Research On Speech Enhancement Algorithms For Digital Hearing Aids

With the gradual aging of population and increasing social focus on theissue of deafness, more and more attention has been paid to the developmentof hearing aids. In recent years, Digital Hearing Aid (DHA) has taken theleading place of hearing aid market due to its desirable features of small size,low power consumption, excellent adjustability, etc. Meanwhile, digital signalprocessor with high speed, multiple functions and low power consumptionoffers an adequate processing core for the vast computational tasks requiredby digital signal processing algorithms in DHA. The thesis primarily studiesspeech enhancement algorithms for DHA with a combination of thecharacteristic of Digital Signal Processor's (DSP) architecture, so as todevelop a real-time, efficient speech enhancement algorithm which isapplicable to DHA.A speech enhancement algorithm for single channel DHA was designedand implemented in real time on a digital hearing aid platform based on TITMS320VC5502 DSP. By a comparison of various algorithms, a voiceactivity detection algorithm based on power envelope was adopted todistinguish the speech-absent parts from speech-present parts in noisy speech.Then with the statistics of speech-absent parts, we performed power spectrumestimation in both speech-absent and speech-present parts. And finally thetarget speech got enhanced through wiener filter. To deal with the annoyingmusical noise of the processed speech, power spectral smoothing andaveraging techniques was adopted by us. The whole algorithm is compact,effective and easy to be implemented and optimized on DSP.To address the problems met in single channel speech enhancement, wefurther studied beamforming algorithms with microphone array for speechenhancement in DHA, with a focus on superdirective beamforming andadaptive beamforming algorithms, and made an analysis of the importantparameters' effect on the performance. Moreover, we imposed somemodifications on the traditional GSC-structured adaptive beamformer toovercome its drawback, i.e., take both noise reduction and speech distortioncaused by the beamforming process into the consideration of the cost functiondesign. As a result, the iterative formula was redefined in NLMS algorithm soas to decrease the speech cancellation. The simulation experiments exhibiteddrastic reduction of background noise and strong interferences, along withobviously less speech distortion after the adoption of the new cost function.And the robustness of the algorithm in an actual environment is also enhanceddue to the modifications. In addition, most of the calculation in the algorithmcan take advantage of the characteristic of DSP architecture and beimplemented efficiently on DSP.Single channel speech enhancement algorithm and the beamformingalgorithm with microphone array suppress the background noise according todistinct noise characteristic, so they complement each other to some extent.We can get a better performance of speech enhancement through acombination of them in DHA.