Abstract # 188:

Scheduled for Friday, August 18, 2006 07:00 PM-09:00 PM: Session 17 (Regency West 1/3 ) Poster Presentation

Data collection using voice recording and speech recognition in demanding environments

C. Ionica1,2, C. Berman2 and J. Li3
1National Institute of Health (LCE/NICHD), PO Box 529, Poolesville, MD 20837, USA, 2Department of Anthropology, State University of New York at Buffalo, 14216 USA, 3School of Life Sciences, Anhui University, Hefei, Anhui Province, China 230039
     Behavioral data collection has benefited from computer technical advances. However, field conditions pose additional demands for observers. We present and demonstrate a flexible technique that we successfully employed to record social behavior in Tibetan macaques (Macaca thibetana) in rough mountainous terrain at Huangshan, China. We recorded ~235 hours of data on Psion®? handhelds and ~1020 hours on tape, the later being transferred to computers using DragonDictate®?, a speech recognition program. Unlike handhelds, which usually require two hands, eye coordination, and restricted code, tape recording allowed one free hand, flexible dictation, and the opportunity to keep subjects constantly in view. We customized DragonDictate®? by replacing its 30,000 words English stock dictionary with our ~200 codes (macro-commands sending unique strings to a text file, later imported in a Noldus Observer®? project). Each user trained the customized dictionary to his/her voice in about 1 week. Regardless of the native language, 5 observers consistently reached recognition accuracy above 96-98%. Data recorded by CI using both techniques showed that voice recording allowed improvements over handhelds in the frequency of events per sample (focal animal: 130%; ad libitum: 836%), number of samples per day (post-conflict follow-ups: 216%; focal animal: 121%), and fewer aborted samples. DragonDictate®? versions 2.5 and 3 function under all Windows®? operating systems and require modest PC hardware. Newer voice recognition programs (Naturally Speaking®?, ViaVoice®?), proved more difficult to configure.