The practical use of a system that supports table conversations, such as in restaurants and community rooms, requires the separation of target conversations near microphones from ambient acoustic noise in the background. Based on acoustic analysis, we confirmed that the difference between a conversation near a microphone and ambient sounds is represented in the standard deviation (SD) of the power levels of each utterance. In the case of noisy ambient conversation in a restaurant, we calculated the boundary between the SD of power values of target utterances near a microphone and that of ambient conversation using the Linear Discriminant Analysis (LDA) method. Using the calculated boundaries, we evaluated the performance of the system in distinguishing target conversations near microphones from ambient acoustic noise caused by extraneous speakers. Experimental results using four speakers demonstrated an average identification rate of 83.5%.
[1]
H. Sebastian Seung,et al.
Algorithms for Non-negative Matrix Factorization
,
2000,
NIPS.
[2]
Taras Butko,et al.
Detection and Positioning of Overlapped Sounds in a Room Environment
,
2012,
INTERSPEECH.
[3]
Tatsuya Kawahara,et al.
Bayesian Multichannel Audio Source Separation Based on Integrated Source and Spatial Models
,
2018,
IEEE/ACM Transactions on Audio, Speech, and Language Processing.
[4]
Yumi Wakita,et al.
F0 Feature Analysis of Communication Between Elderly Individuals for Health Assessment
,
2017,
HCI.
[5]
Yumi Wakita,et al.
Communication Smoothness Estimation Using F0 Information
,
2016
.