Research On Performance Comparison Between Human And Machine For Emotional Speaker Recognition

The ear’s ability to discriminate speech is people’s significant perception. The desire to make an Automatic Speaker Recognition (ASR) system like the ear has always been human’s dream. Since the research of speech, auditory mechanism and pattern classification, the ASR system has reached a good performance. In practical environment, people have emotions when speaking, which greatly affect the speech. The speaker recognition of emotional speech is called Emotional Speaker Recognition. The mismatch between train and test speech as a result of emotion has deeply decreased the performance of ASR system. At the same time, Human’s ability to discriminate speech also decreases for the emotion. Through comparison between the false recognition sentences, we find they are not the same for Human and ASR system.The paper makes a research of emotion speech, design and implement a website platform for ASR. Through spectrogram, feature analysis and hearing perception, compare the ASR and human in emotional speaker recognition.The contributions of this thesis are as follows:1. Design and implement a website platform for ASR, which makes the speaker recognition experiments is more easily and convenient and meets the needs oflarge quantity of experiments in speaker recognition.2. Design and accomplish the emotional speaker recognition task for the hearing of human.3. Make research of comparison of ASR and human in emotional speaker recognition, analysis the performance in change of emotion, speaker, text and listener. Make research of the emotion speech under the effect of breath, cry, whisper and stress, and divide the effect into two categories, time continuous and time discrete. At last, put forward a strategy of combination of ASR system and Human.