Docs: install, record, import and export

The app is available for Windows, macOS and Linux. The Windows and Linux versions run on 64-bit Intel and AMD computers. The macOS version runs on Apple silicon. Get the installer from the download page. Each link on that page is the current release.

The installer includes the transcription engine, but not a model. If no model is ready when the app starts, the app tells you and offers the download.

Windows

Run the installer. The installer is not signed yet, so Windows can warn that it comes from an unknown publisher. If it does, choose More info, then Run anyway. The installer installs the app for your user account, and you can choose the folder.

macOS

Intel Macs are not supported. Open the disk image and drag Amlistening into Applications.

The app has a Developer ID signature, and Apple notarized it, so it opens like any other app. The first time you open it, macOS tells you that it came from the internet. Choose Open to confirm. You do not need to run a command in Terminal.

Linux

The download is an AppImage. This is one file that contains the whole app, so there is nothing to install. Make the file executable, then run it:

chmod +x amlistening-*.AppImage
./amlistening-*.AppImage

The app opens with no project. The work area shows a welcome view with two buttons: New project and Open project. If no model is downloaded, the welcome view also says that transcription is not ready, with a link to Settings → Transcription.

Setting up a project and a model takes a few minutes. Only the model download needs a network.

1. Download a model. Press Ctrl+,⌘Command, to open Settings, then go to Transcription. Choose a model and press Download. The app stores the model on this computer. You can pause the download. If the download stops, it continues from where it stopped. Until a model is ready, the app opens Transcription and offers this download each time it starts. To stop this, turn it off in Settings → Transcription.
2. Create a project. Choose File → New Project…, or press Ctrl+N⌘CommandN. A project is a folder that holds its takes and its keyword library. You cannot change the name of a project after you create it. To record, you must have a project open.
3. Choose the language. With the project open, go to Settings → Transcription and choose the language you will speak. The default model, Nemotron 3.5, knows about 80 languages. Every take in the project uses this language. The app never detects the language by itself.
4. Check the input. Choose the microphone in the input menu at the right of the top bar, or in Settings → Capture. Press Check input on the waveform card, then speak and watch the level meter above the waveform. The first time the app uses the microphone, your computer asks for permission. Allow it.
5. Run the pre-flight check. Choose Record → Run Pre-flight Check, or press Ctrl+Shift+K⌘Command⇧ShiftK. It checks these items, and shows the result in a message at the bottom of the window:
   • The app can write to storage.
   • How much recording time the free disk space allows.
   • An input device is chosen.
   • The microphone hears a signal, if it is open.
   • Transcription and the licence are ready.

You can record before a model is ready. The app writes the audio as usual, but it marks the speech as failed and does not transcribe it. When a model is ready, select the take. Then press Retry beside the message that says how many utterances failed.

Press Record, or press Ctrl+Shift+R⌘Command⇧ShiftR. You do not fill in a form or a name first. The app names the take with the date and time it started, such as 2026-09-12 14:05. Speak, and the transcript adds your words line by line. A Listening row shows where the next line will appear.

Record runs the pre-flight check first. Some problems stop a take from starting, such as no input device, or storage that the app cannot write to. If there is such a problem, no take starts. A red banner shows Recording didn't start and names each problem. Fix the problems, then press Try again.

A quiet microphone does not stop a take. If the input is silent for three seconds, a warning shows in the take until sound returns. The take continues to record.

The app's icon in the menu bar also shows the take. On Windows, the icon is in the notification area. It is red with a record dot while the take records. It is amber with an exclamation mark while the input is silent.

To stop, press Ctrl+Shift+R⌘Command⇧ShiftR again. The take is already on disk.

The transcript shows the oldest line first. To show the newest line at the top, press Oldest first in the header of the transcript, or choose View → Transcript Order. The app remembers your choice.

The Recordings list on the left puts takes in groups under Today, Yesterday and a header for each earlier date. To fold a day, click its header. If you have not renamed a take, its row shows only its start time, because the header already shows the day. Each row also shows the number of keyword hits in the take, with a dot for each keyword colour that has a hit.

To name a take, double-click its row in Recordings. Or right-click the row and choose Rename… or Tags…. You can do this while the take records. To find a take later, type in Search recordings above the list. The search finds names and tags.

