[feature] shortcut to toggle transcription languages/models

[rusak47]

It's quite annoying to have to go into the settings tab just to specify the transcription language. A shortcut for this would be very handy.

For a simple toggle, there should be a way to select the languages the user is interested in—similar to how keyboard layouts can be switched.

However, not all languages are well supported by all models. Latvian, for example, has limited support. To address this, I see two possible approaches:

• a) When toggling languages, automatically switch to the appropriate model as well.
  The downside is that this would make it harder to switch between small and large models for languages that support multiple options.

• b) Add a separate shortcut to toggle the model, and allow the user to specify, per model, which languages it should support.
  This requires more configuration and multiple shortcuts, but it gives fine-grained control—allowing the user to switch models based on language.

Example Use Case
The user configures for example three languages: EN, LV, PL
And some models: small, big, big-lv (amount of models not tied to configured languages)

For small and big, the user enables only EN and PL

For big-lv, they enable only LV (assuming it doesn't perform well with other languages)

When toggling between EN or PL, the user can switch between small and big models depending on whether they prioritize speed or accuracy.
When LV is selected, big-lv model is automatically selected and toggling the model would have no effect, since only one model supports that language.

[cbarange]

Hello,

I’ve implemented two shortcuts: one for switching the model, and one for switching the language. For now, this seemed like the most flexible solution for different user needs.

You can assign a list of supported languages to each model. The currently selected language is shown with a background image of the corresponding country flag.

At the moment, there is no permanent visual indicator for the active model, but a tooltip briefly appears for a few seconds after using the shortcut, showing the selected model name.

Please note that some shortcut combinations may not work because they are reserved by the operating system. If a shortcut does not trigger correctly, try changing the key mapping.

This feature is available in release v0.1.6: https://github.com/AccessDevops/S2Tui/releases/tag/v0.1.6

The builds are currently in progress.

I will keep this issue open and close it once I have tested this version for a few days.

[rusak47]

Nice! I'll give it a test once the binary is available.

[rusak47]

Some observations on this one:

• Strangely, the portable version lacks the country flags under the microphone that were shown in your last screenshot. Maybe images should be packaged separately in an archive for portable versions? Or perhaps log this on app startup to indicate whether images are loaded or missing.

• Feature worth discussing: For settings where the user has multiple "favorite languages" configured—for example, a small model supporting only a few of them and a large model supporting all—I believe it would be worthwhile to cycle the model when switching to the next language.

  In other words, the basis for language switching should be the configured favorite languages, similar to how keyboard layouts work. If the currently selected model does not support the desired language, it should automatically switch to a model that does—going from a less capable model to a more capable one as needed.

Small fails to process Latvian and Russian in auto mode:

• And sorry for repeating—I'm sure you're already aware, but just so it doesn't get buried behind previous comments: it would be great to have all languages available in the language configuration. On the other hand, listing them all as checkboxes might be overwhelming. Perhaps add them on demand—for example, Russian, Latvian, and Latgalian—until you rework the interface into a dropdown with checkboxes.

Overall, this feature is already great. Thanks for the excellent work!

[cbarange]

Hello, thanks again for the review :). I did have the same issue about the flag, It seems like the build package has an issue rendering the flag while the development version of the application did work really well so I applied a fix last night but not published yet I'm waiting to implement the custom model support for the next version and I will publish the fix in this version soon I hope.

I'm also really agreeing with you about the language missing and the language interface so that's why I work on it and you can see the new UI that will be available in the next version :

I will look for implementing a way to auto switch between models based on the language switch I wanted to have both behavior Some users may prefer to have a model first while some other users may want to have a language first switch mechanism So I will implement a little checkbox That will allow the user to Choose between these two different behaviors Like that This will cover All needs

I will work on that soon :), thanks again for the sharing

[cbarange]

Hello,

I succeeded in implementing the automatic switch to the English language model in this release: https://github.com/AccessDevops/S2Tui/releases/tag/v0.1.7

This release also comes with more languages, such as Latvian. If some are still missing, I can add them.

Thanks for your interest in this app :), May I ask if you use it often? What kind of usage do you have for it ?

I personally made it for writing my emails and AI prompts (I'm lazy and was tired of typing on the keyboard).

[rusak47]

Hi @cbarange!

The app is a lot slimmer this time and downloaded very quickly—what a sweet bonus!

However, upon startup, it immediately began downloading Whisper models with no way to stop it. I see that you've changed the directory for model storage. Since this is a one‑time configuration, it's not a huge issue, but maybe it shouldn't happen without user consent—or at least there should be a "Stop" button (especially nowadays, when security can be easily breached even at the delivery stage).

Just a thought: since there's a configuration file, perhaps add an option to configure where models are actually stored. It's not critically important, as it can be easily symlinked. For example:

ln -s /opt/S2Tui/models/ ~/.local/share/com.accessdevops.s2tui/models

I realized why my large model transliterates Russian—it's a fine‑tuned version for Latvian, so it hallucinates in favor of Latvian. This leads me to think that multilingual users need more than just two models. Ideally, we'd have at least:

• One specialized model for the user's local language
• A small standard model
• The original Whisper model, which performs well with popular languages
• And potentially even more, depending on needs.

Offtopic

Long story short: my use case is quite similar to yours.

My journey with speech‑to‑text started after I watched this impressive video:
https://peertube.tv/w/wnv9auiwEzk7MS8mJvv6n5

I really want to get rid of my keyboard—or at least minimize its use—but I'm not ready to program the way that guy does (https://numenvoice.org).

Nerd Dictation provided an easy typewriter mode with custom scripting capability (which I haven't used yet, and after migrating to Wayland I'm having problems even with typewriter mode). But anyway, right now I'm mainly using Nerd Dictation with VOSK models. (everyday)

VOSK models aren't extremely accurate for speech‑to‑text, but they still provide decent quality—and more importantly, Nerd Dictation works in typewriter mode (#3).

However, VOSK models are limited in language support, and Latvian is not on the list. That's when I found a fine‑tuned Whisper model for Latvian, and also discovered your great project, which works seamlessly with GGML models.

So I'd really like to see your project evolve into something bigger.