EasyScript v2.0 — Real-time Live Transcription (macOS)

EasyScript v2.0 — a major upgrade focused on real-time live transcription, plus a redesigned settings/waveform experience.

macOS · Apple Silicon (M1–M5)

✨ Live transcription (real-time)

• Word-by-word at the speaker's rhythm — each word appears at its actual spoken timecode (hesitation shows as a pause, fast speech reveals fast), with automatic catch-up when the model falls behind.
• Zero-flicker live line — once a word is shown it never rewrites; corrections only surface after a line is finalized.
• Near-instant latency — words appear right after they're spoken (hypothesis streaming + faster/denser decoding + early commit + direct PCM, no temp WAV).
• Natural line grouping — up to two complete sentences per line; no breaking at commas or short pauses.
• 🎧 Live recording playback — the session audio is recorded; after Stop you can replay it, scrub the timeline, click any line to jump to its timecode, and Save the audio. The matching subtitle highlights and auto-scrolls as it plays.
• Smart ordering — newest-first while live, chronological (oldest-first) for review.

🎛️ UI & Editor

• Settings popup opened from a gear icon, with clear dividers between Transcribe / Speakers / Translation.
• Waveform zoom up to 64× while keeping your current view position (no more jumping to start/end).
• Speaker-colored waveform after diarization.
• "Skip silence" playback toggle to review only speech.

🗣️ Speakers

• Cuts only at speaker changes (no extra cuts at pauses within one speaker).
• A complete sentence is never split across two speakers.
• Clear, actionable error when your HuggingFace token can't access the gated pyannote models.

📦 Packaging

• Self-contained .app with bundled ffmpeg — no extra installs.

Install

1. Download EasyScript-v2.0-macOS.dmg below.
2. Open it and drag EasyScript into Applications.
3. First launch: right-click the app → Open → Open (the app isn't notarized, so this bypasses Gatekeeper once).

Notes: Requires macOS on Apple Silicon. The first use of a model downloads it (needs internet). Speaker diarization needs a free HuggingFace token (Settings → Speakers) and accepting the pyannote model terms.