| 摘要 |
PROBLEM TO BE SOLVED: To provide an efficient real-time voice section detection method with rich practicality a wide application range, by which a starting point and an end point of an object source voice section including a consonant which is hard to detect, are accurately detected under various living noise environment such as, not only normal noise but music and television sound, and by which object source voice is output in real time without distortion. SOLUTION: At first, according to frequency characteristics of a microphone for voice recording, a vowel of the object source voice is detected by using a specified frequency group. Next, the consonant preceding the vowel of the object source voice is detected by using the specified frequency group. Then, a plurality of characteristics such as prosody, frequency and power intensity, an order, a section, a statistic value, or the like are observed. In a comprehensive method in consideration of balance and weighting of each element, according to a level characteristic of the object source voice and environment sound/non-object sound, the starting point and the end point of the voice section are detected by determining an amplification and attenuation rate and by applying an appropriate determination condition, and the object source voice and the environment sound/non-object sound are distinguished. COPYRIGHT: (C)2007,JPO&INPIT
|
