voiceio vs nerd-dictation
nerd-dictation is a single-file Python script that does offline speech to text on Linux with the VOSK engine. It has no background process, types as you speak with xdotool, ydotool, dotool or wtype, and can be customized with a Python config script. voiceio is a larger tool: a background service built on Whisper, with punctuation and casing from the model, a live IBus preview, vocabulary hotwords and fine-tuning on your voice. nerd-dictation is the smaller, more hackable choice; voiceio does more out of the box and costs more to install.
Which should you use?
Choose nerd-dictation if…
- You want something tiny: one Python file, minimal dependencies, nothing running in the background.
- You like to script it yourself: a Python hook lets you rewrite text any way you want.
- You want a restricted grammar: VOSK can be limited to a fixed set of phrases for higher accuracy on commands.
- Your hardware is modest; VOSK's small models are light.
Choose voiceio if…
- You want punctuation and capitalization; nerd-dictation's README notes that VOSK text is all lower-case.
- You want Whisper-level accuracy on long, natural sentences.
- You want proper nouns right, through vocabulary hotwords, corrections and fine-tuning on your voice.
- You want an in-place, corrected live preview rather than raw keystrokes.
Feature comparison
| voiceio 1.5 | nerd-dictation | |
|---|---|---|
| Platforms | Linux: Wayland and X11 (GNOME, KDE, Hyprland, sway, i3). Untested best-effort path on Windows/macOS | Desktop Linux (X11 and Wayland) |
| Price | Free, MIT | Free, GPL-3.0 |
| Where speech is decoded | On your computer. Optional text-only cloud cleanup, off by default | On your computer (VOSK) |
| Speech models | faster-whisper small (default), medium, large-v3-turbo, distil-large-v3; experimental Parakeet and whisper.cpp server | VOSK models |
| Open source | Yes (MIT) | Yes (GPL-3.0) |
| Live typing into apps | Yes: underlined preview in the focused app through IBus, corrected in place, committed when you stop | Yes: types as you speak by default, via xdotool, ydotool, dotool or wtype |
| Custom vocabulary | Hotwords ranked by use (about 35 per decode) plus find-and-replace corrections | Python config hook; optional VOSK grammar file |
| Learns from you | Yes: optional LoRA fine-tune on your own recordings, on CPU, kept only if it wins on held-out clips | Not stated |
| Voice commands | Dictation commands only ("new line", "scratch that", "correct that") | None built in; scriptable through the config hook |
nerd-dictation facts are from its official site, docs or repository, read on 2026-10-10 (sources below). Prices are as listed on that date and can change. voiceio facts are from its README. Spotted something out of date? Open an issue.
Questions
Is voiceio a good nerd-dictation alternative?
If you want Whisper accuracy, punctuation and casing, a live preview and fine-tuning, yes. If you want the smallest possible tool with no background process, nerd-dictation remains a good fit.
Does nerd-dictation work on Wayland?
Yes, using ydotool, dotool or wtype; its default, xdotool, is X11 only (per its README, checked 2026-10-10).
Does nerd-dictation use Whisper?
No. It uses the VOSK API and VOSK models. voiceio uses Whisper through faster-whisper.
Which uses fewer resources?
nerd-dictation, very likely: it runs only while you dictate and VOSK's small models are light. voiceio keeps a Whisper worker loaded so it can respond immediately.
Try voiceio
Free and MIT licensed. Install it with pipx on Linux, or try the dictation in your browser first: the demo on the home page runs a small speech model inside the tab.