🚀 Beta Test – Murmure 1.9.0

[Kieirra]

Hello everyone,

The beta testing phase for Murmure 1.9.0 is now open. As with the previous round, this is an open beta, anyone who wants to test the pre-release version is welcome, and we want to gather as much feedback as possible before the official release.

📦 Binaries to test

All binaries are available on the pre-release page: https://github.com/Kieirra/murmure/releases/tag/1.8.100

• 🐧 Linux

  • Murmure_amd64.deb (.deb, Debian/Ubuntu)
  • Murmure_amd64.rpm (.rpm, Fedora)
  • Murmure_amd64.AppImage (AppImage)

• 🍎 macOS

  • Murmure_aarch64_darwin.dmg (Apple Silicon)
  • Murmure_x86_64_darwin.dmg (Intel)

• 🪟 Windows

  • Murmure_x64.msi (.msi)
  • Murmure_x64-setup.exe (.exe)

✅ How to test

The full test plan, with what's new in 1.9.0 and the checklist of things to try, is available in the docs:

• 🇬🇧 English: https://docs.murmure.app/beta-testing/
• 🇫🇷 Francais: https://docs.murmure.app/fr/beta-testing/

No need to open a GitHub issue, just reply directly in this discussion with your feedback, what worked, what did not, and the usual info (OS, version, steps to reproduce, logs if you have them).

❤️ Thank you

Thanks to everyone who takes the time to test and help make this release rock-solid.

[laurentkempe]

Hey @Kieirra
Thanks you very much for this new version. It is incredible how you can still surprise me with some of the new features you are bringing to Murmure. A truly fantastic job 👍🏼

OS: Windows 11 Pro 25H2
Version: 1.8.100 (beta)

Transcription

• Record and transcribe in push-to-talk
• Record and transcribe in toggle-to-talk
• Test on a short phrase (1 to 2 words)
  Has some difficulties with the language sometime between French and English, or even on the words.
• Test on a longer dictation (2 to 3 sentences)
• Test a transcription with LLM post-processing

Overlay

• Verify the overlay appears during recording
• Enable streaming mode and verify it works correctly
• Customize the overlay and verify the settings are properly applied
  I had all settings to the maximum, then I use it for 2-3 phrase so that it overflow the height and the text was not scrolling.

Let's call it when you use an agent in VS Code: a tool-calling loop happens, a think-acts-observe-think-again cycle. On each iteration, the agent builds the prompt: system instruction plus context plus story plus all tool results so far, sends it to the model, and checks the response. If the response includes tool calls, execute those tools, capture the result, and loop back. If there are no tool calls, the loop can finish and the system text becomes the final response.

Voice Mode

• Enable voice mode from the Extensions menu
• Trigger a recording by saying the wake word
• Test auto-send with "Thanks alix" after a voice transcription
  I would not expect to have the Thanks Alix at the end of the text. I have the feeling it is more the timeout that work than the auto-send

Ceci est un test pour voir la compatibilité de Murmure avec ma machine Windows. Thanks Alix.

• Test with the silence delay set to Indefinite

Thanks Alix was detected but the session was not ended. I did set silence timeout to Indefinite.

• Verify voice mode toggles correctly via Ctrl+Shift+0
  Works for 0 on the main part of the keyboard, but not the one on the numpad.

Smart Speech Mic

• Enable Smart Mic and scan the QR code with your phone
  Works very nicely. I would expect to have the text not only on the phone but also on murmure running on the PC in the Recent Activity.
• Verify that audio from the phone reaches Murmure
• Run a transcription using the phone microphone
• Test the left-click, delete and enter actions from the phonpour voir si ça marche correctement. Qu'est-ce que c'est que le

Settings Import/Export

• Export all settings
• Export only a selection of settings
• Change a setting, then re-import the exported file
• Verify that settings are properly restored
• Test CLI import: murmure import

Other

• (macOS) Verify Escape no longer blocks other apps from closing when no transcription is running
• Verify Escape cancels the ongoing recording
• Rename a custom formatting rule

Another fantastic release, I could not find anything really important. I will test it more next week and if I find something come back to log it here.

Have a great weekend @Kieirra

[Kieirra]

Thank you very much for the tests and the kind words.

I had all settings to the maximum, then I use it for 2-3 phrase so that it overflow the height and the text was not scrolling.

I will investigate this issue. It should indeed scroll automatically.

Thanks Alix was detected but the session was not ended. I did set silence timeout to Indefinite.

Could you check if the Submit wake word is "Thanks Alix" rather than "Thank you Alix" ?