To hide the list and get more space, click the folder icon at the top of the rail. The rail is the narrow bar at the left edge of the window. Click the icon again to show the list. View → Show / Hide Recordings, or Ctrl+Alt+S⌘Command⌥OptionS, does the same.

When the transcript is correct, export it.

You can use any microphone that your computer can use: a built-in microphone, a USB microphone or a headset. Choose it in Settings → Capture.

If you move around while you talk, a wireless kit keeps the cable out of the way. Examples are DJI Mic, DJI Mic Mini and RØDE Wireless GO. Clip the transmitter on, and connect its receiver to the computer. Then choose the receiver as the input. Amlistening is not affiliated with DJI or RØDE.

Between takes, the microphone is off, and your computer shows that no app uses it. To see the input level before a take, press Check input on the waveform card. The microphone stays open for one minute, or until you press Stop checking.

During a take, the level meter shows the input. Your words show in the transcript while you speak. When you pause, the app finishes the line.

To mark a moment that you want to find again, press MM. Ctrl+Shift+M⌘Command⇧ShiftM also adds a marker, and it works while you type in a field. The waveform shows the markers.

To pause, press Ctrl+Shift+P⌘Command⇧ShiftP. The app writes nothing until you resume, and the take stays one take. To resume, press Ctrl+Shift+P⌘Command⇧ShiftP again. To end the take, press Ctrl+Shift+R⌘Command⇧ShiftR.

Nothing interrupts a take. An update never restarts the app during a take. You cannot download or change a model until the take ends.

On Free, a take stops automatically at 5 minutes. The status bar counts down the last seconds.

You can import a recording from a phone or another recorder into a project. It becomes a take. You cannot import a file while a take records. To import, do one of these:

• Drop the files on the Recordings list.
• Choose File → Import Audio…, then select the files.
• Choose Import audio in the command palette, then select the files.

If you import more than one file, the app imports them one at a time. Each file becomes a take of its own, with the name of the file.

The app reads WAV, MP3, M4A and MP4 files. This includes voice memos and the sound of a video. The app reads what is in the file, not its extension. The sound in an M4A or MP4 file must be AAC.

The app cannot read FLAC, Ogg, Opus, AIFF, raw AAC, or WAV files over 4 GB yet. Convert them to one of the four formats first.

The app reads the file but never changes it. It decodes and transcribes the file on this computer, and uploads nothing.

When the import ends, the take opens. The app transcribes it in the background, and the transcript fills in line by line. A take that you record always comes first. After that, an imported take is the same as any other take. The app marks its keywords, you can correct its lines, and you can export it in every format.

A GoPro video also contains GPS data. On Pro, the take gets the video's trace in the same step. See places.

On Free, an imported file counts toward the 10 takes. If a file is longer than 5 minutes, the app keeps only the first 5 minutes.

To review a finished take, select it in Recordings. To play or pause, use the round play button at the top of the waveform. The volume control is next to it.

To move the playhead, click the waveform. Or give the waveform focus and use the Left and Right arrow keys. To play only one line, click the time at the start of the line. Playback moves by line, not by word.

To correct a line, double-click its text. Press EnterReturn to save, or EscapeEscEscape to cancel. The app marks the line edited. To get the original text back, click revert. The app writes your correction beside the text of the model, never over it.

The app highlights keyword hits in the text.

A take can show where you said each line. The app does not record your position. A place comes from a trace that you import. A trace is a file of positions with times, from a GPS logger, a watch, a phone app or a GoPro. Record it at the same time as the take.

The app aligns the trace with the audio by time. Then every transcript line and every marker has a place.

Import a trace

• A GPX or NMEA file: choose File → Import Trace…, then select the file. The app matches the trace to the takes of the project by time. Each take that the trace overlaps gets the part recorded during that take. To add the trace to only some takes, select them in Recordings, right-click, and choose Import Trace….
• A GoPro video: import it as audio. Choose File → Import Audio…, or drop it on Recordings. Its sound becomes a take, and its GPS data becomes the trace of that take in the same step.

