ISOLATED WORD RECOGNITION AND SPEAKER IDENTIFICATION USING TWO-DIMENSIONAL HALF-PLANE LATTICE PARAMETERS OF SPECTROGRAMS.


Shawcross C., Parker S., Kayran A. H.

Digital Signal Processing - 84, Proceedings of the International Conference., Florence, Italy, pp.449-453 identifier

  • Publication Type: Conference Paper / Full Text
  • City: Florence
  • Country: Italy
  • Page Numbers: pp.449-453
  • Istanbul Technical University Affiliated: Yes

Abstract

The two dimensional image of a speech spectrogram or voiceprint is modelled using the asymmetric autoregressive half plane lattice model to obtain a set of lattice parameters which form the observation vector of a word or feature recognizing algorithm. Both FFT and Wigner transforms were used to process the raw speech data into a two dimensional form suitable for lattice modelling.