Does "Thanks Alix" sound smoother and more natural than "Thank you" ? Maybe I should change the default wake word then.

Works very nicely. I would expect to have the text not only on the phone but also on murmure running on the PC in the Recent Activity.

I didn't think about it because since I put the visualizer on the smartphone, I thought it was enough, but in reality, users probably look at the computer screen. Not the smartphone screen when talking. I will think about it.

[Zabalero]

Awesome work @Kieirra
Thanks for it!! ❤️

[laurentkempe]

Could you check if the Submit wake word is "Thanks Alix" rather than "Thank you Alix" ?

You are right, it is "thank you alix" and it works 👍🏼

Does "Thanks Alix" sound smoother and more natural than "Thank you" ? Maybe I should change the default wake word then.

No, I think it is better like that. I should have been more careful!

[thiswillbeyourgithub]

Something small but when giving a try to the ctrl+space shortcut in the "push as long as you talk" configuration, I noticed that, although it worked the record, the space bar was still captured by the UI. So because what I did JUST BEFORE was ctrl+space was clicking on the "overlay size selector" picker in the settings window, I saw it flicker back and forth very quickly. This could be a bit confusing to noobies maybe? This might be caused by my DE though (ubuntu using i3)

[Kieirra]

Thanks for the report, I'll have a look to see if I can do something about it.

[anouar-coscientia]

Hello @Kieirra

This is my first beta testing on GitHub ever.
I'm thrilled because I love Murmure and use it everyday.
Hopefully, you'll find useful feedbacks.
A second test on Windows 11 is on the run.

Environment

Operating System: Ubuntu 24.04.4 LTS
Kernel: Linux 6.17.0-29-generic
Architecture: x86-64
Desktop env: GNOME 46.0 + Wayland

Transcription¶

• Record and transcribe in push-to-talk (I've tried... it failed, because of Wayland)
• Record and transcribe in toggle-to-talk
• Test on a short phrase (5 to 6 words)
• Test on a longer dictation (2 to 3 sentences)
• Test a transcription with LLM post-processing [1]

[1] LLM Connect

Who's who ?

This was confusing on Linux... As far as I understood, I think...

Ctrl + Alt + Space = murmure --transcription-llm
Ctrl + Shift + X = murmure --transcription-command

It could use some disambiguation.

Cannot use Ctrl + Shift + X

After binding the shortcut, I've needed to restart my computer in order to use the transcription-command. My use case was "take the text in the clipboard and translate it to German".

However, I could not run your example (to select text and to ask the LLM to translate it). I've tried from the terminal, note editor, web browser...

Am I missing something ? Could you provide a wider set of examples, and actions we could perform with this feature ?

About the shortcuts...

It's a detail, but yet it bugs me.
If Ctrl + Alt + Space is supposed to run murmure --transcription-llm, there might be a missed opportunity.

Because this shortcut pairs with Ctrl + Shift + 1, Ctrl + Shift + 2... So it'd better be Ctrl + Shift + X. All of which share Ctrl + Shift. Thus leaving Ctrl + Alt + Space to execute commands.

UX-wise now...

1. Adding the corresponding shortcut to prompt text boxes gave me guidance to get this feature. It's been appreciated.
2. Models' recommendations during the onboarding is also a good idea.
3. The most clever was the various color tones (blue, orange, purple) as a visual feedback for each use cases (transcript, LLM connect + prompt, command).
4. And thanks for your prompt suggestions (Cursor Developer, Translation etc.) !

Overlay¶

• Verify the overlay appears during recording [2]
• Enable streaming mode and verify it works correctly [2]
• Customize the overlay and verify the settings are properly applied

[2] OVERLAY APPEARANCE AND POSITION

What worked

The "overlay position" (top / bottom) worked relatively to the preview position.

If "streaming mode" is Text insertion mode set to "Direct (type text)", then it worked. Otherwise, I don't know what you mean.

What didn't work

The overlay is stuck at a surprising position.
Neither restarting Murmure nor rebooting my computer solved this.
In case it help, this overlay position may be linked to where my mouse cursor was, in the terminal, the first time I ran the transcription shortcut with this release.

Voice Mode¶

• Enable voice mode from the Extensions menu
• Trigger a recording by saying the wake word [3]
• Test auto-send with "Thanks alix" after a voice transcription [3]
• Test with the silence delay set to Indefinite [4]
• Verify voice mode toggles correctly via Ctrl+Shift+0 [5]

[3] VOICE MODE

What worked

Tests were done in a quiet environment (home), and a noisy environment (running train). It worked fine both times.

