Neural network based adaptive microphone array system for speech enhancement

Presents a microphone array system for use in handsfree mobile telephone equipment. The array is based on a fast and efficient "on-site" and "self-calibration" scheme. The performance in suppressing the interior car cabin noise and the far-end speech is approximately 17 dB, respectively, while maintaining the near-end speaker level. The near-end signal is almost undistorted. The performance of two different algorithms, normalized least-mean-square (NLMS) and fully connected backpropagation supervised neural network (MLP-NN) are evaluated. The proposed microphone array calibration scheme can also be used in other situations such as speech recognition devices.