mazinger 1.9.2

End-to-end video dubbing pipeline: transcribe, translate, and voice-clone.

Mazinger Dubber

End-to-end video dubbing pipeline. Download a video, transcribe it, translate the subtitles, clone a voice, and produce a fully dubbed audio or video file — in one command.

▶️ Watch Demo Video (with audio)

What It Does

Mazinger chains ten stages into a single pipeline:

1. Download — fetch a video from a URL or ingest a local file, extract the audio track
2. Transcribe — convert speech to SRT subtitles (OpenAI Whisper API, faster-whisper, or WhisperX)
3. Thumbnails — use an LLM to pick key frames from the video for visual context
4. Describe — analyze the transcript and thumbnails to produce a structured summary (title, key points, keywords)
5. Review — optionally refine ASR output: fix typos, reshape punctuation, and convert technical terms to English
6. Translate — translate the SRT into another language with duration-aware word budgets
7. Re-segment — merge fragments and split oversized subtitles for readability
8. Speak — synthesize voice-cloned speech for every subtitle entry (Qwen3-TTS or Chatterbox), with 16 pre-defined voice themes or your own voice sample
9. Assemble — place each audio segment on the original timeline with optional tempo adjustment, loudness matching, and background audio mixing
10. Subtitle — burn styled subtitles into the video and/or mux the new audio track

Every stage can run independently or as part of the full pipeline. Interrupted runs resume automatically — completed stages and individual TTS segments are cached and skipped.

Prerequisites

• Python 3.10 or later
• ffmpeg installed and on PATH (apt install ffmpeg / brew install ffmpeg)
• An OpenAI API key for LLM-powered stages (transcription, translation, thumbnails, description)
• A CUDA GPU for local transcription and TTS (not needed for cloud-only workflows)

Installation

The base install covers download, transcription (cloud), thumbnails, description, translation, re-segmentation, and subtitle embedding. No GPU needed.

pip install mazinger

Add local transcription or TTS as optional extras:

# Local transcription
pip install "mazinger[transcribe-faster]"      # faster-whisper (Chatterbox-compatible)
pip install "mazinger[transcribe-whisperx]"    # WhisperX (best word-level alignment)

# Voice synthesis
pip install "mazinger[tts]"                    # Qwen3-TTS (voice sample + transcript)
pip install "mazinger[tts-chatterbox]"         # Chatterbox (voice sample only, emotion control)

# Full bundles
pip install "mazinger[all-qwen]"              # WhisperX + Qwen3-TTS
pip install "mazinger[all-chatterbox]"        # faster-whisper + Chatterbox

Qwen and Chatterbox require different transformers versions and cannot share an environment. WhisperX also conflicts with Chatterbox — pair it with Qwen, or use faster-whisper with Chatterbox.

See the Installation Guide for venv recipes, Colab setup, and uv overrides.

Quick Start

Dub a video in one command

mazinger dub "https://youtube.com/watch?v=VIDEO_ID" \
    --voice-sample speaker.m4a \
    --voice-script speaker_transcript.txt \
    --target-language Spanish \
    --base-dir ./output

Use a voice profile instead of local files

Voice profiles are hosted on HuggingFace and downloaded automatically. Several ready-made profiles are available out of the box:

abubakr · daheeh-v1 · 3b1b · italian-v1 · morgan-freeman · trump-v1

See the full list with descriptions in the Available Voice Profiles doc.

mazinger dub "https://youtube.com/watch?v=VIDEO_ID" \
    --clone-profile abubakr \
    --target-language Arabic

Use a voice theme (no files needed)

Choose from 16 pre-defined voice themes — no voice sample or profile download required:

narrator-m/f · young-m/f · deep-m/f · warm-m/f · news-m/f · storyteller-m/f · kid-m/f · teen-m/f

mazinger dub "https://youtube.com/watch?v=VIDEO_ID" \
    --voice-theme narrator-m \
    --target-language Spanish

List all themes with mazinger profile list. Generate a reusable profile with mazinger profile generate. See Voice Profiles for details.

Auto-clone the original speaker's voice

When no voice option is provided, Mazinger automatically clones the speaker directly from the source audio. The pipeline picks the best 20–60 s segment from the transcription and uses it as the cloning reference — no files or configuration needed.

mazinger dub "https://youtube.com/watch?v=VIDEO_ID" \
    --target-language Spanish

proj = dubber.dub(
    source="https://youtube.com/watch?v=VIDEO_ID",
    target_language="Spanish",
)

Produce a video with burned subtitles

mazinger dub "https://youtube.com/watch?v=VIDEO_ID" \
    --clone-profile abubakr \
    --output-type video \
    --embed-subtitles \
    --subtitle-google-font "Noto Sans Arabic" \
    --subtitle-font-size 24

Run a single stage

Every stage has its own sub-command:

mazinger download   "https://youtube.com/watch?v=VIDEO_ID" --base-dir ./output
mazinger slice      "https://youtube.com/watch?v=VIDEO_ID" --start 00:01:00 --end 00:04:00
mazinger transcribe ./output/projects/my-video/source/audio.mp3 -o subs.srt
mazinger translate  --srt subs.srt --target-language French -o translated.srt
mazinger subtitle   video.mp4 --srt translated.srt -o output.mp4

Python API

from mazinger import MazingerDubber

dubber = MazingerDubber(openai_api_key="sk-...", base_dir="./output")

# With a voice theme (simplest)
proj = dubber.dub(
    source="https://youtube.com/watch?v=VIDEO_ID",
    voice_theme="narrator-m",
    target_language="Spanish",
    output_type="video",
)

# Auto-clone the speaker's voice (no voice option needed)
proj = dubber.dub(
    source="https://youtube.com/watch?v=VIDEO_ID",
    target_language="Spanish",
)

# Or with explicit voice files
proj = dubber.dub(
    source="https://youtube.com/watch?v=VIDEO_ID",
    voice_sample="speaker.m4a",
    voice_script="speaker_transcript.txt",
    target_language="Spanish",
    output_type="video",
    embed_subtitles=True,
)

print(proj.final_video)   # ./output/projects/<slug>/tts/dubbed.mp4

Documentation

Full documentation lives in the docs/ directory:

Chapter	Contents
Installation	All install methods, extras, compatibility matrix, Colab and venv recipes
Quick Start	Common workflows with copy-paste examples
Pipeline Overview	How the nine stages connect, data flow, and resume behavior
CLI Reference	Every command, flag, and default value
Python API	Classes, functions, and parameters for programmatic use
Voice Profiles	Using, creating, and uploading voice profiles
Subtitle Styling	Fonts, colors, positioning, RTL support, Google Fonts
Configuration	Environment variables, caching, tempo control, LLM usage tracking
Project Structure	Output directory layout and file naming conventions
YouTube Cookies	How to export and pass cookies for age-restricted or region-locked videos

License

MIT