Time-varied pitch gain predictor for low bit rate speech coders

Jar Ferr Yang, Rong San Lin, Seric Hu

Research output: Contribution to journalArticle

4 Citations (Scopus)


The long-term predictor (LTP) can effectively reduce the redundancy of the quasi-periodicity of speech signals. Traditional pitch predictors in linear predictive coding systems are developed by assuming that pitch gains are constant in each processing subframe. In this paper, we introduce a time-varied prediction gain to achieve a better representation of speech periodicity than the traditional approach. Due to the non-stationary variation of speech, the proposed LTP method can trace the detailed fluctuation of speech amplitude in both transient and stationary periods. Simulation results show that the proposed first-order varied-gain pitch predictor obtains a near speech quality as the fifth-order pitch predictor of the G.723.1 coder does; however, the former requires much lower computation than the latter.

Original languageEnglish
Pages (from-to)751-758
Number of pages8
JournalIEICE Transactions on Information and Systems
Issue number4
Publication statusPublished - 2002 Apr

All Science Journal Classification (ASJC) codes

  • Software
  • Hardware and Architecture
  • Computer Vision and Pattern Recognition
  • Electrical and Electronic Engineering
  • Artificial Intelligence

Fingerprint Dive into the research topics of 'Time-varied pitch gain predictor for low bit rate speech coders'. Together they form a unique fingerprint.

  • Cite this