Algorithm Of Speech Enhancement Based On Auditory Masking And Implementation Based On DSP

Noise exists everywhere in real situations, it impairs the quality of speech, especially in high noise situations, human ear is unable to distinguish the speech, so a system for speech enhancement is needed strongly. The primary purpose of speech enhancement is to reduce the noise disturbance from noisy speech so that the speech will be as close to pure speech as possible and the tiredness will be decreased.In recent years, the technology of speech enhancement has been paid great attention and many algorithms have been appeared. This paper analyses characteristics of speech and noise, then researches several popular algorithms about speech enhancement(self-adaptive filter, wiener filter, spectral subtraction and its improved method), and discusses the merit and shortcoming of them, especially analyses the causes of the musical noise of spectral subtraction.Based on auditory masking properties enhancement algorithm is mainly studied in this paper to solve the musical noise problem. First, the paper analyses characteristics of human hearing—time-frequency analysis and auditory masking, researches several auditory models, and uses the Johnston model finally. Then on the basis of hearing characteristics and auditory models, profound research has been made on algorithms which one combines spectral subtraction with auditory masking and the other combines MMSE with auditory masking, then the paper carries on contrast simulation in different background noise. The result shows that, error of speech estimate will effect error of auditory masking threshold, so this paper proposes an improved algorithm. The algorithm combines noise upgrades in real time, multi-band analysis and auditory masking in speech estimate, then calculates threshold through the speech, after that, the coefficients are adjusted according to perception. Finally, the algorithm is simulated, and the results are analyzed.In the realization, the paper studies the structure functions of the platform, carries on software design and program, and realizes the algorithm on CCS and runs on the system successfully on the ICETEK-VC5416-A. The simulation results shows its excellent performance in background noise and musical noise, and it can be realized.