Local app for Windows, macOS, and Linux

Alibi poster: Clone. Shape. Perform. Local voice production for dialogue and narration.

Use case

Dialogue and narration work involves cloning a voice from a recording, designing a voice from a description, cleaning up takes, arranging multiple voices on a timeline, and exporting a mixed WAV or a narrated video.

How it works

I built Alibi as a local voice production app for dialogue and narration. The shipped interface includes Voices, Generate, Record & Edit, Arrange, and Video workspaces. Clone a voice from a reference recording, design a synthetic voice from a description, generate speech, clean up takes, arrange clips on a timeline, export a mixed WAV, or add narration to a video. Everything runs on the host machine. Audio, profiles, sessions, and generated files remain local by default.

Inspect the implementation

The public repository includes the application, Windows installer source, setup for Windows, macOS, and Linux, and documentation.

Scope

The installer can use an NVIDIA GPU when one is available. The editor is a lightweight timeline for arranging clips and exporting a mixed WAV.

Production workflow

Setup creates an isolated Python environment, checks out the pinned Qwen3-TTS source, selects CPU or CUDA PyTorch, installs the locked dependencies, builds the frontend, and downloads the models. VoiceDesign weights are optional. Audio, profiles, sessions, and generated files remain local by default.

A typical session starts in Record & Edit: record or import a reference, trim and clean it without changing the original file, and save it with the exact transcript and a profile name. In Generate, select the profile, enter text, and generate speech. Add selected takes to Arrange, position clips, adjust tracks, save the session, and export a mixed WAV.

Review procedure

Follow the repository README for setup and the documented checks.

Content reviewed against repository documentation on November 15, 2026.