摘要 |
PROBLEM TO BE SOLVED: To detect a voice section for performing suitable voice recognition even when a relatively large noise exists just after a voice input is started. SOLUTION: A feature extraction part 2 extracts a time series of a feature vector containing voice power inputted by a voice input part 1. A voice section detection part 3 detects a minimum value among the voice power obtained by the feature extraction part 2, and calculates normalized power from the extracted voice power and the detected minimum value, and compares the normalized power with the prescribed threshold value to detect the voice section.
|