Skip to content

Voice-to-text dictation app with local (Nvidia Parakeet/Whisper) and cloud models (BYOK). Privacy-first and available cross-platform.

chore: bump to v1.5.3 with changelog and README updates

Gabriel Stein764612a504ec
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 427b229bf..590252837 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,6 +7,32 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
## [Unreleased]
+## [1.5.3] - 2026-02-24
+
+### Added
+
+- **Unified GPU Banners**: Replaced dual CUDA/Vulkan banners on the home screen with a single GPU acceleration banner; added GPU banners to Transcription Settings and AI Text Enhancement Settings
+- **GpuStatusBadge Redesign**: Auto-retry flow (download → activating → GPU active) with 15s timeout, replacing confusing "CPU Only" and "Re-detect GPU" states; swapped hardcoded hex colors for `bg-success`/`bg-warning` design tokens
+- **Streaming Usage Tracking**: Wired up the previously-uncalled `/api/streaming-usage` endpoint so Deepgram streaming transcriptions report word counts to the server
+- **Cloud API Telemetry**: Forward STT metadata (`sttProvider`, `sttModel`, processing time, audio duration/size/format) and `clientVersion`/`clientType`/`appVersion` to all cloud API requests
+- **Internationalization**: Added 15 missing i18n keys (`app.mic.*`, `app.commandMenu.*`, `app.toasts.*`, `app.oauth.*`, `notes.enhance.title`) across all 10 locale files
+
+### Fixed
+
+- **Windows Blank Screen**: Fixed blank screen on return from sleep/minimize by adding `render-process-gone` handler, `isCrashed()` health checks on show/tray/second-instance paths, `backgroundColor` and `backgroundThrottling` to window config, and `disable-gpu-compositing` for win32
+- **IPC Echo Loop**: Broke infinite IPC bounce in floating icon auto-hide toggle by guarding the setter with an early return when the value hasn't changed
+- **GPU Banner Navigation**: GPU banner "Enable GPU" button now navigates to the correct `"intelligence"` settings section instead of invalid `"reasoning"` ID
+- **AI CTA Deep Link**: Replaced legacy `"aiModels"` alias with canonical `"intelligence"` section ID in the AI enhancement CTA button
+- **Custom Endpoint Routing** (#311): Moved `reasoningProvider === "custom"` check to the top of `getModelProvider()` so custom endpoint models are never misrouted through built-in providers; custom models now show a neutral Globe icon
+- **KDE Wayland Terminal Detection**: Detect Konsole via `kdotool` (fast path) or KWin `supportInformation` via `qdbus` (zero-install fallback) so terminals receive `Ctrl+Shift+V` instead of `Ctrl+V`
+- **RAM Leak on Provider Switch**: Whisper, Parakeet, and llama-server processes now stop when switching to cloud providers, freeing loaded models from RAM
+- **Streaming Usage Session Refresh**: Wrapped `cloudStreamingUsage` in `withSessionRefresh` so expired sessions auto-refresh instead of silently dropping word counts
+- **Duplicate Transcription Logs**: Skip telemetry logging in streaming-usage and transcribe endpoints when reasoning is enabled (the `/api/reason` endpoint already creates the combined row)
+- **Usage Cache Invalidation**: `useUsage` hook now listens for `usage-changed` events to invalidate its cache and refetch immediately after transcription
+- **macOS Binary Architecture**: Added Mach-O header verification to globe-listener and fast-paste build scripts; force rebuild when architecture-specific hash file is missing; runtime architecture check before spawning binary
+- **Globe Key Listener Resilience**: Auto-restart globe key listener on unexpected exit code 0 (sleep/wake invalidation); reset restart counter after sustained uptime; only treat "Failed to create event tap" as fatal
+- **Parakeet Long Recordings**: Lowered max segment duration from 30s to 15s for more reliable chunked transcription; downgraded reasoning failure log from error to warn
+
## [1.5.2] - 2026-02-24
### Fixed
diff --git a/README.md b/README.md
index 065bf816c..e42a5e38a 100644
--- a/README.md
+++ b/README.md
@@ -730,7 +730,7 @@ A: OpenWhispr supports 58 languages including English, Spanish, French, German,
## Project Status
-OpenWhispr is actively maintained and ready for production use. Current version: 1.5.2
+OpenWhispr is actively maintained and ready for production use. Current version: 1.5.3
- ✅ Core functionality complete
- ✅ Cross-platform support (macOS, Windows, Linux)
diff --git a/package-lock.json b/package-lock.json
index 8632690c6..a59071e08 100644
--- a/package-lock.json
+++ b/package-lock.json
@@ -1,12 +1,12 @@
{
"name": "open-whispr",
- "version": "1.5.2",
+ "version": "1.5.3",
"lockfileVersion": 3,
"requires": true,
"packages": {
"": {
"name": "open-whispr",
- "version": "1.5.2",
+ "version": "1.5.3",
"hasInstallScript": true,
"license": "MIT",
"dependencies": {
diff --git a/package.json b/package.json
index 9a62e0e07..06b0ba6ef 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
{
"name": "open-whispr",
- "version": "1.5.2",
+ "version": "1.5.3",
"description": "A desktop dictation application using whisper.cpp for speech-to-text transcription",
"main": "main.js",
"private": true,