Skip to content

ROADMAP.md

Latest commit

 

History

History
93 lines (80 loc) · 4.84 KB

ROADMAP.md

File metadata and controls

93 lines (80 loc) · 4.84 KB
Raw

Roadmap

User-facing progress for Rescript. Items are marked done when they shipped to main (web and/or desktop). Internal refactors and CI-only work are omitted.

Done

Core editor

  • Offline transcript-based editor in the browser (no auth, no uploads)
  • Upload → local Whisper transcription with per-word timestamps
  • Speaker diarization (pyannote) with speaker-grouped transcript
  • Cut / restore by selecting words (⌫, floating toolbar); undo / redo
  • Live preview that skips cut ranges in real time
  • Timeline with waveform, ruler, word labels, cut regions, playhead, and zoom
  • In-browser export to MP4 (ffmpeg.wasm, word-accurate cuts)
  • Continuous selection highlight and Descript-style timeline word pills
  • Cache-aware model loading labels; speaker model preloaded during transcription
  • Monotonic download / transcription progress

Editing & transcript tools

  • Correct misrecognized words (selection → Correct popover; timings redistributed)
  • One-click Remove fillers (um / uh / etc.)
  • One-click Remove silences (pauses / dead air ≥ 0.3s)
  • Import your own transcript (SRT / VTT / JSON) instead of running Whisper
  • Import cancel UX (no stuck selector / blocked media drop)
  • Manual speaker controls (rename, change, add, move label, merge, remove)

Transcription quality & models

  • Whisper Base and Whisper Small selectable on upload
  • Silence skip (VAD) and hallucination mitigations for longer files
  • Multi-speaker clips no longer drop the second speaker after silence-skip
  • Smoother transcription progress across Whisper windows
  • Better word↔audio alignment (speech-onset anchoring / lag correction)
  • Multilingual transcription (language selector; English + German, with DE fillers)
  • WebGPU → WASM fallback when the GPU is lost (e.g. Windows screen lock)
  • Graceful handling of media with no audio track

Audio & media kinds

  • Edit audio files (mp3 / wav / m4a / …) the same way as video
  • Audio-only workspace (hide empty video preview; transcript gets full width)
  • Audio export (M4A and additional formats via the export dialog)

Timeline editing (Descript-style)

  • Drag word start / end to fix ASR bleed
  • Split at playhead (S / Split) into selectable clips
  • Clip trim handles (manual cuts; covered words marked cut in the transcript)
  • Draggable cut edges; scrub while dragging; double-click to reset
  • Scroll to zoom / side-scroll to pan
  • Timeline ↔ transcript selection sync; splits reflected in the transcript
  • Word chips visible at all zoom levels
  • Waveform peak clamping on the timeline

Layout, mobile & chrome

  • Mobile-friendly stacked editor layout
  • Upload screen scrollable on short / mobile viewports
  • Resizable transcript / preview split on desktop (persisted)
  • Transcript scroll rail with tick marks and edge fades
  • Light / Dark appearance in Settings (persisted; default light)
  • Settings menu (appearance + community / support links)
  • Social links (Discord, X, GitHub)
  • Mobile menus stay on-screen (Floating UI flip / shift)
  • Smoother transcript text selection while dragging
  • Spacebar play / pause reliability and media-control polish

Projects & persistence

  • Auto-save projects to IndexedDB
  • Recent projects list on the home screen (open / delete)

Export

  • Tabbed export dialog: Video, Audio, Transcript, Subtitles
  • Video: format (MP4 / WebM) and resolution options
  • Audio: M4A / MP3 / WAV (including from video projects)
  • Transcript: plain text or Markdown (speaker turns, cuts removed)
  • Subtitles: SRT / VTT (edited timeline) or JSON (full words for re-import)

Desktop & distribution

  • Live web app on GitHub Pages
  • Electron desktop app for macOS, Windows, and Linux
  • Signed / notarized Mac builds; auto-update from GitHub Releases
  • Desktop chrome polish (traffic lights, window sizing, animated logo / top bar)

Next

  • Parakeet TDT v3 as an optional transcription backend (faster / more accurate word timings)
  • Fix remaining drift / lag and push word timestamps closer to true speech boundaries
  • In-app transcript editor (richer inline editing beyond the Correct popover)
  • Export the timeline itself to video editing tools (e.g. project / EDL / FCPXML-style interchange)
  • Regenerate — text-to-speech with accurate voice cloning so rewritten lines can be spoken in the original voice
  • Native macOS SpeechAnalyzer as an optional transcription backend
  • More languages beyond English / German; local model import for air-gapped first runs
  • Faster export (stream-copy for keyframe-aligned segments, WebCodecs rendering)
  • Multi-clip projects (reorder scenes), captions burn-in
  • Gap clips / insert silence between words