Determining signal-to-noise ratios of distorted speech signals involves determining probability of characteristic speech signal component with characteristic speech signal parameter(s)

摘要

The method involves determining (7) the probability of occurrence of a characteristic speech signal component using at least one characteristic speech signal parameter, whereby the distorted speech signal is filtered (3) to reduce the noise component and a frequency distribution is produced depending on the filtered values. When performing averaging (9,10) to derive the signal-to-noise ratio the parameter(s) to be averaged is(are) assessed using the probability of occurrence for the speech components. Independent claims are also included for an arrangement for implementing the method and for a use of the method for assessing and validating speech databases, esp. for automatic speech recognition systems.