Employing Sparsity for Joint Sound Source & Acoustic Channel Estimation
Youngmoo E. Kim and Travis M. DollAll audio is experienced through an acoustic channel, which imposes some amount of alteration upon the "true" source signal. Most often, this is due to the characteristics of the environment (e.g., the physical configuration of a room or space). Assuming this to be a linear process, the sound we hear is a convolution of the "true" source and the impulse response of the acoustic channel. In prior work [1, 2] we have investigated the use of sparse methods for blind room channel (impulse response) estimation, which tend to be sparse in the time domain [3]. Currently, we are pursuing an extension of this work to incorporate a sparse model for describing the acoustic source as well, particularly for musical sources, and to estimate these source and channel model parameters jointly. In doing so, we hope to obtain an improved estimate of the underlying source signal simultaneously with the impulse response of the channel. This method has applications for musical instrument recognition and coding, and an accurate estimate of the acoustic channel could be potentially beneficial for solving problems of music transcription from live (reverberant) sound data.