Skip to main content

Add Ukrainian language support for speech-to-text

Features Request & Bugs

It would be great if Seam’s speech-to-text feature supported Ukrainian transcription in addition to English.

Ukrainian-speaking users currently cannot use voice transcription effectively, even though Ukrainian is widely spoken in Ukraine and around the world. Support for Ukrainian would make Seam much more accessible and useful for Ukrainian users.

It would also be helpful to let users choose the AI model used for transcription, for example, from several available models. Different models may offer better accuracy for different languages, accents, or use cases.

If Seam added reliable Ukrainian transcription, I would definitely buy the application.


Thank you for considering this!

Posted by seftoner ·
Status: Completed5 comments

Log in to comment and vote

Comments5

  • Ramzi

    Team•

    Jul 25

    Hi,

    Good news: Ukrainian transcription already works in Seam today. Nothing to wait for.

    Seam runs Parakeet TDT v3 on device, which is the multilingual version of the model and covers 25 European languages, Ukrainian included. There is no language setting to flip, because the model handles all of them at once.

    On letting users pick the transcription model: I am going to pass on that one. I would rather keep Seam simple and opinionated than ship a dropdown that makes everyone benchmark models before they can dictate a sentence. If something better comes along, I will just move everyone to it.

    Give the trial a run with your Mac set to Ukrainian and let me know how it goes. If the output is still off, send me a sample of what you said next to what came out and I will look into it.

    Thanks for writing in.

    • seftoner

      •

      Jul 28

      Yes, I can confirm that Ukrainian is supported, but the current behavior is not what I expected.

      The Speech-to-Text implementation does not support dynamic language detection or multilingual recognition.

      The recognition language is determined solely by the application's default locale (the primary system language). Changing the current input language or keyboard layout while dictating has no effect on speech recognition.


      For example, if the system's primary language is set to Ukrainian, the STT engine will only recognize Ukrainian speech. If I dictate a sentence that mixes Ukrainian and English, the English words are not recognized correctly because the recognizer does not switch languages dynamically.

      Ideally, the application should either support automatic language detection during dictation or use the currently selected input language instead of always relying on the system's default locale.

    • seftoner

      •

      Jul 29

      @Ramzi Hi. I noticed this ticket has been moved to Complete Resolution, but unfortunately the workaround doesn't work well for me.


      For example, I'm currently using Handy with Parakeet TDT 0.6B v3—the same model Seam uses. However, in Handy I can speak both Ukrainian and English naturally, and it recognizes both languages without any issues.


      So from my perspective, this doesn't seem to be a model limitation. It looks like Seam could support the same bilingual behavior as well.

      • Ramzi

        Team•

        Jul 29

        Hi Roman, the fix is out today in Seam 1.10.7. You can update from inside the app, or with brew upgrade seam-app.

        Thanks for pushing back on the original resolution, you were right to. Please let me know if it works you want now.

        • seftoner

          •

          Aug 6

          @Ramzi, thanks a lot! As I promised, I’ve purchased your awesome app. 😊