AudioLCM needs only 2 iterations to synthesize high-fidelity audios, while it maintains sample quality competitive with state-of-the-art models using hundreds of steps!<p>The methodological unification of AI studies across different modalities is unavoidable.