Vibe

Vibe is an open-source web app and server that provides offline audio and video transcription. It uses OpenAI Whisper to convert speech to text locally on the user's machine, ensuring that data does not leave the device. The application is available as a desktop installation for macOS, Windows, and Linux platforms.

Users can transcribe local media files, capture system audio in real time, or record directly from a microphone. The software also supports extracting audio from various popular websites, including YouTube, Vimeo, Facebook, and Twitter, to facilitate transcription of web content. It is designed for researchers, journalists, and privacy conscious users who require local control over their transcription workflows without relying on cloud services.

Key features

• Offline transcription of audio and video files using OpenAI Whisper
• Batch transcription of multiple files simultaneously
• Export options for SRT, VTT, TXT, HTML, PDF, JSON, and DOCX formats
• Local AI analysis and batch summaries via Ollama integration
• GPU optimization for Nvidia, AMD, and Intel hardware using Vulkan and CoreML
• Speaker diarization to distinguish between different voices
• Command line interface support for direct system interaction
• Translation of any supported language into English

The application is built using the Tauri framework and leverages whisper.cpp for the AI model interface. For advanced users, it provides a command line interface and an HTTP API with Swagger documentation for programmatic access. It allows for the customization of model arguments in the settings and the integration of custom models via specific download URLs. The software includes a stable timestamps mode backed by Voice Activity Detection (VAD) to provide movie grade timing for subtitles and reels. Additionally, users can generate multilingual summaries of their transcripts using the Claude API.

Vibe is a privacy focused tool for transcription and local AI analysis.