Demonstrations are designed to go well. They use good audio, a clear speaker, a quiet room and a language pair that performs. Your Tuesday evening will not be that. These are the questions that tell you what Tuesday will be like.
Ask all nine. A provider who answers them straight is telling you something about how the rest of the relationship will go.
1. What am I charged for, exactly?
Per hour, per language, per listener, per month, or some combination. The distinction matters enormously at forty attendees and is invisible at four.
Ask specifically what happens to the bill if a hundred people join instead of twenty, and whether a second language doubles anything. "Per language per hour" and "per hour" sound similar and differ by a factor of five in a room with five languages.
Then ask the two questions nobody thinks of: do unused hours roll over, and is paused time billed? Neither answer changes the headline price and both change what you actually pay.
2. What happens if I use it twice this month and not at all next month?
Seasonal use is normal — a conference season, a festival, a term, a summer of visitors. A provider whose only shape is an annual contract is charging you for the months you do not run.
The related question: what happens at the end of an allowance? Does the broadcast stop mid-sentence, or does it continue and bill the overflow? Both are defensible. Being surprised on a Sunday morning is not.
3. Does a listener have to install anything?
If the answer is an app, ask what proportion of a first-time audience installs it. The honest answer is small.
Think about what you are asking of a visitor: download an unfamiliar app, over a connection they may not have, in a doorway, while people file past them. Every install is a person who gave up before they started, and they will not tell you they did.
4. How does the audio get in?
The single largest quality factor, and the one demonstrations quietly control for. If the answer is "a laptop microphone", quality will vary with the room, and it will vary most in the rooms that need it most.
Ask whether it takes a feed from a mixing desk. That is what separates a good result from a hopeful one, and a provider who has not thought about it has not deployed into many real buildings.
5. How far behind the speaker will the text be?
Ask for a number, not an adjective. "Real time" and "instant" are not numbers.
Then ask to watch it on your own audio rather than their demo file. The difference between those two is the whole answer, and a provider unwilling to do it has told you what the difference is.
6. Which languages, and how good is the worst one?
Every provider lists an impressive number. The list is not the question.
Ask about the specific languages in your room. Be sceptical of uniform confidence across very different scripts and language families — quality genuinely varies, and a provider claiming it does not is either not measuring or not telling you. The useful follow-up: "how do you know?" A provider who has actually tested will describe how. One who has not will describe the model.
7. What happens when the audio is poor?
This is the question almost nobody asks and it is the most revealing one.
Speech recognition given a weak signal does not go quiet. It invents — confident, fluent sentences that were never said. A provider who acknowledges this and describes what they do about it has worked on the problem. One who says it simply does not happen has not measured it, and their system will put words in your speaker's mouth in a language nobody present can check.
8. Who owns the transcripts, and where are they kept?
You are recording what people said in your building. Ask where it is stored, how long it is kept, whether you can delete it on request, how fast, and whether it is used to train anything.
Ask also whether you can choose per session — public, internal, or not stored at all — because a single global setting is not enough for a room that sometimes contains a testimony. If these answers are vague, that is itself an answer. See recording and translating without a privacy problem.
9. What happens when the connection drops, and can I stop?
Ask whether listeners reconnect on their own or have to be told to reload, and whether you are billed while nothing is being broadcast.
Then the exit question: notice period, contract length, and whether your saved material comes with you. The easiest provider to start with is usually the one that is easy to leave, and the answer to this question predicts the others.
The test that outranks all nine
Run it once, on your own audio, in your own room, with your own speaker, and a bilingual member of your community reading along and telling you honestly what it was like.
Not a demo file. Not a recording of a podcast. Your building, your microphone, your Tuesday. Then read the transcript properly rather than skimming it — fluent text that is wrong in places is the failure mode that matters, and skimming will not find it.
Any provider unwilling to let you do that before paying has answered every question above at once.
Our answers
InMyTongue meters by the hour, needs no app, takes a desk feed, lets you decide storage per session, and gives you your transcripts. New accounts include free credit so the real test costs nothing. Run it on your own room.