# Voice Dictation / Voice Control: A Market Survey, Mid-2026 **Date:** 2026-08-07 · **Tier:** `design/` — public engineering research **Companion:** [local/offline STT engine SoA](2026-08-07-stt-engine-sota.md) · [gaze / EMG / BCI / multimodal input SoA](2026-08-07-hci-input-sota.md) · [the roadmap](https://github.com/MSKazemi/yazses/blob/main/ROADMAP.md) A survey of the voice-dictation and voice-control tools on the market as of 2026-08-07, gathered to understand where an offline-first project sits in the wider landscape. This is a snapshot, not an endorsement or a criticism of any of the products named — pricing and feature claims below are as published by each vendor at the time of writing and may have changed since. ## 1. Per-product summaries ### Commercial, cloud-based, AI-native dictation - **Wispr Flow** — Mac/Windows/iOS/Android. $15/mo, with a free 2,000-words/week tier. Cloud-only (AWS plus third-party LLMs). Notable features: auto-edits as you speak, a Command Mode (highlight text, say "make this more concise", get an in-place rewrite), and per-app tone matching (Slack vs. Gmail vs. an editor). The vendor reports 184 WPM effective throughput. Publicly discussed concerns: it captures screenshots of the active window for cloud processing, with Privacy Mode off by default; a 2.7/5 Trustpilot rating as of 2026-04; and it is a comparatively heavy Electron application. - **Superwhisper** — macOS/iOS. Free tier; $8.49/mo; $249.99 lifetime. Local-capable (Whisper/Parakeet on Apple Silicon) with an optional cloud LLM. Notable features: custom "modes" (per-task prompt presets) and context capture from selection/clipboard/app. No Windows support as of this survey. - **Aqua Voice** — Mac/Windows. $8/mo. Cloud-only, using a proprietary STT model (Avalon, 2025-08) trained on prompt-style speech, code, and email. Notable features: reported <50 ms start latency and ~1 s insertion, per-surface style matching, voice-driven editing, and an 800-term custom dictionary. - **Willow Voice** — four platforms, ~$15/mo, cloud-first, with style-matching as its central pitch and an optional offline mode on Mac/iOS. - **Monologue** — $10/mo, with a downloadable model for on-device transcription. - **Spokenly** — Mac/iOS, free, local models with bring-your-own API keys for cloud options. ### Open-source dictation - **VoiceInk** — macOS, GPL-3, $25 one-time. Local whisper.cpp with a "Power Mode" (automatic per-app settings) and AI enhancement via a local Ollama instance. - **Handy** — Mac/Windows/Linux, MIT, free, fully offline, built on Tauri (Rust+React). The largest open-source project in this space by star count as of this survey. Does not ship AI cleanup, and users report a 2–5 s post-speech wait. - **Speed of Sound** — Linux, on-device (Whisper/Parakeet/Canary backends), X11 and Wayland, packaged for Flathub/Snap/AppImage/deb/rpm, with primary/secondary language switching. - Others surveyed: Vocalinux (Vulkan-accelerated), VOXD (Wayland-first), OpenWhispr, nerd-dictation, Whispering. Ito (GPL-3) shut down in 2026-01 after a Show HN launch — a data point on how quickly an open-source dictation project can lose momentum. ### Voice control - **Talon + Cursorless** — free tier, with a $25/mo beta tier. A local Conformer/ Whisper hybrid "mixed mode" handles commands and dictation in one utterance; Cursorless adds structural, AST-aware code editing. Reported downsides: a weeks-long learning curve and vocal fatigue from its dense command grammar. - **Numen** — Linux, libre, Wayland-native, syllable-based full voice control. ### Incumbents / built-in dictation - **Dragon Professional v16** — Windows, $699, no major release since 2023; Dragon Home has been discontinued and Dragon Anywhere mobile was retired on 2026-07-01. - **Windows Voice Access + Fluid Dictation** (2026-06 update) — on-device grammar and filler correction as you speak, gated to Copilot+ NPU hardware. - **macOS Tahoe Dictation** — on-device on Apple Silicon; Apple states it is 55% faster than Whisper on its hardware. Reported to be weaker on technical vocabulary and proper nouns. - **Otter.ai** — cloud meeting transcription, $8.33–$19.99/user/month, English/ French/Spanish only. ## 2. Feature landscape, condensed Fully-offline operation, Linux support, and a single cross-platform daemon architecture are comparatively rare in this survey — most local-capable competitors are Mac-only or Mac-first. Offline diarized meeting transcription, a local encrypted personalization loop, and gaze/EMG/SSH-remote activation sources were not observed in any competing product surveyed. Areas where the cloud-based tools above are ahead of what a fully-offline tool typically offers: voice-editing of already-typed or selected text, per-app tone/ format matching, custom AI prompt modes, and sub-second perceived latency as a default (rather than an opt-in). ## 3. White space — capabilities not observed in any product surveyed 1. Offline, cross-platform voice-editing of selected text using a local LLM — every product with this feature in this survey is cloud-only, and none run on Linux. 2. Offline per-app tone-matching on an ordinary CPU — Microsoft's version is NPU-gated; no product surveyed does this without either the cloud or specialised hardware. 3. A single tool combining everyday dictation with offline, diarized meeting minutes. 4. A local, self-improving personalization loop that gets better over time without any data leaving the device — only Dragon personalizes locally among the products surveyed, and it has had no major release since 2023. 5. Pipeline reliability (automatic recovery from a dropped microphone, a device change, or a silent failure) marketed as a feature in its own right. 6. Gaze-routed dictation, an EMG hotkey, and SSH-forwarded remote dictation — no product surveyed offers any of these. ## 4. What users report valuing most, across reviews surveyed 1. Instant, everywhere text insertion — sub-second latency is a headline claim across nearly every commercial product. 2. AI cleanup that makes spoken text read like careful writing — cited repeatedly as the main reason people pay for a subscription tool. 3. Per-app tone and context matching. 4. Voice-editing of existing text — a standout feature in Wispr Flow reviews specifically. 5. Privacy and ownership — local processing, a one-time or free price, or an open-source license. ## 5. Why users report abandoning these tools Recurring themes in reviews and forum discussion: a sense of privacy shock (Wispr Flow's screenshot capture, in particular); subscription fatigue; post-speech latency (Handy's 2–5 s wait is called out repeatedly); the manual cleanup burden on long dictations without an AI pass; instability in heavier Electron-based apps; and outright platform abandonment (Dragon's stalled release cadence, Ito shutting down). --- Key sources: efficient.app/apps/wispr-flow · docs.wisprflow.ai (Command Mode) · getvoibe.com (Wispr privacy discussion, Handy, Willow, Dragon pricing) · spokenly.app (Superwhisper pricing, Aqua review, Handy) · metawhisp.com (VoiceInk review) · omgubuntu.co.uk 2026-04 (Speed of Sound) · talon.wiki · numenvoice.org · support.microsoft.com (Fluid Dictation) · tldv.io (Otter pricing) · afadingthought.substack.com (a differentiators essay) · onresonant.com (a Reddit discussion) · vibetyper.com (2026 Linux dictation roundup) · alternativeto.net (Ito status).