A real-time speech-music discriminator

A real-time circuit is described which automatically discriminates between speech and music signals. An output of the circuit gives, by means of a fuz zy feature combiner, an estimate of the probability that the input is speec h. The discriminator is tested (for both sexes) for various languages, such as English, Danish, Dutch, French, German, and Japanese, against various t ypes of music, such as pop, opera, romantic, baroque, and various solo musi cal instruments. The discriminator is found to be extremely reliable, the f alse alarm probability (inferring speech while the input is music) being vi rtually zero.