Inter-Channel Prediction to Prevent Unmasking of Quantization Noise in Beamforming

This paper studies the use of inter-channel prediction for the purpose of preventing or reducing the risk of noise unmasking when beamforming type of processing is applied to quantized microphone array signals. The envisaged application is re-use and post-processing of user-created content. Simulations with an AAC coder and real-world recordings using two microphones are performed to study the suitability of two existing coding tools for this purpose: M/S stereo coding, and the AAC Long Term Predictor (LTP) tool adapted for inter-channel prediction. The results indicate that LTP adapted for inter-channel prediction often gives more coding gain than mere M/S stereo coding, both in terms of Signal-to-Noise Ratio and Perceptual Entropy.