Gilune

Capture & voice

The quickest way into Gilune is to say the thing out loud. Your words are turned into text on your own machine, and the recording itself is kept.

A new note from speech

The microphone button sits towards the right-hand end of the top bar, beside the refresh and settings buttons.

  1. Click it. The status bar confirms it is recording, and the microphone turns red.
  2. Say what you want to keep.
  3. Click it again to stop.

Gilune saves the recording, transcribes it, and opens the note it made. The words are the body of the note and the audio stays beside it as an attachment, so you can always go back to what you actually said. Fix the text however you like — it is just a note from that point on.

While a recording is being transcribed, the line at the bottom of the window says which model is listening and how far through it is — transcribing with Parakeet TDT 0.6B v2 · 42% — and then labelling speakers… if speaker labels are on. When the note is written, the model is put away again, so the memory it needed is given back straight away.

Dictating into a note you already have

Open a note and look at the row along the top of the page, beside the ⋯ menu: there is a second microphone there. This one types for you.

Put your cursor where the words should go, click the microphone and speak. The words appear at the cursor as you talk. Words Gilune is still unsure of are shown grey and may change as you carry on; once it is sure, they turn into ordinary text in your note. Grey words are never saved: if you close the app mid-sentence, the file holds only the words that had settled, and the recording itself is kept either way. Click the microphone again to stop, and the last words settle within a second or two. If you had text selected, the dictated words replace it. You can keep typing elsewhere in the note while you dictate; the words keep arriving where you started.

Live words come from the engine chosen under Settings ▸ Transcription ▸ While you speak. Choose Off there and the words all arrive when you stop instead. Live words ask a lot of the processor; if your computer cannot keep up, Gilune says so in the status bar and the rest of the words arrive when you stop.

While either microphone is recording, a few small bars beside it move with your voice, so you can see it is hearing you before any words appear. If they stay flat for a few seconds, Gilune tells you the microphone is not hearing anything: check it is not muted and that the right input device is chosen.

You can only dictate into one note at a time; if you try to start a second one, Gilune says so and leaves the first alone. And if you close the note before the transcription comes back, the words are not lost — they appear in the status bar so you can copy them.

Dictation needs the speech model to be in place (see below). Until it is, the microphone is greyed and won't start: hover over it, or click it, and it says what is missing — usually permission to download the model.

The recording is always saved before anything else happens. Transcription is a convenience laid on top. If the microphone fails, or the transcription is unavailable, or the text comes out wrong, the audio you recorded is already safely on disk.

Capturing on the phone

The phone app opens straight onto a composer, so the first thing you can do is tap it and type. While the keyboard is up the row of tabs along the bottom steps aside to make room. The tick beside the composer keeps what you have written and puts the keyboard away, which brings the tabs back. If you want the keyboard down without keeping anything yet, the keyboard button next to the tick does that and leaves your words where they are.

Transcription runs on your machine

Your speech is not sent anywhere. Gilune uses a speech model that lives on your computer, and once it is in place everything happens offline.

That model is about 57 MiB on the desktop and smaller on the phone, and it is not fetched behind your back. Nothing downloads until you open Settings ▸ Plugins, open the transcription plugin, and allow it network access. The download starts as soon as you allow it. Until you do, the status bar tells you plainly that transcription is waiting for permission, and both microphones say so when you hover over them — and everything else in the app carries on working. Once the model is there, the download never happens again.

The model only takes up memory while it is being used: Gilune loads it for the first recording and lets it go again after ten quiet minutes. The recording that wakes it takes about a tenth of a second longer than the ones after it.

Choosing what hears you (desktop)

Transcription does two jobs, and Settings ▸ Transcription asks one question for each:

The two can come from different plugins: a fast live model for dictation and a slower, more accurate one for recordings. Each row names its plugin and says whether it is Ready, still needs a download, needs network access, or belongs to a plugin that is switched off (choosing it switches that plugin on). If you choose another plugin's model for recordings, the recordings still waiting for words go to it, and the status bar says how many. Whisper uses one model at a time, so if both answers are Whisper they use the same model.

The desktop app comes with two transcription plugins:

Choosing a model that is not on your computer yet opens its plugin's own screen, where you allow network access and see the download and the model's licence. Each model downloads once and is checked against its published checksum before it is used. A plugin's own screen also says what it is used for at the moment.

