Page  63 ï~~Pulsed Noise in Mu Techniques for Extraction, An Perry R. Cook Christopher PRC, CC, JOS @CCRMi Stanford Center For Computer Rese (CCRMA) Stanford, Calif ABSTRACT: The causes and effects of per instruments are investigated. Three classes o string (in the form of a cello), the single re human voice. Three methods of extracting along with new methods of analyzing and vi The effects of pulsed noise on instrument osc NTfT5PSfTTPCP N WTN hT ANTh T flWFTh 'T1

Page  64 ï~~certain signals where the noise bursts are clear enough to b quantitative techniques extraction i results involve which can used for comparison identifying the periods the ori done the period similarity method). detection of some time domain feature using a method suc detection. Each detected period yields a pointer into the ti noise 'periods' can be subdivided (into sections for behavior. In the period-synchronous noise power analysis methc computed in each of the sub-sections. These powers can t height representing power, one axis representing the perio position within the period. Inspection of this noise per valleys (running in the direction of the period number) i duty cycle of the noise pulses can be deduced from the dynamic range of the noise can be deduced from the heigh The third analysis technique involves performing I each of the subdivisions Each DFT can be used to con spectra corresponding to a particular period position can b

Page  65 ï~~Male JAJ@ 100 Hz Fahstb /AI/@ 273 Pe ri od r Open Pos iion (10sed LogM a 0dB6 -20dB Period Posilion LogMq OdD -20dB "i Closed Fm tq. QCHz) Il Figure 3. Power Surfaces (above) and Spectral Periods (belo the tnnmc are! male rhesct voice IAI nt 100 T-W male falsetno/