Paraspeech Changelog

What's New

• iOS app beta - Paraspeech for iPhone and iPad is now available in beta and included with your subscription. Open the Paraspeech sidebar to download it.
• Intel Mac support - Paraspeech now runs on Intel Macs using cloud models.
• macOS 14 support - Minimum macOS support is now macOS 14.0+.
• Pause recordings - Pause from the recording pill and resume when you're ready.
• Rewrite improvements - Better rewrite behavior and more reliable cleanup.
• Better Recording Pill - Less intrusive when minimized, more functionality and better design.

Fixes

• Fixes and reliability improvements across recording, updates, access, downloads, and diagnostics.

What's New

• Faster Japanese and Mandarin dictation - New dedicated local speech models improve single-language setup.
• Sharper dictated emails - Paraspeech better preserves recipients, language, tone, and sign-offs.
• Cleaner local rewrites - Filler, corrections, repeated details, and short follow-ups come out more polished.
• More resilient recordings - Recording sessions handle completion and partial audio issues more cleanly.
• Easier permission setup - Accessibility setup is clearer when Paraspeech needs help typing into other apps.
• Smoother access recovery - Sign-in and already-purchased flows are steadier after relaunch, reinstall, or checkout.

What's New

• Cleaner first-run setup - Onboarding and access routing are tighter, so it is easier to get from install to usable state without bouncing through the wrong recovery path.
• Better local writing output - The local rewrite model set and prompt tuning were refreshed for stronger email cleanup and follow-up drafting.
• Sharper access recovery - Direct sign-in and already-purchased flows behave more predictably when local auth state needs to be restored.

Fixes

• Less onboarding friction - Welcome and recovery screens were simplified so the next step is clearer when access changes mid-setup.
• More reliable sign-in restore - Restoring access after reinstall or account recovery is more dependable.

What's New

• Smoother setup and recovery — Onboarding, access recovery, and sign-in flows behave more predictably in edge cases so it is easier to get back into the app after install, reset, or purchase changes.
• Sharper writing models — The local rewrite prompt set was expanded and tuned for cleaner email, cleanup, and follow-up output.
• Easier first install flow — Fresh installs now move into the app more predictably with fewer setup detours.

Fixes

• More reliable account and license recovery — Account and purchase access come back more consistently after install, reset, and purchase changes.
• Clearer recovery paths — Blocked-account, pending purchase, confirmed access, and linked-local recovery flows now stay in the right lane more consistently.

What's New

• Accounts across your devices — In preparation for the iOS app launch, accounts have been introduced so you can access the app for free. Existing users do not have to set up an account to continue using the Mac app - only if they want to use the iOS app.
• A more capable Dictionary and Smarter Rules — Dictionary and word replacement settings are easier to understand, with examples, info popovers, and better validation for custom vocabulary.
• Better rewrite models — Paraspeech now supports newer Qwen 3.5 rewrite models, plus model prewarming so AI features feel more responsive when you open them. The average rewrite now takes <3 seconds
• Better navigation and settings — The app UI was reorganized with clearer sections, a new Stats tab, easier access to Dictionary and Rewrite settings, and improved account controls.
• Faster local setup — Model downloads and permission steps are clearer, with better progress reporting and a smoother path to getting the Mac app ready.

Fixes

• More reliable purchase and access recovery — Direct purchases, legacy licenses, and account-linked access now recover more consistently after checkout, reinstalls, and device changes.
• Better crash recovery and diagnostics — Recovery flows were simplified to make troubleshooting more dependable.
• Small quality-of-life improvements — Includes an untranscribed history filter, better legacy-model handling, and a set of interface polish fixes across settings and overlays.

What's new?

• Added new Qwen3 Beta speech model.
• Fixes

What's New

• Minor stability and UX updates

What's New

• Fixed keyboard shortcuts not working after waking from sleep
• Fixed auto-updater issues for some users

What's New

• Fixed Rewrite not enabling when Apple Intelligence is not available

What's New

• Removed persistent UI artifact on screen
• Audio Engine now remains stable no matter if input/output changes unexpectedly
• General stability/UX updates

What's New

• Stability Improvements and Bug Fixes

Fixes

• Stability Improvements and Bug Fixes

Fixes

• Crash recovery UI composition fixed
• "Always show recording pill" toggle fixed
• Slow Startup fixes

Fixes

• Stability improvements — Reduced startup hangs and UI stalls in key workflows.
• Audio reliability fixes — Better resilience when audio devices change or the engine restarts.
• Background cleanup safety — Fewer interruptions from history reconciliation work.
• Crash and hang fixes across the menu bar and transcription history.

