Voicebox tutorial: local voice cloning and speech setup

This tutorial introduces Voicebox as an open source ElevenLabs alternative for local voice generation. The presenter walks through installation, model downloads and creating voice profiles. He describes processing as local to the user's machine; the video does not independently establish his broader claim of complete data privacy.

Voice profiles can use a microphone recording, an uploaded audio file or captured system audio. The walkthrough transcribes a recording before saving the profile, then generates speech with Chatterbox TTS. If transcription fails on first use, the presenter advises waiting for the model download or installing a transcription model manually. He recommends Qwen3 1.7B TTS and Chatterbox, describing the latter as more expressive and supporting 23 languages.

Examples use the presenter's voice and samples for Jarvis and Ultron. A robotic effect changes the Jarvis output, while the presenter finds some generated delivery flat. The Custom Voice model accepts emotional delivery instructions, but in the demonstrated version it works only with built-in voices, not cloned profiles.

The Stories editor arranges generated or existing clips on a timeline, including overlapping audio, for longer pieces such as audiobooks or podcasts. The tutorial also covers audio transcription, adding reference samples to profiles, permanent voice effects and custom effect presets.