Beware of gain control

Users should note their microphone should be set with sufficient input gain (if such option is available).
On Ubuntu, my integrated laptop mic was set at 1/3rd of gain. Surprinsgly, it was...

• Enough to transcribe once Murmure was listening...
• But not enough to detect trigger words.
  I still can't make sense of this.

UX speaking now...

1. It took me time to finds words combination which felt natural, memorable, non-trivial. It was not a feature I expected... and now that it's shipped, it is a nice-to-have.
2. I don't understand the difference between "Validate" and "Submit" trigger words. In a text editor or terminal, both resulted in the text being pasted. What's the difference supposed to be?

[4] SILENCE DELAY

I stopped my step after 60s of silence. I'll let you know if I take the time to leave my computer 8h during sleep time with Murmure listening to the sound of silence.

[5] VOICE MODE SHORTCUT

The command "murmure --voice-mode" worked in the terminal and once binded to a shortcut.

Smart Speech Mic¶

• [/] Enable Smart Mic and scan the QR code with your phone [6]
• Verify that audio from the phone reaches Murmure
• Run a transcription using the phone microphone
• Test the left-click, delete and enter actions from the phone

[6] SMART MIC

Flashing the QR code lead me to a page with an en error (ERR_CONNECTION_REFUSED).
I've shut down my VPN from mobile and laptop, tried to open the link from 2 different mobile browsers (Brave, Vanadium) with 3 different QR codes...

Couldn't get any further for now. Let me know if I can help in any way.

Settings Import/Export¶

• Export all settings
• Export only a selection of settings
• Change a setting, then re-import the exported file
• Verify that settings are properly restored [7]
• Test CLI import: murmure import

[7] SETTINGS RESTORATION

I've set some wrong values on purpose to watch Murmure behaviour.
It's nice that the wrong section without blocking the valid settings sections.

Other¶

• (macOS) Verify Escape no longer blocks other apps from closing when no transcription is running
• Verify Escape cancels the ongoing recording [8]
• Rename a custom formatting rule

[8] ESCAPE

Esc. key don't stop Murmure. I've tried hard to make this shortcut worked.
I've tried to pair it with Ctrl, or Alt but Ubuntu won't let me create Ctrl + Esc.
Only Ctrl + Space (to start / stop) Murmure works. No Esc.

Misc.

Love the redesigned menus. It's slick and neat!

Linux support is awesome. I use both Win11 and Ubuntu on a daily basis. It's nice to have some common tools and workflows between both. And Murmure is definitely among those.

Feel free to ping me for future tests on both present and future releases.

[Kieirra]

Hi @anouar-coscientia,

Thanks a lot for the very thorough feedback, super useful.

A few answers point by point:

• Push-to-talk on Wayland: yes, known limitation. The XDG portal doesn't reliably propagate press/release events, so hold-to-talk can't really work today. I'll disable it on Wayland for the next beta so it's not selectable anymore, and toggle-to-talk stays the recommended mode.
• LLM Connect vs Select Command: they're two different things, I'll make this clearer in the UI.
  • LLM Connect works like the standard transcription, except the result goes through a preset prompt (translation, reformulation, custom...).
  • Select Command is live: you select some text, trigger the shortcut, and dictate what you want done with it. Example: select an English paragraph, trigger the shortcut, say "translate to French", and the selection gets replaced by the French version. No preset, you tell it what to do each time.
• Overlay stuck position: known issue on GNOME Wayland. The lib I use to position the overlay (gtk-layer-shell) isn't supported by Mutter, so the overlay is actually set to "hidden" by default on GNOME for that exact reason. If you enabled it manually, that's why it ends up misplaced. There's not much I can do until GNOME picks up wlr-layer-shell or I find another path.
• Voice Mode wake words: Validate stops and transcribes, Submit transcribes and presses Enter. Submit is particularly useful in chat apps (Claude Code, ChatGPT, etc.) to send the message hands-free and get the reply right away.
• Smart Mic ERR_CONNECTION_REFUSED: most likely a Wayland compatibility issue, the required authorisations aren't there. I'll disable Smart Mic on Wayland for the next beta until I find a proper fix.
• Escape: noted, I'll change the default Cancel key on Wayland.

Thanks again, I'll ping you for the next rounds.

[anouar-coscientia]

Your feedback on LLM Connect vs Select Command and Voice Mode wake words is what I was looking for! Thanks! Let's keep in touch.

