A Hybrid of Fractal Code Descriptor and Harmonic Pattern Generator for Improving Speech Recognition of Different Sampling Rates

Abstract

Currently, the different sampling rate for speech recognition is a grand challenge due to supporting applications of divergent platform devices, such as mobile device interaction, interactive voice response system, voice search, voice dictation and voice identification. Furthermore, such applications require efficient speech features to represent input signals. However, the different sampling rates of speech signals lead to the different features. This phenomenon comes from speech harmonic signal lost. It becomes a key factor that decreases the speech recognition rate. Therefore, this paper proposes a hybrid of fractal code descriptor and harmonic pattern generator to convert all different sampling rate signals to standardized signals. In this method, an independent resolution property of fractal code descriptor is applied to training and testing speech signals. Then, the pitches of such signals are used to recover harmonic pattern of lost signals. This method can effectively reconstruct speech signals at any sampling rates. When its performance is evaluated with AN4 corpus of CMU Sphinx speech recognition engine, the experimental results show that the proposed method can significantly improve the speech recognition rate, even if the sampling rate of testing speeches differs from that of training speeches.