Lucas Lee’s Ranking: Transcription Services
A comparison of available tools based on privacy ethics, security and handiness
WITH THE ADVANCEMENT of artificial intelligence and machine learning, one task performed by many journalists has changed: transcription. Artificial Intelligence makes the processing of speech-to-text faster and simpler than ever. But which method of transcription is most secure? Which platform protects the conversations shared by journalists and sources? Here’s my (worst to best) ranking of the more popular transcription services available:
7. Rev (human transcription)
Rev is a San Francisco-based company that offers transcription at a rate of $1.25 (U.S.) per minute, done by human freelancers (called Revvers). Although Rev promises TLS 1.2 encryption, the fact that Rev employs people to listen to recordings means the company lands at the bottom of the list. It is recommended that journalists refrain from granting access to interviews and recordings to strangers, despite the fact that Rev claims that all Revvers are vetted and have signed NDAs. Rev does offer AI-based transcription, which is the better option to process information divulged by sources.
6. Temi
Temi, another San Francisco-based company, charges 25 cents per minute for transcription performed by its proprietary algorithm. Temi’s low ranking on this list is due to several issues I have with its privacy policy. The first is its vague description of what it considers “personal data.” Temi defines personal data as: “when you voluntarily provide such information such as when you register for access to the Services (for example, your name and mailing address), use certain Services, contact us with inquiries or respond to one of our surveys.” The lack of definition of “certain Services” lands Temi at the sixth spot.
Another area of concern is that by using Temi, the user agrees that their “Personal Data” may be transferred in the event of corporate sale, merger, reorganization, dissolution, or a similar event. The vague nature of both descriptions is why I recommend against using Temi.
5. Otter.ai
Otter.ai, Los Angeles-based, provides a more comprehensive privacy policy than that of Temi in my view. Otter.ai lists the privacy regulations by which they abide—EU customers are granted rights under the General Data Privacy Regulation (GDPR), which is one of the most stringent pieces of privacy legislation ever passed. But Otter.ai is far from perfect. In regard to customer data, users grant the company “a worldwide, non-exclusive, royalty-free, license to access, process, copy, export, and display Customer Data.” The broadness of this language is one of the reasons I do not recommend the Otter.ai service. Another issue is this: Otter.ai claims they encrypt many of their services, not all.
4. Descript
Descript’s security policy is comprehensive and transparent. The company claims to employ a Data Protection Officer accountable for the implementation of its security practices. The company claims it is compliant with the GDPR and the California Consumer Privacy Act. But I do not recommend Descript either—far from it. Descript does not have a proprietary machine-learning algorithm—it uses Google Cloud Speech-to-Text for its AI transcription, and for its human-powered transcription. This means that Descript is sharing user data with Google and Rev. Trusting Descript with information means trusting Google and Rev as well.
3. Trint
U.K.-based Trint has a lot going for it as the transcription service of choice for journalists. In fact, it was founded by a journalist—an award-winning war correspondent named Jeff Kofman. Trint claims it is ISO/IEC 27001 certified, which is a standard of security requirements created by the International Organization for Standardization. Trint claims it encrypts data with TLS 1.2 during upload, and uses AES 256, a very secure algorithm, while the data is at rest. But I don’t recommend Trint either.
Trint, like all other transcription services listed before it, is legally compelled to disclose user data if need be. Their policy states: they may disclose your personal information to third parties, the courts and/or regulators or law enforcement agencies in connection with proceedings or investigations anywhere in the world where we are compelled or believe it is reasonable to do so. This clause means relinquishing any control the journalist has over source information.
2. oTranscribe
One transcription service I do recommend is oTranscribe, because oTranscribe is not really a transcription service at all. It is a web page that allows users to upload their recordings and start typing away, themselves. Manually. The benefit is that the platform allows users to avoid switching from window to window when transcribing. And it’s secure too—the recording and transcription files never leave the user’s computer, and oTranscribe can prove it, because unlike all the aforementioned companies, oTranscribe is open source.
1. Sitting your behind down, plugging headphones in, and going all-in
The absolute best way to transcribe interviews is the old-fashioned way. Sources should feel confident that journalists will keep the information they share, whether or not confidential, away from the hands of companies and governments. The only way to fulfill that duty to sources is to keep all information away from third-party companies. My recommendation is this: avoid using TapeACall. Refrain from storing any source information using cloud services, like Google Drive, iCloud, or Dropbox. Never use online word processors like Google Docs. If using Microsoft Word, disable auto-syncing. And, for the love of God, never use AI transcription.
This list and therefore opinions stated within are purely the author’s.
—Lucas Lee