Releases: idanyekutiel/wispah
Releases · idanyekutiel/wispah
Wispah Flow Version 2026.04.20
Installation
- Download the DMG below
- Open it and drag Wispah to Applications
- Launch Wispah Flow and follow the setup wizard
Changelog
- Rewrote recording error banners to one deterministic panel — Recording errors now use a single top-of-screen banner with one dismiss path instead of the old pill-plus-drop animation flow that could leave stale UI stranded onscreen.
- Added reliable manual and timed error dismissal — Error banners can now always be cleared with
xor by tapping the banner, and they auto-dismiss after12s. - Fixed automatic transcription mismatching manual retranscribe — When the first automatic upload fails or comes back empty after speech-boundary trimming, the app now retries the full saved audio before falling back to recovered chunk audio.
- Treats
"Thank you for watching"endings as a suspect transcript — The recorder now mistrusts that known hallucinated outro and retries automatically instead of accepting a truncated transcript as final. - Caps hallucination-triggered retries to one extra attempt — Known suspicious outro detection can only trigger one automatic retry, so it cannot spin through repeated fallback loops.
- Prevents duplicate recovered-audio retries inside one transcription task — The app now assembles fallback audio at most once per recording and reuses that recovery source instead of re-entering destructive chunk recovery with stale file paths.
- Fixed the stuck recording error-banner race — Error overlays now cancel their delayed label-drop work item when dismissed, so they cannot reappear after dismissal and get stranded onscreen.
- Preserves chunk files when the master recording is chosen — The recorder no longer deletes chunk checkpoints as soon as the primary
.caffile wins, so the automatic fallback retry still has real audio files left to assemble if the primary transcription comes back empty. - Fixed updater checks when GitHub’s REST API is rate-limited — Update checks now send proper GitHub API headers and fall back to the public latest-release page if the API returns a
403rate-limit response, so the updater no longer dead-ends on that error. - Moves recovered chunk audio into a separate retry file — Automatic fallback transcription now assembles chunk recovery into its own temporary audio file instead of mutating the original capture path in place, which avoids the missing-file failure that could still abort recovery.
- Automatically retries transcription with recovered chunk audio — If the fast primary recording returns an empty transcript or fails during transcription, the app now assembles the preserved fallback chunks and retries automatically instead of leaving recovery to manual retranscribe.
- Keeps fallback chunks alive until transcription finishes — Recording cleanup no longer destroys chunk checkpoints immediately after stop, so the app can still recover from a bad primary file during the same transcription pass.
- Fixed stuck error overlays — Error banners now always schedule a managed auto-dismiss cleanup instead of relying on an unmanaged timer path, so file and recording errors no longer get stranded on screen.
- Added a manual dismiss control for recording errors — Error overlays now expose an explicit close button, so even if the automatic cleanup path is delayed you can still clear the banner immediately.
- Rebuilt recording for fast start and fallback recovery — Recording now enters its active state immediately after the mic session and writers are armed, instead of waiting for the first analyzed buffer before the UI can proceed.
- Added master recording plus rolling chunk checkpoints — The recorder now writes one fast primary file and rolling
5scheckpoint chunks from the same live PCM buffers, so normal stops stay fast while long recordings still have validated fallback audio if the primary file goes bad. - Improved long-recording failure detection — The app now validates actual written audio against wall-clock recording time and refuses to send obviously broken one-second files into transcription.
- Fixed intermittent near-empty recordings — Recording startup now waits for both the first live audio buffer and AVFoundation's file-output start callback before the app considers the recorder ready, which closes the race that could occasionally produce
0:01recordings. - Tightened record-start timing on macOS — The recorder now opts into AVFoundation's sample-accurate file-recording start path so stop events cannot beat the file writer as easily on short or fast interactions.
- Fixed stop-to-transcribe hangs on macOS — Pressing the hotkey to stop recording now hands file finalization off asynchronously instead of blocking the app while AVFoundation finishes the recording file, so transcription can begin reliably after stop.
- Removed main-thread audio finalization stalls — The recorder now finalizes long recordings in the background and only returns to the app once the audio file is actually ready, which avoids the blank stopped state where the waveform froze but transcription never started.
