mpv-AI-sub-fastwhisper

Project Overview

mpv-AI-sub-fastwhisper is a subtitle script running inside mpv that transcribes a video's audio with faster-whisper to produce SRT subtitles. Once generated, the script calls the GPT API to translate the content into a target language (Chinese by default), overlaying both as bilingual subtitles. It is forked from dyphire's project of the same name and adds a detailed illustrated tutorial.

Press one shortcut and let the video grow bilingual subtitles on its own.

Key Features

• Speech transcription: uses faster-whisper to recognize the current video's audio and produce SRT subtitles
• Automatic translation: calls the GPT API after transcription to translate into the target language, Chinese by default
• Bilingual display: the translation on top and the original below, color-coded for easy comparison
• Shortcut trigger: press Alt + F during playback to start recognition, generation, and translation
• Illustrated tutorial: walks through downloading the recognizer, placing models, and configuring script options with images
• Configurable options: paths, models, target language, and the GPT API are all set in the script options

Target Audience

• Users watching foreign-language videos without subtitles
• People who want bilingual subtitles directly inside mpv
• Users with an NVIDIA GPU willing to run recognition locally
• Enthusiasts who generate subtitles in bulk

Quick Start

1. Prepare an NVIDIA GPU (6GB+ VRAM recommended) and install 7-Zip or WinRAR
2. Create target folders such as D:\000\ai-whisper\ and D:\000\mpv\
3. Download Faster-Whisper-XXL from Purfview's release page (choose r245.4) and extract it into ai-whisper
4. Create a _models folder under ai-whisper and put the faster-whisper-large-v3 model there
5. Place this script and its config into mpv's scripts and script-opts directories
6. Fill in the recognizer path and GPT API settings in the script options, then press Alt + F during playback

Core Advantages

• Local recognition: transcription happens on your machine, so no video upload
• Bilingual in one step: recognition, translation, and overlay in a single chain
• Detailed tutorial: every step is illustrated, so beginners can follow along
• Flexible options: model, target language, and endpoint are all adjustable

System Requirements

Item	Requirement
OS	Windows
Hardware	NVIDIA GPU, 6GB+ VRAM recommended
Dependencies	faster-whisper engine, FFmpeg, mpv
Optional	GPT API (for subtitle translation)

Notes

• You must supply your own GPT API settings, or only the original text is produced without translation
• Model files are large, so both the download and recognition need time and disk space
• Recognition speed and quality depend on GPU performance and the chosen model
• Never share a config file that contains your keys