Transcription was once a manual, tedious process. Doctors, journalists, and a swath of assorted professionals would record their notes and conversations on scratchy Dictaphone tape, and then sit down in front of their computers to type them up.
Fast-forward to 2020, and there are a variety of services you can use to turn an audio recording into text on a computer screen. However, one question remains: Are they safe? After all, you might be uploading voice recordings of sensitive conversations and private voice mails.
Let’s take a look at these services, and how you can protect your information.
How Audio Transcription Services Work
Audio transcription services tend to fall into three camps. The first is entirely computer-driven and uses existing AI and machine learning models to process the conversation. The second is the most expensive because people do the heavy lifting. The third is a combination of computer processing and humans.
Odds are you’re probably most familiar with the first category. Voice transcription services—like those offered by Google, Apple, and Otter.ai—convert the analog waves your voice creates into a digital representation. It then breaks them into small (sometimes, one-thousandth of a second) segments and matches them to known “phonemes,” or elements of a language.
These algorithms then try to examine them within the context of other phonemes and put them through statistical and AI models that ultimately produce text. Because these transcription services are entirely computer-driven, they tend to be the most inexpensive to run. However, accuracy isn’t always on-point, particularly when it comes to extracting text from noisy or multi-person environments.
Human-powered transcription involves dedicated platforms, like Rev, that connect customers with a pool of pre-approved transcribers. You can also hire someone from a freelance marketplace, like Upwork or Fiverr, to transcribe for you.
Finally, there’s the mix of the two. To expedite the transcription process, some sites allow AI to do the preliminary work, and then someone tidies up the output and fixes any mistakes.
Transcription Services Behaving Badly
In recent years, many transcription services have been subject to breaches and scandals.
Perhaps the oldest (and, arguably, most shocking) was SpinVox, which, in the ’00s, offered a service that turned voice mails into SMS messages. At the time, this was regarded as nothing short of a technological breakthrough. The firm quickly attracted positive press, customers, and vast moats of funding.
The problem? Unbeknownst to customers, their voice mails were processed by people working from offices in places like Pakistan, Mauritius, and South Africa. One company insider claimed that only 2 percent of voice mails were machine-processed, and the rest were handled by around 10,000 exploited workers.
When staffers at a Pakistani SpinVox office didn’t get paid, they started sending messages directly to customers to protest. Eventually, the truth came out, and SpinVox lost much of its value. Ultimately, the remnants of the company were sold to Nuance, one of the biggest voice recognition providers in the world.
More recently, cybersecurity journalist, Brian Krebs, discovered a major breach that occurred at MEDantex, a Kansas-based provider of voice transcription services for medical professionals. Predictably, data (some of which dated back to 2007) containing sensitive medical records was leaked. The contents could be downloaded from an insecure portal as Microsoft Word files.
Even digital transcription services aren’t safe. After all, when you use an entirely computer-driven service, the firm might use human contractors to perform quality-control.
In 2019, Belgian news site, VRT NWS, discovered Google contractors were listening to conversations between individuals and their Google Home smart assistants. One contractor even provided VRT NWS with access to conversations, many of which were of a profoundly sensitive (and, in some cases, sexually intimate) nature.
Amazon, Apple, and Microsoft were also using contractors in this way. In other words, someone might be listening to voice recordings from your virtual assistant.
Are Online Transcription Services Safe?
The most pertinent issue is whether online transcription services are safe. Unfortunately, the answer is a bit complicated.
The voice transcription space is, at this point, largely mature. The most egregious bad actors have been weeded out.
Nevertheless, when you entrust your data (in this case, private conversations) to a third-party, you rely on it to protect it. This is just as true for online services as it is for human transcribers.
Ultimately, you have to ask yourself two things: Do you trust the service, and how sensitive are your conversations?
As we mentioned previously, AI-driven services frequently rely on employees and third-party contractors to do quality-control checks. While these checks represent a fraction of all transcriptions, there’s always the chance someone will be listening to your conversation.
In many cases, though, this isn’t a deal-breaker. However, if your conversation is deeply private or commercially-sensitive, you might want to consider opening a text editor and transcribing the old-fashioned way.