Video Translator Pro

Windows desktop application for translating and dubbing videos with speech recognition, machine translation, neural text-to-speech and FFmpeg-based media processing.

The project is designed not only to generate a translated audio track, but also to preserve speech quality when the translated phrase is longer than the original timing slot.

Highlights

• Speech recognition with OpenAI Whisper.
• Translation through deep-translator / Google Translator.
• Neural dubbing with Edge TTS, with controlled gTTS fallback.
• Pause Sync mode: long translated phrases are not aggressively compressed or cut; the video can insert a short freeze/pause so the phrase can finish naturally.
• Configurable voice-speed limit: x1.15 / x1.20 / x1.25.
• FFmpeg/FFprobe pipeline for audio extraction, processing, muxing and final validation.
• Optional original-audio background mix.
• Multi-language target support.
• Persistent TTS cache for recovery after failures or cancellation.
• Translation checkpoints and batch recovery after interrupted processing.
• Verification of completed MP4 files before they are treated as successful.
• Structured diagnostic sessions intended for troubleshooting with ChatGPT / Codex.

Reliability and recovery

The application contains several mechanisms for long-running jobs:

• atomic translation checkpoints;
• recovery of unfinished batches after an abnormal shutdown;
• reuse of successfully generated TTS segments;
• bounded retries and network-storm backoff;
• separate protection for translation, Edge TTS and gTTS network failures;
• periodic heartbeat events during long Whisper and FFmpeg stages;
• validation that the final MP4 contains both video and audio streams;
• cleanup and size limits for recovery/TTS cache data.

Pause Sync

Translated speech can be substantially longer than the original speech. Instead of forcing every phrase into the original slot with extreme speed-up, Video Translator Pro uses the following strategy:

1. Use available natural silence after the original phrase.
2. Apply moderate TTS/tempo acceleration within the configured limit.
3. If the phrase still does not fit, insert a video freeze/pause until the translated phrase finishes.

This can make the final video slightly longer than the source, but preserves the full translated phrase and reduces robotic-sounding speech.

Requirements

• Windows 10/11.
• Python 3.10–3.13.
• FFmpeg and FFprobe available in PATH, or ffmpeg.exe and ffprobe.exe placed next to video_translator.py.
• Python dependencies from requirements.txt.

Installation

py -m venv .venv
.\.venv\Scripts\Activate.ps1
py -m pip install --upgrade pip
py -m pip install -r requirements.txt

Then install FFmpeg or place ffmpeg.exe and ffprobe.exe next to the program.

Run

py video_translator.py

On Windows you can also use Запустить.bat after installing the dependencies.

Main Python dependencies

• openai-whisper
• edge-tts
• gTTS
• deep-translator
• moviepy<2
• numpy
• requests

Runtime data

During normal operation the application can create diagnostic logs, translation checkpoints, TTS cache data, temporary media and user settings. These runtime artifacts are excluded from Git through .gitignore and should not be committed to the repository.

Diagnostics

Each diagnostic session is designed to keep useful context for debugging while avoiding huge repetitive logs. The logging system separates informational events from warnings/errors, groups repeated failures by stable signatures, keeps traceback/context for real failures and produces compact summaries for analysis with ChatGPT or Codex.

Verification

The repository includes a Windows GitHub Actions workflow that compiles the main application source on every push and pull request:

python -m py_compile video_translator.py

This CI check catches Python syntax regressions without downloading Whisper models or starting a translation job. Full Whisper, TTS, networking and FFmpeg behavior still requires runtime testing.

Notes

• Whisper can only translate speech that it successfully recognizes; speech hidden by noise/music may be missed.
• Edge TTS is the preferred speech provider. gTTS is used only as a controlled fallback when Edge TTS remains unavailable.
• The final video may be longer than the original when Pause Sync inserts pauses to preserve complete translated speech.

Documentation

The original Russian documentation and changelog are included in the repository for additional implementation and recovery details.

License

No open-source license is currently granted. The source code is published for portfolio and code-review purposes.