Contact
Back to portfolio

Partner · Epex Digital

Epex Video Dubber

A desktop app that dubs foreign-language video into Khmer on its own — transcribing, translating and re-voicing every line in matching male and female voices, then laying the result back over the original picture.

Partner
Epex Digital
Type
Desktop app
Platform
Windows & macOS
Year
2026
The task queue: paste links or add files and folders, then follow each episode's progress — with the API manager and engine settings alongside.
The task queue: paste links or add files and folders, then follow each episode's progress — with the API manager and engine settings alongside.

The challenge

Epex Digital runs content for Facebook pages with Khmer-speaking audiences, and much of the video worth sharing — dramas, clips and explainers — is in Chinese, English and other languages. Dubbing it by hand means transcribing, translating, recording voices and syncing every line to the picture: hours of work per episode, for a queue that never stops growing.

What we built

We built Epex Video Dubber. Paste a list of links or drop in a folder, and it works through the queue on its own. Each video is downloaded, transcribed with word-level timestamps, translated into Khmer with a length budget for every line so the dub fits the original timing, spoken in male or female Khmer voices, and mixed back over the original music. It installs as a single app on Windows and macOS, with everything — ffmpeg included — bundled inside.

Key features

  • 01

    Batch dubbing from links or files

    Paste YouTube, TikTok and other links one per line, or add videos and whole folders. Jobs run from a queue with progress, retry, pause and resume — and a paused job picks up where it stopped, even after a restart.

  • 02

    Translation that fits the timing

    Speech is transcribed with word timestamps, then translated in batches by GPT-4o or Gemini with a length budget for every line, so the Khmer lands where the original speech did.

  • 03

    Male and female voices, detected automatically

    The app reads the pitch of each line in the original audio and routes it to a male or female Khmer neural voice — no manual tagging of who is speaking.

  • 04

    Keeps the music and effects

    Demucs separates the soundtrack from the dialogue, so the original score and sound effects stay, ducked under the new Khmer voices.

  • 05

    Line-by-line sync

    Every line starts on time. Lines whose translation runs long are spoken faster within a set limit or cut short by the next line, and the log reports the worst overrun for every video.

  • 06

    Branding and Khmer subtitles

    Place up to three logos with a drag-and-drop editor over a real preview of the video, and add Khmer subtitles — burned in or as a selectable track — with .srt transcripts written alongside.

  • 07

    GPU-accelerated encoding

    The final encode uses NVIDIA NVENC or Apple VideoToolbox when the machine really supports it — proven with a real one-frame test encode — and falls back to the CPU automatically if the GPU fails.

  • 08

    Cost control and key rotation

    Choose OpenAI, Gemini or both — Gemini's free tier first, with OpenAI taking over any step it fails. Several API keys per provider rotate past rate limits, and on OpenAI a typical 3-minute video costs about $0.10.

Screenshots

Click any screenshot to open it full size.

Live engine telemetry for every job — voice clips rendered, music kept under the dialogue and a sync report for each video — next to the hardware, cookie, branding and output settings.
Live engine telemetry for every job — voice clips rendered, music kept under the dialogue and a sync report for each video — next to the hardware, cookie, branding and output settings.

Want something like this for your business?

Tell us how your business runs today — spreadsheets, group chats, paper receipts and all. We will show you what it could look like.

Start a project