Skip to content

Dry speech is non-point-like with high bitrate binaural rendering of MASA

Basic info

Bug description

With acoustically dry clean speech items having speech at left or right in MASA format, the parametric binauralizer may render the panning as less point-like than is desireable which is perceived as somewhat ambiguous direction of speech. This was noticed to be caused by the low regularization factor on the highest bitrates of MASA which seems to be a bit too aggressive for this signal type. The solution is to increase the lowest possible regularization factor slightly to still obtain the benefits with generic signal types while avoiding this problem.

Fix via tuning the value will be provided soon.

Ways to reproduce

Simply have dry clean speech directly to the left or right and use high bitrate in MASA to binaural.