A new adaptive summing technique for audio teleconferencing

Simultaneous speech from two test subjects were used as the source data to test the effectiveness of a proposed algorithm. The audio results indicate that, as predicted, the direct summation of μ-law and adaptive PCM signals creates severe distortion. When an adaptive voice summing technique was applied, the distortion was dramatically decreased. Our results show that the resultant intelligibility at overlapping regions is reasonably high in all cases. The speech quality from μ-law compandors and feedback adaptive PCM compandors is only slightly inferior to the direct summation of analog voice signals. Compared to a non-adaptive scheme developed by the author in 1984, the weighting factors for each speaker in this technique can be automatically determined at each sampling time. Also, there is no need to use the strategy of frame delay described in the 1984 paper. This technique has also shown promise for future audio teleconferencing application.