Open-Source AI Dubbing: Voice-Pro, pyVideoTrans & More

Voice-Pro, pyVideoTrans and VideoLingo are worth shortlisting for open-source AI dubbing, but they solve different problems. Voice-Pro combines speech and voice tools; pyVideoTrans offers configurable pipeline stages; VideoLingo emphasizes subtitle translation and segmentation. SoniTranslate, YouDub and Linly-Dubbing offer other approaches, while OpenDub needs careful identification because multiple projects share the name.

Choose by your operating system, language pair, editing needs and willingness to configure models. GitHub popularity helps identify projects worth investigating; it does not establish translation quality or reliability.

Reviewed 24 October 2026. This is a comparison of official repositories, documentation and license files, with GitHub API metadata checked on that date. We have not run a common audio-quality or speed benchmark across these projects. Features below are documented capabilities, not independently measured results. Development-branch documentation can differ from a downloadable release.

Quick Comparison

Start with the workflow you need. The licenses below describe each application’s repository code; model weights and external services have their own terms.

Popularity and maintenance are different signals

These are GitHub stars observed on 24 October 2026, rounded to one decimal thousand where appropriate. “Last push” is repository activity, not a release date or evidence that every feature works.

Voice-Pro: a substantial omission from many dubbing lists

Voice-Pro by ABUS brings transcription, translation, vocal separation, TTS and voice cloning into a Gradio interface. Options include Whisper, F5-TTS, CosyVoice, Kokoro and Edge-TTS.

Installation centers on Windows with NVIDIA graphics. The README lists 4GB+ VRAM, preferably 8GB+, and 20GB+ free storage; requirements vary with the models selected. It explicitly says macOS and Linux operation is unverified and development is paused. Use current installation guidance rather than historical subscription notes.

Check whether the selected translation and TTS services need a connection. The LICENSE file identifies GPL-3.0.

Full-featured pipelines

pyVideoTrans: configurable stages and manual review

pyVideoTrans combines speech recognition, subtitle translation, multi-role dubbing and video synthesis. It supports pausing for corrections at individual stages, which is useful when a name or translation needs fixing before generating audio.

Windows users have a packaged application; macOS and Linux users can follow source installation instructions. The documented backends include local recognition and translation alongside online services. Voice cloning depends on the selected integration.

Start here if: you need different engines for different stages. Export a short clip before processing a batch. Hardware requirements depend on the backend; follow its current instructions.

Trade-off: flexibility means more configuration. A free application can still use paid APIs, and a local transcription stage does not make the remaining stages private or offline.

VideoLingo: subtitle-led translation and dubbing

VideoLingo combines recognition, terminology handling, subtitle segmentation and dubbing in Streamlit. Its current documentation describes local Qwen recognition/alignment, configurable translation providers and several TTS integrations. This differs from older guides that describe WhisperX as its default installation.

Windows has a startup script; source instructions cover macOS and Linux. Apple Silicon and Intel Mac recognition paths differ. The project also documents a local HTTP API for processing workflows.

Start here if: subtitle wording and line segmentation matter alongside the new soundtrack. Trade-off: the documented workflow does not automatically assign a different voice to each speaker. Mixed-language speech, alignment and timing require review. Its former “Netflix-level” branding is not an independent quality certification.

OpenCreator, formerly KrillinAI: a broader creator workspace

The original KrillinAI repository now redirects to OpenCreator. Video translation remains part of the product, but its current scope also includes other media and agent-assisted workflows.

The translation workspace documents transcription, subtitle editing and translation, dubbing, portrait/landscape layouts, and SRT, audio or video export. Desktop downloads cover Windows and macOS; real model tasks need a configured sign-in or provider credentials. Local transcription options exist, but that does not make every operation local or free.

Start here if: you want translation within a broader creative workspace. Trade-off: a tutorial for an older KrillinAI release may describe a different installation or interface. Check the release you actually download before following setup instructions.

Speaker controls and voice cloning

SoniTranslate: synchronized audio with Gradio and Colab

SoniTranslate is another established option missing from the earlier comparison. It provides a Gradio interface, speaker-related controls and flexible video, audio and subtitle outputs. Its documentation includes both a Colab notebook and local installation tested on Linux.

The local route involves Python/Conda, FFmpeg and model dependencies. Some diarization models require accepting their access conditions and supplying a Hugging Face token. Different TTS options have different language coverage and terms.

Start here if: you want to experiment with synchronized translation and speaker settings through a notebook or browser interface. Trade-off: Colab resources and local dependency setup both need planning. A supported transcription language does not guarantee a matching cloned voice in the output language.

YouDub: sentence-level voice references and workflow control

YouDub-webui now documents YouDub Desktop. It offers subtitles, dubbed audio, or both, with sentence-level source audio references and optional background-audio preservation. Its workflow includes reviewing progress, retrying failed stages and inspecting outputs.

