Smart meeting room: ubiquitous speech recognition and analysis of mental states of attendees in meetings

效率低下的会议每年给组织造成数十亿美元的损失。造成这一问题的一个重要因素是缺乏关于会议上进行的对话的准确知识，以及对与会者的认知和情感状态的客观分析。准确记录对话并分析与会者的认知和情绪状态将使经理和决策者能够客观地监控和优化会议的持续时间、频率和结构。因此，该项目的目标是开发可靠的方法来监控会议参与者，以便转录对话并量化/分类他们的认知负荷和情绪。此外，准确的会议记录将允许自动生成会议纪要和任务分配/管理。该项目将培训2名博士生、1名硕士生和2名本科生，并首先配置一个能够记录音频、视频和压力数据的智能会议室。将开发新的语音源分离方法，以解决同时对话并实现特定说话风格的个性化。然后将开发个性化和自适应的语音识别方法，用于语音和对话的转录。对于认知和情绪状态的分类和量化，将开发用于检测心率(通过带压力感应椅的心肺功能图)和呼吸频率的可靠方法。最后，将开发新的学习模型，能够准确地将这些模式以及派生的特征映射到情绪和认知状态。该项目将通过开发语音识别和认知/情感计算领域的新算法和智能系统，为加拿大的人工智能生态系统做出贡献。此外，该项目可以通过优化会议为加拿大企业做出重大贡献，这反过来又有可能导致财务增长和吸引顶尖人才。该项目得到了加拿大公司IMRSV Data Labs的支持，该公司总部位于渥太华，活跃于人工智能和机器学习领域。

Inefficient meetings cost organizations billions of dollars every year. An important factor that contributes to this problem is a lack of accurate knowledge regarding conversations taking place at meetings as well as an objective analysis of cognitive and emotional states of attendees. Accurate transcription of conversations and analysis of cognitive and emotional states of attendees will enable managers and decision-makers to monitor and optimize the duration, frequency, and structure of meetings objectively. Accordingly, the goal of this project is to develop robust methods for monitoring attendees in meetings in order to transcribe conversations and quantify/classify their cognitive load and emotions. Additionally, accurate transcription of meetings will allow for automated generation of minutes and task assignment/management. The project will train 2 PhD students, 1 master's student, and 2 undergraduate students, and begin by first configuring an intelligent meeting room capable of recording audio, video, and pressure data. Novel methods for speech source-separation will be developed to address simultaneous conversations and enable personalization for particular speaking styles. Personalized and adaptive speech recognition methods will then be developed for transcription of speech and conversation. For classification and quantification of cognitive and emotional states, robust methods for detection of heart rate (through ballistocardiography with pressure sensing chairs) and respiration rate, will be developed. Finally, novel learned models, capable of accurately mapping these modalities along with the derived features to emotional and cognitive states will be developed. This project will contribute to Canada's AI ecosystem through development of novel algorithms and intelligent systems in the fields of speech recognition and cognitive/affective computing. Additionally, the project can significantly contribute to Canadian businesses by optimizing meetings, which in turn has the potential to result in financial growth and attract top talent. The project is backed by IMRSV Data Labs, a Canadian company based in Ottawa, active in AI and machine learning.