VLSI implementation of the SIFT algorithm for pitch detection

Speech voicing classification and pitch detection are fundamental techniques in speech analysis. Voicing information provides valuable insights into the nature of the excitation source used in speech production, and the pitch information is useful to many speech processing applications. In 1972 John Markel developed a technique which combined the benefits of inverse linear predictive (LPC) analysis and simple short-time autocorrelation to extract essential speech parameters. The research resulted in the simplified inverse filter tracking (SIFT) algorithm for voiced/unvoiced classification of speech signals and pitch period determination [1]. Up until now this algorithm was available in various software implementations only. This paper describes an alternative real-time CMOS hardware implementation of this algorithm that is small enough to be implemented into a mobile communications device.