[tomsquest]

Test Plan

• v1.8.100
• Kubuntu 26.04

Transcription

• Record and transcribe in push-to-talk
• Record and transcribe in toggle-to-talk
• Test on a short phrase (5 to 6 words)
• Test on a longer dictation (2 to 3 sentences)
• Test a transcription with LLM post-processing

Overlay

• Verify the overlay appears during recording
• Enable streaming mode and verify it works correctly
• Customize the overlay and verify the settings are properly applied

Voice Mode

• Enable voice mode from the Extensions menu
• Trigger a recording by saying the wake word
• Test auto-send with "Thanks alix" after a voice transcription
• Test with the silence delay set to Indefinite
• Verify voice mode toggles correctly via Ctrl+Shift+0

Comment:
I did not try the shortcuts (Wayland/KDE, weird binding, more details below)

Smart Speech Mic

• Enable Smart Mic and scan the QR code with your phone
• Verify that audio from the phone reaches Murmure
• Run a transcription using the phone microphone
• Test the left-click, delete and enter actions from the phone

Comment:
Phone did not connect (the connection timed out). Could be related to the firewall of my computer.

Settings Import/Export

• Export all settings
• Export only a selection of settings
• Change a setting, then re-import the exported file
• Verify that settings are properly restored
• Test CLI import: murmure import

Other

• NA (macOS) Verify Escape no longer blocks other apps from closing when no transcription is running
• KO Verify Escape cancels the ongoing recording
• Rename a custom formatting rule

Comment:

• Escape does not work (it's not bound in KDE shortcuts because it's a global shortcut so it is preventing the use of the Escape key in other apps).
• Renaming a rule set the replacement to be removed (I mean, given a rule foo -> bar, when renaming the rule to qix, then the new rule is just qix)

Shortcuts

I got strange behavior with the shortcuts.

1. The shortcut to record is not initially working. I have to restart Murmure to make it work
2. Sometimes the portal window to bind the shortcut is re-opened when Murmure starts. I don't think it's always the case (I have to restart the computer to reproduce, tedious). Maybe it will be fine with the improvements you talked about.

I just tested in CLI mode, and it seems to be a very good alternative, and one I am used to. 👍

[Kieirra]

Hi @tomsquest,

Thanks a lot for the detailed test plan.

A few answers:

• Rename rule: that's actually the intended behaviour. When a rule has no custom name, the summary shows the trigger -> replacement. When you give it a name, the summary shows the name only, and the trigger -> replacement moves to a tooltip on hover. The data itself isn't touched, just the display.
• Escape: that's the second time this comes up (also raised just above by @anouar-coscientia), so clearly a real annoyance. I'll change the default key, noted.
• Smart Mic timeout: same story, two reports now (see @anouar-coscientia's feedback above). It really doesn't work on Wayland today, so I'll disable it on Wayland for the next beta until I find a proper fix.
• Shortcuts not working on first launch: I haven't reproduced this exactly, but it lines up with a known upstream issue in the portal binding library on GNOME 48 (silent BindShortcuts failure). I'll dig into adding a retry on the initial registration so a transient portal hiccup doesn't require a restart.
• Portal dialog re-opening after reboot: that one is more on the compositor side. The portal can re-prompt for binding confirmation after a clean reboot, especially on KDE, and there's not much Murmure can do about it directly. Worth keeping an eye on though.
• CLI mode: glad it works that well. I'm actually wondering if it shouldn't become the default. For 1.9.0 it stays as an experimental option, and I'm keeping the door open to drop the XDG portal path entirely later if it keeps causing too many issues.

Thanks again, feel free to ping me for the next rounds.

[Kieirra]

Beta is now closed. Murmure 1.9.0 is officially out: https://github.com/Kieirra/murmure/releases/tag/1.9.0

Thanks a lot to everyone who tested and reported here, your feedback shaped this release. The major pieces landed: Smart Speech Mic, native Wayland support, and the full overlay overhaul, along with a lot of the smaller fixes you spotted.

You are all credited in the changelog. See you for the next beta round.

[tomsquest]

Many thanks @Kieirra 👏

[anouar-coscientia]

Please was all mine @Kieirra , thanks for this nice release. Kudos to @BorisLord for implementing Wayland support!

[Kieirra]

Hi everyone, and thanks again for testing 1.9.0.

Good news: the beta for 1.10.0 is now open. This round is focused on the core. A new custom Parakeet model and better audio resampling make transcription noticeably more accurate (especially in French), long dictations are now processed in the background while you speak so the result is ready much faster, and dictionary words now change the model's tokens directly for far fewer mistakes on your own jargon.

If you want to keep helping, the new beta and its full test plan are here:
👉 #381

Thanks a lot for your feedback so far, it really makes a difference.