Two papers advance speech-to-text alignment for ASR

First paper proposes gradient-based alignment applicable to any ASR model. Second paper introduces fully differentiable forced alignment via soft dynamic programming.