EasyScript for Premiere

Auto silence-cut, transcription with speaker names, subtitles, 50+ language translation, beat markers and — new in 2.3 — voice / music separation, all directly on your Premiere timeline. Everything runs on your own machine, GPU first. It never touches your source files; every cut and rename is a timeline edit you can undo.

The real panel, on a sample interview.

This is EasyScript 2.3 itself, with Premiere and the AI backend simulated by an 80-second, two-speaker interview. The clip is already loaded — try Detect silence, then Transcribe and Speakers on the Transcript tab, Separate voice / music… at the top, and Detect beats on the Beats tab.

Nothing is uploaded and no audio plays here — the waveform, transcript, translations, stems and beats are canned results. In Premiere the same buttons run Whisper, pyannote, Mel-Band RoFormer and the beat tracker on your own GPU.

From raw timeline to clean, subtitled, translated.

Ten steps, all on the timeline you're already working in. Run only the ones you need.

Everything in one panel.

Auto silence-cut

Detect silences and breaths on the waveform, fine-tune padding, dB threshold and min-silence, then cut on the timeline — frame-exact — or rebuild long videos as a new sequence in one pass.

Whisper transcription

Speech to text with Whisper Turbo by default (Large V3 accuracy, about 3× faster), word-level timing, a vocabulary for names and terms, and a Music / Song mode. Runs on-device.

Speaker diarization

Every word is attributed to its speaker and lines split where the speaker really changes. The model is bundled — no HuggingFace token — and named voices are recognised again in later videos.

Subtitles on the timeline

Generate caption tracks directly in Premiere. Pick the original plus any translations, with timing after the cut or to the original.

50+ language translation

Translate segments into dozens of languages with NLLB-200, Ollama, Hy-MT2 or the Claude API — local or cloud, your choice.

Voice / music separation

Split any clip into Voice only (clean transcription even over loud music) and Music only (the best input for beats), with Mel-Band RoFormer running locally. Switch Original | Voice | Music instantly.

Stems back on the timeline

Import the voice or music at its original timecode, on the first audio track that is free for the whole range — never over existing audio. Files land in an EasyScript Stems folder and bin.

Beat markers

Detect beats and downbeats, choose a marker every 1/4 beat up to 4 bars, review them on the waveform, then place them on the sequence or on the clip selected in the timeline.

Never touches your media

Every cut and rename happens on timeline clips only — your source files stay untouched and a single Undo reverts anything.

Read the waveform, then the words.

Filter the segment list by All / Speech / Silence / Translation, search and replace across the transcript, and jump straight to any moment on the timeline.

Pull the voice and the music apart.

Click Separate voice / music… and the loaded range is split by Mel-Band RoFormer, on Apple Silicon (Metal), NVIDIA (CUDA) or any Windows GPU (DirectML), with a CPU fallback. The two stems add back up to the source exactly, and nothing is uploaded.

The first separation downloads the model once (about 870 MB). Updating from an older EasyScript? Run the installer again — it replaces both the panel and the backend.

One transcript, every language.

Add as many target languages as you like and translate every segment in place. Each language stacks alongside the source so you can review, edit, and burn them all into subtitles.

Tune it once, forget it.

Installs in a couple of clicks.

1. Fully quit Premiere (Cmd + Q).
2. Right-click install-mac.command → Open → Open.
3. It installs the panel and backend automatically — reopen Premiere when it finishes.

1. Fully quit Premiere.
2. Right-click install-win.bat → Run as administrator.
3. Reopen Premiere when it finishes.

First launch

Open it from Window → Extensions → EasyScript. The panel auto-starts the backend (the corner shows Connecting → Connected). The first transcription downloads its model and the first separation another (~870 MB), each once — after that everything works offline. Plan for about 9 GB of disk if you use every feature.

Requirements: Adobe Premiere Pro 2022 or newer · macOS 13+ or Windows 10+ · 8 GB RAM (16 GB for the large models). GPU first — Apple Silicon (Metal), NVIDIA (CUDA) or any Windows GPU (DirectML) — with a CPU fallback. FFmpeg is bundled.

Good to know.

It is a Premiere Pro extension that cuts silence, transcribes speech with speaker names, makes subtitles, translates, places beat markers and separates voice from music — all directly on your timeline. It never modifies your source files; every change is a timeline edit you can Undo (Cmd/Ctrl + Z).

Cut, transcribe and translate — on the timeline.

Free, for Premiere Pro on Windows and macOS.

Frequently asked questions

What does EasyScript do?

It is a Premiere Pro extension that cuts silence, transcribes speech with speaker names, makes subtitles, translates, places beat markers and separates voice from music — all directly on your timeline. It never modifies your source files; every change is a timeline edit you can Undo (Cmd/Ctrl + Z).

What’s new in 2.3?

Voice / music separation with Mel-Band RoFormer (Voice only for transcription, Music only for beats), importing the separated voice or music back to the timeline at its original timecode, beat markers on the clip selected in the timeline, and GPU-first processing on Metal, CUDA and DirectML. 2.2 brought Turbo transcription with word timing, built-in speakers without a token and the voice library.

Which Premiere versions and OS are supported?

Adobe Premiere Pro 2022 or newer, on macOS 13+ or Windows 10+, with 8 GB of RAM (16 GB for the large models). Apple Silicon Macs are fastest; NVIDIA GPUs use CUDA, other Windows GPUs DirectML, and everything falls back to the CPU.

How do I install or update it?

The package contains the .zxp, a backend folder and installers. Quit Premiere, then on macOS right-click install-mac.command → Open, on Windows right-click install-win.bat → Run as administrator. Reopen Premiere and open Window → Extensions → EasyScript. To update, run the new installer the same way — update the whole package, not only the .zxp, or separation reports “Not Found”.

Do I need an internet connection?

Only to download each model the first time you use it (transcription, translation, and about 870 MB for separation). After that everything runs offline on your machine — nothing is uploaded.

Do I still need a HuggingFace token for speakers?

No. Since 2.2 the speaker model (pyannote community-1) ships inside EasyScript. Set a known speaker count or sensitivity if the audio is tricky, and rename speakers once so their voices are recognised in later videos.

How good is the voice / music separation?

It uses Mel-Band RoFormer (Kim vocal model), a state-of-the-art separation model, on a full-quality copy of the loaded range. Voice only removes music for cleaner transcription; Music only removes the voice and is the best input for beat detection. The two stems add back up to the original exactly.

What translation options are there?

Choose a provider in Settings: NLLB-200 (local, fast), Ollama, Hy-MT2, or the Claude API. Add as many target languages as you like, review every line, then include them when you make subtitles.

Will it change my original footage?

No. All cuts and renames apply to clips on the timeline only — your source files are never touched, and a single Undo reverts anything. Stems and cut XMLs are saved as new files next to your project.