A time-domain pitch extraction scheme for noisy speech signals

  • Celia Shahnaz Centre for Signal Processing and Communications, Dept. of Electrical and Computer Engineering, Concordia University, Montreal, Que. H3G 1M8, Canada
  • Wei-Ping Zhu Centre for Signal Processing and Communications, Dept. of Electrical and Computer Engineering, Concordia University, Montreal, Que. H3G 1M8, Canada
  • M. Omair Ahmad Centre for Signal Processing and Communications, Dept. of Electrical and Computer Engineering, Concordia University, Montreal, Que. H3G 1M8, Canada
Keywords: Computer simulation, Error analysis, Harmonic analysis, Speech analysis, Time domain analysis, High-pitched speakers, Pitch detection algorithms, Pitch-errors

Abstract

A new pitch detection algorithm for speech corrupted by a white or a car noise is presented, considering the LP residual of the pre-processed speech as a representation for the GC events. A weighted and harmonically summed AMSF of the LP residual is proposed that is able to effectively quell the pitch-errors in the presence of a noise. The peaks of AMSF at different pitch-harmonic locations are added and weighted by a periodicity dependent weighting factor for every possible pitch period. The resulting weighted and harmonically summed AMSF of the LP residual is globally maximized to extract the desired pitch period. The proposed method is able to reflect its efficacy to a significant extent for extracting pitch of both low and high-pitched speakers in the white or car environmental noise. Simulation results have shown that the proposed method outperforms the pitch detection algorithms implemented in the same noisy environment.
Published
2007-09-01
How to Cite
1.
Shahnaz C, Zhu W-P, Ahmad MO. A time-domain pitch extraction scheme for noisy speech signals. Canadian Acoustics [Internet]. 2007Sep.1 [cited 2019Oct.16];35(3):114-5. Available from: https://jcaa.caa-aca.ca/index.php/jcaa/article/view/1927
Section
Proceedings of the Acoustics Week in Canada

Most read articles by the same author(s)