This paper describes a technique for producing an auto-segmental representation of an utterances intonation. The technique works in a bottom-up manner by using a recurrent neural network to perform a classification of each framen in the input waveform. The technique correctly identifies 87.5% of pitch accents and boundary tones.
Bibliographic reference. Taylor, Paul (1995): "Using neural networks to locate pitch accents", In EUROSPEECH-1995, 1345-1348.