HomeAnalysisWhat to check before your company approves a voice dictation tool

What to check before your company approves a voice dictation tool

Say eight people on one team ask for a dictation tool. The vendor’s page says “secure” in large letters and never says where the audio goes. You can’t approve that as it stands, and you can’t wave it off either, so the homework lands on you.

Start with where the audio goes. The answer settles most of what follows, and the list at the end is something you can send to a vendor.

Where does the audio go, and who can read it?

A cloud tool sends your recording to the vendor’s servers and turns it into text there. A local tool does that work on the computer in front of you. Some tools split the job, handling the speech on your machine and sending the text off to be tidied up. All three kinds exist, and a sales page isn’t always plain about which one you’re looking at.

Your clients’ confidentiality terms probably say who may see their information. A manager reading a client’s draft contract aloud to a microphone is making a recording of that information. Once it sits on a vendor’s server, the vendor’s staff and its suppliers’ staff become possible readers, and your company’s data rules decide whether that’s acceptable. Two vendors can both print “secure” on the homepage and give opposite answers about where the recording is stored. The privacy policy is where they differ. Search it for “retention” first, then “training” and “sub-processors”, which is legal language for the other companies that handle the audio on the vendor’s behalf.

Does it work with the network off?

This is the easiest check on the list. Take a spare laptop that has finished its first-time setup, switch off Wi-Fi, start dictating and say a sentence. If the text appears at normal speed, the speech engine isn’t waiting on a server.

First-time setup is the dull part that trips people up. Some local tools download their speech model the first time you start them, which can come to a few hundred megabytes and needs the internet once. If your firewall blocks unknown downloads, somebody has to write an exception, and it’s better to find that out on day one than in front of the team. After setup, look at what else the app reaches for, such as a license check or an update check. Neither is a problem when you know about it. Your firewall log will show them anyway.

What needs an account?

Accounts help and hurt. With them you get one invoice and a way to switch off a leaver’s access, and the vendor holds a record of every user. Without them the vendor holds less about your people, and working out who has a copy becomes a spreadsheet job.

I lean toward fewer accounts. An IT team that has to revoke access on a leaver’s last day may reasonably disagree. Whichever way you go, find out who owns the account when the person leaves: them, or the company.

How is it licensed, and what does it cost over three years?

Two things get tangled on price pages. One is what’s counted, people or computers. The other is how you pay, monthly or once.

Start with monthly. Take a made-up tool at $10 a month per person. That’s $120 a year, $360 per person over three years, and $7,200 for a team of 20. (Every number in this section is hypothetical. Swap in the vendor’s real quote and redo the sums.) Subscriptions usually stop working, or fall back to a limited version, when you stop paying.

Then the counting. Suppose ten people each use a desktop at the office and a laptop at home. A per computer license counts twenty. A per person license counts ten, if it covers both machines, and that “if” is worth getting in writing.

A one-time license has a different shape. You pay once, so the three year cost equals the first year cost, unless updates are sold separately. Ask whether updates are included and for how long, and whether the quote is before or after tax. That means finding out whether tax goes on top.

A subscription makes sense when the vendor’s servers do ongoing work for you, and a one-time license suits software that runs on your own machine. A vendor with real running costs may need the monthly fee.

What happens to custom words and snippets?

A custom word list holds your clients’ names and your project code names. A snippet, the short phrase that expands into a paragraph, may hold an address or a standard clause. Both are company information, and somebody on your team will spend real time building them.

Ask where they live: on each computer, or in the vendor’s account. Then find out whether they go with a person who leaves, and whether you can export them as a file and hand the same starting list to every new hire. If the list can’t leave the tool, you’ve bought a lock-in.

When a cloud tool is the better choice

Local isn’t automatically better. If most of your team dictates in something other than English, check which languages each tool supports before you commit. A team that wants shared notes or searchable transcripts is also better off in the cloud, since the text sitting on a server everyone can reach is the whole point.

Older laptops can struggle with a local speech engine, so put your oldest machine in the trial. And whether a cloud engine really hears your people better than a local one is something only a test with their voices will settle, whatever a vendor’s average says.

What will staff do if you say no?

Some will find their own. Someone who writes long emails all afternoon won’t give up dictation because a policy says so. The likely replacements are the microphone button on a phone keyboard or a web page that transcribes what you say, and you won’t have read the privacy settings of either.

My preference is one approved tool whose settings I know. The alternative is a ban, and then you inherit whatever people pick for themselves. Say yes with limits: one tool, and a short rule on what stays off the microphone.

Questions to send a vendor

Send them by email and keep the replies. A phone call isn’t a written answer.

  • Does the audio leave the computer in the default setting, and does the app show which mode it’s in?
  • Do you keep recordings? For how long, and can we delete them? Is anything we dictate used to improve your models?
  • Which other companies handle the audio or the text on your behalf?
  • Once it’s set up, does it work with the network off? What does it connect to?
  • Does each person need an account, and who owns it when they leave?
  • Is the license per person or per computer? Are updates included, and what happens if we stop paying?
  • Can we export the custom words and snippets?
  • What is the three year price, in writing, with and without tax?

An example of the local approach

Dictate Safe is one example, and it’s made by Vibhaga Group, which also runs this site, so weigh this paragraph accordingly. You hold a key, talk, and it types into whatever app is open. In its normal mode the speech engine runs on your own computer; an optional cloud setting uses your own key, and the app shows which mode it’s in. It runs on Mac and Windows and works in English only. It’s a one-time purchase, with no subscription: one license covers one person on up to 2 computers and includes updates and fixes to the features sold, while the app is offered. A purchased license is tied to an account on our site, so activating one means signing in. The first run downloads a speech model, which needs the internet for that download. It comes with 25 free dictations, no account and no card, so someone can try it before anything is bought.

Sign Up for Weekly Industry Insights

Note: OEM NEWS functions on a set of solid business ethics with a zero-tolerance policy for any deviations. Please learn more about our ethics framework here.

OEM NEWS - Follow us on Google NewsDownload OEM news android app

RELATED ARTICLES

LEAVE A REPLY

Please enter your comment!
Please enter your name here
Captcha verification failed!
CAPTCHA user score failed. Please contact us!
- Advertisment -

Most Popular

Recent Comments