Fusion of MFCC and IMFCC for Whispered Speech Recognition

Whisper is a low voice spoken in someone’s ear or in private. Whisper speech recognition has always been a hot research topic. Accurate recognition of whispered speech plays an important role in the digital model of speech generation. For example, in the login scene of mobile banking, “dynamic digital voiceprint + face recognition technology” is often used for voiceprint login and identity verification. However, there is no public corpus of Chinese string whispers. To this end, this paper first creates a Chinese digit string whisper corpus, which lays the foundation for the experiments in the paper and future research work; secondly, this paper tries two fusion methods, which are improved by 0.4% and 0.7% respectively on the basis of the baseline.