October 30, 2026
Keep transcripts and subtitle work together in Dashboard
The new Dashboard brings saved transcripts, subtitle projects, and source files together in one place, with a direct path to start another transcription.
- Added Overview, Files, Projects, Transcripts, and Subtitles views under /dashboard.
- Kept projects in the current browser, with an option to save original audio.
- Unified the homepage, transcription, subtitle, and Dashboard colors around black, white, and gray.
October 20, 2026
Automatic language detection now fails gracefully
Language detection is now an enhancement rather than a prerequisite, so recordings can still use Whisper's compatible transcription path when a model or clip cannot provide a reliable language token.
- Short, silent, and model variants without language metadata continue to transcribe instead of failing early.
- Music-only retries respect cancellation and handle both timestamped and plain text outputs.
- The result uses the model-reported language when it is available.
October 20, 2026
Auto-detect keeps the original Whisper compatibility path
Automatic language mode now preserves the transcription prompt that previously worked for songs, while retrying with the detected language only when the first result contains music labels alone.
- Automatic mode no longer forces its first language guess into every recording.
- Song-like music-only results receive one detected-language retry when needed.
- Manual language choices continue to use the selected language directly.
October 20, 2026
Improved transcription for songs and long recordings
Whisper now uses its built-in long-audio merge path again, preserving decoder context across overlapping windows so sung vocals are less likely to be reduced to music labels.
- Restored Transformers.js overlap handling for 30-second Whisper windows.
- Kept automatic and manual language selection without treating each window as a separate transcript.
- Preserved timestamped chunks, cancellation, and transcription progress reporting.
October 20, 2026
Language selection now covers all Whisper languages
EarScribe now supports all 99 Whisper transcription languages, including Persian, with automatic language detection available for recordings whose language is unknown.
- Added Persian and the full set of languages supported by Whisper to the recording language picker.
- Automatic mode detects the language from the first 30 seconds before transcription begins.
- Replaced the browser-native language menu with a searchable, keyboard-friendly picker.
October 16, 2026
Long audio transcription no longer stalls at 95%
One-hour recordings now run through bounded 30-second windows with five seconds of overlap, so lower-memory phones receive real progress updates while transcription continues.
- Added segmented Whisper inference with global timestamps and overlap cleanup.
- Progress now reflects completed audio and reaches 100% only after all windows finish.
- Preserved cancellation support so a new file or a cancelled run stops the active job.