Speech segmentation by variance fractal dimension

This paper describes an implementation of the variance fractal dimension algorithm as a technique for the analysis of speech waveforms. The technique produces a fractal dimension trajectory which can be used for the detection of boundaries of an utterance in noise. The approach is superior to any other energy-based boundary-detection technique. It can also be used to segment speech utterances into sentences, words, or even phonemes. These observations are based on extensive experimental results on speech digitized at 44.1 kilosamples per second, with 16 bits in each sample.<
>