Aligning Meeting Recordings Via Adaptive Fingerprinting

This paper proposes a robust and efficient way to temporallyalign a set of unsynchronized meeting recordings, such as mightbe collected by participants’ cell phones. We propose an adaptiveaudio fingerprint which is learned on-the-fly in a completelyunsupervised manner to adapt to the characteristics of a givenset of unaligned recordings. The design of the adaptive audiofingerprint is formulated as a series of optimization problemswhich can be solved very efficiently using eigenvector routines.We also propose a method of aligning sets of files which usesthe cumulative evidence from previous alignments to help alignthe weakest matches. Based on challenging alignment scenariosextracted from the ICSI meeting corpus, the proposed alignment

Proceedings of the 16th Annual Conference of the International Speech Communication Association (Interspeech 2015), Dresden, Germany, pp. 786-790

Speech