v2.1.5

What's New

This update adds ElevenLabs realtime speech recognition, makes the default hotkeys reachable on common keyboards, hardens toggle and hold hotkey handling, and introduces a new logo with a bilingual product website.

New Features & Improvements

• ElevenLabs Realtime Speech Recognition — ElevenLabs can now be selected for realtime speech recognition. Final output comes from server confirmations, so interim text never leaks into what is pasted, confirmed transcripts finish immediately instead of waiting for a fixed timeout, and failed transcriptions are kept for an explicit retry. Model settings add tooltips and visible paid-feature notices.
• Accessible Default Hotkeys — The default wake key is now left Command on macOS and left Ctrl on Windows, and left Shift finishes a recording through the general-cleanup template — all reachable on common keyboards. Existing custom bindings are preserved.
• New Logo and Product Website — Refreshed the app logo and launched a bilingual product website at voicepaste.dev with interactive demos, setup documentation, a download portal, and release history.
• Updated Model Details — Refreshed pricing and capability details for online recognition models and removed obsolete setup guidance.

Fixed

• Toggle and Hold Hotkey Conflicts — Hands-free wake keys are now independent from raw-input shortcuts. Recordings require an exact chord match and a 500ms hold before they start, invalid shortcuts are cleared with a toast when switching modes, and macOS right-side modifiers keep working, with safe recovery when a key release is lost.
• Windows Shortcut Recording — Fixed shortcut recording in settings not capturing any keys on Windows while the settings window had focus.
• Clearer Settings Guidance — In-app help and the ElevenLabs / Volcengine credential links now open the guides on voicepaste.dev, supported transcript confirmation modes are explained, and ElevenLabs API key setup is documented in both languages.

Under the Hood

• Added a website workspace with the bilingual product site and download portal; desktop and website checks are now split in CI.
• Updated contribution guidelines for the desktop and website codebases.

Downloads

平台	下载
macOS (Apple Silicon)	VoicePaste_2.1.5_aarch64.dmg
macOS (Intel)	VoicePaste_2.1.5_x64.dmg
Windows (x64 NSIS)	VoicePaste_2.1.5_x64-setup.exe
Windows (x64 MSI)	VoicePaste_2.1.5_x64_en-US.msi

Full Changelog: v2.1.4...v2.1.5

v2.1.4

What's New

This update substantially improves voice overlay animations, making appearance, dismissal, and resizing smoother. It also adds polishing failure retry and polishing from history, alongside hotkey improvements.

New Features & Improvements

• Voice Overlay Animation Upgrade — Improved centered fades, scaling, and size transitions on macOS / Windows for smoother appearance, dismissal, and state changes, while keeping the background, text, and indicators aligned throughout.
• Polishing Failure Retry — Failed polishing now shows the reason and holds the raw transcript. Press the global main hotkey to retry, Enter to output the original text, or Esc to discard it. After 5 seconds without input, the original text is delivered automatically, without re-recording. Enter and Esc can also interrupt an ongoing retry.
• Polish from History — Choose a template to polish or re-polish a successful transcription and update the existing entry. New entries retain the template and original transcript, so subsequent polishing uses the original text when available.
• Contextual Shortcut Hints — The voice overlay shows relevant cancel, finish, template, and retry shortcuts. Recording hints prioritize the most recently used eligible template, with the template preference retained across restarts.

Fixed

• Hotkey Event Filtering — Added key filtering on macOS / Windows so starting or stopping a recording and handling polishing failures do not also type into the foreground app or trigger its actions. macOS falls back to listening only when Accessibility permission is unavailable.
• Exact Hotkey Matching — Extra Ctrl / Alt / Shift / Command modifiers no longer trigger shorter bindings. Character keys in a main hotkey chord are filtered only when the chord matches, preserving ordinary presses of those keys.
• Hotkey Recording Protection — Keys are filtered while recording a shortcut in settings to reduce accidental typing and input-method interference. Global main hotkey recordings must start with a function key, Tab, or a modifier; invalid recordings show an explanation and restore the previous setting. Template stop keys still allow ordinary character keys.
• Polishing Errors and Wait Times — Distinguishes missing configuration, invalid credentials, unavailable models, incorrect service URLs, rate limits, network failures, and timeouts. Transient errors retry within bounded attempt and time limits. Reasoning is disabled where supported to reduce responses containing reasoning without answer text; truncated output is no longer retried automatically.
• History Recording Playback — Playback can now be stopped, and starting another recording stops the previous one. Fixed overlapping playback and stale button states after rapid clicks.
• History Action Feedback — Copying, deleting, re-transcribing, and polishing history entries now report success or failure.
• Overlay Layout Fixes — Fixed first-frame text truncation and stale animations interfering with a newly shown overlay.
• Overlay States and Waveform — Clearer preparation, transcription, polishing, and completion feedback, with more readable shortcut hints across light and dark backgrounds. Both platforms now use consistent rounded waveform animations that activate with sound and settle smoothly during silence, reducing low-level noise triggers.

