Speech Recognition & AI
40% Faster STT Inference
Upgraded the STT pipeline, delivering 40% faster inference across all languages.Bilingual Mode Stability Fixes
Multiple critical bilingual mode fixes deployed throughout March:- Fixed language detection being swapped on first recognition on mobile
- Fixed text correction breaking language detection (Japanese speech incorrectly transcribed as English)
- Fixed audio loss when speakers switch languages mid-conversation
- Short utterance accuracy restored with adaptive audio length detection
iOS Audio Quality Improvement
iOS was using a telephony-grade audio mode instead of the measurement mode used on Android. Changed to proper STT audio capture mode, bringing iOS audio quality to parity with Android at 1-2m distances.Malay Language Support
Malay STT added. Bilingual mode support for Malay also added, with the Indonesian/Malay pair intentionally disabled (language switching between the two cannot be reliably detected).Faster Translation
Deployed a new translation model with significantly faster inference, improving real-time translation responsiveness.More Languages Coming
Khmer ASR model training started. Additional languages in preparation: Cantonese, Uzbek, Polish, Dutch, Mongolian, and Bengali.Meeting Log & Transcript Improvements
Customizable AI Meeting Summary Templates
Users can now select a summary type before ending a meeting. The AI generates a structured summary tailored to the meeting type. Six templates are available:- General (default) — Overview, Topics, Action Items
- Customer Interview — Overview, Customer Needs & Signals, Next Steps & Risks
- Regular Team Meeting / Scrum — Progress, Blockers & Risks, Action Items
- HR Recruiting Interview — Candidate Profile, Evidence & Observations, Evaluation & Next Steps
- Human Evaluation Meeting — Evaluation Context, Evidence Discussed, Decision & Follow-Up
- Formal Presentation — Presentation Overview, Key Takeaways, Decisions & Follow-Up
Transcript Full-Text Search (Production Launch)
Full-text keyword search has been significantly upgraded with a new search engine. Major improvements:- Covers 1 year of transcripts (~11M documents)
- Searches title, original STT text, and translated text
- Accurate CJK (Japanese, Chinese, Korean) search — a significant improvement over the previous engine
- Phrase search with boosted relevance scoring
Meeting Log Sharing Permissions
Workspace administrators can now update sharing settings on meeting logs they have read access to. Legacy shared logs are preserved and can be narrowed down.Per-User Summary Template Preference
Each user’s preferred AI summary template is saved and persists across sessions. The meeting end modal on mobile now also supports tags, notes, and summary type selection.Mobile App Updates
v0.5.16 — Major Overhaul
- Fixed initial speech loss bug on mobile with a complete audio pipeline rewrite
- iOS SDK upgraded to Tahoe (iOS 26) compatibility
- Offline translation now available on Android
- System audio translation on iOS: Translate system audio playback (e.g., YouTube) instead of mic input
- Full billing flow revamp matching web app parity — billing log, invoice and receipt download added
- Plan restore flow fixed with full regression testing across all upgrade/downgrade scenarios
Other Mobile Improvements
- Mobile auto-refresh after recording session ends
- Scroll flicker fix during audio playback on shorter transcripts
- Audio highlight and text highlight regression fixes with unit tests
- Creator info (avatar and name) added to search results
- Caption tab support for all new transcription languages
- Summary email notification toggle added (parity with web)
- Google sign-in issue resolved
Desktop App Updates
System Tray Menu Redesign
- Start/stop recording directly from the macOS/Windows system tray
- Deprecated time-tracking and daily-report menu items removed
- Full 6-locale support for tray menu labels
Bug Fixes
- Fixed accidental meeting start from dropdown modal click-through
- Fixed inconsistent persistence of mic/presentation settings across sessions
- Fixed screen/window swap flash when switching modes
Listener Mode Improvements
- Bilingual language settings for guests: Listener guests joining via shared URL now have full bilingual STT/translation language settings, matching workspace users
- High accuracy mode for listeners: Text correction now works correctly for listener sessions
- Chinese localization fixed: Language selectors in listener mode now correctly include Chinese
Subtitle & Caption Improvements
- Caption tab progress bar: Transcription progress bar now visible immediately after file upload and persists after page reload
- Word-level timestamp restoration: Fixed missing word-level timestamps in caption tab
- Non-speech audio handling: Uploaded audio with no speech now shows a clear “no speech detected” message instead of a broken state
- Text highlight fix: Fixed broken word-level highlighting during audio playback
UI/UX Improvements
- Billing/subscription change screens: Improved order confirmation and success screens showing prorated amount, effective date, and new renewal date
- Share QR code UI redesign: New unified sharing dialog deployed to production
- Vocabulary Builder UI fixes: Delete confirmation modal, visible text input borders, fixed pronunciation for alphanumeric combinations
- Plan change page: Clarified billing cycle reset behavior — usage resets on upgrade, downgrades apply from next cycle
Bug Fixes
- Fixed bilingual mode corruption: Japanese speech incorrectly transcribed to English after Malay language addition
- Fixed wrong language statistics: Language variants counted separately in usage dashboard — now properly aggregated
- Fixed meeting log data loss: Concurrent transcript finalization and cleanup could delete data before finalization completed
- Fixed Chinese localization: Missing Chinese in language selector lists across transcription and listener modes