Noise Reduction in the Spherical Harmonic Domain Using a Tradeoff Beamformer and High Resolution DOA Estimates

Daniel Jarrett, Maja Taseska, Emanuël Habets and Patrick Naylor

Audio Examples

The audio examples provided were obtained using measured room impulse responses with a 32 microphone rigid spherical array of radius 4.2 cm. The noise consisted of incoherent sensor noise with an input signal to incoherent noise ratio of 25 dB at the reference microphone, and of coherent interference with a variable input signal to coherent noise ratio (iSCNR).

The tradeoff between noise reduction and speech distortion was controlled by a tradeoff parameter μ, which was a function of the desired speech presence probability (DSPP), and the parameters η (between 0 and 1) and μ'. Small values of μ correspond to low levels of speech distortion and noise reduction, and larger values of μ correspond to higher levels of speech distortion and noise reduction.

Note: You will need Safari (for OS X) or Google Chrome (for OS X, Windows or Linux) to listen to the audio examples using the player below. Alternatively, you can download a WAV file for each audio example.

iSCNR: -5 dB

iSCNR: 5 dB

iSCNR: 15 dB