Under the Hood

• Unified hotkey filtering through keytap to reduce duplicate system keyboard listeners.
• Fixed Rust formatting and cross-platform Clippy checks, and expanded tests for polishing, hotkeys, and overlay animation and layout.

Downloads

平台	下载
macOS (Apple Silicon)	VoicePaste_2.1.4_aarch64.dmg
macOS (Intel)	VoicePaste_2.1.4_x64.dmg
Windows (x64 NSIS)	VoicePaste_2.1.4_x64-setup.exe
Windows (x64 MSI)	VoicePaste_2.1.4_x64_en-US.msi

Full Changelog: v2.1.3...v2.1.4

v2.1.3

What's New

• Hotkey Settings Redesign — Click the input to record (no more per-row record buttons); the page splits into the global main key and per-feature hotkeys. Hands-free mode fixes the start key to the global main key, with a separate stop key.
• Re-transcribe from History — Entries that kept their recording can be re-transcribed from the saved audio without re-recording.
• Decoupled Built-in Libraries — Built-in prompt templates and hotword tables no longer re-merge on startup, so deletions stick; add them on demand via the "常用…" pickers.

Downloads

平台	下载
macOS (Apple Silicon)	VoicePaste_2.1.3_aarch64.dmg
macOS (Intel)	VoicePaste_2.1.3_x64.dmg
Windows (x64 NSIS)	VoicePaste_2.1.3_x64-setup.exe
Windows (x64 MSI)	VoicePaste_2.1.3_x64_en-US.msi

Full Changelog: v2.1.2...v2.1.3

v2.1.2

What's New

Hotfix release focused on Windows hotkey recording reliability.

Fixed

• Windows Hotkey Recording — Fixed hotkey recording intermittently failing on Windows. When the settings window had focus, WebView2 blocked the system keyboard hook so key events never reached the recorder; Windows now polls the physical keyboard state directly, bypassing the issue entirely. Alt / F10 are also blocked from activating the WebView2 menu bar, which otherwise paused page scripts and stalled the recording callback.
• macOS Settings Window — The settings window now comes to the front reliably on launch and when clicking the Dock icon.
• Doubao ASR — Switched the default to non-streaming mode for more stable recognition.

Downloads

平台	下载
macOS (Apple Silicon)	VoicePaste_2.1.2_aarch64.dmg
macOS (Intel)	VoicePaste_2.1.2_x64.dmg
Windows (x64 NSIS)	VoicePaste_2.1.2_x64-setup.exe
Windows (x64 MSI)	VoicePaste_2.1.2_x64_en-US.msi

Full Changelog: v2.1.1...v2.1.2

v2.1.1

What's New

Hotfix release for hotkey reliability.

Fixed

• Hotkey Stops Responding — Fixed the toggle hotkey silently failing when keys like CapsLock or NumLock stayed "held" (their keyup is often dropped by the OS). Non-hotkey keys are now ignored at the source so they can't jam the matcher, and any stuck key is reclaimed automatically after a few seconds idle.

Downloads

平台	下载
macOS (Apple Silicon)	VoicePaste_2.1.1_aarch64.dmg
macOS (Intel)	VoicePaste_2.1.1_x64.dmg
Windows (x64 NSIS)	VoicePaste_2.1.1_x64-setup.exe
Windows (x64 MSI)	VoicePaste_2.1.1_x64_en-US.msi

Full Changelog: v2.1.0...v2.1.1

v2.1.0

What's New

This release focuses on performance and stability: the recording path is fully off WebView (audio capture and cues moved into the Rust backend), the macOS overlay no longer depends on a WebView, and the settings window loads on demand — cutting idle memory by ~80MB. It also adds transcription-failure retry and backend hotkey recording.

New Features

• Native Audio Capture — Microphone capture moved entirely to the backend cpal (CoreAudio on macOS, WASAPI on Windows), replacing WebView getUserMedia. Fixes the quiet-then-loud volume ramp that hurt early-word recognition, and removes the browser audio settle delay for a snappier start.
• Cue Playback — Start/end cues now play via backend rodio, on a separate channel from mic capture. Native capture has no echo cancellation, so the start cue bleeds into the mic — the leading window is skipped so the cue is never mistaken for speech. Playback is also more stable and on-time, immune to WebView teardown.
• Backend Hotkey Recording — Hotkeys are recorded via the backend keytap instead of the DOM, so hardware-level keys the DOM can't see (e.g. macOS Fn) can now be bound.
• Memory — The settings WebView is now lazy (destroyed on close, not hidden) and the macOS overlay dropped its WebView, cutting idle memory by ~80MB.
• Optional Recording Retention — New "Keep Recordings" option in Settings → App Settings; recordings can be played back from the home history, and failed attempts are kept automatically for retry.
• Transcription Failure Retry — On online-ASR connect failure / timeout / empty result, retry by pressing the hotkey again or clicking retry — it re-transcribes the saved audio without re-recording. Commit timeout now returns an error instead of falling back to partial text, so incomplete results are never pasted as success.
• Overlay UI Polish — Refined the recording overlay's status hints; the pill now eases open with a transition and hint text breathes, making input feedback feel smoother.

