VoxCPM 2: Pinokio setup and voice cloning tutorial

This tutorial covers VoxCPM 2 installation through Pinokio and speech generation in its Gradio interface. The presenter uses the app's search and install workflow, leaving the default folder name unchanged while Pinokio installs dependencies. The guide presents the model as supporting 30 languages and running locally on a PC.

The walkthrough explains three modes. Voice design creates speech from a written description without reference audio. Controllable cloning combines a reference recording with style instructions. Ultimate cloning automatically transcribes the reference audio, but the presenter says it does not accept creative style prompts and can produce a brief artifact at the start because it continues from the reference.

Examples test melancholy, anger, laughter and a child voice. The presenter explains CFG as control over adherence to style instructions and describes generation steps as a speed versus quality tradeoff. Several examples use CFG 2 and 10 steps; later tests vary CFG between 1.8 and 2.6.

The multilingual section tests English, Turkish, Spanish, French, Arabic and Farsi, sometimes reusing generated speech as the next reference. The presenter reports consistent voice characteristics across these examples. These are demonstrations and the speaker's assessments, rather than independent measurements of pronunciation or cloning accuracy.