Skip to main content

Add support for nvidia/parakeet-tdt-0.6b-v3 (Windows)

Feature Request

https://huggingface.co/spaces/hf-audio/open_asr_leaderboard

Feedback Suggestion

Please add support for nvidia/parakeet-tdt-0.6b-v3 in the local transcription software.

  • This model is listed on the Open ASR Leaderboard and demonstrates stronger multilingual performance.

  • It achieves higher transcription accuracy than openai/whisper-large-v3.

  • It is also more than 16× faster than openai/whisper-large-v3-turbo, making it highly efficient.

Additionally, since we have already configured the “Configure Your AI – Bring your own API key to power Vowen’s AI features”, we suggest expanding functionality with:

  • Translation feature: Transcribed content could be directly translated into a target language.

  • Real-time transcription + real-time translation: Making the software more suitable for global use cases and multilingual workflows.

  • Voice Activity Detection (VAD): Automatically segment transcription based on natural speech pauses instead of manual hotkeys.

  • Speaker Diarization: Distinguish between different speakers in conversations, improving readability and usability in meetings or multi-speaker recordings.

Posted by dao tian ·
Status: Completed6 comments

Log in to comment and vote

Comments6

  • limbertin

    •

    Apr 6

    For windows 11, my solution was to install the parakeet onnx version as custom server, it uses even less memory. It works faster than the included version in Vowen, when used in conjunction to an NVIDIA graphic card.

    With this setup I tried enabling real-time preview through the config file, but it doesn't work. It doesn't make any difference in the input. It only shows the whole text.

    As an observation, enabling "real-time preview" and also "Enhance transcription with AI", it wouldn't make sense.

    • Vowen

      Team•

      Apr 6

      @limbertin Ah yes, we’re working on real-time preview for Parakeet on Windows, it should be out soon. Also working on making the setup faster and add support for GPUs, right now it runs on CPUs.

      Regarding “real-time preview” yes, it’s the raw state before the text is cleaned / enhanced by AI. Real-time transcription helps with making corrections, and state where I can gather my thoughts, are take a pause to reflect on what was being spoken.

  • Vowen

    Team•

    Mar 2

    Support for Parakeet v3 in Windows is proving to be quite tricky as we’d have to bundle python libraries with the app. In macOS with coreML, the implementation is quite straightforward. Marking this done, as all the features like “Real-time transcription”, “VAD”, “Speaker diarization” have been completed for macOS. We’ll come back to Windows once we are able to make more progress on this meaningfully. Thank you!

  • Vowen

    Team•

    Jan 6

    We’ve added parakeet-v3 support for macOS systems. However, for windows it’s a bit tricky to implement it right now, we will be releasing it for windows by this month if possible. We’ve added the translation features, and VAD already.

    Considering the direction we are trying to take in building this app at the moment, we don’t have the bandwidth to work on adding Speaker Diarization and Real-time transcription. We will consider adding these features in the future when we try to improve the “Take meeting notes” feature further. Thanks for your feedback, we will be adding this in the coming months.

    • Alexander

      •

      Jan 7

      It will be amazing to see the release for windows by the end of the month 🙏

  • Vowen

    Team•

    Dec 13, 2025

    Hi @dao tian We’ll be adding support for the features very soon in the upcoming releases. Thank you for the feedback and helping improve the app.