Fixed

• Waveform — Smoother waveform (per-bar frequency bands + asymmetric envelope), with macOS / Windows height computation unified.
• Settings Always-on-Top — Fixed the settings window not staying on top when shown at launch or reopened mid-session.
• Hotkey Symbols — Fixed Windows showing macOS keyboard symbols; keycaps now render per-platform native glyphs.
• Hotkey Conflicts & Races — Reworked into a chord state machine: toggle fires on keyup cycles, hold on keydown, finalizing with the longest chord held during recording; resolves prefix conflicts (e.g. bare Ctrl stealing Ctrl+Shift) and rapid-press races.

Downloads

平台	下载
macOS (Apple Silicon)	VoicePaste_2.1.0_aarch64.dmg
macOS (Intel)	VoicePaste_2.1.0_x64.dmg
Windows (x64 NSIS)	VoicePaste_2.1.0_x64-setup.exe
Windows (x64 MSI)	VoicePaste_2.1.0_x64_en-US.msi

Full Changelog: v2.0.0...v2.1.0

v2.0.0

⚠️ BREAKING CHANGE: VoicePaste has been completely rewritten from Electron to Tauri v2 (Rust backend). This is a major architecture change. Config from 1.x is automatically migrated on first launch.

Architecture

• Electron → Tauri v2 — Entire app rebuilt with Tauri v2 (Rust + WebView). Installer reduced from ~120 MB to ~20 MB, idle memory from ~500 MB to ~80 MB.
• Apple Signed & Notarized — macOS builds are signed and notarized with an Apple Developer certificate — no Gatekeeper warnings on install.
• React / TypeScript Frontend — Settings UI rebuilt with React + TypeScript + Tailwind CSS. Unified design system with light / dark / system theme support.

ASR Speech Recognition

• Dual Engine (Cloud + Local) — Online: ByteDance Doubao streaming ASR with new API Key authentication. Local: sherpa-onnx with 4 offline models.
• 4 Local Models — SenseVoice (ZH/EN/JP/KO/Cantonese), Zipformer (ZH + EN bilingual), FunASR-Nano (ZH/EN + 7 dialects), Qwen3-ASR-0.6B (30 languages). On-demand loading, 500 MB+ memory, CPU / CUDA / CoreML acceleration.
• VAD + Punctuation — Built-in Silero VAD (Voice Activity Detection) for audio segmentation; optional Punctuation model for post-processing.
• Simulated Streaming — Local models without native streaming get real-time partial results via VAD-based segmented decoding.
• Auto-Reconnect — Online model auto-reconnects on disconnection for reliable long recordings.
• Dual Auth Compatibility — Supports both the new Volcengine API Key flow and the legacy APP ID / Access Token / Secret Key flow.

LLM Text Polishing

• 8 LLM Providers — DeepSeek, OpenAI, Anthropic, Gemini, OpenRouter, SiliconFlow, Ollama, and OpenAI-compatible APIs.
• Custom Prompt Templates — Built-in templates for general cleanup, translation, email drafting, and more. Customizable prompts with per-template hotkey bindings.
• Hotword Boosting — Three modes (Auto / Enabled / Disabled) to append hotwords to the LLM prompt for stronger recognition accuracy.

Hotwords

• Multi-Group Libraries — Create multiple hotword groups and switch between them.
• Weight Parameter — hotword|weight format with 1–10 range.
• Batch Add — Comma-separated input for adding multiple hotwords at once.
• Format Restoration — Automatically restores original formatting (capitalization, punctuation, special characters) after recognition.
• Online Hotword Table — Volcengine online hotwords work alongside local hotwords; local takes priority.

Hotkey System

• keytap Native Engine — Replaces uiohook-napi / tauri-plugin-global-shortcut. Supports modifier-only bindings and precise left/right key distinction.
• Toggle & Hold Modes — Toggle (press to start, press again to stop) and Hold (press and hold to speak, release to stop). ESC to cancel.
• Per-Template Hotkeys — Bind independent hotkeys to different text polishing templates, each with its own trigger mode.

Configuration System

• registry.json Model Registry — All model defaults managed in one place. Users only override what they need in config.yaml.
• Three-Layer Merge — Shared defaults → model-specific defaults → user overrides. Unset fields fall through automatically.
• Hot Reload — Changes take effect immediately on save — no restart required.
• 1.x Config Migration — Automatically imports Electron-era config on first v2.0 launch.

