Voicebox

Voicebox is an open-source desktop application that lets you clone voices, generate speech, and dictate text entirely on your local machine. It works by analyzing short audio samples (as little as 3 seconds) to create voice profiles that can then generate speech across seven different text-to-speech engines. The app runs completely offline using your computer's GPU, making it a privacy-focused alternative to cloud-based voice services.

The tool stands out with its comprehensive feature set including a multi-voice timeline editor for creating conversations, an audio effects pipeline with presets, and system-wide dictation capabilities. It supports various inference backends including Metal, CUDA, ROCm, Intel Arc, and DirectML, giving you flexibility based on your hardware. The dictation feature uses Whisper models for transcription and includes optional local LLM refinement to clean up ums and self-corrections without sending data to the cloud.

Voicebox is particularly useful for content creators producing voiceovers, voice artists experimenting with different vocal styles, and developers working with AI agents. The MCP integration allows AI assistants like Claude Code and Cursor to speak back to you using cloned voices, while the local operation ensures complete data privacy and no subscription costs. It's available for macOS, Windows, and Linux systems.