A study of multimodal recognition for human communication search

我们开发了用于人类使用语音和视频进行交流的多模式模式识别技术。提出了一种基于混合高斯模型和支持向量机的事件抽取统计方法。我们参加了TRECVID2010研讨会，在那里我们的方法在来自世界各地的40名参与者中获得了第四名的表现。我们还开发了新的主动学习方法，用于语音建模和自适应、抗噪语音识别、会议语音识别的信号处理、多模式模式识别、说话人/手势识别、语音风格分析和视频摘要。

We developed multimodal pattern recognition techniques for human communication using speech and video. We proposed a statistical technique using Gaussian mixture models and support vector machines for event extraction. We participated in TRECVID2010 workshop, where our method achieved the 4-th performance among 40 participants from all over the world. We also developed new methods for active learning for speech modeling and adaptation, noise robust speech recognition, signal processing for meeting speech recognition, multimodal pattern recognition, speaker/gesture recognition, speech style analysis and video summarization.