Hiroshi Fujimura, Graduate School of Information Science, Nagoya University, Nagoya
Katsunobu Itou, Graduate School of Information Science, Nagoya University, Nagoya
Kazuya Takeda, Graduate School of Information Science, Nagoya University, Nagoya
We analyze the dependency oj.conversationa1 utterances on the mode of dialogue. A speech corpus of 800 speakers collected utzder three dzperent modes, i.e., talking to a human operutor u WOZ systenz and an ASR systenz, is used for analysis. Characteristics such as sentence conzplexity, loudness of the voice and speaking-rate were found to be sign$cuntly different among the dialogue modes. Linear regression unalysis results also clarlfi the relative importance ofthose churucteristic.s on speech recognition accuracy.
Citation:
Hiroshi Fujimura, Nobuo Kawaguchi, Shigekio Matsubara, Katsunobu Itou, Kazuya Takeda, "Analysis of a large in-car speech corpus," icdew, pp.1209, 21st International Conference on Data Engineering Workshops (ICDEW'05), 2005