Switching a transcription plugin on in Settings ▸ Plugins no longer switches the other off: the one you switched on is simply there to be chosen. Switching off the plugin chosen for recordings leaves recordings waiting, kept safely, until you choose another engine or switch it back on.

Setting the spoken language

On the transcription plugin's own settings screen there is a Spoken language field. Left at auto, Gilune works out the language first, which costs an extra pass over the audio and is noticeably slower on short clips. Set it to a language code you actually speak — en, sv, de and so on — and transcription is quicker. The change takes effect the next time you start Gilune.

When a recording has no words in it

Sometimes there is nothing to transcribe — the microphone did not pick you up, or the recording is a few seconds of a quiet room. Gilune says no speech heard beside that recording rather than leaving it looking like it is still being worked on. The recording itself is kept and still plays; there simply were no words in it.

Transcribing a recording again

If the words came out wrong — you have since switched to a better model, or set the spoken language — you can have the recording heard again. In the sidebar, click the little arrow beside the note to show its recording, right-click the recording and choose Transcribe again. The new words replace the note's text.

If you have changed the note's text since it was transcribed, Gilune asks before replacing it. Your current text is kept in the note's history inside Gilune, but nothing in the app shows that history yet, so copy anything you want to keep before you say yes.

Transcribing a recording made somewhere else

A recording from your phone's voice recorder, a meeting tool or anywhere else can become a note too. Click the audio file button beside the microphone in the top bar and choose the recording. Gilune makes a note named after the file in your Captures folder, transcribes it and opens it. To put the note straight into a particular folder, right-click that folder in the sidebar and choose Transcribe a recording… instead.

Gilune reads WAV and AIFF recordings, whatever their quality or number of channels, and converts them to the form its transcribers use; the converted copy is what sits beside the note, and the file you chose is left where it was. MP3 and M4A files are not read yet — convert them to WAV first. On the phone, this is not available yet.

Who said what (desktop)

When a recording has more than one person speaking — a meeting, an interview, a conversation — Gilune can split its note by speaker:

**Speaker 1:** Right, shall we start with the budget?

**Speaker 2:** Yes — I've got the numbers from last week.

Switch it on under Settings ▸ Plugins ▸ Speaker labels and allow it network access; it downloads two small models (34 MB together) once and runs them on your computer. That screen also lists which transcribers it works with — on the desktop, both Whisper and NVIDIA NeMo, with any of their models — and each transcriber's own screen says whether it supports speaker labels. Labels follow the engine chosen for recordings; Settings ▸ Transcription says whether that engine works with them.

Labels are added when a recording is transcribed, after you stop: the words still appear as you speak, and the speakers are worked out from the whole recording at the end. A recording with one voice reads exactly as before, with no label, and dictating into a note you are writing is never labelled. Gilune works out how many people are speaking; if it keeps finding the wrong number in your recordings, enter it in Number of speakers. To label a recording made before you switched this on, use Transcribe again on it.

Speakers are numbered in the order they first speak; rename them in the note however you like. Voices that sound alike, or people talking over each other, can be split wrongly. On the phone, speaker labels are not available yet.

Catching up on recordings that were missed

If you recorded something while transcription was switched off, or before you had installed it, those recordings are still sitting in your vault untranscribed — they were saved, they simply were not processed.

To deal with them, go to Settings ▸ Plugins and press Catch up missed captures. It works through anything that was recorded before a plugin was available or while it was off. It is safe to press more than once; it will not transcribe the same recording twice.

Where recordings live

Notes made from speech land in a Captures folder in your vault, and the sidebar shows a count beside it so you can see what has come in and not yet been filed. Move them wherever you like — they are ordinary notes.

The audio file sits next to its note and shows up under it in the sidebar when you click the little arrow. Renaming or moving the note in Gilune takes the recording with it. Deleting a note leaves its recording on disk deliberately: the recording is the original, and Gilune does not throw originals away.

When something goes wrong

On the desktop, Gilune writes its warnings and errors to a log file on your computer — on Windows, %LOCALAPPDATA%\Gilune\logs\gilune.log. Nothing in it is sent anywhere. Settings ▸ About ▸ Show the log file opens the folder; if a transcription fails or something else misbehaves, that file says what happened, and it is the thing to attach when you report a problem.

If the microphone button is not there

The microphone is contributed by the transcription plugin. If it has been switched off or removed, the button is not drawn at all — which is the honest thing to do, since nothing would be listening. Turn it back on under Settings ▸ Plugins. More about plugins →