Ibus-speech-to-text

Ask Fedora
1

I have Parksonism (Parkinson’s without the dopamine insufficiency). It would be very helpful if I could get this to work.

Has anyone been able to get this to work? I have set things up correctly. I can choose between ENG and text-to-speech. I restarted ibus. I have opened several applications (kate, etc) and I get nothing. Indeed, were it working, I could probably fill this text box.

2

Hi, David

Could you please provide details about the steps you are following, as well as system on which you are using ibus-speech-to-text ?

3
  1. Change ibus preferences from EN to “speech to text.”
  2. ibus restart

At that point it is supposed to work. No? BTW, I have tested the microphone. Moreover, I have tested using email, kate, tweetdeck and LibreOffice writer.

4

Hi David,

I hope after following the second step of “ ibus restart ” , you have downloaded and selected the required model if not you can follow the below steps

  1. Change ibus preferences from EN to “speech to text”.
  2. ibus restart
  3. From ibus setup tool you can get below options
    1. You can select between “vosk“ and “whisper“ recognition backend engines, for smoother and faster speech recognition you can continue with “vosk” with smaller models and if you want more accuracy then you can select “Whisper“ but it can be slower as compared to “vosk”
    2. You can enable “Preload Models“ & “Active on start“ options for faster speech recognition & smoother experience
    3. You can disable the “System locale“ option to get more locale options
    4. Now from locale section you can select any locale and click on 3 dots next to it to get available language models for selected locale, you can download any model, I would recommend to start with smaller models, larger models are more heavy to system and can get system slower.
    5. Now after downloading & selecting the required language model from selected locale you can do “ibus restart“ again to load the selected model , and open any application to start dictation, sometimes it may take few seconds to start speech recognition.

Let me know if you are facing any problem after following above steps

5

Hi Manish,
how can I change ibus preferences from EN to “speech to text”.?
Running ‘ibus-setup’ does not offer this. Which command must be executed to change this preference? It would be nice and helpful to be more precise in Changes/ibus-speech-to-text - Fedora Project Wiki

Thank you
Jochen

6

Hi @desperado ,
After installing ibus-speech-to-text, you need to add ibus-speech-to-text as a input source by following below steps:
You can do this through
Settings → Keyboard → Input Sources , Now Click on " Add Input Source " and search for " Speech To Text" it should be listed under " Other " . After adding it as a input source do ibus restart

Now after adding ibus-speech-to-text as a input source, you can switch between input methods by

  1. Keyboard shortcut Super + Space OR
  2. GNOME UI : You can switch manually from the top panel:
    1. Click the current input source indicator (e.g. “EN”)
    2. Select Speech To Text

You can open the IBus STT Setup tool by following below steps:

  1. From GNOME UI : Click current input source indicator (e.g. EN), Select Speech To Text from the available input methods and then go to Settings OR
  2. Settings → Keyboard → Input Sources → Speech To Text → Click on 3 dots → Preferences

Tips: After downloading and selecting the required language model, selected model still not loaded properly then close the IBus STT setup window and do the ibus restart , and reopen the IBus STT setup

Let me know if you have any other doubt.

8

Dear Manish,
thank you for trying to help. Now I was able to add “Seech to Text” to “Input Source”. I proceeded with the IBus SST setup, but get an error although I downloaded the required language:

image
image559×1035 42.6 KB

Do I need to store the download in a specific directory?

Thank you, Jochen

9

Hi @desperado ,
that’s not error , that’s just notification about formatting file availability for vosk speech engine’s voice command, It should not be appear for whisper engine, I will fix it in next release, Pls ignore it for now.
As i can see from the attached image, you are able to select the model correctly, if still you are not able to dictate the words, ibus restart command can fix it.

For German locale you can also try vosk’s model, After selecting the vosk as a backend pls do ibus restart to load the engine properly.

Let me know if you are facing any issue,

10

Hi Manish,
great! Together we managed that STT is now working in my Fedora Linux. All I now have to do is to decide whether Vosk or Whisper suits more for me and which size of the models my environment is able to handle.

Thank you very much,

Jochen

11

Great to hear that :slight_smile: