SpeakType v0.17.2

SpeakType v0.17.2 (Windows)

Patch release since v0.17.0:

• Doubao is ready right after sign-in — SpeakType now picks up the App Key automatically from Doubao's own web bundle once you are logged in; you no longer need to trigger Doubao's built-in voice input first. That path still works as a fallback, and a manual App Key can still be entered in Settings (#382)
• Fix: model download no longer crashes the app on a full disk (ENOSPC); the partial file is kept and resumes (#381)
• Fix: narrow-window layouts for the Persona page, hotkey warning and Japanese History header; duration units and built-in persona names follow the UI language (#380, #381)
• Traditional→Simplified normalization now uses the full OpenCC table (#381)
• Plus assorted polish (#377, #378)

Downloads

Platform	File
Windows 10/11 x64 installer	SpeakType-Setup-0.17.2.exe
Windows portable	SpeakType-0.17.2-portable.exe
macOS preview (unchanged)	v0.17.1-mac-preview
Android (unchanged)	SpeakType-0.17.0.apk

Default: 100% offline recognition with SenseVoice (Chinese, English, Japanese, Korean, Cantonese). Optional: Parakeet TDT v3 for English/European languages, Doubao via your own account, or bring your own cloud API key.

SpeakType v0.17.1-mac-preview (macOS 预览版)

SpeakType macOS 预览版（v0.17.1-mac-preview）

预览版，未经 Apple 签名与公证。Apple Silicon 包在 macOS 26.5 的 Apple Silicon 虚拟机上真实构建、安装并跑通「长按右 Option 说话 → SenseVoice 识别 → 落字到 TextEdit 光标处」；Intel 包为同一台机器交叉打包，未在 Intel 硬件上运行过。应用版本号仍为 0.17.0（与 Windows v0.17.0 同源代码 + macOS 适配，PR #379）。

下载

文件	架构	体积	SHA-256
SpeakType-0.17.0-mac-arm64.dmg	Apple Silicon	113.2 MB	834f68483a3e0c91d91d30648ba0ffe3acffb074d97b33a0802f4904192f1213
SpeakType-0.17.0-mac-arm64.zip	Apple Silicon	123.3 MB	10560900161d140076648372e9ba155df675560e490757198f00cee2292ce96b
SpeakType-0.17.0-mac-x64.dmg	Intel（未实测）	119.4 MB	924da7fc90d35a5e587df7070170a1ec439ad132f3d62551d2dda36d1b51cf7f
SpeakType-0.17.0-mac-x64.zip	Intel（未实测）	130.6 MB	7aa288f13c87331619832a8cd548c8259380abf3bf53e892f243f5f7e86a7b76

SHA256SUMS.txt 可用 shasum -a 256 -c SHA256SUMS.txt 校验。

系统要求与首次打开

• macOS 15.5+（随包 sherpa-onnx 1.13.4 的 onnxruntime 最低系统版本）。
• 包是 ad-hoc 签名、未公证，Gatekeeper 会拦；macOS 26 上 Finder 右键「打开」也不行。绕过：xattr -d com.apple.quarantine /Applications/SpeakType.app，或 系统设置 → 隐私与安全性 → 仍要打开（需管理员密码）。
• 首次启动会引导授予 辅助功能 权限（全局热键），并弹 自动化（控制 System Events）授权；首次录音弹 麦克风 授权。
• 默认长按键是 右 Option；免按模式 Option+Q；本地引擎只提供 SenseVoice / Parakeet（whisper.cpp 仅 Windows）。

完整说明、已知限制与逐项验证记录：docs/macos.md · docs/reviews/macos-round1.md

未验证项

物理 Mac（真实麦克风/键盘）、Intel 运行、麦克风 TCC 首次弹窗、录音时静音、Parakeet 推理；签名/公证需 Apple Developer 账号。

SpeakType v0.17.0

SpeakType v0.17.0

Highlights since v0.16.0:

• Hands-free auto paragraphs — long pauses insert a blank line between paragraphs during continuous dictation (#354)
• Voice commands (opt-in) — say "new line", "new paragraph", or "delete last sentence" in hands-free mode (#357, #360, #362)
• Fix: switching local models no longer permanently breaks recognition until restart (#346)
• Fix: buttons scrolled under the top drag strip are clickable again (#352)
• Fix: short-pause sentence starts no longer lose characters in hands-free Silero mode (#358, #359)
• Fix: native dropdowns now always follow the dark theme (#361)
• Plus assorted polish (#342, #347, #350)

Downloads

Platform	File
Windows 10/11 x64 installer	SpeakType-Setup-0.17.0.exe
Windows portable	SpeakType-0.17.0-portable.exe
Android (unchanged)	SpeakType-0.15.0.apk

Default: 100% offline recognition with SenseVoice (Chinese, English, Japanese, Korean, Cantonese). Optional: Parakeet TDT v3 for English/European languages, or bring your own cloud API key.

SpeakType v0.16.0

SpeakType v0.16.0

Feature and reliability release since v0.15.1 (90+ commits).

