After giving an overview of what I use for search and spending some time optimizing my speech recognition setup, I would now like to share some of my tools for it.
My history with voice recognition starts somewhere in 2023, where I started dictating a paper for college. And it really kicked off when I dictated my entire bachelor's thesis in 2024. Back then I used Whisper with its base model using the following script:
#!/bin/sh
echo "Recording. Press C-c to stop..."
ffmpeg -hide_banner -nostats -loglevel quiet -y -f alsa -i default "/tmp/audio.wav"
echo "Transcribing audio..."
whisper "/tmp/audio.wav" --model base --fp16 False --output_format txt -o /tmp/transcription
xclip -selection c < /tmp/transcription/audio.txt
echo "DONE! Check your Clipboard"This dictation helped me greatly alleviate my hand pain and finger pain that is still present if I have days where I type a lot. So for writing anything longer than two or three sentences, I really appreciate the option of dictating it.
What really sparked me to reinvestigate dictation for my workflow now in 2026 was the introduction of a new voice model natively in macOS 27. This model I tried and used since early Q3 of this year, but I have not stopped there.
On windows at work i quickly sought an equivalent and landed on the open source tool Handy. This application provides an interface to many easily downloadable voice models, such as Whisper Large V3, which I now use, and Qwen3 1.7B ASR, which is also quite decent. Then, after having used Handy for a while on the Windows PC, I decided to also use it on my Mac, so I can have the same voice model on both systems and align my speech patterns towards the quirks of each option. On both platforms I have the dictation toggle bound to the forward mouse button.
On iOS, I have spent some time looking for a good app for this and landed on VivaDicta. VivaDicta is a good option because it is free and open source. It's also convenient to import Voice Memos into it and to export the audio and/or the text from it. I dictated a first draft of this post with the app on September 19th.
As for voice models, my recommendation leans towards Qwen3 ASR 1.7B or Whisper Large V3. You can use faster models or even the natively built-in models of macOS and Windows, but if you spend any serious amount of time dictating, this doesn't make much sense. I should also note that if you are a heavy user of AI tools, most of them have a dictation option built into their UI.
I hope this was interesting and you consider trying out a dictation tool, at least for longer writing sessions.