| With the development of multimedia technology, the use of video has increased in many fields, and captions are frequently inserted into video images to aid the understanding of audience. This thesis designs and implements an automatic caption generation system based on Hidden Markov Models. This system can extract speech endpoint from continuous wavform and combined with caption to locate the words in subtitle.This thesis firstly introduces the use of HMM in speech recognition, and studies the construction of acoustic model and the extraction of audio parameters. Considering the features of large vocabulary, speaker-independent continuous Chinese speech model training, we use 39-dimensional MFCC parameter and Chinese phoneme for models training. With all the models and the text information of captions, we combine the phoneme models together and use Token Pass Models to segment the audio and implement the forced alignment of audio and caption text. At last we implement the caption location system which automatically generates the subtitles for videos. |