- Reworked long recording stability on macOS — The recorder now uses
AVCaptureAudioFileOutputfor the actual on-disk recording file instead of manually appending audio buffers throughAVAssetWriterInput, which is a safer fit for longer-running captures. - Separated recording from waveform analysis —
AVCaptureAudioDataOutputis now used only for waveform and speech detection, while file writing is handled by AVFoundation's dedicated file output path. - Improved Studio Display compatibility — External microphones like Studio Display now go through the same dedicated file-recording path, which should avoid the corrupted playback and transcription issues that only showed up on longer recordings.
Requirements
Wispah Flow Version 2026.04.19.1
Installation
- Download the DMG below
- Open it and drag Wispah to Applications
- Launch Wispah Flow and follow the setup wizard
Changelog
- Prevents duplicate recovered-audio retries inside one transcription task — The app now assembles fallback audio at most once per recording and reuses that recovery source instead of re-entering destructive chunk recovery with stale file paths.
- Fixed the stuck recording error-banner race — Error overlays now cancel their delayed label-drop work item when dismissed, so they cannot reappear after dismissal and get stranded onscreen.
Requirements
Wispah Flow Version 2026.04.19
Installation
- Download the DMG below
- Open it and drag Wispah to Applications
- Launch Wispah Flow and follow the setup wizard
Changelog
- Preserves chunk files when the master recording is chosen — The recorder no longer deletes chunk checkpoints as soon as the primary
.caffile wins, so the automatic fallback retry still has real audio files left to assemble if the primary transcription comes back empty.
Requirements
Wispah Flow Version 2026.04.18.4
Installation
- Download the DMG below
- Open it and drag Wispah to Applications
- Launch Wispah Flow and follow the setup wizard
Changelog
- Fixed updater checks when GitHub’s REST API is rate-limited — Update checks now send proper GitHub API headers and fall back to the public latest-release page if the API returns a
403rate-limit response, so the updater no longer dead-ends on that error.
Requirements
Wispah Flow Version 2026.04.18.3
Installation
- Download the DMG below
- Open it and drag Wispah to Applications
- Launch Wispah Flow and follow the setup wizard
Changelog
- Moves recovered chunk audio into a separate retry file — Automatic fallback transcription now assembles chunk recovery into its own temporary audio file instead of mutating the original capture path in place, which avoids the missing-file failure that could still abort recovery.
- Automatically retries transcription with recovered chunk audio — If the fast primary recording returns an empty transcript or fails during transcription, the app now assembles the preserved fallback chunks and retries automatically instead of leaving recovery to manual retranscribe.
Requirements
Wispah Flow Version 2026.04.18.2
Installation
- Download the DMG below
- Open it and drag Wispah to Applications
- Launch Wispah Flow and follow the setup wizard
Changelog
- Automatically retries transcription with recovered chunk audio — If the fast primary recording returns an empty transcript or fails during transcription, the app now assembles the preserved fallback chunks and retries automatically instead of leaving recovery to manual retranscribe.
- Keeps fallback chunks alive until transcription finishes — Recording cleanup no longer destroys chunk checkpoints immediately after stop, so the app can still recover from a bad primary file during the same transcription pass.
Requirements
Wispah Flow Version 2026.04.18.1
Installation
- Download the DMG below
- Open it and drag Wispah to Applications
- Launch Wispah Flow and follow the setup wizard
Changelog
- Fixed stuck error overlays — Recording and file error banners now clean themselves up through a managed auto-dismiss path instead of getting stranded on screen.
- Added a manual dismiss control for recording errors — Error overlays now include an explicit close button, so you can clear them immediately if you do not want to wait for the timeout.
Requirements
Wispah Flow Version 2026.04.18
Installation
- Download the DMG below
- Open it and drag Wispah to Applications
- Launch Wispah Flow and follow the setup wizard
Changelog
- Rebuilt recording for fast start and fallback recovery — Recording now enters its active state immediately after the mic session and writers are armed, instead of waiting for the first analyzed buffer before the UI can proceed.
- Added master recording plus rolling chunk checkpoints — The recorder now writes one fast primary file and rolling
5scheckpoint chunks from the same live PCM buffers, so normal stops stay fast while long recordings still have validated fallback audio if the primary file goes bad. - Improved long-recording failure detection — The app now validates actual written audio against wall-clock recording time and refuses to send obviously broken one-second files into transcription.
