Media
2 items
Description
Live video subtitles with configurable speech and translation models, OpenAI-compatible APIs, languages, and custom prompts. Real-Time AI Subtitles for Videos and Livestreams Stream Translator is a highly customizable Chrome extension that generates real-time translated subtitles for videos and livestreams. It captures audio from the current tab, transcribes speech with a speech-to-text model, translates the transcript with a language model, and displays the result over the webpage or fullscreen player. The extension is not tied to a specific AI provider. You can configure local or remote OpenAI-compatible APIs, model names, API keys, source and target languages, video context, and custom prompt templates to build a subtitle workflow that matches your content. Key Features Real-time audio capture and translated subtitles for the current tab Support for videos, livestreams, iframe players, and fullscreen mode Fully configurable speech recognition and translation models Compatible with local and remote OpenAI-compatible APIs Model discovery through compatible /models endpoints Configurable source language, target language, and video context Fully customizable transcription and translation prompts Five-line transcript history for continuity and incomplete speech Adjustable font size, colors, opacity, position, line count, and original text Draggable subtitle overlay Closing the overlay also stops audio capture and translation Built-in silence, noise, repetition, and Whisper hallucination filtering Session isolation and bounded buffering to prevent stale subtitle backlogs Alt+T keyboard shortcut for quick start and stop Mandarin UI on Chinese systems and English UI on all other systems Recommended Models Recommended STT model: faster-whisper-large-v3-turbo It provides a strong balance between transcription quality and inference speed for videos and livestreams. With an NVIDIA GPU and float16 inference, it can produce accurate transcripts with relatively low latency. For stable real-time transcription, use a fixed source language with task=transcribe. Stream Translator also applies voice activity detection, decoding limits, and repetition filters to reduce hallucinations caused by silence, music, and background noise. Recommended translation model: meta-llama-3.1-8b-instruct Its moderate size and fast response time make it suitable for real-time subtitle translation. Running it locally through LM Studio, Ollama, or vLLM provides a useful balance of latency, privacy, and customization. With custom prompts, it works well for: Gaming and esports terminology Livestream slang, memes, and casual speech Incomplete sentences and frequent pauses Custom translation styles and tones Player names, character names, and specialized vocabulary The extension sends an explicit source-to-target language contract and includes the five most recent transcript lines as continuity context. How to Use Install and pin Stream Translator in Chrome. Open Model & Subtitle Settings. Configure the speech recognition API, model, and source language. Configure the translation API, meta-llama-3.1-8b-instruct, and target language. Add video context or customize the prompt templates when needed. Save the settings and open a video or livestream. Select Test Subtitle Overlay to verify that subtitles can be displayed. Select Translate Current Tab, or press Alt+T. Drag the overlay to reposition it. Select the close button to hide subtitles and stop translation.