Key points are not available for this paper at this time.
Alarms such as phone rings, smoke alarms and sirens are an important component of the acoustic world, designed to convey urgent information in an efficient and unambiguous manner. We are investigating automatic recognition of this class of sounds both because of the practical applications for the hearing impaired, and also because alarm sounds are deliberately constructed to be easily heard, making them a promising target for detection in adverse circumstances. We compare two different approaches to alarm detection, one based on techniques and representations borrowed from speech recognition, and the other more specifically designed to exploit the structure of alarm sounds and minimize the influence of background interference. In this preliminary work, both approaches achieve similarly poor error rates but with different patterns in response to alarm type and background noise.
Daniel P. W. Ellis (Mon,) studied this question.