
Dabri, an open-source Linux speech-to-text application, has released version 2.6.0 with changes that allow its Parakeet Ultra recognition model to run on x86-64 processors without AVX2 support. The update also reduces the model’s download size from 486 MB to 416 MB while removing several software dependencies.
Published on November 20, 2026, the release addresses a hardware compatibility limitation introduced when Parakeet Ultra was added to Dabri in version 2.5.0. The previous implementation required AVX2, an instruction-set extension available on some modern processors but absent from older x86-64 hardware.
The compatibility change comes from moving Parakeet Ultra onto whisper.cpp, which already powers Dabri’s other speech-recognition models. The developer integrated the model through the library’s Parakeet implementation, eliminating the separate transcribe.cpp backend and its OpenBLAS dependency.
Removing those components also reduces the size of Dabri’s distributed packages. The release notes report reductions of approximately 15 MiB for the AppImage, 8 MiB for the Flatpak and 38 MiB for the installed Fedora package.
Dabri uses Whisper and NVIDIA Parakeet models to transcribe speech locally, without requiring a cloud transcription service. Recognized text can be inserted directly into the currently active desktop application, including browsers, text editors and integrated development environments. Users can also send transcriptions to the clipboard.
Version 2.6.0 includes a fix for a command-line issue that could leave users waiting 45 seconds when both automatic typing and clipboard output failed. The dabri stop command now returns the transcription immediately in that situation.
Users who already have Parakeet Ultra installed will receive the updated model the next time Dabri loads it. The developer also advises removing the previous parakeet-ultra-Q4_K_M.gguf file from the models directory after the replacement has been downloaded.
Dabri 2.6.0 is available through the project’s GitHub release page and Flathub.
Discover more from Aree Blog
Subscribe now to keep reading and get access to the full archive.


