High-frequency conversational chunks. Listen, shadow 3 times out loud, then prove it in recall mode.
Shadowing: press play, then repeat OUT LOUD at the same speed β three times per phrase.