3.4 Sound source separation fails

Problem

A mixture of sounds from the microphone input cannot be separated. Upon listening to output, the sound cannot be considered as speech, and the distortion is terrible.

Solution

Confirm the following when performing sound separation using GHDSS node.

Confirm that the source localization has been successfully completed: Since the GHDSS module uses source localization results (number of sound sources) for inputs, the separation performance degrades if localization failed. Modify the network files so that the localization results are displayed and confirm whether the system localizes sounds successfully. For further information, see the recipe, Sound source localization fails.
Confirm that GHDSS separated the sounds properly: To determine whether the GHDSS has been performed properly, save a result of GHDSS using SaveWavePCM and confirm that separation was successful. If separation failed in GHDSS , see Parameter tuning of sound source separation
Confirm that post-processing was performed properly: When PostFilter or HRLE are inserted just after separation processing, separation may fail due to improper parameters. Save the post-processed sound and confirm that the post-processing was successful. If the post-processing failed, see bsec:Separation: 006]Reducing the leak noise by post processing

Discussion

None.