What's New

• Faster and more accurate transcription models — Existing models are now even faster and there are two new models that are a bit slower, but much more accurate.
• 100+ Language Supported — Paraspeech now supports a much broader set of languages across the new audio models.
• Automatic Translation — You can now also talk in your native language and auto-translate to English - all in one go.
• Quality of life improvements — Small polish changes across the app to make everyday dictation smoother.
• Revamped UI for better usability — Quick access to microphone options in the menu bar, more intuitive history item views, and more.
• Smarter Transcription Clean-Up — Paraspeech now applies app-specific post-processing rules. Transcriptions are automatically formatted based on the app you're dictating into—emails get proper formatting, code editors preserve technical terms, and more.
• Interactive Recording Indicator — New option to keep the status overlay visible while recording, so you always know when Paraspeech is listening.
• New Stats View — A fresh look at your transcription statistics with improved layout, better metrics visualization, and a shareable stats card to show off your usage.

Fixes

• Lots of small fixes for edge cases. 1.5.1:
• auto send rules configurable per app
• various UI/UX optimizations
• bug fixes

What's New

• New Recording Sounds - Fully redesign start, stop and error sounds to keep you focused while still giving you feedback.
• UI tidied up — The UI was overhauled to give you the settings you need where you need them, while still giving control to advanced users.
• Better crash recovery — Crash handling now gives clearer recovery guidance after an unexpected quit.

Fixes

• More reliable transcription history — Fixed edge cases where history entries could show incorrect timestamps or appear out of order.
• Cleaner history view — Long transcripts now show a truncation indicator instead of overflowing the list view.

What's New

• Apple Intelligence integration — on macOS 26+, use Apple's built-in AI model to clean up transcripts with no downloads required.
• Redesigned history view with inline actions, AI-processed badges, and a glowing amber indicator when AI mode is active.
• System hotkey conflict warnings for common macOS shortcuts like ⌘Space or ⌘Tab.
• Lifetime license clarity on the checkout screen.

Fixes

• More reliable audio capture with better recovery when microphones disconnect or switch.
• Smoother app launch by resolving onboarding stalls on first launch.
• Reduced repetitive error logs from hotkey monitoring and audio capture retries.

What's New

• Graceful audio drain — recordings now fully flush buffered audio before stopping, so you never lose the last few words of your speech.
• Streamlined App Store onboarding — the EULA step is now skipped for App Store builds since Apple's Terms of Service already covers it.

Fixes

• Fixed AI cleanup prompts not being applied correctly for some local models.
• Fixed checkout session recovery when purchases are interrupted mid-flow.
• Improved subscription handling during the purchase flow.

What's New

• Per-model LLM settings — fine-tune temperature, top-p, min-p, and repetition penalty individually for each AI model instead of using global defaults.
• Improved menu bar reliability — the popover now responds more consistently when clicking the menu bar icon.
• Launch at Login prompt — App Store users are now prompted to enable launch at login after completing onboarding.

Fixes

• Fixed "lost activation" bug where licenses could become deactivated after app restart or system changes.
• Fixed Launch at Login not registering properly on fresh installs.

What's New

• All new UI
• AI-powered transcription cleanup — automatically fixes formatting, removes filler words, and polishes punctuation.
• Ultra-reliable audio capture engine with automatic recovery when your microphone disconnects or changes mid-recording. We're preparing for a big release with this.
• Bluetooth headset pre-warming reduces the delay when starting recordings with wireless audio devices.
• Voice note retention settings let you control how long your audio recordings are stored.
• Transcription history now supports pagination and search.
• Redesigned licensing and activation experience with easy license recovery.
• Further improved number conversion algorithm (more updates for this coming soon)

Fixes

• Fixed occasional audio cutoff at the end of recordings.
• Fixed accidental double-trigger when quickly tapping the record key.
• Improved stability when switching audio devices during recording.
• Better handling of system wake from sleep.
• Launch on Login fixed
• ...and 30+ other small fixes.

