The circuit you link to uses back-to-back diodes that will limit ALL audio above a certain level - but the 'limiting' will be in the form of clamping and will severely distort the resultant output.
Since the speech is actually part of the music you can't reasonably or easily separate them -...