A clean-account comparison that points to the right layer.

Diagnosing Voice Input with a second user account is a way to separate a personal profile fault from a device fault. A fresh account starts with default settings, no saved permissions, and none of your usual startup tools. If voice input works there, the fault travels with your everyday profile. If it fails there too, the fault belongs to the device or to something shared by every account. Decide in advance which answer would change your repair plan, then create the account. This one comparison is often worth more than an hour of changing settings you cannot easily undo. Keep it.
A second account gives you a controlled environment without buying hardware. It removes your saved microphone permission, your add-ons, your startup helpers, and your customized input layout in one step. That is a large change, which is why the result must be read carefully. If the fresh account works, you still have to find which personal setting matters. If it fails, you can stop editing your profile and look at the device. The value of the test is not the answer alone but the direction it sends your next hour of work. Write the direction clearly on the sheet.
Understand what the test cannot prove. A second account shares the same microphone, the same ports, the same cable, and the same page. A fault in any of those will appear in both accounts, and the test will correctly point away from your profile. But it cannot tell you which shared part is faulty, only that the shared layer is worth testing. Write that limit on the sheet before you start, so a result of fails in both does not get misread as the device is definitely broken. It means the shared layer is the place to look next. Keep that.
Plan the comparison before you create anything. Write the sentence you will speak, the page you will open, the field you will use, and the four possible outcomes. Decide what you will do for each outcome, including the outcome where both accounts fail. Keep the plan to one page. A planned test produces a decision; an unplanned one produces a story. When the result arrives, read the plan rather than improvising, because the improvised version is usually the one that fits what you hoped to see. Read the plan aloud to a colleague before you create the test account, and note the date.
Create the second account with the smallest setup that can run the test. A guest session or a standard user account is usually enough. Do not copy your documents, your bookmarks, or your settings into it, because that reintroduces the variables you are trying to remove. Name it clearly, such as voice test, and note the date. Sign in once, wait for the first-run setup to finish, and restart before testing. A rushed account can still be completing background setup when you open the page, and that noise will confuse the result. Write the setup time on the sheet.
Install only what the test requires. If the page needs a browser, use the same browser build as your everyday account so the comparison stays fair. If the voice tool is a system-level app, install the same version. Do not add your usual helpers, because a helper is exactly the kind of variable the second account is meant to remove. Write the versions on the sheet beside the account name. A fair test uses the same software in both accounts and different personal settings, and that is the only difference you want. Keep both version lines on one sheet.
Check the basic device state inside the new account before you open the page. Open the system input list, confirm the microphone appears, and watch the level meter while you speak. If the device is missing in the fresh account, stop and fix that first, because the test result would be meaningless. Write the device name and the meter result. A second account that cannot see the microphone is telling you something about how the device is shared, and that is a different diagnosis with its own steps. Write the device name down before you open the test page every time.
Record the environment in one block: account name, date, device name, browser version, app version, and the page address. Add a note about anything you could not make identical, such as a different default zoom or a missing font. Those differences are small but they matter when you compare two runs. Keep the block at the top of the sheet so every later row has context. A test account without a written environment is just a second place where something worked or did not, and that is not enough to act on. Keep the block at the very top of the page.
Run the script in your everyday account first, then in the test account, on the same day and within the same hour if you can. Speak the same sentence, use the same distance, and open the same field. Write one row per account with the result, the time, and the device name. Do not switch the order between runs, because the first run warms up the page and the microphone. A fixed order, a fixed sentence, and a fixed field are what make the two rows comparable. Without them, the second account is just another anecdote. Write both rows at once.
Keep the two runs separate. Do not leave the everyday account signed in and then open the test account in another window, because shared focus and shared audio can blur the result. Sign out, wait for the desktop to settle, then sign in to the test account and run the script once. Write the sign-out time. If the first attempt in the test account fails, repeat it once before recording, since first-run permission prompts are common and are part of what the account is testing. Record both attempts if they differ. Keep both attempt times on the sheet.
Watch the permission prompt in each account and write down what you see. The everyday account may skip the prompt because it already has a saved answer, while the test account should ask. That difference is expected and it is useful evidence. If the test account never asks, the page or the browser default is blocking requests before the account layer matters. If it asks and you allow it, note the device name shown in the prompt. A prompt that names a different device than the one you expect is a finding, not a detail. Write it down exactly as shown.