What's New

• Pick your speech model (Multilingual vs. English Only): English only will perform better and more accurate, especially if you talk in english with an accent. Multilingual supports 100+ languages.
• Privacy Mode Toggle: keeps history empty and wipes existing recordings the moment you flip it on
• Recording overlay now has Default, Compact, and Hidden layouts so you can shrink it or hide it entirely and also add a timer if you prefer
• Word replacements support case-sensitive rules for names, brands, and acronyms that need exact casing
• Recording shortcuts accept combinations with normal letters, Hyper-style combos, and the double-tap is now totally optional
• Audio Dampening: additionally to muting audio while recording, you can now make Paraspeech reduce volume by 50%

Fixes

• Fixed a bug that prevented some users from recording longer sessions
• General stability fixes

New Features

• File based Transcription and VTT Export (beta): transcribe existing audio files and export subtitles.
• New recording indicator for better visual feedback during recording.
• Auto-mute during Recording: automatically mutes system audio while recording.
• New Text Processing experiments: Smart filler removal, better number converter, and spoken punctuation support.
• Fully customize word replacements/removals without any limitations.

Small Things

• Light mode compatibility improvements.
• New settings panel with more configuration options.
• Paraspeech now also lives in your Dock for easier access.

Fixes

• Small changes for App Store Release.

What's New

• Expanded Hotkey system for power users: You can now switch to 'split mode' to set two different hotkeys for toggle and hold-to-record as well as choosing if you want to double or single-tap for toggle recordings.
• Clear History button added.
• Double tap ESC to cancel a recording.

Fixes

• App icon updated for proper scaling.

What's New

• You can now set the same key for hold-to-record and toggle-to-record but I had to tighten the hotkey system a bit: only modifier keys for now, otherwise I'll have to ask for input monitoring permission, which I'd like to avoid for now.
• Simplified and more robust microphone and bluetooth headset handler.
• Added new forum for bugs and feature requests: https://feedback.paraspeech.com
• Introduced opt-in usage stats through PostHog (EU Servers) helping me with debugging or trying to understand which features are worth maintaining/expanding. This is fully optional and even if opted-in, it never ever sends audio, transcripts, clipboard content or personal info. It's purely focused on usage stats, but it helps me a lot.
• The three most recent recordings are now saved on your mac in case something goes wrong and you can retranscribe the files from the menu.
• Added an indicator if your mic input is muted or not working, so you don't speak into the void.

Fixes

• Robust handling of mic states and changes.
• Full compatibility with all bluetooth headsets.
• Much better recording stability in edge cases or on weird macOS/hardware combinations.
• Clipboard repopulation now happens about half a second after paste. The delay is necessary for some weird edge cases so I can make sure your previous clipboard content is always restored.
• Status Indicator should now stay where it's supposed to be and will not yield to macOS trying to reorder it below the notch.
• Lot's of small ones: if you reached out to me with a bug, there is a very high chance I fixed it in this version.

What's New

• Bluetooth headset alerts now explain the trade-offs, let you choose a recommended mic, and you can turn the warnings off if you don't need them.
• The audio menu refreshes itself, shows which mic macOS is using, and keeps your saved headset choices tidy.

Fixes

• Switching between AirPods, wired mics, and the Mac microphone is smoother, so recordings keep rolling without dropouts.

Highlights

• Bluetooth mic chooser keeps you in control when AirPods or headsets connect, so audio never downgrades without asking.
• Model upgrader walks everyone through the latest FluidAudio bundle on first launch for better accuracy.
• Completely reworked hotkeys: any combo you like, hyper combos, Fn/Globe, with smarter conflict warnings and a tuned double-tap delay.
• Left/right modifier detection has been retired so shortcuts treat modifiers as side-agnostic for App Store compliance.
• Onboarding is calmer with a better tryout step, recovery paths, and no duplicate model downloads.
• Now officially supported on macOS 14.6 and newer (Sonoma and beyond).

Fixes & refinements

• Clipboard content is preserved reliably, including images and rich data.
• Light mode UI, overlays, and menu icons render crisply without artifacts.
• Status overlay no longer glitches on specific macOS builds.
• Settings received a clean sweep: fast mode ships off by default and audio input controls are surfaced.
• AirPods connect and disconnect gracefully without crashing Paraspeech.
• Resolved rare engine warm-up crashes during launch and transcription start.
• First recording after long idle now transcribes every time.
• Audio engine cooperates with other apps and no longer steals input focus.
• Device switching is stable on macOS 14.6+.
• Model downloads persist after onboarding to avoid repeat fetches.
• Single-key triggers now respect modifier presses so you do not fire commands by accident.
• Hyper, Globe, and sided modifier combinations are honoured everywhere.
• Double-tap toggle latency trimmed to a swift 250 ms.
• Clipboard restore survives other apps touching the pasteboard mid-flight.
• Onboarding, settings separators, and scrollbars all feel lighter and smoother.
• Free tier now includes 50 transcriptions; licensing checks run once a day, stay silent offline, and fall back gracefully if a license is revoked.
• Fast recording is temporarily removed while I rework the flow.