Voicebox setup: local voice cloning with Qwen3-TTS

This tutorial covers Voicebox, an open source voice synthesis app, through a Windows setup and voice cloning workflow. The speaker says it runs on the user's computer and works offline after the required downloads. ElevenLabs provides a brief point of comparison, though the speaker qualifies their recollection of its trial and restrictions.

The setup starts with the Windows setup.exe installer, then moves to model management. The speaker recommends downloading the 1.7B TTS model and Whisper Base before creating a voice. The synthesis model is Qwen3-TTS. Whisper transcribes the reference recording so the user does not have to enter its spoken text manually.

Voice samples can come from an audio file, a microphone or system audio capture. Each clip has a 30-second limit, and a profile can contain multiple clips. The demonstration creates a profile from a microphone recording, then generates speech from typed text. The speaker also reports a stalled model download that completed after restarting the app.

Later examples use system audio. One produces a weaker result, which the speaker attributes to noises and impressions in the reference material. The final section introduces Story for combining several voices into one conversation, but does not demonstrate that workflow.