Transcription technology has advanced a lot over the past decade, thanks to breakthroughs in machine learning and voice recognition. The great strength of automation solutions is that (when they work), they can transcribe speech in real-time and, potentially, even faster if you are working with recorded audio.
Even the best human transcribers would struggle to beat the latest automated transcription tools for speed, but they would excel where the technology struggles most: accuracy. The question we want to pose is: can you rely on the automated transcription tech available today for professional work?
Automated transcription is already being used on professional projects but the plot twist is that this technology is actually used by professional transcribers. While it’s popular to talk about advances in AI, machine learning and similar technology in a humans vs machines scenario, this couldn’t be further from the truth when it comes to the latest language technology.
Much like professional translators use translation technology every day, professional transcribers benefit most from using automated transcription.
In the intro of this article, we said that humans can’t match the speed of modern algorithms but also stated that automated tech isn’t ready to match humans for accuracy yet (the big question is whether they ever will be).
As things stand, the humans aren’t competing with the machines at all – in fact, they complement each other wonderfully. Automated transcription can transcribe a certain percentage of audio in any given project instantly and, even if it only achieves 20% accuracy, this is still a fifth of the workload taken off the shoulders of the professional transcriber.
This results in faster turnaround times, lower cognitive fatigue and a more efficient transcription process – savings that can be passed on to customers.
This question is more difficult to answer and it varies from one project to the next. If you are working with high-quality audio where speakers clearly pronounce every word and only one person speaks at a time, the latest transcription software does an impressive job. In some instances, there has been an accuracy rate of around 70% when using automated transcription on these types of files.
The issue is, this accuracy rate can drop significantly if any of the following exist in the audio being transcribed:
- Poor audio quality
- Background noise
- Multiple speakers (interruptions)
- Fast talkers (individuals)
- Fast conversations (groups)
- Speech impediments
- Conversational nuances (e.g. “uh”, pauses, etc.)
To achieve an accuracy rate in the 70% region, you almost need to work with the perfect audio file with no interruptions and participants speaking in very neutral languages. Any interferences make it difficult for algorithms to interpret language and separate speech from other audio – although they are constantly improving at this.
The biggest challenge is how humans actually talk, especially when they are not closely following a script. In everyday conversations, we tend to pause, stutter, go off on tangents and trample all over grammar rules when we string improvised sentences together, which is very difficult for a rules-based algorithm to work with. Again, these are improving, but the challenge is still there for a software programme.
Unfortunately, accent variations and any speech impediments make it increasingly difficult for algorithms and this is an area where the technology has a lot of ground to make up.
Using automated transcription as a standalone solution for transcribing content for anything that leaves the office is not recommended. For personal or internal use, the technology may be good enough to give you a close representation of what is said – and you can get a good feel for this by using YouTube’s automated transcripts.
For anything that is published or used externally, you should always seek professional transcription services so the final version of your content is 100% accurate. Your transcriber will almost certainly use automated transcription tools to speed up their workflow but they will heavily edit the results to ensure the accuracy is there and, in some cases, the source material may be too challenging for the technology to offer much help at all.
As with all technology of this nature, automated transcription is a tool best used by professional humans, not in place of them.
Find out more about our transcription services via this page.