Highlights

• Hands-free dictation is more robust: audio arriving in the gap between sentences is buffered and re-fed to the next session, so mid-speech session switches no longer drop words or mix languages (#338). Rule-based sentence endings in hands-free mode now get a proper period (#327, #328).
• File transcription quality: long recordings are now sliced at natural speech pauses instead of being decoded as one block, restoring sentence punctuation (#336). Cloud transcription gets a 60s client timeout, Esc-to-cancel while processing, and clearer 401 errors (#305).
• Memory: the local ASR worker now explicitly reclaims native decoder memory after each utterance, eliminating linear memory growth in long hands-free sessions (#333, #334).
• Config export/import: one-click migration of preferences, dictionary, personas and per-app rules, with import validation and visible feedback for dropped fields (#295, #299, #301, #321).
• Traditional Chinese & Korean: Traditional-Chinese numbers join ITN normalization (#317), Traditional-heavy and Hangul input skip the Simplified-only punctuation model (#315, #319), and dictionary search understands Simplified/Traditional variants (#325, #332).
• Hotword safety: multi-character homophone replacement now requires at least one original character to match, preventing false corrections (#310, #323).
• Plus: desktop-phone relay heartbeats and busy-state fixes (#290, #291, #292), amount/time restoration after enhanced punctuation (#294), "3.30 pm" → "3:30 pm" (#308), F8 rewrite prompt follows the command language (#298), and many copy/UI fixes.

Downloads

• SpeakType-Setup-0.16.0.exe - Windows installer (recommended)
• SpeakType-0.16.0-portable.exe - portable, no install
• Android phone-microphone APK: unchanged since v0.15.0 - download SpeakType-0.15.0.apk

SpeakType v0.15.1

SpeakType v0.15.1

Patch release: privacy hardening, cleaner uninstall, and a truly self-contained portable build.

Fixes

• Privacy - no silent connections: in Local (offline) mode, a leftover Doubao login cache no longer causes any network connection at startup (#232). Switching the recognition provider away from Doubao or ChatGPT at runtime now also closes their hidden bridge windows, so no background connections linger (#234, #236).
• Cleaner uninstall: the uninstaller now removes the launch-at-login registry entry, so uninstalling never leaves a dead autostart pointing at a removed exe (#238).
• Portable build is truly clean: first run of the portable exe no longer inherits an old installation's config (including login caches) - it starts fresh in its own SpeakType-data folder next to the exe (#240).
• Better Chinese default: configs from older versions without an explicit local model now default to SenseVoice when the recognition language is Chinese/Japanese/Korean/Cantonese, instead of a model that latinizes CJK speech (#240).

Downloads

Platform	File
Windows 10/11 x64 installer	SpeakType-Setup-0.15.1.exe
Windows portable	SpeakType-0.15.1-portable.exe
Android (phone as microphone)	unchanged - use SpeakType-0.15.0.apk from v0.15.0

Website: https://speaktype.zalize.com

SpeakType v0.15.0

SpeakType v0.15.0

Quality-of-life release: history that scales to hour-long transcripts, an undo for every delete, keyboard-first accessibility, and a cleaner Settings layout.

Highlights

• History built for long transcripts: transcription results persist across restarts and land in History — searchable by file name, exportable, and now collapsed to 8 lines by default with Show all / Show less. The Correct editor auto-grows with content (up to 14 lines) and extra-long unbroken strings wrap instead of breaking the layout.
• Undo for single deletes: deleting a history entry shows an Undo pill (10 s, pauses while hovered) that restores the entry in its original position.
• Keyboard accessibility: the transcription drop zone is focusable and Enter/Space-activated, focus jumps to Undo after a delete, and toggles expose proper switch semantics for screen readers.
• Cleaner Settings: microphone and phone-as-microphone controls moved to the top of the Speech tab; the model tab is now called "AI polish" in all five languages.
• Human-readable errors: network-level failures ("fetch failed" and friends) now say "Cannot reach the speech recognition service — check your network or switch provider in Settings" everywhere: toast, history entry, and Retry.

Downloads

Platform	File
Windows 10/11 x64 installer	SpeakType-Setup-0.15.0.exe
Windows portable	SpeakType-0.15.0-portable.exe
Android (phone as microphone)	SpeakType-0.15.0.apk

Default engine is fully offline (SenseVoice for Chinese/CJK, Parakeet for English & European languages). Optional cloud channels: any OpenAI-compatible ASR, ChatGPT web transcription, and more.

Website: https://speaktype.zalize.com

SpeakType v0.14.0

SpeakType v0.14.0

File transcription lands: SpeakType is no longer just live dictation — drop in a recording and get a transcript.

Highlights

• File transcription (new page): drag in an audio or video file (mp3, wav, m4a, ogg/opus, mp4, webm... up to 3 hours) and get an offline, segmented transcript with timestamps — powered by the same local engines (SenseVoice / Parakeet / whisper.cpp), so nothing leaves your machine. Live progress, cancel anytime, copy all, and export TXT or SRT subtitles. Hotword correction and Simplified-Chinese conversion apply just like live dictation.
• Details that respect your click: right after a transcription finishes, the progress row lingers briefly so the export buttons don't jump under your cursor; the missing-model banner's Download button now opens the correct Speech settings tab; error banners are readable in dark mode.

