A Bark-scale filter bank approach to independent component analysis for acoustic mixtures

Uniform filter bank approach can be considered to perform independent component analysis (ICA) for convolved mixtures. It achieves better separation performance than the frequency domain approach and gives faster convergence speed with less computational complexity than the time domain approach. However, when the uniform filter bank approach is applied to natural audio signals, it provides slower convergence for low frequency subbands and gives inferior separation performance for high frequency subbands. Owing to spectral characteristics of natural signals, we present a filter bank approach that employs a Bark-scale filter bank. In the Bark-scale filter bank, low frequency region is minutely divided, whereas high frequency region has much wider subbands. The Bark-scale filter bank approach shows faster convergence speed than the uniform filter bank approach because it has more whitened inputs in the low frequency subbands. It also improves the separation performance as it has enough data to train adaptive parameters exactly in the high frequency subbands.