An Acoustic Model Adaptation Method Based on Pronunciation Variability Analysis for Foreign Speech Recognition and apparatus thereof
暂无分享,去创建一个
The present invention is to provide a sound model conversion techniques to improve the recognition performance for the other sprachraum speaker voice. First, analyze the other sprachraum speaker sound to suit the acoustic model to investigate the mutation pronunciation. Then it converts the acoustic model while sharing stages of the learning process of the acoustic model based on a variation of the other sprachraum speaker audio pronunciation. By combining the present invention with the existing acoustic model adaptation scheme of converting an acoustic model it can be obtained a good recognition performance gains over. The present invention is not only recognized performance improvement for other sprachraum speaker voice, it seems the effect of reducing the degradation of the recognition performance for the native speaker audio. Speech recognition, speech recognition, speaker sprachraum other data based pronounce analysis, decision trees, acoustic model transformation, the acoustic model adaptation, the state share