Multi-Stage codebook optimization for LSF quantization with inter-stage prediction
Tang Kun · Journal of Tsinghua University(Science and Technology) · 2010
Line spectral frequency(LSF) parameters play an important role in speech coding algorithms.A multi-stage codebook optimization algorithm was developed for the LSF quantization with inter-stage prediction.Training vectors are clustered according to the codebook vector during every iteration step.The codebook is updated with the criterion to minimize the square error at each stage with other stage codebooks fixed.The iteration is stopped at a fixed number of iterations or when the quantization error between the current and previous iterations is less than a threshold.Tests show that the algorithm efficiently reduces the LSF quantization error and improves the synthetic speech quality on a 300 b/s vocoder.