English | 简体中文

Whisper48 is an easy-to-use tool for automatically generating accurate subtitles from videos and audio files. It uses advanced AI models to:

• ✅ Convert speech to text automatically
• ✅ Generate accurate word-level timestamps (when each word is spoken)
• ✅ Support multiple languages (Japanese, Chinese, English, French, German, Spanish, Italian, Portuguese, Russian, and more)
• ✅ Work completely in the cloud (no installation required!)

Perfect for: Video creators, translators, content producers, podcast editors, and anyone who needs to transcribe audio without tedious manual work.

The easiest way to use Whisper48 is through Google Colab - a free cloud service that runs the software for you. No installation needed!

1. Prepare Your Files

• Have your audio or video file ready (MP3, WAV, MP4, MKV, etc.)
• Upload it to your Google Drive (if you don't have one, create a free Google account)
• Place your file in any folder in Google Drive - the notebook will help you find it

2. Open the Notebook

• Click this link: WhisperX48 on Google Colab
• Google will open the notebook in your browser

3. Configure GPU (Important!)

• At the top of the page, click Runtime → Change runtime type
• Select GPU as the hardware accelerator
• Click Save
• This gives you free computing power to process your files faster

4. Run Each Step in Order

• Each cell (code block) has clear instructions in both Chinese and English
• Click the ▶ button (or press Ctrl+Enter) to run each cell one by one
• Wait for it to finish before moving to the next one
• The notebook will guide you through:
  1. Connecting to Google Drive
  2. Installing required software
  3. Selecting your media file
  4. Configuring settings (language, model size, etc.)
  5. Processing and downloading your subtitles

5. Download Your Subtitles

• After processing completes, the subtitle file (.srt format) will automatically download
• Use this file in any video editor (Adobe Premiere, DaVinci Resolve, CapCut, etc.)

• Audio Quality: Clearer audio = better results. If possible, use the original audio track rather than compressed versions
• Model Size: Choose large-v2 or large-v3 for best accuracy. Larger models take longer but are more accurate
• Language Selection: Select the correct language of your content for better accuracy
• Chunk Size: Leave at 5 seconds (default) unless you know what you're doing

Requirements:

• Advanced technical knowledge
• Python programming experience
• Powerful GPU (NVIDIA recommended)
• 30+ GB of free disk space

Instructions: Try running this script on your own computer. You will need to install PyTorch, WhisperX, FFmpeg, and other dependencies first.

⚠️ Note: Local setup is complex and not tested thoroughly. We recommend using Google Colab instead.

WhisperX Technology: Unlike standard Whisper, WhisperX provides word-level timestamps:

• Standard Whisper: "Hello world" (0:00-0:05)
• WhisperX: "Hello" (0:00-0:02) "world" (0:02-0:05)

This precision allows for better subtitle synchronization and more professional results.

• 🇯🇵 Japanese (日本語)
• 🇨🇳 Chinese (中文)
• 🇬🇧 English
• 🇫🇷 French
• 🇩🇪 German
• 🇪🇸 Spanish
• 🇮🇹 Italian
• 🇵🇹 Portuguese
• 🇷🇺 Russian
• And many more...

• Audio: MP3, WAV, M4A, AAC, FLAC
• Video: MP4, MKV, TS, FLV

For a comparison between WhisperX and other models please have a look here: https://www.bilibili.com/video/BV1RFa5zhEeG/

Whisper48 uses a 3-step process:

1. Speech Recognition: OpenAI's Whisper model converts audio to text
2. Timestamp Alignment: WhisperX uses wav2vec2.0 to align text with precise word timings
3. Subtitle Generation: Automatic subtitle formatting and download

The entire process runs on Google Colab's free GPU servers - no installation needed on your computer!

• Website: Detailed guides and FAQ: ifeimi.github.io/whisper48
• Email: Contact the developer at: yfwu0202 AT gmail DOT com
• GitHub Issues: Report bugs or request features on the project's issue page

This project is based on excellent open-source software:

• WhisperX: Improves OpenAI's Whisper with accurate word-level timestamps
• OpenAI Whisper: State-of-the-art speech recognition model
• faster-whisper: Optimized Whisper implementation
• wav2vec2.0: Speech model for timestamp alignment

Whisper48 started as a fork from N46Whisper and has been significantly modified to:

• Use more accurate Whisper-based models (WhisperX)
• Improve Japanese language support
• Provide better timestamp accuracy
• Add more language options
• Simplify the user interface

We encourage contributions to both this project and the upstream projects!

This project is released under the MIT License. See LICENSE.md for details.

• Original concept: N46Whisper by Ayanaminn
• Improvements: WhisperX integration and optimization
• Code improvements: Bilingual documentation and interface enhancements

You are free to:

• ✅ Use for personal and commercial projects
• ✅ Modify and redistribute
• ✅ Study and learn from the code

You must:

• ✅ Include the license text
• ✅ Credit the original authors

Have questions or suggestions? Get in touch!

• Email: ifeimi48 AT gmail DOT com
• GitHub Issues: Report bugs or request features
• Website: ifeimi.github.io/whisper48

I would love to hear your feedback and help with any questions!

Last updated: January 2026