A Low-complexity 4.0 Kb/s Speech Coding Algorithm
Tang Kun · Audio Engineering · 2009
The 4.0 Kb/s low-complexity speech coding algorithm based on SELP (Sinusoidal Excitation Linear Prediction) is proposed. The algorithm employs global optimum scalar quantization with inter-frame prediction technology for LSP parameters,not only guaranteeing the quality of speech and the rate of coding,but also lowing down the calculation and storage requirements. Meanwhile,the variation of speech is depicted more actually by the parameters extraction of energy and UV (Voice/Unvoice) in sub-frame. The fractional pitch exerts a great of benefits for describing the tone of speech. And the multi-stage vector quantization of excitation parameters compensates the disadvantage of SELP on speech' s naturalness. The simulation results show that the informal subjective MOS score of this algorithm reaches 3.5,having the same speech quality of AMR 4.75 Kb/s,and also maintain low-complexity comparing with those algorithms based on CELP around 4.0 Kb/s.