of the transcription. 

Inaccuracies – The experts at McGowan Transcriptions recently compared themselves to the top three leading AI transcription services. They sent numerous files to be transcribed and found numerous inaccuracies. Here are a few: 

  • ‘we were established in 1987’ was transcribed to ‘yes english 1987’
  • ‘Kulala Lumpa’ was transcribed to ‘Columbo’
  • ‘My life history’ was transcribed to ‘the industry’
  • ‘My questions are short’ was transcribed to ‘My friends say that shot’ 
  • ‘Overall responsibility for staffing’ was transcribed to ‘overall responsibility for stuffing’
  • ‘It was covid’ was transcribed to ‘it was confirmed’
  • ‘Yeah, basically a lot’ was transcribed to ‘yeah, basically everything a lot of cool’

The most worrying of all was the word ‘kitchen’ was transcribed to ‘terrorist’. This kind of mistake can have huge repercussions on a company and could potentially damage their reputation.

Which One Is Right For You?

The best transcription for you will depend on a number of things like your budget, your timeframe, your industry and just how accurate you need the document. If you have a small budget and the accuracy of the document is not that important to you then ASR is perfect, you will get your document delivered very quickly but very inaccurately. 

McGowan Transcriptions have been in the transcriptions industry for over 29 years. 

When thinking about which is right for you, think about these 5 simple questions:

What is the quality of your recording like? – The accuracy of an ASR transcription is based on how good the audio quality is. The software needs extremely clear audio in order to decipher exactly what is being said. A human transcriber can much more accurately decipher from poor audio. If your audio quality is poor then avoid ASR.  

Does your content have multiple speakers? – If your audio file has two or more speakers then ASR will struggle to identify the number of speakers and who exactly is speaking at what time. A human transcriber will be able to tell exactly how many speakers there are and exactly who is speaking and when. 

Does your audio contain regional accents? – Again ASR will struggle to decipher regional accents and dialects. This is because the software is not sophisticated enough to understand the hundreds of different accent variations and dialects. A human transcriber will have been dealing with numerous different accents and dialects and will be highly experienced when it comes to dealing with them. 

What type of transcription do you need? –  The main thing to think about is exactly what you want from your transcriptions. Do you need a quick, semi-accurate transcription? Or do you need a more accurate, higher quality piece of work? ASR will transcribe every single word. On the other hand, a human transcriber will be able to identify and remove unnecessary content to offer better clarity.

How accurate do you need your transcription to be? – If you need 99% accuracy then the only option for you is human transcriptionists.

To conclude, you can go with the cheaper ASR, but you will be sacrificing accuracy and quality. Computer software is never going to be able to outperform a human that has been in the industry for decades. For legal, medical, educational and many more industries that use transcription services, supreme accuracy is non-negotiable. So the question is not which is better out of ASR and human transcribers, there is no real competition when comparing the two. The question is in fact what do you want? Accuracy, reliability or a budget option.