From a GPX file, the app reads the track points that have a time. An NMEA file must have RMC sentences, because they contain the date. The app skips sentences with a bad checksum or no fix, and tells you how many it skipped. The app does not add a file to a take that already has it.

The app aligns a trace by the start time of the take, so the take must have a start time that the app can trust. Some imported recordings have only the time when their file last changed. These are untimed takes. A trace import skips them and tells you which takes it skipped.

Align the trace

If the clock of the logger was wrong, every place is early or late by the same amount. To correct this, set the trace offset:

• In the trace view, step the offset by 10 s or 1 s in either direction, or type the seconds. The reset button sets it back to 0 s.
• In Recordings, right-click a take and choose Trace Offset….

The offset belongs to the trace file. So it moves the places on every take that has the file. The app saves the offset when you change it.

A take can have more than one trace, for example one from a logger and one from a GoPro. One of them is the active trace. The active trace gives the lines and markers their places. The newest trace that you import becomes the active trace. To choose a different one, go to Traces in the Details tab of the inspector.

The trace view

To open the trace view, review a take and press the Trace view button at the top of the take. The trace view shows the trace in two ways:

• The distance strip shows the take along the distance travelled.
• The map shows the trace from above.

The map uses no map tiles. So it needs no network, and no request tells anyone where you were. While the take plays, a point moves along the trace. To go to a line, click its stretch of the trace, or its pin.

Each transcript line shows its place as latitude and longitude, with a button to copy it. To see the place of a marker, point to the marker on the waveform. Its tooltip shows the place.

Geo export

To export places, choose GPX or GeoJSON from the arrow beside Export:

• GPX writes a waypoint for each line and each marker, and the active trace as a track.
• GeoJSON writes the trace as a path, each transcript line as the stretch where you said it, and each marker as a point.
• JSON export also includes the place of each line and each marker.

If you export the whole project as GPX or GeoJSON, you get one file. The app skips the takes with no trace, and tells you their names.

If you open one of these files in a web map tool, the tool sends the file to its own server. The app tells you this when it exports.

What needs Pro

Trace import is part of Pro, from a file and from a GoPro video. On Free, a GoPro video imports as audio, without its trace.

A trace that a take already has stays at every tier. Its places, the trace view, the trace offset and geo export all work on Free.

Each project has its own keyword library. A new project starts with no keywords. The library is in the Keywords tab of the inspector. The inspector is the panel on the right of the window.

To add a keyword, press Add keyword at the bottom of the tab. The number beside each keyword is its hits in the take on screen.

The Details tab, next to Keywords, shows facts about the take. It shows when you recorded the take, its length, the device and format, the number of lines, and its folder on disk.

A keyword has a name and a list of phrases. Each phrase that the transcript contains is a hit. The app marks hits in the transcript while you record and when you review. If you change the library, the app matches the takes in the project again.

Each keyword has a colour. Choose one of five colours, or any colour you want. The app adjusts your colour so that it shows in every appearance. It also keeps the colour away from red, because red means recording.

When a line matches a keyword, the app also adds a marker on the waveform at that line. So you can mark a take by saying a keyword. If you change the library, the app adds markers to older takes for the new hits.

To use the library in another project, export it as keywords.json and import it there. Use the two buttons beside Add keyword.

The keywords.json file

A library file is plain JSON. You can write one by hand, or make one from a list that you already have. Export writes the list under keywords. Import also accepts the list alone, as a JSON array.

{
  "keywords": [
    {
      "id": "handling",
      "label": "Handling",
      "phrases": ["turn-in", "understeer"],
      "color": 1
    },
    {
      "id": "safety",
      "label": "Safety",
      "phrases": ["near miss", "hazard"],
      "color": "#3a7bd5"
    }
  ]
}

id

Required. A unique name for the keyword. An import uses it to find the same keyword in the library.

label

Required. The name that the Keywords tab shows.

phrases

Required. A list of the phrases that count as a hit. Matching ignores case.

color

Optional. 1 to 5 for Green, Amber, Violet, Blue and Slate, or your own colour as "#rrggbb". Any other value, or no value, gives Slate.

An import adds to the library and removes nothing. If a keyword in the file has the same id as one in the library, it replaces that keyword. The app adds the other keywords. It skips a keyword that has no id, label or phrases. If it skips every keyword in the file, it imports nothing and tells you why.

