Towards Automatic Intoxication Detection from Speech in Real-Life Acoustic Environments

摘要

In-car intoxication detection from speech is a highly promising non-intrusive method to reduce the accident risk associated with drunk driving. However, in-car noise significantly influences the recognition performance and needs to be addressed in practical applications. In this paper, we investigate how seriously the intrinsic in-car noise and background music affect the accuracy of intoxication recognition. In extensive test runs using the official speech corpus of the INTERSPEECH 2011 Intoxication Challenge, realistic car noise and original popular music we conclude that stationary driving noise as well as music introduce a significant downgrade when acoustic models are trained on clean speech only, which can partly be alleviated by multi-condition training. Besides, exploiting cumulative evidence over time by late decision fusion appears to be a promising way to further enhance performance in noisy conditions.