There is definitely a way to make this MUCH faster

[Boruch-Baum]

Several YEARS ago, I installed a package called 'nerd-dictation' and I'm still using it. It is much faster and more responsive than vocal linux. In fact, it almost instantaneously transcribes audio, and is more accurate. When I evaluated vocal linux, I began with the recommended whisper.cpp back-end and suffered seconds-long delays. I then went "model shopping" for better results, but to no avail. I've tried vocal linux with larger whisper models. I've tried setting the config for "English-only". I've even tried using the same model that I'm using for 'nerd-dictation', namely VOSK. Vocal linux is always so much agonizingly slower. This indicates to me that the issue is unrelated to the model being used, but is rather the slowness of 'vocal linux' post-processing. Clearly, 'nerd-dictation', for all its faults and limitations, is doing something right that vocal linux has gotten terribly wrong.

Do you all think that you're "not allowed" to study and learn from the nerd-dictation source code? You are allowed.

I like the vocal linux GUI. It's much better than the extremely primitive one that I wrote for my 'nerd-dictation' fork, but that's not enough to make me keep vocal linux installed. BTW, the uninstall script missed deleting the ~/.local/share/vocallinux-ibus directory.

[jatinkrmalik]

Hi @Boruch-Baum ,

Thank you for trying out Vocalinux and for the feedback regarding the GUI. I also appreciate the heads-up about the uninstall script missing the ~/.local/share/vocallinux-ibus directory. I will look into getting that fixed.

Regarding the performance issues and your comments on the post-processing: it is important to remember that we are all learning here. I am not a specialist in dictation software or audio processing. Vocalinux was originally built purely as a personal project, and it has only grown to this point because of community interest and support. (:

Since you are familiar with nerd-dictation's source code and have a clear idea of what it is doing right that Vocalinux is getting wrong, please don't let anything stop you from submitting a patch or a pull request to help improve it?

Open-source projects thrive on active collaboration rather than just comparison. If you see a clear path to making the post-processing much faster, I would gladly welcome your contributions so the whole community can benefit.

[Boruch-Baum]

On 2026-07-20 22:17, Jatin K Malik wrote: Since you are familiar with nerd-dictation's source code

Where did I write that?

and have a clear idea of what it is doing right that Vocalinux is getting wrong,

Where did I write that?

please don't let anything stop you from submitting a patch or a pull request to help improve it?

Statement or question? Don't bother answering. It's not important. I have no interest submitting a patch or pull request to your project. What I would like is for this discussion to focus upon its stated subject, speeding up your project, instead of focusing on *me*.

…

-- hkp://keys.gnupg.net CA45 09B5 5351 7C11 A9D1 7286 0036 9E45 1595 8BC0

[jatinkrmalik]

@Boruch-Baum Yeah, I pointed that last reply at you instead of at the lag. Wrong move. You never said you knew nerd-dictation's code.

Let me try again:

On our side the loop is basically: record into a buffer -> wait for silence (default silence_timeout is 2 seconds) -> then decode the whole chunk. For VOSK that means AcceptWaveform on the full buffer and only FinalResult() after. No partial text while you are still talking. whisper.cpp is the same shape. So you can use the same VOSK model as nerd-dictation and still feel seconds slower, because we sit on silence (VAD filter) and then batch-decode (then inject).

nerd-dictation is a thinner VOSK path: record and recognize together, type as it goes. Different design, not just a prettier tray.

If you still have Vocalinux around, try dropping Silence timeout in Settings to something like 0.5-1s on VOSK. That only shortens the wait, it is not real streaming. But if most of the pain is the 2s gate, it should help.

One number that would help if you have it: from when you stop talking to the first character, VOSK, both tools, same box.

---

p.s: As I said in the last post, I'm still learning here. I have been building this for over one year and I still learn something new every day.

[Boruch-Baum]

On 2026-07-24 01:13, Jatin K Malik wrote: ***@***.*** Yeah, I pointed that last reply at you instead of at the lag. Wrong move. You never said you knew nerd-dictation's code. Let me try again:

No worries. Best wishes for the success of your project. Be sure that there are many anonymous people out there silently benefiting from your good free-sharing nature. Do feel welcome to browse and use any of my own repos.

…

-- hkp://keys.gnupg.net CA45 09B5 5351 7C11 A9D1 7286 0036 9E45 1595 8BC0