A real-time implementation for computing line spectral frequencies in speech processing application

The split Levinson algorithm has been shown to be more efficient than the classical algorithm for computing line spectral frequencies (LSFs) in complexity. In addition to computation efficiency, the architecture of the split algorithm also enables the use of fast numerical procedures for real-time applications constrained by low processing delay. More precisely, a very efficient LSF computation can be implemented on a single-chip processor optimized for digital signal processing. The proposed method of implementation takes advantage of the reduced complexity, recurrence property, and parallelism of the formulation based on the split algorithm and numerical procedure