Hi, I see this RFP from 2023 never got anyone to do it, but the need to have a nice voice-to-text or voice input or dictation thing in Debian is still there.
It seems Mozilla DeepSpeech stopped development already in 2021, and its successor, https://github.com/coqui-ai/TTS also stopped development 2 years ago. Now, with the rise of LLMs and various related technologies, there are a list of options both for the speech recognition part and for the UI integration part. A few years back users had to choose between legacy speech-to-text systems with high error rates or new LLM things that were heavy to run, mainly Whisper and Parakeet. Now in 2026, seems we have lightweight LLM-based things that seem superior, e.g. https://github.com/kyutai-labs/pocket-tts and https://github.com/SYSTRAN/faster-whisper Voxtype seems to have a nice overview at https://voxtype.io/compare/ Voxtype itself also seems nice, and apparently it is packaged in Arch and NixOS. However, programs like Voxtype, Vocalinux (https://github.com/jatinkrmalik/vocalinux), Handy (https://github.com/cjpais/Handy) or Blurt (https://extensions.gnome.org/extension/6742/blurt/) are apps or extensions. Another approach entirely would be to hook into the iBus input system which is the approach taken in e.g. https://github.com/PhilippeRo/IBus-Speech-To-Text and https://github.com/GitJuhb/voice-typing-linux. Seems Fedora included ibus-speech-to-text (see https://fedoraproject.org/wiki/Changes/ibus-speech-to-text). Perhaps we should start a discussion with the Debian Input Method Team <[email protected]> to see if they have any opinions on this.
