Amsterdam UMC, Vrije Universiteit Amsterdam, Otolaryngology-Head and Neck Surgery, Ear and Hearing, Amsterdam Public Health Research Institute, De Boelelaan 1117, Amsterdam, Netherlands.
J Acoust Soc Am. 2021 Feb;149(2):1371. doi: 10.1121/10.0003580.
Probabilistic models to quantify context effects in speech recognition have proven their value in audiology. Boothroyd and Nittrouer [J. Acoust. Soc. Am. 84, 101-114 (1988)] introduced a model with the j-factor and k-factor as context parameters. Later, Bronkhorst, Bosman, and Smoorenburg [J. Acoust. Soc. Am. 93, 499-509 (1993)] proposed an elaborated mathematical model to quantify context effects. The present study explores existing models and proposes a new model to quantify the effect of context in sentence recognition. The effect of context is modeled by parameters that represent the change in the probability that a certain number of words in a sentence are correctly recognized. Data from two studies using a Dutch sentence-in-noise test were analyzed. The most accurate fit was obtained when using signal-to-noise ratio-dependent context parameters. Furthermore, reducing the number of context parameters from five to one had only a small effect on the goodness of fit for the present context model. An analysis of the relationships between context parameters from the different models showed that for a change in word recognition probability, the different context parameters can change in opposite directions, suggesting opposite effects of sentence context. This demonstrates the importance of controlling for the recognition probability of words in isolation when comparing the use of sentence context between different groups of listeners.
概率模型已被证明在听力领域中对语音识别中的语境效应具有量化作用。Boothroyd 和 Nittrouer [J. Acoust. Soc. Am. 84, 101-114 (1988)] 引入了一种以 j 因子和 k 因子为语境参数的模型。后来，Bronkhorst、Bosman 和 Smoorenburg [J. Acoust. Soc. Am. 93, 499-509 (1993)] 提出了一种更详细的数学模型来量化语境效应。本研究探讨了现有的模型，并提出了一种新的模型来量化句子识别中的语境效应。语境效应通过参数来建模，这些参数代表句子中一定数量的单词被正确识别的概率的变化。使用荷兰句子噪声测试的两项研究的数据进行了分析。当使用依赖信噪比的语境参数时，得到了最准确的拟合。此外，将语境参数的数量从五个减少到一个，对当前语境模型的拟合优度只有很小的影响。对不同模型的语境参数之间的关系进行分析表明，对于单词识别概率的变化，不同的语境参数可以朝着相反的方向变化，这表明句子语境具有相反的作用。这表明，当比较不同组听众对句子语境的使用时，控制孤立单词的识别概率非常重要。
