Vibevoice like Karpathy

Vibevoice is a cutting-edge local speech-to-text tool that leverages the power of the faster-whisper model to provide fast and efficient dictation across any application on your system. Designed for productivity enthusiasts, Vibevoice allows users to seamlessly integrate voice commands without the need for costly online tokens.

Key Features

• Dictation Anywhere: With Vibevoice, transcribe spoken words effortlessly in any text field across applications such as text editors, browsers, and chat platforms.

• Easy Operation: Simply run the application and activate dictation by holding down the right control key. As you speak, your words will automatically be typed out, enhancing your workflow.

Quick Start Guide

1. Start the application by executing the following command:

   vibevoice
   

2. Hold down the right control key (Ctrl_r) while dictating.
3. Release the key to transcribe, and see your text appear instantly.

Configuration Options

Customize your experience by changing the trigger key:

export VOICEKEY="ctrl_l"  # Use left control instead

Requirements

For optimal performance, ensure the following:

• Python Dependency: Python 3.12 or higher.
• System Requirements: A CUDA-capable GPU is recommended; however, CPU usage can be enabled in the application settings. Make sure to have CUDA 12.x, cuBLAS, and cuDNN 9.x installed to run the software efficiently.

Acknowledgments

Inspired by the original work of whisper-keyboard by Vlad, and built utilizing the optimized Whisper implementation from Faster Whisper. Developed by Marc Päpper.

Experience fast, local speech-to-text functionality with Vibevoice, where your voice is the key to productivity.