Suppression of Multiple Noises with Different Directions of Arrival Using Instantaneous Spectral Subtraction
Koki Takenaka, Kenji Ozawa · 2021 International Conference on Electrical, Communication, and Computer Engineering (ICECCE) · 2021
The aim of this study is to suppress multiple noise signals using a microphone array and two-dimensional (2D) spectrum. The output from the microphone array is regarded as a spatio-temporal sound pressure distribution image, and the noise spectrum is estimated based on the 2D amplitude and phase spectra obtained by the 2D Fourier transform of the image. This estimation is performed using a deep neural network (DNN) for every 32-ms temporal segment, and the noise is suppressed by subtracting the estimated results from the observed spectrum. The system was realized by observing the characteristics of 2D spectra obtained for signals from multiple sources, and using the 2D spectra of two sources as training data of the DNN. Results of the computer simulation experiments show that the noise suppression values are -17 to -13 dB for two noise sources and -17 to -10 dB for three noise sources.