| Summary: | Speakerphones are an essential part of making remote meetings with groups of people possible. An important ingredient to achieve a satisfying teleconferencing experience is good audio pickup quality of the speakerphone. Limes Audio supports a multi-microphone speakerphone reference design with daisy chain connectivity. This implies a structure where several microphone signals need to be mixed down to one output channel for transmission.This thesis is focused on the challenges of automatically mixing down these multiple microphone signals to one output with minimum amount of reverberation and extraneous noise. In this thesis, I thoroughly analyze the automatic microphone mixing problem and the existing implementation from Limes Audio. I review a selection of alternative mixing methods, propose a new method, and compare the performance of them using objective speech quality and intelligibility measures. One significant part of the process has been to define a suitable test protocol for assessing the performance of different mixer implementations, and the project has resulted in a test procedure which could be used for evaluating the automaticmixing performance of competing products in the future. The results from my evaluation show that the best quality is achieved by using a single microphone per speaker. The current method tries to do this and achieves relatively good performance compared to the alternatives. The proposed method with somewhat different characteristics performs worse in some scenarios, but better in others.
|
|---|
