Detection and Recognition of Key Words in Isolated and Continuous Speech Under Noisy, Stressful Conditions

该研究启动奖将支持汉森博士在噪音和压力条件下进行的语音识别研究。初步模拟表明，应力补偿可以将基于马尔可夫模型的识别率提高高达42%(由于噪声和压力，识别率降低了60%以上)。汉森博士已经收集了一个超过1.6万个重音和非重音的数据库，其中包括精神病患者和承受高速加速力的人的讲话。他的数据还包括隆巴德效应的例子，在这种效应中，个人修改自己的语音来克服环境噪音。

This Research Initiation Award will support Dr. Hansen's research in speech recognition under conditions of noise and stress. Initial simulations show that stress compensation can improve Markov-based recognition rates by as much as 42% (over a 60% decrease due to noise and stress). Dr. Hanson has collected a database of over 16,000 stressed and unstressed utterances, including speech from psychiatric patients and from people enduring high acceleration forces. His data also include examples of the Lombard effect, in which individuals modify their speech to overcome environmental noise.