Downloads

Platform	File
Windows 10/11 x64 installer	SpeakType-Setup-0.14.0.exe
Windows portable	SpeakType-0.14.0-portable.exe
Android (phone as microphone)	SpeakType-0.14.0.apk

Default engine is fully offline (SenseVoice for Chinese/CJK, Parakeet for English & European languages). Optional cloud channels: any OpenAI-compatible ASR, ChatGPT web transcription, and more.

Website: https://speaktype.zalize.com

SpeakType v0.13.0

SpeakType v0.13.0

Terminal-friendly dictation and a smarter, more reliable phone-as-microphone experience.

Highlights

• Terminal-aware dictation: when the target is a terminal (cmd, PowerShell, Windows Terminal, WSL, ...), SpeakType strips the trailing period and undoes automatic sentence capitalization, so Get process foo. lands as a runnable get process foo — dictionary proper nouns like SpeakType keep their casing. Notepad and other apps are untouched.
• Phone-mic pairing that tells you what's going on: a stale pairing page (after the desktop rotated its pairing token or room code) now stops retrying after ~12 s and asks you to re-scan the QR / re-enter the code, instead of silently reconnecting forever. The relay page's Hold-to-talk button stays disabled until your PC is actually in the room.
• Mode switching just works: toggling LAN/relay no longer silently fails when an old pairing page is still open.
• Polished details: config-error toasts jump straight to the right Settings tab; word stats count real words (not characters); long-sentence live captions keep scrolling past 20 s; dictionary .txt export; history export respects your UI language; two-step persona deletion with rule cleanup; F8 rewrite instructions get hotword correction.

Downloads

Platform	File
Windows 10/11 x64 installer	SpeakType-Setup-0.13.0.exe
Windows portable	SpeakType-0.13.0-portable.exe
Android (phone as microphone)	SpeakType-0.13.0.apk

Default engine is fully offline (SenseVoice for Chinese/CJK, Parakeet for English & European languages). Optional cloud channels: any OpenAI-compatible ASR, ChatGPT web transcription, and more.

Website: https://speaktype.zalize.com

SpeakType v0.12.0

SpeakType v0.12.0

Clearer setup guidance and rock-solid long dictation sessions.

Highlights

• Setup problems are now visible: if dictation can't start (cloud channel not signed in, endpoint not configured, local model not downloaded), a toast tells you exactly why with a one-click Open Settings shortcut — no more silent failures.
• Toasts you can actually read: configuration errors stay for 12 s, and hovering any toast pauses its auto-hide.
• Test transcription for web-session channels: verify connectivity right from Settings; failures show the exact reason.
• Privacy: clearing history now also deletes failed-recording audio saved on disk.
• Smoother hold-to-talk: consecutive English sentences dictated within 15 s are joined with a proper space instead of "fine.Tap".
• Stability, verified: 30-minute hands-free soak (137/137 sentences, no leaks), idle engine auto-releases ~1 GB after 10 min and re-warms in ~2 s.

Downloads

Platform	File
Windows 10/11 x64 installer	SpeakType-Setup-0.12.0.exe
Windows portable	SpeakType-0.12.0-portable.exe
Android (phone as microphone)	SpeakType-0.12.0.apk

Default engine is fully offline (SenseVoice for Chinese/CJK, Parakeet for English & European languages). Optional cloud channels: any OpenAI-compatible ASR, ChatGPT web transcription, and more.

Website: https://speaktype.zalize.com

SpeakType v0.11.0

SpeakType v0.11.0

Self-improving dictation, polished.

Highlights

• Self-learning auto-corrections (EN + ZH): after text lands, fix a misrecognized word in place and SpeakType learns it — next time it's corrected automatically. Multi-word edits, single-character Chinese fixes, and instant learning when you switch windows all work; pure capitalization edits are never learned.
• Undo for learned words: the "New word learned" toast now has an Undo button — one click removes the word and restores the history entry.
• English hotwords: dictionary entries now correct English words case/space-insensitively (e.g. "speak type" -> "SpeakType").
• Phone-as-microphone entry on Home: the remote mic feature is now discoverable right from the home screen.
• Faster model housekeeping: switching local models immediately frees the old engine (incl. whisper-server); update check is prefetched at startup.

Downloads

Platform	File
Windows 10/11 x64 installer	SpeakType-Setup-0.11.0.exe
Windows portable	SpeakType-0.11.0-portable.exe
Android (phone as microphone)	SpeakType-0.11.0.apk

Default engine is fully offline (SenseVoice for Chinese/CJK, Parakeet for English & European languages). Optional cloud channels: any OpenAI-compatible ASR, ChatGPT web transcription, and more.

Website: https://speaktype.zalize.com