NexTalk

NexTalk is a narrow tool that solves a narrow problem well, and the honesty of that framing is part of its appeal. Strengths. The architecture is the differentiator. A C++ Fcitx5 plugin talking to a Flutter UI over a Unix domain socket means NexTalk inserts text at the input method layer, the same place your keyboard layout lives. That is why the vendor can credibly claim it works in Wayland, X11, terminals and IDEs without the clipboard workarounds or ydotool keystroke simulation that break on newer compositors. The latency claim of under 20ms is achievable precisely because Sherpa-onnx (Zipformer) runs on-device — there is no round trip to a server. The overlay is a transparent capsule rendered at 60fps in Flutter that auto-submits when you stop speaking, which is the right interaction for dictation: you never reach for a mouse. Push-to-talk and voice activity detection cover the two control styles people actually want. Weaknesses. Language coverage is the big one. The site states recognition is currently optimized for Mandarin Chinese and English, with more models promised. If you dictate in French, Spanish or Japanese, check before you commit. Offline Zipformer models will generally trail large cloud transcription services on difficult audio — accents, noisy rooms, jargon — so treat NexTalk as excellent-for-clean-speech rather than a transcription engine for archival recordings. There is no post-processing, formatting or editing layer beyond the raw recognized text: what you speak is what lands in the field. And the Fcitx5 dependency is hard. If your desktop runs IBus or another input method framework, NexTalk is out of scope until you switch. Where it fits. Linux developers and writers who already run Fcitx5, work on Wayland or X11, and want hands-on-keyboard dictation without sending audio anywhere. Air-gapped and restricted-network machines are a natural home — this is one of the few dictation tools with no network dependency in the pipeline at all. Where it doesn't. Windows and macOS users, mobile and web users, teams that need cloud sync or shared transcripts, and anyone who needs a broad multilingual model today. If those describe you, a cross-platform or cloud dictation service is the better purchase.