Design Of Speech Recognition System Based On CFCC And Application In The Internet Of Things

Feature extraction is one of the key technology in Automatic Speech Recognition(ASR). In recent years, feature extraction combined with auditory characteristics is a hotspot. With the extensive popularization of the Internet ofThings (IoT), the design and realization of ASR for IoT is attracting more and more attention from researchers.This paper summarizes the research and development of feature extraction and its appli-cation on IoT. And a feature based on auditory model called Cochlea Feature Cepstral Coeff-ieients(CFCC) is implemented. The CFCC is applied to the speaker-independent and voice interactive smart home system which is one of the important branch of IoT. The main work are as follow:First, the mathematical definition and realization method of CFCC is introduced in this paper, and the simulation study in MATLAB is implemented. MFCC and GFCC are also extracted to compared with CFCC under different level of Signal to Noise Ratio(SNR). The clean testing condition recorded under a quiet environment in the laboratory has a high SNR, the accuracy of all features are more than 90%. After adding white noise, the SNR of the testing condition drops to 6dB, the performance of MFCC drops to 67.5%, GFCC is 86.8%, and CFCC is 90.4%. If the SNR is OdB, the MFCC accuracy is less than 20%, GFCC is 45.0%, and CFCC is 65.2%, but it is still higher than the other two. The experiments shows that CFCC has better noise immunity.Second, design an ASR based on CFCC which is integrating CFCC into the Sphinx to replace its default feature MFCC. Then ported it to embedded platform. The embedded plat-form based on OMAP3530 and installed Linux. What’s more, the QT graphical library, ALSA library, NRF24L01 wireless module and its driver are added which can guarantee the imple-mentation of ASR effectively.Finally, applied the ASR based on CFCC into smart home system. This system liberate users from the traditional way of hands and eyes interaction. Users could operate the smart home system in non-contact way without fixd place and interrupting their current behavior through voice.