Section: New Results

Online Methods for Audio Segmentation and Clustering

Audio segmentation is an essential problem in many audio signal processing tasks, which tries to segment an audio signal into homogeneous chunks. Rather than separately finding change points and computing similarities between segments, we focus on joint segmentation and clustering, using the framework of hidden Markov and semi-Markov models. We introduced a new incremental EM algorithm for hidden Markov models (HMMs) and showed that it compares favorably to existing online EM algorithms for HMMs. Early experimental results on musical note segmentation and environmental sound clustering are promising and will be pursued further in 2015.

Theoretical results were published in [11] in collaboration with the SIERRA project-team, and experimental results were further extended in [32] . Early experimental setups show that our algorithms out perform state-of-the-art supervised methods for Percussion Sound classification. In collaboration with IRCyNN (Nantes) we are currently studying algorithmic extensions to complex environmental sounds.