You can start, stop and mark a take without using the window. Use keys that work when another app is in front, or use your voice. Set both in Settings → Capture. Both are off until you turn them on.

Keys

In the Keyboard shortcut column, click Set shortcut beside Start a take, Stop the take or Add a marker. Then hold Ctrl⌘Command or Alt⌥Option and press a key.

A key alone, or a key with only Shift, is not a shortcut, and the row tells you so. A function key such as F5 works alone.

If another app already uses the combination, the row shows Not registered, another app is holding that combination. Choose a different combination. To clear a shortcut, click the × beside it.

Voice

Turn on Voice activation. Between takes, the app then listens for a start phrase. The default start phrase is “start recording”. During a take, a stop phrase ends the take. The default stop phrase is “stop recording”.

You can add your own phrases beside each action. Each phrase must have at least two words, so that usual speech does not start or end a take by accident.

The app listens only on this computer, and it needs a ready model. It writes nothing that it hears between takes to disk.

Start and stop are separate actions. If you say a phrase or press a key twice to be sure, the second one does not undo the first. To mark a moment by voice, say a phrase of a keyword.

A take is already plain files on disk. See what a take is on disk. Use export to give a transcript to a different tool. Choose the format for the job:

• SRT and WebVTT, for captions.
• JSON, for your own tools.
• Plain text, to read it or paste it into another tool.
• GPX and GeoJSON, for map tools, if the take has a trace. See places.

The app uses one format for all exports. The default is plain text. To change it, choose a format from the arrow beside the Export button of a take, or in Settings → App → Export format. The button shows the format, such as Export TXT.

• To export one take: press Export at the top of the take, or press Ctrl+E⌘CommandE (File → Export Take…). Then choose where to save the file. If no take is open, Ctrl+E⌘CommandE exports the latest take. You can also right-click a row in Recordings and choose Export….
• To export the whole project: press the export button at the top of Recordings. Or press Ctrl+K⌘CommandK to open the command palette, and choose the export command. JSON, plain text, GPX and GeoJSON write the whole project as one file. SRT and WebVTT write one file for each take, with times that start at zero.

Plain text contains only the transcript. To start each line with its time as [mm:ss], turn on Timestamps in plain text in Settings → App.

Export includes only transcribed lines. So a take with no transcript has nothing to export.

The model transcribes each line once, and the transcript includes your corrections. If you want a cleaned-up version or a summary, export plain text or JSON. Then give it to the AI tool that you already use. You do this step yourself: the app never sends a transcript anywhere.

This list shows the shortcuts for your computer. On a Mac, press Command where other computers use Ctrl. For more commands, press Ctrl+K⌘CommandK to open the command palette.

Ctrl+K⌘CommandK

Open the command palette

Ctrl+Shift+R⌘Command⇧ShiftR

Start or stop recording

Ctrl+Shift+P⌘Command⇧ShiftP

Pause or resume

MM

Add a marker while recording

Ctrl+Shift+M⌘Command⇧ShiftM

Add a marker, even while typing in a field

Ctrl+Shift+K⌘Command⇧ShiftK

Run the pre-flight check

Ctrl+E⌘CommandE

Export the take on screen, or the latest one

Ctrl+Alt+S⌘Command⌥OptionS

Show or hide the recordings list

Ctrl+N⌘CommandN

Create a project

Ctrl+O⌘CommandO

Open a project

Ctrl+Shift+O⌘Command⇧ShiftO

Open the library folder

Ctrl+,⌘Command,

Open Settings

Ctrl+Shift+D⌘Command⇧ShiftD

Change the appearance: day, dusk or night

←Left arrow←Left arrow

Move the playhead back, when the waveform has focus

→Right arrow→Right arrow

Move the playhead forward, when the waveform has focus

EnterReturn

Save a correction

EscapeEscEscape

Cancel a correction

A take is a folder of plain files. No file uses a proprietary format. The take is in a project folder, which you can copy, archive or give to another person.

