40 likes | 54 Views
Explore cutting-edge polyaural processing methods for improved speech recognition systems. Learn about bandpass filtering, nonlinear rectification, cross-correlation, and recombination techniques. Enhance individual speaker separation and overall system performance.
E N D
EXAMPLES OF POLYAURAL PROCESSING Richard M. Stern, Evandro Gouvêa Robust Speech Recognition Group Carnegie Mellon University Telephone: (412) 268-2535 Fax: (412) 268-3890 rms@cs.cmu.edu http://www.cs.cmu.edu/~rms Examples from Interspeech 2007 Paper
Some implementation details • 11 simulated mics in Flanagan-type logarithmic array • Two speakers: • One arrives “on axis” (zero delay between sensors) • Other arrives with delays of 1, 2, 4, and 8 samples: • No reverberation added • Polyaural processing: • Bandpass filtering followed by nonlinear rectification • Cross-correlation with lag at “look direction” • Bandpass filtering followed by recombination across frequency
Examples of polyaural processing Brian Stef • Individual speakers • Combined at zero dB SIR • Delay-and-sum beamforming • Polyaural separation, no straightness wtg • Polyaural separation with straightness wtg