Laboratory for Recognition and Organization of Speech

The Laboratory for Recognition and Organization of Speech and Audio investigates novel applications of signal processing and machine learning to extract information from audio signals including speech, music and the general sound environment. I will describe some recent and ongoing projects in LabROSA, including:

  • Eigenrhythms – subspace analysis of drum patterns extracted from pop music http://www.ee.columbia.edu/~dpwe/pubs/tr04-eigenrhythm.pdf
  • Temporal envelope modeling via linear prediction applied to spectra, and applications in speech recognition and analysis/synthesis http://www.ee.columbia.edu/~dpwe/pubs/sapa04-plp2.pdf
  • Segmenting and classifying long-duration `personal audio’ recordings http://www.ee.columbia.edu/~dpwe/pubs/carpe04-minimpact.pdf
  • Analysis and segmentation of recorded meetings using acoustic and lexical cues http://www.ee.columbia.edu/~dpwe/pubs/nist04-turnid.pdf http://www.ee.columbia.edu/~dpwe/pubs/nist04-laughs.pdf

Speaker Details

Dan Ellis is an Assistant Professor in the Electrical Engineering Dept. at Columbia where he leads the Laboratory for Recognition and Organization of Speech and Audio (LabROSA). His background is in signal processing and audio analysis, and his dissertation, completed at the MIT Media Lab, described a prediction-driven approach to Computational Auditory Scene Analysis. His research interests include sound modeling, mixture organization and separation, models of human sound perception, robust speech recognition, and music analysis.

Date:
Speakers:
Dan Ellis
Affiliation:
Columbia University

    • Portrait of Jeff Running

      Jeff Running