Speech Recognition & AI
Multilingual ASR Model Upgrade
We’ve trained a new unified multilingual speech recognition model that covers Japanese, English, Chinese, Korean, and Vietnamese in a single model — replacing separate per-language models. Key improvements:- Short utterance accuracy improved from 92% to 96%, now surpassing OpenAI Whisper large v3 (95%)
- Language detection accuracy improved from 71% to 96%, dramatically reducing incorrect language identification
- Improved noise robustness — tested with background noise, music, and environmental sounds
Translation Quality Improvement
Our self-hosted translation engine has been significantly upgraded:- Higher translation accuracy across all supported language pairs
- Faster translation speed, especially for Chinese language pairs
- Unified multilingual architecture for more consistent quality across languages
STT Correction Refinements
Updated our real-time speech correction to be less aggressive — the system now preserves fillers and natural speech patterns (e.g., “um,” “あの~”) instead of removing them. This reduces text flickering during live transcription and provides a more natural reading experience.Korean Word-Level Highlight
Korean language transcripts now support word-by-word highlight playback, enabling precise navigation through recorded meeting content.New Features
Business Intelligence Dashboard
- Removed the “beta” label — the BI dashboard is now a standard feature
- Fixed mindmap graph generation to work consistently across Japanese and English
- Restored zoom state persistence so your preferred view is remembered between sessions
Meeting Notes
- Default templates now always appear when no custom template exists
- All users (including members) can now switch between note templates
- Fixed an issue where notes taken without a template could not be edited afterward
Listener Mode Improvements
- Listeners can now join Audio Extraction sessions even before the host connects
- Listener count is now accurately tracked in session logs, including cases where the host doesn’t speak
- Password change during an active session now properly notifies and disconnects listeners with a clear message
Mobile App Updates
iOS & Android Release (v0.5.10+)
- Apple Sign-In: Added automatic account creation via Apple Sign-In, matching the existing Google Sign-In experience
- Meeting summary recreation: You can now regenerate AI meeting summaries directly from the mobile app
- Word highlight playback: Fixed text flickering during transcript replay for smooth, accurate word-by-word highlighting
- 3x playback speed: Added to match the web app’s playback speed options
- Offline improvements: Better handling of network interruptions and offline-to-online transitions
- Android deep link fix: Opening transcript URLs from outside the app no longer shows a black screen
Bug Fixes
- Fixed false “plan limit exceeded” notification that could appear when the server was temporarily unreachable, even though usage was well within limits
- Fixed duplicate meeting logs where mobile and desktop sessions could create two separate records for the same meeting
- Fixed plan cancellation button that was blocked when the reason text was shorter than 15 characters (especially problematic for Japanese input)
- Fixed Stripe plan display showing incorrect usage minutes on the payment screen
- Fixed translation progress indicator that appeared incorrectly when no translation language was selected
- Fixed transcript mode cleanup where the microphone status could become corrupted after stopping a session
- Resolved duplicate login modal issue across web and mobile platforms
- Fixed Chinese dictionary translations where entries were not being applied correctly due to a key format mismatch
UI/UX Improvements
- Listener mode entrance screen: Redesigned with a scrollable workspace list (previously limited to 4 visible items) and improved visual styling
- Toast notification cleanup: Removed distracting edit/delete notifications during live events — important for presentations where VoicePing is displayed on screen
- Dashboard locale fixes: Corrected display formatting issues in the Japanese interface
- Meeting summary display: Removed character limits on summary text for both mobile and desktop, showing the full AI-generated summary
Other
- Deprecated the remote control feature due to licensing restrictions. Affected users have been notified.