Abstract: A multiple tone detector includes n tone detectors, each detecting one of n distinct tones, where n.gtoreq.2, and a background detector which generates a measure of accumulative background energy E.sub.avg in a frequency band or bands which do not include at least a subset of the n tones. The output of the background detector is applied to a smoothing filter, which generates the accumulative background energy measure E.sub.avg for a current frame as a weighted sum of the background detector output for the current frame and the background energy measure E.sub.avg from a previous frame. A parameter controlling response time of the smoothing filter is varied depending upon whether or not speech is determined to be present in the background portion of the input signal. A processor uses the energy measures from the n tone detectors and the background detector to compute n ratios, where a given ratio is the ratio of the energy measure of the ith tone to the accumulative background energy measure E.sub.avg.