Maybe transcribing the audio and then offering a text translation? That sounds more doable then audio to audio.