What it does

VideoLingo combines speech recognition, subtitle translation, segmentation and dubbing in one Streamlit interface. It can download a YouTube video with yt-dlp, recognize speech word by word with Qwen3-ASR and a forced aligner, split subtitles within configurable length limits, translate them with custom and AI-generated terminology, and produce subtitle files and optionally subtitled or dubbed videos. Dubbing runs through text-to-speech back ends including GPT-SoVITS, OpenAI, Fish Audio, Edge TTS and F5-TTS, with a custom adapter option. Translation uses an LLM through an OpenAI-compatible API that you set up with a URL, key and model in the sidebar. Speech recognition runs locally by default, or you can choose a hosted provider, in which case audio is sent to that provider and may incur charges. It runs on Windows, macOS and Linux. Windows has a double-click starter, other systems run from source with uv, and an optional Docker setup targets Linux with an NVIDIA GPU. A local HTTP API lets scripts and agents drive the same pipeline. The README lists limits: background noise and mixed-language speech can hurt recognition and timing, and dubbing does not assign a separate voice to each speaker.

Capabilities

Word-level speech recognition and alignment
Subtitle segmentation and translation with custom terminology
Dubbing through several text-to-speech back ends
Pause, resume or stop at any step, with progress logging
Local HTTP API for scripts and agents
Windows, macOS and Linux, with an optional Docker setup

When to use it

You need subtitles for a video in another language
You want to dub a video with a text-to-speech voice
You prefer to run the pipeline yourself with your own model keys