Generate speech in a chosen voice from reference audio