Other

• Beta Update Channel — Opt in via settings to receive beta release notifications. Stable users are unaffected.
• Structured Logging — Leveled, module-prefixed logging with file rotation (300KB, gzip) for easier troubleshooting.
• Custom Sound Effects — Replace the default start/end sounds with your own audio files.
• Keep Clipboard — Optionally keep results in the clipboard without auto-pasting.
• macOS Liquid Glass Overlay — Native AppKit Liquid Glass transparent overlay with light/dark/system theme support.
• Configuration Guide — New GUIDANCE.md covering quick start, model setup, hotwords, LLM, and all features.

---

Downloads

Platform	Download
macOS (Apple Silicon)	VoicePaste_2.0.0_aarch64.dmg
macOS (Intel)	VoicePaste_2.0.0_x64.dmg
Windows (x64 NSIS)	VoicePaste_2.0.0_x64-setup.exe
Windows (x64 MSI)	VoicePaste_2.0.0_x64_en-US.msi

Full Changelog: v1.3.0...v2.0.0

v1.3.0

⚠️ Please upgrade to VoicePaste 2.0.0

This is the final Electron release of VoicePaste. It exists to bring the built-in auto-updater to existing 1.2.0 installs so they can reach this version, after which you are guided to the new major release.

VoicePaste 2.0.0 has been rebuilt on a brand-new architecture (Tauri) for better performance and a much smaller app size. Because the architecture changed, 2.0.0 cannot be auto-updated from 1.x — please download it manually:

👉 https://github.com/that-yolanda/voicepaste/releases/latest

On the About page, a "前往下载" button now opens the GitHub Releases page directly.

What's New

• macOS Liquid Glass Overlay — New overlay appearance that adapts to light/dark mode with vibrancy fallback for older macOS versions.
• Notification Sounds — Configurable start and end sounds with custom file support; playback moved to the main process for reliability.
• History Improvements — Copy and delete individual history entries.
• LLM Prompt Update — Refined default text-structuring prompt for better polishing.
• Update Redirect to GitHub — Last Electron release before the Tauri migration. The update button now links to GitHub Releases for manual download instead of using the built-in auto-updater.

Fixes

• Fix ASR session race condition when recording stops quickly
• Fix built-in sound file loading path in packaged macOS app
• Skip notarization step when code signing is disabled in build
• Show older history records when recent days have no entries instead of empty state

Downloads

• VoicePaste-1.3.0-arm64.dmg — macOS (Apple Silicon)
• VoicePaste-1.3.0-x64.dmg — macOS (Intel)
• VoicePaste-1.3.0-win-x64.exe — Windows (x64 NSIS installer)

Full Changelog: v1.2.0...v1.3.0

v1.2.0

What's New

• LLM Text Polishing — Post-process ASR output via 8 LLM providers (DeepSeek, OpenAI, Anthropic, Gemini, OpenRouter, SiliconFlow, Ollama, custom), with multiple prompt templates and per-template hotkey bindings.
• Prompt Template Management — Create and manage multiple polishing templates in settings, each with its own hotkey and trigger mode.
• Real-time Audio Waveform — Live waveform visualization in the floating overlay during recording.
• Notification Sounds — Distinct start and end sounds to indicate recording readiness and recognition success.
• DMG Build Output — Added DMG format for macOS builds, fixing auto-update failures on read-only volumes.
• Hotkey & Settings Polish — Multi-provider selector, prompt template editor, ESC to unbind hotkeys, refined keycap display for left/right modifiers.

Downloads

• VoicePaste-1.2.0-arm64.dmg — macOS (Apple Silicon)
• VoicePaste-1.2.0-x64.dmg — macOS (Intel)
• VoicePaste-1.2.0-win-x64.exe — Windows (x64 NSIS installer)

Full Changelog: v1.1.0...v1.2.0

v1.1.0

What's New

• Settings Home Page — Added a home page with usage statistics, activity heatmap, and input history for a quick overview of your voice input activity.
• Unified Pack Command — Build macOS arm64 and x64 in a single pnpm run pack command, simplifying the release process.
• Settings Redesign — Reorganized settings page with sidebar navigation and auto-save for a cleaner editing experience.
• Auto-Update Fix — Prevented the auto-update check from re-triggering when saving config, avoiding unnecessary network requests.
• CI & Release Overhaul — Unified the pack script, added CI pipeline, and revamped the release skill for a more reliable build process.

Downloads

• VoicePaste-1.1.0-arm64.zip — macOS (Apple Silicon)
• VoicePaste-1.1.0-x64.zip — macOS (Intel)
• VoicePaste-1.1.0-win-x64.exe — Windows (x64 NSIS installer)

Full Changelog: v1.0.8...v1.1.0