In order to observe vocal cord vibration during running speech, a new system of high-speed digital imaging was developed. The system consists of a fiberscope, a CCD image sensor and a digital image memory. The number of pixels is 128x32 and the frame rate is 2000 frames/sec. A special large size image memory (64Mbyte) was constructed which enables data recording for 6 seconds utterance. By using the system, analysis of vocal cord vibrations during the release of /h/ and /s/ were conducted. For /h/, vibration starts with wider separation of vocal cords. Adductive movement of the glottis continues over several pitch periods during the following vowel and there is a corresponding change in the voice source spectrum which are estimated through inverse filtering of the speech wave.