The repository links Windows and macOS packages. Models and inference dependencies are prepared for the selected workflow, while translation requires provider configuration. The current development documentation lists English, Chinese and Japanese in its language catalog, with actual coverage depending on providers.

Start here if: sentence-level review fits your workflow. Trade-off: the project is changing substantially; development documentation and packaged releases can differ. Confirm the installed version’s workflow, language support and model requirements before choosing it for a longer project.

Linly-Dubbing: a model integration toolkit

Linly-Dubbing integrates separation, recognition, translation and speech generation, including CosyVoice and links to Linly-Talker lip-sync work. These are separate pipeline capabilities; generating translated speech does not automatically mean a video has been lip-synced.

Its installation guide documents Python/Conda, model submodules and CUDA environments. Qwen and online translation options are described, and optional speaker diarization requires model access.

Start here if: you are comfortable assembling and maintaining a model stack. Trade-off: the repository’s last recorded push was March 2025. That is not proof the software fails, but it makes checking dependency versions and unresolved installation issues especially worthwhile before adopting it.

Which OpenDub? Identify the repository first

“OpenDub” is not a unique project identifier. Do not assume a search result, browser extension and desktop application share the same code, license or capabilities. Two distinct options relevant here are:

estebanstifli/OpenDub: a desktop preview

This OpenDub documents local analysis, translation, speech generation and segment editing. It can regenerate individual lines and export video or audio. The default workflow is described as local after downloading models.

The project labels its packages early preview. Windows has received the most testing; Linux and macOS packages are offered for testing. Setup requires roughly 8 GiB of model weights, plus application and media storage. Its application code is MIT, but models and bundled components retain their own terms.

Consider it for: evaluating local segment editing. Limitation: it has a very small community, and its documentation explicitly distinguishes duration fitting from lip synchronization. Treat it as an evaluation candidate rather than assuming production maturity.

Softcatala/open-dubbing: a separate CLI project

Softcatala/open-dubbing is an experimental command-line system with configurable recognition, translation and TTS components. It can run local models, while some engine choices use online services.

Its associated public Catalan dubbing service combines this pipeline with separate editor and service projects. That service’s language scope should not be confused with every possible engine configuration in the CLI.

Consider it for: scripting and studying a configurable dubbing pipeline. Limitation: its maintainers explicitly describe it as experimental. It is not the desktop OpenDub above, and engine language lists are not a guarantee of tested end-to-end language pairs.

Archived projects and components that are not complete dubbing tools

Chenyme-AAVT has about 3.1k stars but is archived. It can be a reference for an existing installation; it should not be presented as an actively maintained first choice for a new workflow.

There are also useful speech models, lip-sync models and API examples on GitHub. A TTS model alone does not transcribe, translate, preserve background sound and export your finished video. Before treating a repository as an alternative to the tools above, check which pipeline stages it actually implements and whether its source contains the product itself or only a client for a hosted service.

Which Tool Should You Choose?

Use this short evaluation before committing to a tool or processing a full library:

1. Define the output. Do you need a translated audio file, selectable subtitles, a finished MP4, or visual lip-sync? These are different deliverables.
2. Check the entire language pair. Recognition, translation and TTS must all support the source/target combination. A large translation-language list alone is insufficient.
3. Run one representative clip. Use a short excerpt from your own material containing a name, a number, background sound and a speaker change if relevant. Keep the same clip when comparing tools.
4. Inspect intermediate text. Correct transcription and translation before synthesizing the final voice. Record whether the interface lets you fix one sentence without rerunning everything.
5. Measure your actual costs. Include model downloads, API calls, failed attempts and editing time. Record setup time separately from processing time.
6. Check the export. Listen to the start, middle and end. Confirm speaker assignment, background sound, subtitle timing and file compatibility in the destination player.

For a reproducible comparison, keep a small record like this:

These are evaluation criteria, not benchmark results from this article. For publishing the resulting audio, continue with the YouTube multi-language audio setup guide.

What “free” and “local” actually mean

Open-source code removes a software purchase barrier, but a workflow may still consume paid translation or speech APIs, cloud compute, storage and your time. A model can also have different terms from the application that loads it. The repository license column is a starting point for checking the exact components you intend to use, not a blanket commercial-use clearance.

For offline work, follow the data through recognition → translation → speech generation → export. Running a browser interface on your computer proves only where that interface runs. Check the backend settings at every stage; a network-based translation or voice service still receives the material it processes.

Before comparing pricing, decide whether you need a reusable local setup or a one-off finished video. Our AI dubbing cost guide covers the categories to budget for, while the free AI dubbing guide compares other starting routes.

When a hosted workflow fits better

If environment setup and model configuration take more time than the project warrants, try the same short clip in a hosted workflow. Compare how easily you can correct the translation, change a voice and export the result. Use the actual output and editing effort to decide.

Related Guides