LocalText2Voice: Windows setup and audiobook tutorial

This tutorial covers LocalText2Voice 1.2.1, which the presenter describes as a free, open source Windows application under the MIT license. It splits long texts into speech segments, supports different character voices, and combines narration with music before export. The demonstration uses OmniVoice, though the interface also supports other local engines and cloud TTS services.

Setup includes a CPU option using Piper and a GPU option. During OmniVoice installation, the application recommends at least 30 GB of free disk space and estimates 30-60 minutes for setup; the presenter reports a shorter wait. Windows displays a signature warning during installation. macOS and Linux packaging appear as roadmap experiments, rather than demonstrated installation options.

The first example turns more than 44,000 characters of Russian verse into a 52-minute, 32-second recording. A second prose example uses Whisper review to compare generated speech with the source text. Users can edit a segment's text and regenerate it individually. The tutorial also covers voice references, normalization dictionaries, output folders, and markup for pauses, voice changes and sound effects.

Language support and voice cloning depend on the selected model. Generation speed also varies with the model and hardware. The presenter points out pronunciation errors in the sample and reports being unable to stop playback from the review panel's listen button.