中国科学院心理研究所机构知识库

Speech recognition based on speed difference and psychological effect on the speakers' performance is a complex problem involving speech processing and psychological analysis. In this study, we aimed to recognize the main speaker in a meeting by analyzing the speaker's speech characteristics, speed difference, and sense of agency (SoA). SoA refers to an individual's perception of control or agency over their actions and outcomes. To achieve this, we proposed a technology that detects the main speaker by analyzing the speech rate consistency among speakers. Typically, the main speaker in a meeting maintains a consistent speech rate, while other speakers may vary their speech rate. We employed a sliding window approach to segment the continuous speech stream and estimated the speed of each segment to generate a speed curve. By identifying the local minimum on this curve, we determined the changepoints used by different speakers in turn. Next, we considered the segments between two adjacent changepoints whose speed was lower than a certain threshold. These segments were accurately identified and classified as belonging to the main speaker. Experimental results demonstrated that our method improved the recognition of speakers by analyzing speech characteristics, speed difference, and SoA, as well as assessing the perception of control or agency over speaking actions and outcomes. The performance of our method was improved by approximately 30% compared to traditional methods. In conclusion, our research contributes to the field of speech recognition by incorporating psychological factors such as SoA and speed difference in the identification of main speakers in meetings. This approach enhances the accuracy and efficiency of speaker recognition systems, leading to improved performance in various applications.