Research And Implementation Of Speech Enhancement Based On ARM Platform

Speech is an important medium for the exchange of human communication, is one of the most important people face to face communication form. However, in the actual environment, speech is often disturbed by the noise, the speech with noise,reduce the speech intelligibility, and even more, completely submerged voice signal, makes communication can not be achieved. With the development of science and technology, the emergence of various intelligent devices, exchange of people and machines are more dependent on the speech, speech noise will seriously affect the quality of voice interaction.Noise reduction of speech is the speech signal processing is a promising technology, its purpose is to remove the noise in speech signal. The method is the use of modern digital signal processing from the speech signal with noise to obtain pure speech signal, restrain the background noise, thus to improve the speech intelligibility and signal to noise ratio.The speech denoising algorithm a lot, but due to various noises in real circumstance, still not a generic algorithm. I first introduce the background, current situation and significance of speech denoising algorithm, after discussing the characteristics of the speech and noise signals, introduces in detail the subjective and objective evaluation criterion of speech denoising algorithm, carefully studied all kinds of speech denoising algorithm, such as statistical model, signal subspace, Wiener filtering and the generative model, compared the effect of noise reduction the advantages and disadvantages of the various algorithms. Mainly introduces the classical spectral subtraction, analysis the advantages and disadvantages of the spectral subtraction method, and the improved spectral subtraction method is discussed in detail, and put forward the weighted energy spectrum entropy spectral subtraction based on the noise reduction effect, can be in and the complexity of the algorithm to achieve the system design requirements.This paper introduces the ARM processor and Linux operating system and Python, discussed in detail the spectral subtraction based on weighted energy spectrum entropy in ARM platform to achieve the effect of noise was tested. The test results show that the noisy speech by the system after processing, can restrain the background noise to a certain extent, to improve the speech signal to noise ratio, improve the speech intelligibility of speech, makes possible in strong noise calls, voice interaction enhances the possibility for rear end.