- Fixed intermittent near-empty recordings — Recording startup now waits for both the first live audio buffer and AVFoundation's file-output start callback before the app considers the recorder ready, which closes the race that could occasionally produce
0:01recordings. - Tightened record-start timing on macOS — The recorder now opts into AVFoundation's sample-accurate file-recording start path so stop events cannot beat the file writer as easily on short or fast interactions.
- Fixed stop-to-transcribe hangs on macOS — Pressing the hotkey to stop recording now hands file finalization off asynchronously instead of blocking the app while AVFoundation finishes the recording file, so transcription can begin reliably after stop.
- Removed main-thread audio finalization stalls — The recorder now finalizes long recordings in the background and only returns to the app once the audio file is actually ready, which avoids the blank stopped state where the waveform froze but transcription never started.
- Reworked long recording stability on macOS — The recorder now uses
AVCaptureAudioFileOutputfor the actual on-disk recording file instead of manually appending audio buffers throughAVAssetWriterInput, which is a safer fit for longer-running captures. - Separated recording from waveform analysis —
AVCaptureAudioDataOutputis now used only for waveform and speech detection, while file writing is handled by AVFoundation's dedicated file output path. - Improved Studio Display compatibility — External microphones like Studio Display now go through the same dedicated file-recording path, which should avoid the corrupted playback and transcription issues that only showed up on longer recordings.
Requirements
Wispah Flow Version 2026.04.16
Installation
- Download the DMG below
- Open it and drag Wispah to Applications
- Launch Wispah Flow and follow the setup wizard
Changelog
- Fixed intermittent near-empty recordings — Recording startup now waits for both the first live audio buffer and AVFoundation's file-output start callback before the app considers the recorder ready, which closes the race that could occasionally produce
0:01recordings. - Tightened record-start timing on macOS — The recorder now opts into AVFoundation's sample-accurate file-recording start path so stop events cannot beat the file writer as easily on short or fast interactions.
- Fixed stop-to-transcribe hangs on macOS — Pressing the hotkey to stop recording now hands file finalization off asynchronously instead of blocking the app while AVFoundation finishes the recording file, so transcription can begin reliably after stop.
- Removed main-thread audio finalization stalls — The recorder now finalizes long recordings in the background and only returns to the app once the audio file is actually ready, which avoids the blank stopped state where the waveform froze but transcription never started.
- Reworked long recording stability on macOS — The recorder now uses
AVCaptureAudioFileOutputfor the actual on-disk recording file instead of manually appending audio buffers throughAVAssetWriterInput, which is a safer fit for longer-running captures. - Separated recording from waveform analysis —
AVCaptureAudioDataOutputis now used only for waveform and speech detection, while file writing is handled by AVFoundation's dedicated file output path. - Improved Studio Display compatibility — External microphones like Studio Display now go through the same dedicated file-recording path, which should avoid the corrupted playback and transcription issues that only showed up on longer recordings.
Requirements
Wispah Flow Version 2026.04.14.4
Installation
- Download the DMG below
- Open it and drag Wispah to Applications
- Launch Wispah Flow and follow the setup wizard
Changelog
- Fixed stop-to-transcribe hangs on macOS — Pressing the hotkey to stop recording now hands file finalization off asynchronously instead of blocking the app while AVFoundation finishes the recording file, so transcription can begin reliably after stop.
- Removed main-thread audio finalization stalls — The recorder now finalizes long recordings in the background and only returns to the app once the audio file is actually ready, which avoids the blank stopped state where the waveform froze but transcription never started.
- Reworked long recording stability on macOS — The recorder now uses
AVCaptureAudioFileOutputfor the actual on-disk recording file instead of manually appending audio buffers throughAVAssetWriterInput, which is a safer fit for longer-running captures. - Separated recording from waveform analysis —
AVCaptureAudioDataOutputis now used only for waveform and speech detection, while file writing is handled by AVFoundation's dedicated file output path. - Improved Studio Display compatibility — External microphones like Studio Display now go through the same dedicated file-recording path, which should avoid the corrupted playback and transcription issues that only showed up on longer recordings.