More Reliable Exports, Speaker Matching, and Auto Sync
Sequence 1.4.1 improves export reliability, speaker identification, and external-audio synchronization. It also adds Italian, Brazilian Portuguese, Korean, and Simplified Chinese interface translations.
Export
- Fixed audio dropouts caused by tiny rounding differences where adjacent edge fades meet.
- Reduced Direct Export memory use on edits with many cuts, including silence, filler-word, and profanity edits.
- Added clearer recovery messages when the system terminates an export encoder, and improved the error details shown when Direct Export fails.
- Localized export dropdowns.
Speaker Matching and Editing
- Improved voice-profile matching and taught profiles to learn from manual speaker corrections for future matching.
- Added New Speaker support for persistent custom speakers. Copying an editing variant preserves its custom speakers.
- Added sentence-level speaker indicators with VoiceOver support. Indicators stay hidden when a sentence cannot be clearly attributed to one speaker.
- Improved speaker colors, appearance settings, the Unassigned label, and speaker identity consistency when rerunning detection.
- Preserved speaker history after deleting a person from the library.
Auto Sync and Transcript Timing
- Improved Auto Sync accuracy when transcript hints disagree with clear waveform matches, and fixed failures with short or unmatched transcripts.
- Made waveform matching faster and refreshed the inspector immediately after synchronization.
- Improved Update Timing after automatic or manual synchronization changes. Dismissing the prompt leaves the action available, and repeated timing adjustments are retained.
- Fixed Update Timing on canonical transcripts and automatically updated transcript timing after Re-Sync External Audio.
- Added Update Timing to Audio Sync, and Re-Sync External Audio to transcript actions and menus.
- Fixed normalized projects becoming unavailable after Auto Sync. Affected libraries are repaired when opened, with a notice when an unverifiable normalized copy must be retired.
- Fixed duplicate silence ranges after transcript timing changes and improved audio-meter source selection and invalid-value handling.
Transcript and Timeline Fixes
- Fixed trimming or restoring a recording unexpectedly undoing filler-word, profanity, or manually hidden-word edits.
- Kept Jump to Word available for beeped or silenced profanity.
- Fixed stale trim controls and normalization previews, and retried failed timeline waveform loads when the timeline is configured again.
- Restored support for silence decisions saved by Sequence 1.2.x without overwriting newer saved decisions.
Model Downloads and File Access
- Made speech-model installation atomic and verified model files before allowing transcription to start.
- Improved download recovery after expired download tokens or interrupted transfers, reusing verified files and resuming partial downloads.
- Preserved the previous complete model pack when an installation is interrupted, and improved recovery on the next attempt.
- Added a file-access prompt when Sequence needs permission to read existing media, with retries after access is granted. Missing media opens the relink flow instead.
Reliability
- Fixed a crash caused by resolving interface colors off the main thread.
- Added validated, encrypted recovery backups before database upgrades.
- Improved consent-aware diagnostics for model download, readiness, and loading failures, and stopped repeated reports for malformed legacy silence data.