pyVideoTrans

Appearance

πŸ—£οΈ Build Your Own Real-Time Speech Transcription Tool ​

Real-time speech-to-text, such as for meeting minutes, class notes, or interview transcripts, is now very common and a hot topic that interests many people.

So, would you like to deploy an open-source, fun real-time transcription project β€” WhisperLiveKit? It allows you to easily set up a real-time speech recognition system on your own computer!

πŸ’‘ First, the conclusion: Who is it for? ​

WhisperLiveKit is perfect for learning and experiencing the principles and workflow of AI real-time speech recognition. However, a heads-up: It cannot yet fully replace professional commercial products, but it's already very interesting and quite powerful.

Advantages:

  • πŸš€ Super simple to deploy
  • πŸ’» Comes with a web interface to directly experience cutting-edge technology

Points to note:

  1. Latency Issues Models with high Chinese recognition accuracy (like large-v2/v3) are relatively slow. The speech-to-text delay might be over 10 seconds, or even longer. If your computer has an NVIDIA GPU (recommended 12GB VRAM or more), it will be much faster. Smaller models are faster, but Chinese recognition is not accurate enough.

  2. Network Environment The program needs to download a very large core model file, which is hosted outside the firewall. πŸ‘‰ Therefore, you need to have your "scientific internet access" tool ready in advance.

🧰 Step 1: Preparation (Sharpen the axe before chopping the wood) ​

Before starting, please make sure your computer has the following ready:

  1. Install uv This is a modern Python package management tool. It can install all dependencies with "one command," which is extremely convenient.

    If you haven't installed it yet, you can check my previous tutorial.

  2. Install ffmpeg It's the "Swiss Army knife" of audio/video processing. Our program relies on it to read microphone audio.

    Similarly, if not installed, you can also refer to the previous article.

  3. Enable Network Proxy ⚠️ This is very important! Because the model file needs to be downloaded from servers outside the firewall, please be sure to enable "scientific internet access," and set it to "global proxy" or "system proxy" mode.

βš™οΈ Step 2: Install the Core Program ​

  1. Create a new folder, for example: D:/python/livekit

  2. Open this folder, type cmd in the address bar, and press Enter. You will see a black command-line window πŸ‘‡

  3. Copy the command below into it, then press Enter to execute:

uv init && uv add whisperlivekit faster-whisper --index https://pypi.tuna.tsinghua.edu.cn/simple

πŸ’‘ This command will:

  • Use uv to automatically install WhisperLiveKit and the acceleration dependency faster-whisper
  • And accelerate downloads via the Tsinghua mirror source

Enter the command, press Enter to execute

Waiting for installation... ⏳ Installing, please wait a moment

When you see the interface below, it means the installation is successful! πŸŽ‰ Installation completed successfully

πŸš€ Step 3: Start the Real-Time Transcription Service ​

Continue executing the following command in the command-line window:

uv run whisperlivekit-server --audio-max-len 10 --frame-threshold 20 --model large-v3-turbo --language zh

Parameter Explanation: ​

  • --model large-v3-turbo: Use the faster large-v3-turbo model (much faster than large-v2/v3, with slightly lower accuracy)
  • --language zh: Specify recognition for Chinese

⚠️ The first run will automatically download the model file, which is large. Please ensure a stable network connection and be patient.

Starting up, the first time will download the model, which takes a long time

When the window shows the URL in the image below, congratulations! πŸŽ‰ The service started successfully! Note: When this address is displayed, it means startup was successful

🌐 Step 4: Start Using It! ​

Open a browser (Chrome or Edge recommended) and visit the address:

πŸ‘‰ http://localhost:8000/

You will see a clean web interface πŸ‘‡ Web interface

Click the big red button to allow the browser to access the microphone. Then start speaking, wait a few seconds, and the recognized text will appear on the screen! Listening and transcribing

🧩 Common "Crash" Scenarios and Solutions ​

Don't worry, here are the most common errors:

  • ❌ Model Download Failed If the error message contains words like "huggingface," "download," "timeout," it's almost always because the proxy is not enabled or not set to global mode.

  • ❌ uv Not Found Means uv is not installed properly or not added to the system environment variables.

  • ❌ ffmpeg Not Found Similarly, check if it's installed correctly and the environment variables are configured.

πŸ’€ Lazy Person's Bonus: One-Click Startup Script! ​

Typing commands every time is too troublesome? Then here's a "one-click startup"!

  1. In the project folder (D:/python/livekit), create a new text document.
  2. Copy the following content into it:
@echo off
call uv run whisperlivekit-server --audio-max-len 10 --backend faster-whisper --frame-threshold 20 --model large-v3-turbo --language zh
pause
  1. Click "File" β†’ "Save As," select All Files for the save type, name it start.bat, and then save.
  2. ⚠️ Make sure the filename ends with .bat (not .bat.txt)!

From now on, you just need to double-click the start.bat file to start the service with one click～ No more typing long commands every time, easy and efficient!

πŸŽ‰ Congratulations on completing the deployment! From now on, you can achieve real-time speech recognition on your own computer. WhisperLiveKit is a project very suitable for learning and demonstration. Feel free to try different models and parameters to explore more of its capabilities!