Research And Implementation On Single-Channel Speech Enhancement Algorithm In Digital Hearing Aid Chips

The hearing aid is the class II medical electronic devices for hearing impaired person. In 90's, every famous hearing aid companies begin to invest and build factories in China, but there is a large gap between the foreign and domestic hearing aid business, particularly in the speech processing chips area, the foreign chips have nearly moponolized the domestic market. Therefore, developing high performance and affordable speech processing chips is a task to be sloved now.Speech enhancement is one of the core functions in the speech processing chips in digital hearing aid, its purpose is to remove the background noise in the environment, and enhance the robustness of the input to the noise for flowing processes. Its performance is directly related to the digital hearing aids'accuracy of hearing loss compensation. When speech enhancement algorithm removed the noise, it must minimize the loss of the target speech signal. At this stage, most of the volume of hearing aids is relatively small, from the speaker to the microphone acoustic feedback is relatively large, which will lead to "whistle" in the production. Therefore, the speech enhancement algorithm using in digital hearing aids need to introduce effective echo cancellation algorithm to control this kind of acoustic feedback.The hearing aids have the characteristic of small size, low power consumption and real-time, so firstly we designed a single-channel speech enhancement algorithm using for digital hearing aid. In this algorithm, we used a short-term energy voice activity detection algorithm. This algorithm has the characteristics of small computation and obvious effect, so it is suitable for using for digital hearing aid. We used it to distinguish the signal from speech zone and non-speech zone, so we use the non-speech zone to take advantage of statistical properties of the noise power spectrum. Then we use the Wiener method to enhance the speech. And, we bring the LMS adaptive echo cancellation to this algorithm, make it have the function of eliminating whistle. Then, we using C language modeling this algorithm, make the fix-point process, and verify it successfully in the TMS320C5509A DSP development board. After this, the average signal to noise ratio is 42.5db. Then based this single-channel speech enhancement algorithm, we proposed a hardware structure. Finally, we make the real-time verification in the Xilinx V2p FPGA board.