New projects go in the amlistening folder in your Documents folder. To open that folder, choose File → Open Library Folder, or press Ctrl+Shift+O⌘Command⇧ShiftO.

proving-ground/
  project.json
  sessions/
    2026-08-05_1042_damper-sweep/
      session.json
      audio/000000.wav
      events.jsonl
      transcript.jsonl
      gps.jsonl

session.json

The name, project and state of the take, and its list of audio chunks.

audio/000000.wav

The audio, in append-only chunks.

events.jsonl

The markers, stored as sample offsets.

transcript.jsonl

The transcript lines, stored as sample offsets. A correction is a new row beside the original row.

gps.jsonl

The traces imported onto the take, their offsets, and which trace is active. The file appears when you import a trace.

The files are the source of truth. The app rebuilds everything else from them. If the app stops during a take, see recovery after a crash.

The app measures time in a take by audio samples, not by the clock. So markers and transcript lines stay at the right moment, even if the clock of the computer changes suddenly.

When you delete a take or a project, the app moves it to the Trash or the Recycle Bin.

If the app or the computer stops during a take, the take is already on disk, except for its last few seconds. The next time the app starts, an alert lists each interrupted take with a Recover button.

Recover repairs the header of each audio chunk from the length of its file. Then it rebuilds the take from its chunks. This works even if session.json is missing or incomplete. Recover reads nothing from the project file, so a damaged project file cannot stop the recovery of a take.

Then the app finishes the transcript in the background, from the audio on disk. It transcribes the lines that were still waiting. It also transcribes speech that has no line because of the stop, such as the sentence you were saying.

If you quit while a take has lines still waiting, the app finishes them the same way the next time you open the take.

Capture, transcription, review and export need no network, at any tier. You need a network for only three things:

• To download the transcription model, on the first run.
• To sign in and activate a Pro licence.
• To get updates.

Your recordings, transcripts, markers, keywords and exports stay on your computer. The app uploads none of them, at any tier. The app has no telemetry.

Four things use a network. None of them can stop you from recording a take:

Model download

This happens when you press Download in Settings → Transcription. The app downloads the model file. The server that sends the file sees the connection data that it needs to answer, including your IP address. The app sends no information about you or your recordings.

Sign-in

This happens when you press Sign in. You sign in on amlistening.app in your browser, with your email and password. Where it is available, you can use your Google account instead. The app then gets a licence. The licence contains a random ID that the app made for this computer. The ID is not the name of your computer or your user name. The licence also contains the email that you signed in with, so the app can show which account it belongs to. The app keeps the licence in a file on this computer. It sends the licence only to amlistening.app, when the licence refreshes.

Licence refresh

This happens only after you sign in. When the app starts, and every few hours, it checks whether its licence is past half its life. If it is, and a network is available, the app replaces the licence with a new one.

Update check

This happens when the app starts, and when you check manually. To check manually, use Settings → Updates, or Check for Updates in the app menu on a Mac. The app asks the release server whether a newer version exists. If there is no network, the check fails and shows no message. The app downloads an update only when you tell it to. If you turn on Update automatically, the app downloads updates itself and also checks every six hours.

Free needs no sign-in. So a Free installation uses the network only for the model download and the update check.

The app makes the update check and the licence refresh without you asking. To stop them, turn on Work offline under Network in Settings → App. An organisation can set this for every user of a computer. For an IT review, the security page lists every host that the app can contact.

Free gives you 10 takes of 5 minutes each. It needs no sign-in and does not expire.

Free counts the Free takes that are in the library now. If you delete one, you can record another. Takes that you record on Pro never count. When the library has 10 Free takes, Record opens an upgrade prompt and does not start a take. A Free take stops automatically at 5 minutes.

Pro has neither limit, and it can import traces. See places. The pricing page shows the plans and prices.

Playback, transcripts, corrections, tags and export work at every tier. The places from a trace that a take already has also work at every tier.

You sign in to get a Pro licence into the app. Free needs no account.

1. Buy Pro on the website.
2. In the app, click the account button at the bottom of the left rail, above Settings.
3. Press Sign in. Your browser opens amlistening.app.
4. Sign in there. The app then gets a licence for this computer.

To sign out, use the same account button. Signing in never uploads a recording.

