Detection of repeated signal components and applications to audio analysis

Our work in the last two years was mainly concerned with the detection of structured audio components within source signals. In this, an important type of structure are repetitions such as repeating bird calls or percussive elements in music. A few months ago, we have proposed a novel technique for detecting multiply (i. e., more than once) repeated signal components within a target signal. For such cases, we were able to improve classical autocorrelation techniques. In our experiments, we up to now have successfully considered applications in bioacoustics and in speech processing. It was interesting to discuss the topic within an interdisciplinary community as it was present at the Dagstuhl seminar and to learn about further possible applications — and existing solutions — from other domains, especially when dealing with noisy or distorted signals. For me, related interesting questions are both how to automatically separate, or even extract, all structured signal parts from the residual signal and how to do this efficiently for large scale signal scenarios. As a first follow-up activity to the Dagstuhl seminar, I am organizing a special session on “Audio Signal Detection and Classification” covering topics such as audio monitoring, signal detection, segmentation and classification, audio fingerprinting, matching techniques, and audio information retrieval. The special session, which will be held at the IEEE Workshop on Cloud Computing for Signal Processing, Coding and Networking (IWCCSP) on March 11, 2014, aims at bringing together experts from the audio signal processing area with the cloud computing community