The Research Of Speech Enhancement Algorithm And Implementation Based On DSP

In general, speech is often corrupted acoustically by ambient noise, the result is the degradation of communication quality and even the system of digital voice processing can not work normally, therefore filtering background noise is essential. The primary purpose of speech enhancement is to extract more pure speech from speech signal with noise. Noise is generated stochastically and the reason for generation and characteristic is quite complicated, therefore eliminating the noise completely is impossible. The essential target of actual speech enhancement is improving the quality of speech and making it more clear and comprehensible so as to be accepted easily or improve the performance of the speech processing system. This articles has conducted thorough research on speech enhancement, the main works are as follows:1. This paper analyses and researches several popular voice activity detection methods, such as spectrum entropy, frequency band variance and statistical model, carries on contrast simulation with signal-to-noise ratio scope from -15dB to 15dB, and decides that VAD method based on statistical model is used as main method for the following research after analysis.2. A variety of methods about speech enhancement are researched systematically, including self-adaptive filter, wiener filter, minimum mean square error, spectral subtraction and so on, and the respective merit and shortcoming of them are analyzed. Under several kind of noise environment, computer simulation for these methods is conducted with signal-to-noise ratio scope from -10dB to 15dB.3. Spectral subtraction becomes most popular technology for being simple and effective among multitudinous speech enhancement algorithms. But the traditional spectrum subtraction often is accompanied by "music noise" and the distortion of speech is big simultaneously. In view of flaw of traditional spectrum subtraction, this paper proposes an improved algorithm based on spectral subtraction of multi-band. In order to verify the practical application value, this paper uses different background noise (noise base and noise of actual gathering). MATLAB simulation result indicates that the proposed method can suppress the background noise and residual "musical noise" more effectively, maintains good intelligibility, and improves the speech quality.4.The principle and principal characteristic of DSP are introduced, especially TMS320VC5416. Software design of the improved algorithm is carried on and the improved algorithm is performed on the DSK which is produced by ICETEK in Beijing firstly. Then, hardware design of the algorithm is carried on according to characteristic of improved method and system work requirement, selecting and designing of outside equipment is completed, a speech enhancement system platform is designed and manufactured by the author, and the proposed algorithm runs on the system successfully and realizes speech enhancement.