The app checks the licence on your computer, with no network. Each licence lasts 30 days from when it was issued. After half that time, the app renews the licence automatically when a network is available. If the licence expires while you are offline, the app works as Free until it can renew the licence.

If your subscription ends, the app changes to Free when its licence expires. Nothing is locked. You can still read, play, correct and export every take that you recorded. The Free limits apply only when you record new takes.

Updates are free at every tier. All the update settings are in Settings → Updates. When the app starts and a network is available, it checks for a newer version. If there is no network, it shows nothing.

You can also check manually at any time in Settings → Updates. On a Mac, you can also use Check for Updates in the app menu. If Work offline is on, the app never checks, and the check button is disabled.

When a new version is available:

• A dot shows on the Settings button at the bottom of the left rail.
• The status bar tells you.
• Settings → Updates shows Download update.

Nothing downloads until you press Download update. Restart to update waits while a take records. If you do not restart, the downloaded update installs when you next quit.

If you do not want to check yourself, turn on Update automatically. The app then checks when it starts and every six hours. It downloads a new version in the background when a network is available, and installs it when you next quit. The app never restarts itself, so it never interrupts a take. Update automatically is off until you turn it on.

Installing

Windows warns that the installer is from an unknown publisher.

The installer is not signed yet. Choose More info, then Run anyway.

The AppImage does not open.

The file is not executable yet. Run chmod +x on the file, then open it again.

Recording

Record shows Recording didn't start.

The red banner names each problem that stopped the take. Fix each problem, then press Try again. If the banner says there is no input device or no signal, see “The level meter does not move” below. If no project is open, Record waits for one, because a take belongs to a project. Create a project with Ctrl+N⌘CommandN, open one with Ctrl+O⌘CommandO, or choose one from File → Open Recent.

Record opens an upgrade prompt.

The library already has 10 Free takes. Delete one to record another, or upgrade to Pro.

The level meter does not move.

The meter moves only while the microphone is open: during a take, or after you press Check input. Check the microphone in Settings → Capture. If you connected a microphone after the app started, choose Record → Rescan Input Devices. Then check that Amlistening has permission to use the microphone. On a Mac, look in System Settings → Privacy & Security → Microphone. On Windows, look in Settings → Privacy & security → Microphone, and allow desktop apps.

A wireless microphone does not appear.

Turn on the kit and connect its receiver to the computer. Then choose Record → Rescan Input Devices, and choose the receiver in Settings → Capture.

A hands-free key says it is not registered.

Another app already uses that combination. Choose a different combination in Settings → Capture. When the system gives the key to Amlistening, the warning goes away.

Saying the start phrase does nothing.

Check that Voice activation is on in Settings → Capture. Check that a model is ready, because the app listens with the model. The app listens only between takes, and only for a full phrase of at least two words. Check that the level meter moves when you speak.

Transcription

You recorded speech, but no transcript appears.

Open Settings → Transcription. Check that a model is downloaded and selected. The app marks speech recorded with no model as failed. To transcribe it, select the take and press Retry. If a model is ready and no speech is marked as failed, check the input. A take with a silent input shows a No signal warning. Between takes, press Check input to see the level.

A model download stopped.

Press Resume. If the download failed, press Retry download. Both continue from where the download stopped.

Something else

The app closed during a take.

Open the app again, and press Recover on the alert. See recovery after a crash.

Nothing here solves the problem.

Open About Amlistening. On a Mac, it is in the Amlistening menu. On Windows and Linux, it is in the Help menu. Press Copy details to copy the version, platform, model and library location to your clipboard. Paste them into an email to support@amlistening.app and tell us what happened. Nothing leaves the app until you send the email.

The app has no sync, no web playback and no collaboration. It is for one user, with no seats, no server and no sharing. Recordings stay on your computer.

The app has no forms, templates or metadata fields. A take has a name, tags, and the traces that you import. The app does not track where you are. A place comes only from a trace.

The app has no speaker labels, so it does not show who said a line. It records only a microphone input. It does not record the sound of other apps, and it does not join a call or a meeting.

The app has no second pass and no summary. The model transcribes each line once: while you speak, or in the background for an imported file. Nothing changes the line after that. You can correct any line yourself, or export the transcript and give it to another tool.