🎙 TR Altyazı — AI Transcript & Auto-Cut for Premiere Pro

Local AI transcription, SRT export, and silence-based auto-cut — all inside Premiere Pro. No API keys. No cloud. Completely free.

---

✨ What It Does

TR Altyazı is a Premiere Pro CEP panel that runs OpenAI Whisper locally on your GPU. Transcribe speech, generate subtitles, and remove silences — all without leaving Premiere.

Feature	Description
🎙 AI Transcription	Powered by faster-whisper with word-level timestamps
✂ Auto-Cut	Detect and remove silences with 4 precision controls
📄 SRT Export	One-click subtitle file to your desktop
🎬 Timeline Insert	Captions directly into your Premiere sequence
🔒 100% Local	Your footage never leaves your machine
💸 Zero Cost	No API keys, no subscriptions, no usage limits

Optimized for Turkish with large-v3, but works with any language Whisper supports (English, Spanish, Japanese, Arabic, etc.)

---

🖥 Preview

┌─────────────────────────────────────┐
│  TR  Altyazı & Transcript    ● cut  │
├──────────────┬──────────────┬───────┤
│  Transcript  │   Auto-Cut   │ Ayar  │
├─────────────────────────────────────┤
│  ▶ Transkript Al          large-v3  │
│  Kaynak: [Aktif Sequence      ▼]    │
│                                     │
│  › Sequence: my_video (165s)        │
│  › Timeline audio export ediliyor  │
│  › Export tamam: tr_tl_177.wav      │
│  › Whisper'a gönderiliyor...        │
│  ████████████████░░░░  85%          │
│                                     │
│  0:02.1 → 0:05.8  Merhaba arkadaşlar│
│  0:06.0 → 0:09.2  Bugün sizinle...  │
│  0:09.5 → 0:14.1  Bu videoyu...     │
│                                     │
│  [ Timeline'a At ]                  │
│  [ Düzenle ]  [ SRT İndir ]  [ ✕ ] │
└─────────────────────────────────────┘

---

⚡ Quick Start

Requirements

• Adobe Premiere Pro 2022–2026
• Python 3.9+
• NVIDIA GPU with CUDA (for GPU acceleration — CPU works too, just slower)
• ffmpeg in PATH (required by pydub for audio processing)

1. Clone & Install

git clone https://github.com/eryukselaskar/premiere-tr-plugin.git
cd premiere-tr-plugin

# Run the installer (copies extension + sets registry keys)
Set-ExecutionPolicy Bypass -Scope Process
.\install.ps1

Or manual install — copy the repo folder to:

C:\Users\<YOU>\AppData\Roaming\Adobe\CEP\extensions\tr-subtitle\

2. Enable Unsigned Extensions (one-time)

# Run in PowerShell as Administrator
reg add "HKCU\Software\Adobe\CSXS.11" /v PlayerDebugMode /t REG_DWORD /d 1 /f
reg add "HKCU\Software\Adobe\CSXS.12" /v PlayerDebugMode /t REG_DWORD /d 1 /f
reg add "HKCU\Software\Adobe\CSXS.13" /v PlayerDebugMode /t REG_DWORD /d 1 /f

3. Start the Whisper Server

cd whisper-server
pip install -r ../requirements.txt

python whisper_server.py
# Server starts at http://127.0.0.1:5123
# large-v3 model (~3 GB) downloads on first run

4. Open in Premiere Pro

Window → Extensions → TR Altyazı & Transcript

---

🔧 How It Works

Premiere Pro Panel (HTML/JS)
        │
        ├── evalScript() ──► ExtendScript (JSX)
        │                         │
        │                    exportAsMediaDirect()
        │                    (exports current timeline audio)
        │
        └── fetch() ────────► Flask Server (Python)
                                    │
                              faster-whisper
                              (GPU transcription)
                                    │
                              word timestamps
                              silence detection
                              ◄─── segments ─────┘

Why export the timeline first? If you've already used Auto-Cut to remove silences, the source file is longer than your timeline. The panel exports the current timeline state as audio before transcribing, so timestamps always match what's on screen.

---

🎛 Auto-Cut Parameters

Parameter	Default	Description
Cutoff (dBFS)	-50 dB	Volume threshold — below this = silence
Min. Silence	0.5 s	Minimum silence duration to cut
Min. Segment	0.5 s	Keep segments longer than this
Padding	0.2 s	Buffer to keep around speech

The panel detects silences server-side with pydub, then applies cuts via Premiere's QE API (razor + ripple delete).

---

📁 Project Structure

premiere-tr-plugin/
├── CSXS/
│   └── manifest.xml          ← CEP extension definition
├── HTML/
│   ├── index.html            ← Panel UI + all JavaScript
│   └── CSInterface.js        ← Adobe CEP bridge library
├── jsx/
│   └── hostscript.jsx        ← ExtendScript (Premiere timeline API)
├── whisper-server/
│   └── whisper_server.py     ← Flask + faster-whisper server
├── install.ps1               ← One-click Windows installer
├── requirements.txt
└── README.md

---

🧠 Model Performance

Tested on RTX 5070 (12 GB VRAM):

Model	VRAM	Speed	Turkish Quality
tiny	1 GB	~10x realtime	Poor
base	1 GB	~7x realtime	OK
small	2 GB	~5x realtime	Good
medium	5 GB	~3x realtime	Very Good
large-v2	10 GB	~2x realtime	Excellent
large-v3	12 GB	~2x realtime	Best ⭐

Change model in the Settings tab. CPU fallback works automatically if no GPU is detected.

---

🛠 Troubleshooting

Extension not showing in Premiere's Extensions menu

1. Make sure PlayerDebugMode = 1 (DWORD, not String) for CSXS.11, CSXS.12, CSXS.13
2. Verify the folder is at: %APPDATA%\Adobe\CEP\extensions\tr-subtitle\
3. Fully restart Premiere Pro (not just reload)
4. Open manifest.xml in a browser to verify it's valid XML

Whisper server won't start

# Check CUDA
nvidia-smi

# Reinstall with CUDA support
pip install faster-whisper --upgrade

# Run with explicit CUDA device
$env:CUDA_VISIBLE_DEVICES=0; python whisper_server.py

ffmpeg not found (pydub error)

# Install via winget
winget install ffmpeg

# Or via chocolatey
choco install ffmpeg

Then restart the Whisper server.

Auto-cut misses some silences

Try raising the Cutoff threshold (e.g., -50 dB → -45 dB). Background noise can prevent pydub from detecting silence. Also reduce Min. Silence if short pauses are being missed.

---

🤝 Contributing

PRs welcome! Ideas for contribution:

• macOS support (different CEP paths)
• Multi-track Auto-Cut
• Caption styling via Essential Graphics API
• Speaker diarization
• Streaming progress from Whisper to the panel
• Batch transcription for multiple sequences

git clone https://github.com/eryukselaskar/premiere-tr-plugin.git
code premiere-tr-plugin

For ExtendScript debugging, use the ExtendScript Debugger VS Code extension.

---

📄 License

MIT — do whatever you want, attribution appreciated.

---

If this saved you time, a ⭐ star helps others find it.

Made with ☕ for video editors who hate paying for transcription APIs.