The Study Of Real-time Speech Enhancement Algorithm For Broadcasting Station

The shortwave half-duplex FM(Frequency Modulation) station is widely used in the millitary wireless communication of voice and data. Because the electromagnetic wave is easily be polluted by the complicate electromagnetic environment, the voice signal received is generally have a lot of noise, which reduce the comfortableness and clearness, and causes auditory tiredness. The thesis concentrates on the issue of real-time speech enhancement for broadcasting station.The main work in this thesis is as follows.First, based on the discussing of the voice transmission and noise properties of half-duples FM station, the ETSI(European Telecommunications Standards Institute) standard Wiener filtering based speech enhancement algorithm is dressed, the principle and key-points of implementation are analyzed. This algorithm works not well in non-stationary noise environment and low signal-noise ratio, because it uses time-region energy based voice activity detection to distinguish voice and noise, which is not reliable enough and will be confusing in low SNR environment. And the algorithm is complicated and computation consuming is another problem.Second, the spectrum subtraction speech enhancement based on noise minimum statistics is investigated, which does not need voice activity detection to remove the noise and have improvement on the stability to the ETSI algorithm. But due to the used criteria of minimum mean square error, the processed signal introduces a lot of musical noise which is not comfortable to hear.Third, in order to deal with this problem, the thesis studied the critical band and masking properties of human hearing, and combined it with the minimum statistics to remove the musical noise.At last, Based on a fixed point DSP chip BF533, the embedded speech platform circuit is designed and debuged, the uClinux operating system is imported into the circuit, device drivers are developed, a complete embedded speech signal processing platform is constructed. The fixed-point C program based on the improved algorithm is developed and bettered for BF533, and is imported into the speechsignal processing platform successfully. And the performance of speech enhancement system is evaluated.