Detecting speaker change in background audio streams
Hilary Si Yin Toh, Pei Xuan Lee, Boon Pang Lim, Nancy F. Chen · The Journal of the Acoustical Society of America · 2013
The cocktail party effect refers to the ability of humans to selectively focus on one speech stream among multiple human-generated background conversations and noise sources. This ability is influenced by various factors including directionality, speaking rates, different speech accents and speaking styles. It is displayed most clearly as a binaural effect, requiring listeners to be able to identify and differentiate between multiple sound sources simultaneously. Past studies [Cherry, J. Acoust. Soc. Am. 25(5), 975–979 (1953)], have shown that listeners can detect a change of speaker voice from male to female in unattended audio streams, while focusing their attention on repeating speech in an attended audio stream. Our work further investigates this by focusing on speaker change across the same gender: a more difficult task given a smaller variance in pitch across speakers. Experiments were conducted on 24 listeners and findings suggest that they are able to recognize speaker changes of the same gender. Two factor ANOVA shows an interaction between listener and speaker gender (F(1, 22) = 7.688, p = 0.01), suggesting listeners perform better when detecting speaker changes within their own gender.