Build a four-cell table before you decide anything. The rows are everyday account and test account, and the columns are works and fails. Mark one cell with the observed result and leave the others empty. Add the two attempt times and the device name under the table. This tiny table prevents the most common mistake in this kind of diagnosis, which is to remember one result and reinterpret it later. The table is the evidence, and the decision should be a sentence that points to one cell. Keep the table and the decision sentence together on the same page.
Read the four-cell table against the plan you wrote. If the everyday account fails and the test account works, the fault is in your personal profile: a saved permission, an add-on, a startup helper, or a customized input setting. If both fail, the shared layer is the suspect: the device, the cable, the port, the page, or the browser installation. If both work, the original fault was probably temporary or tied to a condition you have not reproduced, and you need more runs rather than a repair. If the everyday account works and the test account fails, the test itself is flawed and should be repeated.
For a personal profile fault, repair in the everyday account only. Re-grant the microphone permission, remove and re-add the suspect add-on, and reset the one input setting you changed. Change one item at a time and rerun the script after each change. Write the row in the same table. Do not rebuild the whole profile, because the table already told you the fault is local and a rebuild destroys the evidence you need. A targeted repair is faster, safer, and easier to explain to the person who inherits the desk. Keep the old table beside the new rows.
For a shared-layer fault, leave both accounts alone and move to the physical chain. Swap the cable, change the port, and test the device in the other account again. Write each result in a new row. If the fault follows the cable, the account test has done its job by sending you to the right layer. If it follows the device across both accounts, the device is the better suspect. The second account is not a repair tool; it is a compass that points at the layer where the repair belongs. Write the layer name clearly on the sheet.
Write the decision in one sentence and attach the table. For example: the test account records cleanly, so the everyday profile is the layer to repair. Date it and sign it. Then give the sentence to anyone who asks what happened, instead of retelling the whole afternoon. A decision tied to a table is easy to revisit if the fault returns, and it stops the next person from repeating the same comparison. The sentence is the deliverable, and the table is its proof. Keep both together in the written note you leave for the next person at the desk today.
Retest with the same script in the same order after the repair. Run the everyday account first, then the test account, and write two new rows with the date. A repair is confirmed when the everyday account passes twice and the test account still passes. If the everyday account passes once and fails once, the repair is unproven and you should repeat the run before closing the issue. Two clean rows on different days are a stronger claim than one, and they cost only a few minutes of dictation. Write the two new rows directly under the old table today.
Use a system-level path as a cross-check across both accounts. Voiskey is an AI voice typing app for macOS and Windows that works at the system level: press the shortcut, speak, and the text appears at the insertion point in the app you are using. It removes filler words, stutters, and false starts, and it adapts punctuation to the task. You can download Voiskey and run the same sentence in both accounts. If the system path is clean in both, the device and the shared layer are healthy, and any remaining difference belongs to the profile. Write that down.
Keep the test account for a week before removing it. A short-lived account lets you repeat the comparison if the fault returns, and it saves you from rebuilding a clean environment under pressure. Note the removal date on the sheet. If you share the device with a team, tell them the account exists and what it is for, so nobody mistakes it for an abandoned profile. A labeled, temporary account is a reusable diagnostic tool rather than clutter, as long as it is removed on schedule. Put the removal date on the sheet and also on the account label.
Close the loop with a short record that the next person can follow. Include the account names, the four-cell table, the decision sentence, the repair, and the retest dates. Add the versions you used and any part you replaced. Keep the record with your desk notes. When a similar fault appears, read the record before creating a new account, because the old result may still apply. A documented comparison turns a one-time trick into a repeatable method for the whole team. Keep it where the team already looks for desk notes and then review it once every quarter.
Voiskey works at the system level, so you can press the shortcut in either account and see text at the insertion point in the app you are using. If it works in both accounts, the shared device layer is healthy and the remaining fault belongs to your everyday profile.
Run the same short passage in the same app and field that you used in your everyday account. Voiskey learns names, jargon, and acronyms, so a repeated technical term is a useful signal. Keep the device, distance, and pace identical between the two runs.
Yes. Voiskey is an AI voice typing app that works at the system level after you press the shortcut. There is no page permission to grant and no add-on to install. That makes it a fair control when you compare a clean account with your everyday one.
Because the app works at the system level rather than inside a browser profile, the account differences that affect a page do not gate it. It places text at the insertion point in any text field. That consistency is what lets you tell a profile fault from a device fault.

Download Voiskey for Mac and Windows and compare accounts in minutes.
Use it across macOS, Windows, iOS, and Android