Bandwidth extension of narrowband speech based on Hidden Markov Model
Zhang Yong, Yi Liu · 2014
This paper presents a novel algorithm for restoration of the missing bandwidth of narrowband speech signals. The proposed algorithm improved the performance of the traditional line spectral frequencies (LSF) based extension algorithm by exploiting a Hidden Markov Model (HMM) to indicate the proper representatives of different frames, and by applying a minimum mean square criterion to estimate the wideband LSF values. Moreover, a new fuzzy mapping algorithm was proposed to estimate the gain factor. When compared to the conventional state-of-the-art bandwidth extension algorithm, the average PESQ score is increased by up to 0.35. Furthermore, in a subjective preference evaluation with 24 experienced listeners, the results show that the proposed algorithm outperforms the traditional method and completely eliminates the undesired whistling sounds.